1 /*
2 * Copyright (c) 2003, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "ci/ciUtilities.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/barrierSetNMethod.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "nativeInst_x86.hpp"
34 #include "oops/instanceOop.hpp"
35 #include "oops/method.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/methodHandles.hpp"
39 #include "runtime/frame.inline.hpp"
40 #include "runtime/handles.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubCodeGenerator.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/thread.inline.hpp"
45 #ifdef COMPILER2
46 #include "opto/runtime.hpp"
47 #endif
48 #if INCLUDE_ZGC
49 #include "gc/z/zThreadLocalData.hpp"
50 #endif
51
52 // Declaration and definition of StubGenerator (no .hpp file).
53 // For a more detailed description of the stub routine structure
54 // see the comment in stubRoutines.hpp
55
56 #define __ _masm->
57 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
58 #define a__ ((Assembler*)_masm)->
59
60 #ifdef PRODUCT
61 #define BLOCK_COMMENT(str) /* nothing */
62 #else
63 #define BLOCK_COMMENT(str) __ block_comment(str)
64 #endif
65
66 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
67 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
68
69 // Stub Code definitions
70
71 class StubGenerator: public StubCodeGenerator {
72 private:
73
74 #ifdef PRODUCT
75 #define inc_counter_np(counter) ((void)0)
76 #else
77 void inc_counter_np_(int& counter) {
78 // This can destroy rscratch1 if counter is far from the code cache
79 __ incrementl(ExternalAddress((address)&counter));
80 }
81 #define inc_counter_np(counter) \
82 BLOCK_COMMENT("inc_counter " #counter); \
83 inc_counter_np_(counter);
84 #endif
85
86 // Call stubs are used to call Java from C
87 //
88 // Linux Arguments:
89 // c_rarg0: call wrapper address address
90 // c_rarg1: result address
91 // c_rarg2: result type BasicType
92 // c_rarg3: method Method*
93 // c_rarg4: (interpreter) entry point address
94 // c_rarg5: parameters intptr_t*
95 // 16(rbp): parameter size (in words) int
96 // 24(rbp): thread Thread*
97 //
98 // [ return_from_Java ] <--- rsp
99 // [ argument word n ]
100 // ...
101 // -12 [ argument word 1 ]
102 // -11 [ saved r15 ] <--- rsp_after_call
103 // -10 [ saved r14 ]
104 // -9 [ saved r13 ]
105 // -8 [ saved r12 ]
106 // -7 [ saved rbx ]
107 // -6 [ call wrapper ]
108 // -5 [ result ]
109 // -4 [ result type ]
110 // -3 [ method ]
111 // -2 [ entry point ]
112 // -1 [ parameters ]
113 // 0 [ saved rbp ] <--- rbp
114 // 1 [ return address ]
115 // 2 [ parameter size ]
116 // 3 [ thread ]
117 //
118 // Windows Arguments:
119 // c_rarg0: call wrapper address address
120 // c_rarg1: result address
121 // c_rarg2: result type BasicType
122 // c_rarg3: method Method*
123 // 48(rbp): (interpreter) entry point address
124 // 56(rbp): parameters intptr_t*
125 // 64(rbp): parameter size (in words) int
126 // 72(rbp): thread Thread*
127 //
128 // [ return_from_Java ] <--- rsp
129 // [ argument word n ]
130 // ...
131 // -60 [ argument word 1 ]
132 // -59 [ saved xmm31 ] <--- rsp after_call
133 // [ saved xmm16-xmm30 ] (EVEX enabled, else the space is blank)
134 // -27 [ saved xmm15 ]
135 // [ saved xmm7-xmm14 ]
136 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
137 // -7 [ saved r15 ]
138 // -6 [ saved r14 ]
139 // -5 [ saved r13 ]
140 // -4 [ saved r12 ]
141 // -3 [ saved rdi ]
142 // -2 [ saved rsi ]
143 // -1 [ saved rbx ]
144 // 0 [ saved rbp ] <--- rbp
145 // 1 [ return address ]
146 // 2 [ call wrapper ]
147 // 3 [ result ]
148 // 4 [ result type ]
149 // 5 [ method ]
150 // 6 [ entry point ]
151 // 7 [ parameters ]
152 // 8 [ parameter size ]
153 // 9 [ thread ]
154 //
155 // Windows reserves the callers stack space for arguments 1-4.
156 // We spill c_rarg0-c_rarg3 to this space.
157
158 // Call stub stack layout word offsets from rbp
159 enum call_stub_layout {
160 #ifdef _WIN64
161 xmm_save_first = 6, // save from xmm6
162 xmm_save_last = 31, // to xmm31
163 xmm_save_base = -9,
164 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
165 r15_off = -7,
166 r14_off = -6,
167 r13_off = -5,
168 r12_off = -4,
169 rdi_off = -3,
170 rsi_off = -2,
171 rbx_off = -1,
172 rbp_off = 0,
173 retaddr_off = 1,
174 call_wrapper_off = 2,
175 result_off = 3,
176 result_type_off = 4,
177 method_off = 5,
178 entry_point_off = 6,
179 parameters_off = 7,
180 parameter_size_off = 8,
181 thread_off = 9
182 #else
183 rsp_after_call_off = -12,
184 mxcsr_off = rsp_after_call_off,
185 r15_off = -11,
186 r14_off = -10,
187 r13_off = -9,
188 r12_off = -8,
189 rbx_off = -7,
190 call_wrapper_off = -6,
191 result_off = -5,
192 result_type_off = -4,
193 method_off = -3,
194 entry_point_off = -2,
195 parameters_off = -1,
196 rbp_off = 0,
197 retaddr_off = 1,
198 parameter_size_off = 2,
199 thread_off = 3
200 #endif
201 };
202
203 #ifdef _WIN64
204 Address xmm_save(int reg) {
205 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
206 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
207 }
208 #endif
209
210 address generate_call_stub(address& return_address) {
211 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
212 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
213 "adjust this code");
214 StubCodeMark mark(this, "StubRoutines", "call_stub");
215 address start = __ pc();
216
217 // same as in generate_catch_exception()!
218 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
219
220 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
221 const Address result (rbp, result_off * wordSize);
222 const Address result_type (rbp, result_type_off * wordSize);
223 const Address method (rbp, method_off * wordSize);
224 const Address entry_point (rbp, entry_point_off * wordSize);
225 const Address parameters (rbp, parameters_off * wordSize);
226 const Address parameter_size(rbp, parameter_size_off * wordSize);
227
228 // same as in generate_catch_exception()!
229 const Address thread (rbp, thread_off * wordSize);
230
231 const Address r15_save(rbp, r15_off * wordSize);
232 const Address r14_save(rbp, r14_off * wordSize);
233 const Address r13_save(rbp, r13_off * wordSize);
234 const Address r12_save(rbp, r12_off * wordSize);
235 const Address rbx_save(rbp, rbx_off * wordSize);
236
237 // stub code
238 __ enter();
239 __ subptr(rsp, -rsp_after_call_off * wordSize);
240
241 // save register parameters
242 #ifndef _WIN64
243 __ movptr(parameters, c_rarg5); // parameters
244 __ movptr(entry_point, c_rarg4); // entry_point
245 #endif
246
247 __ movptr(method, c_rarg3); // method
248 __ movl(result_type, c_rarg2); // result type
249 __ movptr(result, c_rarg1); // result
250 __ movptr(call_wrapper, c_rarg0); // call wrapper
251
252 // save regs belonging to calling function
253 __ movptr(rbx_save, rbx);
254 __ movptr(r12_save, r12);
255 __ movptr(r13_save, r13);
256 __ movptr(r14_save, r14);
257 __ movptr(r15_save, r15);
258
259 #ifdef _WIN64
260 int last_reg = 15;
261 if (UseAVX > 2) {
262 last_reg = 31;
263 }
264 if (VM_Version::supports_evex()) {
265 for (int i = xmm_save_first; i <= last_reg; i++) {
266 __ vextractf32x4(xmm_save(i), as_XMMRegister(i), 0);
267 }
268 } else {
269 for (int i = xmm_save_first; i <= last_reg; i++) {
270 __ movdqu(xmm_save(i), as_XMMRegister(i));
271 }
272 }
273
274 const Address rdi_save(rbp, rdi_off * wordSize);
275 const Address rsi_save(rbp, rsi_off * wordSize);
276
277 __ movptr(rsi_save, rsi);
278 __ movptr(rdi_save, rdi);
279 #else
280 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
281 {
282 Label skip_ldmx;
283 __ stmxcsr(mxcsr_save);
284 __ movl(rax, mxcsr_save);
285 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
286 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
287 __ cmp32(rax, mxcsr_std);
288 __ jcc(Assembler::equal, skip_ldmx);
289 __ ldmxcsr(mxcsr_std);
290 __ bind(skip_ldmx);
291 }
292 #endif
293
294 // Load up thread register
295 __ movptr(r15_thread, thread);
296 __ reinit_heapbase();
297
298 #ifdef ASSERT
299 // make sure we have no pending exceptions
300 {
301 Label L;
302 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
303 __ jcc(Assembler::equal, L);
304 __ stop("StubRoutines::call_stub: entered with pending exception");
305 __ bind(L);
306 }
307 #endif
308
309 // pass parameters if any
310 BLOCK_COMMENT("pass parameters if any");
311 Label parameters_done;
312 __ movl(c_rarg3, parameter_size);
313 __ testl(c_rarg3, c_rarg3);
314 __ jcc(Assembler::zero, parameters_done);
315
316 Label loop;
317 __ movptr(c_rarg2, parameters); // parameter pointer
318 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
319 __ BIND(loop);
320 __ movptr(rax, Address(c_rarg2, 0));// get parameter
321 __ addptr(c_rarg2, wordSize); // advance to next parameter
322 __ decrementl(c_rarg1); // decrement counter
323 __ push(rax); // pass parameter
324 __ jcc(Assembler::notZero, loop);
325
326 // call Java function
327 __ BIND(parameters_done);
328 __ movptr(rbx, method); // get Method*
329 __ movptr(c_rarg1, entry_point); // get entry_point
330 __ mov(r13, rsp); // set sender sp
331 BLOCK_COMMENT("call Java function");
332 __ call(c_rarg1);
333
334 BLOCK_COMMENT("call_stub_return_address:");
335 return_address = __ pc();
336
337 // store result depending on type (everything that is not
338 // T_OBJECT, T_VALUETYPE, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
339 __ movptr(c_rarg0, result);
340 Label is_long, is_float, is_double, exit;
341 __ movl(c_rarg1, result_type);
342 __ cmpl(c_rarg1, T_OBJECT);
343 __ jcc(Assembler::equal, is_long);
344 __ cmpl(c_rarg1, T_VALUETYPE);
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 __ 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 __ 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 __ 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 __ 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 __ 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, in_Java_heap;
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::zero, in_Java_heap);
1048 // Not in Java heap, but could be valid if it's a bufferable value type
1049 __ load_klass(c_rarg2, rax);
1050 __ movbool(c_rarg2, Address(c_rarg2, InstanceKlass::extra_flags_offset()));
1051 __ andptr(c_rarg2, InstanceKlass::_extra_is_bufferable);
1052 __ testbool(c_rarg2);
1053 __ jcc(Assembler::zero, error);
1054 __ bind(in_Java_heap);
1055
1056 // set r12 to heapbase for load_klass()
1057 __ reinit_heapbase();
1058
1059 // make sure klass is 'reasonable', which is not zero.
1060 __ load_klass(rax, rax); // get klass
1061 __ testptr(rax, rax);
1062 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1063
1064 // return if everything seems ok
1065 __ bind(exit);
1066 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1067 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1068 __ pop(c_rarg3); // restore c_rarg3
1069 __ pop(c_rarg2); // restore c_rarg2
1070 __ pop(r12); // restore r12
1071 __ popf(); // restore flags
1072 __ ret(4 * wordSize); // pop caller saved stuff
1073
1074 // handle errors
1075 __ bind(error);
1076 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1077 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1078 __ pop(c_rarg3); // get saved c_rarg3 back
1079 __ pop(c_rarg2); // get saved c_rarg2 back
1080 __ pop(r12); // get saved r12 back
1081 __ popf(); // get saved flags off stack --
1082 // will be ignored
1083
1084 __ pusha(); // push registers
1085 // (rip is already
1086 // already pushed)
1087 // debug(char* msg, int64_t pc, int64_t regs[])
1088 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1089 // pushed all the registers, so now the stack looks like:
1090 // [tos + 0] 16 saved registers
1091 // [tos + 16] return address
1092 // * [tos + 17] error message (char*)
1093 // * [tos + 18] object to verify (oop)
1094 // * [tos + 19] saved rax - saved by caller and bashed
1095 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1096 // * = popped on exit
1097
1098 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1099 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1100 __ movq(c_rarg2, rsp); // pass address of regs on stack
1101 __ mov(r12, rsp); // remember rsp
1102 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1103 __ andptr(rsp, -16); // align stack as required by ABI
1104 BLOCK_COMMENT("call MacroAssembler::debug");
1105 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1106 __ mov(rsp, r12); // restore rsp
1107 __ popa(); // pop registers (includes r12)
1108 __ ret(4 * wordSize); // pop caller saved stuff
1109
1110 return start;
1111 }
1112
1113 //
1114 // Verify that a register contains clean 32-bits positive value
1115 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1116 //
1117 // Input:
1118 // Rint - 32-bits value
1119 // Rtmp - scratch
1120 //
1121 void assert_clean_int(Register Rint, Register Rtmp) {
1122 #ifdef ASSERT
1123 Label L;
1124 assert_different_registers(Rtmp, Rint);
1125 __ movslq(Rtmp, Rint);
1126 __ cmpq(Rtmp, Rint);
1127 __ jcc(Assembler::equal, L);
1128 __ stop("high 32-bits of int value are not 0");
1129 __ bind(L);
1130 #endif
1131 }
1132
1133 // Generate overlap test for array copy stubs
1134 //
1135 // Input:
1136 // c_rarg0 - from
1137 // c_rarg1 - to
1138 // c_rarg2 - element count
1139 //
1140 // Output:
1141 // rax - &from[element count - 1]
1142 //
1143 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1144 assert(no_overlap_target != NULL, "must be generated");
1145 array_overlap_test(no_overlap_target, NULL, sf);
1146 }
1147 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1148 array_overlap_test(NULL, &L_no_overlap, sf);
1149 }
1150 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1151 const Register from = c_rarg0;
1152 const Register to = c_rarg1;
1153 const Register count = c_rarg2;
1154 const Register end_from = rax;
1155
1156 __ cmpptr(to, from);
1157 __ lea(end_from, Address(from, count, sf, 0));
1158 if (NOLp == NULL) {
1159 ExternalAddress no_overlap(no_overlap_target);
1160 __ jump_cc(Assembler::belowEqual, no_overlap);
1161 __ cmpptr(to, end_from);
1162 __ jump_cc(Assembler::aboveEqual, no_overlap);
1163 } else {
1164 __ jcc(Assembler::belowEqual, (*NOLp));
1165 __ cmpptr(to, end_from);
1166 __ jcc(Assembler::aboveEqual, (*NOLp));
1167 }
1168 }
1169
1170 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1171 //
1172 // Outputs:
1173 // rdi - rcx
1174 // rsi - rdx
1175 // rdx - r8
1176 // rcx - r9
1177 //
1178 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1179 // are non-volatile. r9 and r10 should not be used by the caller.
1180 //
1181 DEBUG_ONLY(bool regs_in_thread;)
1182
1183 void setup_arg_regs(int nargs = 3) {
1184 const Register saved_rdi = r9;
1185 const Register saved_rsi = r10;
1186 assert(nargs == 3 || nargs == 4, "else fix");
1187 #ifdef _WIN64
1188 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1189 "unexpected argument registers");
1190 if (nargs >= 4)
1191 __ mov(rax, r9); // r9 is also saved_rdi
1192 __ movptr(saved_rdi, rdi);
1193 __ movptr(saved_rsi, rsi);
1194 __ mov(rdi, rcx); // c_rarg0
1195 __ mov(rsi, rdx); // c_rarg1
1196 __ mov(rdx, r8); // c_rarg2
1197 if (nargs >= 4)
1198 __ mov(rcx, rax); // c_rarg3 (via rax)
1199 #else
1200 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1201 "unexpected argument registers");
1202 #endif
1203 DEBUG_ONLY(regs_in_thread = false;)
1204 }
1205
1206 void restore_arg_regs() {
1207 assert(!regs_in_thread, "wrong call to restore_arg_regs");
1208 const Register saved_rdi = r9;
1209 const Register saved_rsi = r10;
1210 #ifdef _WIN64
1211 __ movptr(rdi, saved_rdi);
1212 __ movptr(rsi, saved_rsi);
1213 #endif
1214 }
1215
1216 // This is used in places where r10 is a scratch register, and can
1217 // be adapted if r9 is needed also.
1218 void setup_arg_regs_using_thread() {
1219 const Register saved_r15 = r9;
1220 #ifdef _WIN64
1221 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1222 __ get_thread(r15_thread);
1223 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1224 "unexpected argument registers");
1225 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1226 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1227
1228 __ mov(rdi, rcx); // c_rarg0
1229 __ mov(rsi, rdx); // c_rarg1
1230 __ mov(rdx, r8); // c_rarg2
1231 #else
1232 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1233 "unexpected argument registers");
1234 #endif
1235 DEBUG_ONLY(regs_in_thread = true;)
1236 }
1237
1238 void restore_arg_regs_using_thread() {
1239 assert(regs_in_thread, "wrong call to restore_arg_regs");
1240 const Register saved_r15 = r9;
1241 #ifdef _WIN64
1242 __ get_thread(r15_thread);
1243 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1244 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1245 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1246 #endif
1247 }
1248
1249 // Copy big chunks forward
1250 //
1251 // Inputs:
1252 // end_from - source arrays end address
1253 // end_to - destination array end address
1254 // qword_count - 64-bits element count, negative
1255 // to - scratch
1256 // L_copy_bytes - entry label
1257 // L_copy_8_bytes - exit label
1258 //
1259 void copy_bytes_forward(Register end_from, Register end_to,
1260 Register qword_count, Register to,
1261 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1262 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1263 Label L_loop;
1264 __ align(OptoLoopAlignment);
1265 if (UseUnalignedLoadStores) {
1266 Label L_end;
1267 // Copy 64-bytes per iteration
1268 __ BIND(L_loop);
1269 if (UseAVX > 2) {
1270 __ evmovdqul(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1271 __ evmovdqul(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1272 } else if (UseAVX == 2) {
1273 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1274 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1275 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1276 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1277 } else {
1278 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1279 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1280 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1281 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1282 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1283 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1284 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1285 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1286 }
1287 __ BIND(L_copy_bytes);
1288 __ addptr(qword_count, 8);
1289 __ jcc(Assembler::lessEqual, L_loop);
1290 __ subptr(qword_count, 4); // sub(8) and add(4)
1291 __ jccb(Assembler::greater, L_end);
1292 // Copy trailing 32 bytes
1293 if (UseAVX >= 2) {
1294 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1295 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1296 } else {
1297 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1298 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1299 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1300 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1301 }
1302 __ addptr(qword_count, 4);
1303 __ BIND(L_end);
1304 if (UseAVX >= 2) {
1305 // clean upper bits of YMM registers
1306 __ vpxor(xmm0, xmm0);
1307 __ vpxor(xmm1, xmm1);
1308 }
1309 } else {
1310 // Copy 32-bytes per iteration
1311 __ BIND(L_loop);
1312 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1313 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1314 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1315 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1316 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1317 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1318 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1319 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1320
1321 __ BIND(L_copy_bytes);
1322 __ addptr(qword_count, 4);
1323 __ jcc(Assembler::lessEqual, L_loop);
1324 }
1325 __ subptr(qword_count, 4);
1326 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1327 }
1328
1329 // Copy big chunks backward
1330 //
1331 // Inputs:
1332 // from - source arrays address
1333 // dest - destination array address
1334 // qword_count - 64-bits element count
1335 // to - scratch
1336 // L_copy_bytes - entry label
1337 // L_copy_8_bytes - exit label
1338 //
1339 void copy_bytes_backward(Register from, Register dest,
1340 Register qword_count, Register to,
1341 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1342 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1343 Label L_loop;
1344 __ align(OptoLoopAlignment);
1345 if (UseUnalignedLoadStores) {
1346 Label L_end;
1347 // Copy 64-bytes per iteration
1348 __ BIND(L_loop);
1349 if (UseAVX > 2) {
1350 __ evmovdqul(xmm0, Address(from, qword_count, Address::times_8, 0), Assembler::AVX_512bit);
1351 __ evmovdqul(Address(dest, qword_count, Address::times_8, 0), xmm0, Assembler::AVX_512bit);
1352 } else if (UseAVX == 2) {
1353 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1354 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1355 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1356 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1357 } else {
1358 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1359 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1360 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1361 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1362 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1363 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1364 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1365 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1366 }
1367 __ BIND(L_copy_bytes);
1368 __ subptr(qword_count, 8);
1369 __ jcc(Assembler::greaterEqual, L_loop);
1370
1371 __ addptr(qword_count, 4); // add(8) and sub(4)
1372 __ jccb(Assembler::less, L_end);
1373 // Copy trailing 32 bytes
1374 if (UseAVX >= 2) {
1375 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1376 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1377 } else {
1378 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1379 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1380 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1381 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1382 }
1383 __ subptr(qword_count, 4);
1384 __ BIND(L_end);
1385 if (UseAVX >= 2) {
1386 // clean upper bits of YMM registers
1387 __ vpxor(xmm0, xmm0);
1388 __ vpxor(xmm1, xmm1);
1389 }
1390 } else {
1391 // Copy 32-bytes per iteration
1392 __ BIND(L_loop);
1393 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1394 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1395 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1396 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1397 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1398 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1399 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1400 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1401
1402 __ BIND(L_copy_bytes);
1403 __ subptr(qword_count, 4);
1404 __ jcc(Assembler::greaterEqual, L_loop);
1405 }
1406 __ addptr(qword_count, 4);
1407 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1408 }
1409
1410
1411 // Arguments:
1412 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1413 // ignored
1414 // name - stub name string
1415 //
1416 // Inputs:
1417 // c_rarg0 - source array address
1418 // c_rarg1 - destination array address
1419 // c_rarg2 - element count, treated as ssize_t, can be zero
1420 //
1421 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1422 // we let the hardware handle it. The one to eight bytes within words,
1423 // dwords or qwords that span cache line boundaries will still be loaded
1424 // and stored atomically.
1425 //
1426 // Side Effects:
1427 // disjoint_byte_copy_entry is set to the no-overlap entry point
1428 // used by generate_conjoint_byte_copy().
1429 //
1430 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1431 __ align(CodeEntryAlignment);
1432 StubCodeMark mark(this, "StubRoutines", name);
1433 address start = __ pc();
1434
1435 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1436 Label L_copy_byte, L_exit;
1437 const Register from = rdi; // source array address
1438 const Register to = rsi; // destination array address
1439 const Register count = rdx; // elements count
1440 const Register byte_count = rcx;
1441 const Register qword_count = count;
1442 const Register end_from = from; // source array end address
1443 const Register end_to = to; // destination array end address
1444 // End pointers are inclusive, and if count is not zero they point
1445 // to the last unit copied: end_to[0] := end_from[0]
1446
1447 __ enter(); // required for proper stackwalking of RuntimeStub frame
1448 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1449
1450 if (entry != NULL) {
1451 *entry = __ pc();
1452 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1453 BLOCK_COMMENT("Entry:");
1454 }
1455
1456 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1457 // r9 and r10 may be used to save non-volatile registers
1458
1459 // 'from', 'to' and 'count' are now valid
1460 __ movptr(byte_count, count);
1461 __ shrptr(count, 3); // count => qword_count
1462
1463 // Copy from low to high addresses. Use 'to' as scratch.
1464 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1465 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1466 __ negptr(qword_count); // make the count negative
1467 __ jmp(L_copy_bytes);
1468
1469 // Copy trailing qwords
1470 __ BIND(L_copy_8_bytes);
1471 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1472 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1473 __ increment(qword_count);
1474 __ jcc(Assembler::notZero, L_copy_8_bytes);
1475
1476 // Check for and copy trailing dword
1477 __ BIND(L_copy_4_bytes);
1478 __ testl(byte_count, 4);
1479 __ jccb(Assembler::zero, L_copy_2_bytes);
1480 __ movl(rax, Address(end_from, 8));
1481 __ movl(Address(end_to, 8), rax);
1482
1483 __ addptr(end_from, 4);
1484 __ addptr(end_to, 4);
1485
1486 // Check for and copy trailing word
1487 __ BIND(L_copy_2_bytes);
1488 __ testl(byte_count, 2);
1489 __ jccb(Assembler::zero, L_copy_byte);
1490 __ movw(rax, Address(end_from, 8));
1491 __ movw(Address(end_to, 8), rax);
1492
1493 __ addptr(end_from, 2);
1494 __ addptr(end_to, 2);
1495
1496 // Check for and copy trailing byte
1497 __ BIND(L_copy_byte);
1498 __ testl(byte_count, 1);
1499 __ jccb(Assembler::zero, L_exit);
1500 __ movb(rax, Address(end_from, 8));
1501 __ movb(Address(end_to, 8), rax);
1502
1503 __ BIND(L_exit);
1504 restore_arg_regs();
1505 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1506 __ xorptr(rax, rax); // return 0
1507 __ vzeroupper();
1508 __ leave(); // required for proper stackwalking of RuntimeStub frame
1509 __ ret(0);
1510
1511 // Copy in multi-bytes chunks
1512 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1513 __ jmp(L_copy_4_bytes);
1514
1515 return start;
1516 }
1517
1518 // Arguments:
1519 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1520 // ignored
1521 // name - stub name string
1522 //
1523 // Inputs:
1524 // c_rarg0 - source array address
1525 // c_rarg1 - destination array address
1526 // c_rarg2 - element count, treated as ssize_t, can be zero
1527 //
1528 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1529 // we let the hardware handle it. The one to eight bytes within words,
1530 // dwords or qwords that span cache line boundaries will still be loaded
1531 // and stored atomically.
1532 //
1533 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1534 address* entry, const char *name) {
1535 __ align(CodeEntryAlignment);
1536 StubCodeMark mark(this, "StubRoutines", name);
1537 address start = __ pc();
1538
1539 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1540 const Register from = rdi; // source array address
1541 const Register to = rsi; // destination array address
1542 const Register count = rdx; // elements count
1543 const Register byte_count = rcx;
1544 const Register qword_count = count;
1545
1546 __ enter(); // required for proper stackwalking of RuntimeStub frame
1547 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1548
1549 if (entry != NULL) {
1550 *entry = __ pc();
1551 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1552 BLOCK_COMMENT("Entry:");
1553 }
1554
1555 array_overlap_test(nooverlap_target, Address::times_1);
1556 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1557 // r9 and r10 may be used to save non-volatile registers
1558
1559 // 'from', 'to' and 'count' are now valid
1560 __ movptr(byte_count, count);
1561 __ shrptr(count, 3); // count => qword_count
1562
1563 // Copy from high to low addresses.
1564
1565 // Check for and copy trailing byte
1566 __ testl(byte_count, 1);
1567 __ jcc(Assembler::zero, L_copy_2_bytes);
1568 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1569 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1570 __ decrement(byte_count); // Adjust for possible trailing word
1571
1572 // Check for and copy trailing word
1573 __ BIND(L_copy_2_bytes);
1574 __ testl(byte_count, 2);
1575 __ jcc(Assembler::zero, L_copy_4_bytes);
1576 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1577 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1578
1579 // Check for and copy trailing dword
1580 __ BIND(L_copy_4_bytes);
1581 __ testl(byte_count, 4);
1582 __ jcc(Assembler::zero, L_copy_bytes);
1583 __ movl(rax, Address(from, qword_count, Address::times_8));
1584 __ movl(Address(to, qword_count, Address::times_8), rax);
1585 __ jmp(L_copy_bytes);
1586
1587 // Copy trailing qwords
1588 __ BIND(L_copy_8_bytes);
1589 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1590 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1591 __ decrement(qword_count);
1592 __ jcc(Assembler::notZero, L_copy_8_bytes);
1593
1594 restore_arg_regs();
1595 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1596 __ xorptr(rax, rax); // return 0
1597 __ vzeroupper();
1598 __ leave(); // required for proper stackwalking of RuntimeStub frame
1599 __ ret(0);
1600
1601 // Copy in multi-bytes chunks
1602 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1603
1604 restore_arg_regs();
1605 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1606 __ xorptr(rax, rax); // return 0
1607 __ vzeroupper();
1608 __ leave(); // required for proper stackwalking of RuntimeStub frame
1609 __ ret(0);
1610
1611 return start;
1612 }
1613
1614 // Arguments:
1615 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1616 // ignored
1617 // name - stub name string
1618 //
1619 // Inputs:
1620 // c_rarg0 - source array address
1621 // c_rarg1 - destination array address
1622 // c_rarg2 - element count, treated as ssize_t, can be zero
1623 //
1624 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1625 // let the hardware handle it. The two or four words within dwords
1626 // or qwords that span cache line boundaries will still be loaded
1627 // and stored atomically.
1628 //
1629 // Side Effects:
1630 // disjoint_short_copy_entry is set to the no-overlap entry point
1631 // used by generate_conjoint_short_copy().
1632 //
1633 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1634 __ align(CodeEntryAlignment);
1635 StubCodeMark mark(this, "StubRoutines", name);
1636 address start = __ pc();
1637
1638 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1639 const Register from = rdi; // source array address
1640 const Register to = rsi; // destination array address
1641 const Register count = rdx; // elements count
1642 const Register word_count = rcx;
1643 const Register qword_count = count;
1644 const Register end_from = from; // source array end address
1645 const Register end_to = to; // destination array end address
1646 // End pointers are inclusive, and if count is not zero they point
1647 // to the last unit copied: end_to[0] := end_from[0]
1648
1649 __ enter(); // required for proper stackwalking of RuntimeStub frame
1650 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1651
1652 if (entry != NULL) {
1653 *entry = __ pc();
1654 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1655 BLOCK_COMMENT("Entry:");
1656 }
1657
1658 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1659 // r9 and r10 may be used to save non-volatile registers
1660
1661 // 'from', 'to' and 'count' are now valid
1662 __ movptr(word_count, count);
1663 __ shrptr(count, 2); // count => qword_count
1664
1665 // Copy from low to high addresses. Use 'to' as scratch.
1666 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1667 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1668 __ negptr(qword_count);
1669 __ jmp(L_copy_bytes);
1670
1671 // Copy trailing qwords
1672 __ BIND(L_copy_8_bytes);
1673 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1674 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1675 __ increment(qword_count);
1676 __ jcc(Assembler::notZero, L_copy_8_bytes);
1677
1678 // Original 'dest' is trashed, so we can't use it as a
1679 // base register for a possible trailing word copy
1680
1681 // Check for and copy trailing dword
1682 __ BIND(L_copy_4_bytes);
1683 __ testl(word_count, 2);
1684 __ jccb(Assembler::zero, L_copy_2_bytes);
1685 __ movl(rax, Address(end_from, 8));
1686 __ movl(Address(end_to, 8), rax);
1687
1688 __ addptr(end_from, 4);
1689 __ addptr(end_to, 4);
1690
1691 // Check for and copy trailing word
1692 __ BIND(L_copy_2_bytes);
1693 __ testl(word_count, 1);
1694 __ jccb(Assembler::zero, L_exit);
1695 __ movw(rax, Address(end_from, 8));
1696 __ movw(Address(end_to, 8), rax);
1697
1698 __ BIND(L_exit);
1699 restore_arg_regs();
1700 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1701 __ xorptr(rax, rax); // return 0
1702 __ vzeroupper();
1703 __ leave(); // required for proper stackwalking of RuntimeStub frame
1704 __ ret(0);
1705
1706 // Copy in multi-bytes chunks
1707 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1708 __ jmp(L_copy_4_bytes);
1709
1710 return start;
1711 }
1712
1713 address generate_fill(BasicType t, bool aligned, const char *name) {
1714 __ align(CodeEntryAlignment);
1715 StubCodeMark mark(this, "StubRoutines", name);
1716 address start = __ pc();
1717
1718 BLOCK_COMMENT("Entry:");
1719
1720 const Register to = c_rarg0; // source array address
1721 const Register value = c_rarg1; // value
1722 const Register count = c_rarg2; // elements count
1723
1724 __ enter(); // required for proper stackwalking of RuntimeStub frame
1725
1726 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1727
1728 __ vzeroupper();
1729 __ leave(); // required for proper stackwalking of RuntimeStub frame
1730 __ ret(0);
1731 return start;
1732 }
1733
1734 // Arguments:
1735 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1736 // ignored
1737 // name - stub name string
1738 //
1739 // Inputs:
1740 // c_rarg0 - source array address
1741 // c_rarg1 - destination array address
1742 // c_rarg2 - element count, treated as ssize_t, can be zero
1743 //
1744 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1745 // let the hardware handle it. The two or four words within dwords
1746 // or qwords that span cache line boundaries will still be loaded
1747 // and stored atomically.
1748 //
1749 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1750 address *entry, const char *name) {
1751 __ align(CodeEntryAlignment);
1752 StubCodeMark mark(this, "StubRoutines", name);
1753 address start = __ pc();
1754
1755 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1756 const Register from = rdi; // source array address
1757 const Register to = rsi; // destination array address
1758 const Register count = rdx; // elements count
1759 const Register word_count = rcx;
1760 const Register qword_count = count;
1761
1762 __ enter(); // required for proper stackwalking of RuntimeStub frame
1763 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1764
1765 if (entry != NULL) {
1766 *entry = __ pc();
1767 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1768 BLOCK_COMMENT("Entry:");
1769 }
1770
1771 array_overlap_test(nooverlap_target, Address::times_2);
1772 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1773 // r9 and r10 may be used to save non-volatile registers
1774
1775 // 'from', 'to' and 'count' are now valid
1776 __ movptr(word_count, count);
1777 __ shrptr(count, 2); // count => qword_count
1778
1779 // Copy from high to low addresses. Use 'to' as scratch.
1780
1781 // Check for and copy trailing word
1782 __ testl(word_count, 1);
1783 __ jccb(Assembler::zero, L_copy_4_bytes);
1784 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1785 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1786
1787 // Check for and copy trailing dword
1788 __ BIND(L_copy_4_bytes);
1789 __ testl(word_count, 2);
1790 __ jcc(Assembler::zero, L_copy_bytes);
1791 __ movl(rax, Address(from, qword_count, Address::times_8));
1792 __ movl(Address(to, qword_count, Address::times_8), rax);
1793 __ jmp(L_copy_bytes);
1794
1795 // Copy trailing qwords
1796 __ BIND(L_copy_8_bytes);
1797 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1798 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1799 __ decrement(qword_count);
1800 __ jcc(Assembler::notZero, L_copy_8_bytes);
1801
1802 restore_arg_regs();
1803 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1804 __ xorptr(rax, rax); // return 0
1805 __ vzeroupper();
1806 __ leave(); // required for proper stackwalking of RuntimeStub frame
1807 __ ret(0);
1808
1809 // Copy in multi-bytes chunks
1810 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1811
1812 restore_arg_regs();
1813 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1814 __ xorptr(rax, rax); // return 0
1815 __ vzeroupper();
1816 __ leave(); // required for proper stackwalking of RuntimeStub frame
1817 __ ret(0);
1818
1819 return start;
1820 }
1821
1822 // Arguments:
1823 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1824 // ignored
1825 // is_oop - true => oop array, so generate store check code
1826 // name - stub name string
1827 //
1828 // Inputs:
1829 // c_rarg0 - source array address
1830 // c_rarg1 - destination array address
1831 // c_rarg2 - element count, treated as ssize_t, can be zero
1832 //
1833 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1834 // the hardware handle it. The two dwords within qwords that span
1835 // cache line boundaries will still be loaded and stored atomicly.
1836 //
1837 // Side Effects:
1838 // disjoint_int_copy_entry is set to the no-overlap entry point
1839 // used by generate_conjoint_int_oop_copy().
1840 //
1841 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1842 const char *name, bool dest_uninitialized = false) {
1843 __ align(CodeEntryAlignment);
1844 StubCodeMark mark(this, "StubRoutines", name);
1845 address start = __ pc();
1846
1847 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1848 const Register from = rdi; // source array address
1849 const Register to = rsi; // destination array address
1850 const Register count = rdx; // elements count
1851 const Register dword_count = rcx;
1852 const Register qword_count = count;
1853 const Register end_from = from; // source array end address
1854 const Register end_to = to; // destination array end address
1855 // End pointers are inclusive, and if count is not zero they point
1856 // to the last unit copied: end_to[0] := end_from[0]
1857
1858 __ enter(); // required for proper stackwalking of RuntimeStub frame
1859 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1860
1861 if (entry != NULL) {
1862 *entry = __ pc();
1863 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1864 BLOCK_COMMENT("Entry:");
1865 }
1866
1867 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1868 // r9 is used to save r15_thread
1869
1870 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
1871 if (dest_uninitialized) {
1872 decorators |= IS_DEST_UNINITIALIZED;
1873 }
1874 if (aligned) {
1875 decorators |= ARRAYCOPY_ALIGNED;
1876 }
1877
1878 BasicType type = is_oop ? T_OBJECT : T_INT;
1879 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1880 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1881
1882 // 'from', 'to' and 'count' are now valid
1883 __ movptr(dword_count, count);
1884 __ shrptr(count, 1); // count => qword_count
1885
1886 // Copy from low to high addresses. Use 'to' as scratch.
1887 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1888 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1889 __ negptr(qword_count);
1890 __ jmp(L_copy_bytes);
1891
1892 // Copy trailing qwords
1893 __ BIND(L_copy_8_bytes);
1894 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1895 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1896 __ increment(qword_count);
1897 __ jcc(Assembler::notZero, L_copy_8_bytes);
1898
1899 // Check for and copy trailing dword
1900 __ BIND(L_copy_4_bytes);
1901 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1902 __ jccb(Assembler::zero, L_exit);
1903 __ movl(rax, Address(end_from, 8));
1904 __ movl(Address(end_to, 8), rax);
1905
1906 __ BIND(L_exit);
1907 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
1908 restore_arg_regs_using_thread();
1909 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1910 __ vzeroupper();
1911 __ xorptr(rax, rax); // return 0
1912 __ leave(); // required for proper stackwalking of RuntimeStub frame
1913 __ ret(0);
1914
1915 // Copy in multi-bytes chunks
1916 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1917 __ jmp(L_copy_4_bytes);
1918
1919 return start;
1920 }
1921
1922 // Arguments:
1923 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1924 // ignored
1925 // is_oop - true => oop array, so generate store check code
1926 // name - stub name string
1927 //
1928 // Inputs:
1929 // c_rarg0 - source array address
1930 // c_rarg1 - destination array address
1931 // c_rarg2 - element count, treated as ssize_t, can be zero
1932 //
1933 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1934 // the hardware handle it. The two dwords within qwords that span
1935 // cache line boundaries will still be loaded and stored atomicly.
1936 //
1937 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1938 address *entry, const char *name,
1939 bool dest_uninitialized = false) {
1940 __ align(CodeEntryAlignment);
1941 StubCodeMark mark(this, "StubRoutines", name);
1942 address start = __ pc();
1943
1944 Label L_copy_bytes, L_copy_8_bytes, L_exit;
1945 const Register from = rdi; // source array address
1946 const Register to = rsi; // destination array address
1947 const Register count = rdx; // elements count
1948 const Register dword_count = rcx;
1949 const Register qword_count = count;
1950
1951 __ enter(); // required for proper stackwalking of RuntimeStub frame
1952 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1953
1954 if (entry != NULL) {
1955 *entry = __ pc();
1956 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1957 BLOCK_COMMENT("Entry:");
1958 }
1959
1960 array_overlap_test(nooverlap_target, Address::times_4);
1961 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1962 // r9 is used to save r15_thread
1963
1964 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
1965 if (dest_uninitialized) {
1966 decorators |= IS_DEST_UNINITIALIZED;
1967 }
1968 if (aligned) {
1969 decorators |= ARRAYCOPY_ALIGNED;
1970 }
1971
1972 BasicType type = is_oop ? T_OBJECT : T_INT;
1973 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1974 // no registers are destroyed by this call
1975 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1976
1977 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1978 // 'from', 'to' and 'count' are now valid
1979 __ movptr(dword_count, count);
1980 __ shrptr(count, 1); // count => qword_count
1981
1982 // Copy from high to low addresses. Use 'to' as scratch.
1983
1984 // Check for and copy trailing dword
1985 __ testl(dword_count, 1);
1986 __ jcc(Assembler::zero, L_copy_bytes);
1987 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1988 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1989 __ jmp(L_copy_bytes);
1990
1991 // Copy trailing qwords
1992 __ BIND(L_copy_8_bytes);
1993 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1994 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1995 __ decrement(qword_count);
1996 __ jcc(Assembler::notZero, L_copy_8_bytes);
1997
1998 if (is_oop) {
1999 __ jmp(L_exit);
2000 }
2001 restore_arg_regs_using_thread();
2002 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2003 __ xorptr(rax, rax); // return 0
2004 __ vzeroupper();
2005 __ leave(); // required for proper stackwalking of RuntimeStub frame
2006 __ ret(0);
2007
2008 // Copy in multi-bytes chunks
2009 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2010
2011 __ BIND(L_exit);
2012 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2013 restore_arg_regs_using_thread();
2014 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2015 __ xorptr(rax, rax); // return 0
2016 __ vzeroupper();
2017 __ leave(); // required for proper stackwalking of RuntimeStub frame
2018 __ ret(0);
2019
2020 return start;
2021 }
2022
2023 // Arguments:
2024 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2025 // ignored
2026 // is_oop - true => oop array, so generate store check code
2027 // name - stub name string
2028 //
2029 // Inputs:
2030 // c_rarg0 - source array address
2031 // c_rarg1 - destination array address
2032 // c_rarg2 - element count, treated as ssize_t, can be zero
2033 //
2034 // Side Effects:
2035 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2036 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2037 //
2038 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2039 const char *name, bool dest_uninitialized = false) {
2040 __ align(CodeEntryAlignment);
2041 StubCodeMark mark(this, "StubRoutines", name);
2042 address start = __ pc();
2043
2044 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2045 const Register from = rdi; // source array address
2046 const Register to = rsi; // destination array address
2047 const Register qword_count = rdx; // elements count
2048 const Register end_from = from; // source array end address
2049 const Register end_to = rcx; // destination array end address
2050 const Register saved_count = r11;
2051 // End pointers are inclusive, and if count is not zero they point
2052 // to the last unit copied: end_to[0] := end_from[0]
2053
2054 __ enter(); // required for proper stackwalking of RuntimeStub frame
2055 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2056 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2057
2058 if (entry != NULL) {
2059 *entry = __ pc();
2060 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2061 BLOCK_COMMENT("Entry:");
2062 }
2063
2064 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2065 // r9 is used to save r15_thread
2066 // 'from', 'to' and 'qword_count' are now valid
2067
2068 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2069 if (dest_uninitialized) {
2070 decorators |= IS_DEST_UNINITIALIZED;
2071 }
2072 if (aligned) {
2073 decorators |= ARRAYCOPY_ALIGNED;
2074 }
2075
2076 BasicType type = is_oop ? T_OBJECT : T_LONG;
2077 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2078 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2079
2080 // Copy from low to high addresses. Use 'to' as scratch.
2081 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2082 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2083 __ negptr(qword_count);
2084 __ jmp(L_copy_bytes);
2085
2086 // Copy trailing qwords
2087 __ BIND(L_copy_8_bytes);
2088 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2089 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2090 __ increment(qword_count);
2091 __ jcc(Assembler::notZero, L_copy_8_bytes);
2092
2093 if (is_oop) {
2094 __ jmp(L_exit);
2095 } else {
2096 restore_arg_regs_using_thread();
2097 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2098 __ xorptr(rax, rax); // return 0
2099 __ vzeroupper();
2100 __ leave(); // required for proper stackwalking of RuntimeStub frame
2101 __ ret(0);
2102 }
2103
2104 // Copy in multi-bytes chunks
2105 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2106
2107 __ BIND(L_exit);
2108 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2109 restore_arg_regs_using_thread();
2110 if (is_oop) {
2111 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2112 } else {
2113 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2114 }
2115 __ vzeroupper();
2116 __ xorptr(rax, rax); // return 0
2117 __ leave(); // required for proper stackwalking of RuntimeStub frame
2118 __ ret(0);
2119
2120 return start;
2121 }
2122
2123 // Arguments:
2124 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2125 // ignored
2126 // is_oop - true => oop array, so generate store check code
2127 // name - stub name string
2128 //
2129 // Inputs:
2130 // c_rarg0 - source array address
2131 // c_rarg1 - destination array address
2132 // c_rarg2 - element count, treated as ssize_t, can be zero
2133 //
2134 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2135 address nooverlap_target, address *entry,
2136 const char *name, bool dest_uninitialized = false) {
2137 __ align(CodeEntryAlignment);
2138 StubCodeMark mark(this, "StubRoutines", name);
2139 address start = __ pc();
2140
2141 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2142 const Register from = rdi; // source array address
2143 const Register to = rsi; // destination array address
2144 const Register qword_count = rdx; // elements count
2145 const Register saved_count = rcx;
2146
2147 __ enter(); // required for proper stackwalking of RuntimeStub frame
2148 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2149
2150 if (entry != NULL) {
2151 *entry = __ pc();
2152 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2153 BLOCK_COMMENT("Entry:");
2154 }
2155
2156 array_overlap_test(nooverlap_target, Address::times_8);
2157 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2158 // r9 is used to save r15_thread
2159 // 'from', 'to' and 'qword_count' are now valid
2160
2161 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2162 if (dest_uninitialized) {
2163 decorators |= IS_DEST_UNINITIALIZED;
2164 }
2165 if (aligned) {
2166 decorators |= ARRAYCOPY_ALIGNED;
2167 }
2168
2169 BasicType type = is_oop ? T_OBJECT : T_LONG;
2170 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2171 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2172
2173 __ jmp(L_copy_bytes);
2174
2175 // Copy trailing qwords
2176 __ BIND(L_copy_8_bytes);
2177 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2178 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2179 __ decrement(qword_count);
2180 __ jcc(Assembler::notZero, L_copy_8_bytes);
2181
2182 if (is_oop) {
2183 __ jmp(L_exit);
2184 } else {
2185 restore_arg_regs_using_thread();
2186 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2187 __ xorptr(rax, rax); // return 0
2188 __ vzeroupper();
2189 __ leave(); // required for proper stackwalking of RuntimeStub frame
2190 __ ret(0);
2191 }
2192
2193 // Copy in multi-bytes chunks
2194 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2195
2196 __ BIND(L_exit);
2197 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2198 restore_arg_regs_using_thread();
2199 if (is_oop) {
2200 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2201 } else {
2202 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2203 }
2204 __ vzeroupper();
2205 __ xorptr(rax, rax); // return 0
2206 __ leave(); // required for proper stackwalking of RuntimeStub frame
2207 __ ret(0);
2208
2209 return start;
2210 }
2211
2212
2213 // Helper for generating a dynamic type check.
2214 // Smashes no registers.
2215 void generate_type_check(Register sub_klass,
2216 Register super_check_offset,
2217 Register super_klass,
2218 Label& L_success) {
2219 assert_different_registers(sub_klass, super_check_offset, super_klass);
2220
2221 BLOCK_COMMENT("type_check:");
2222
2223 Label L_miss;
2224
2225 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2226 super_check_offset);
2227 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2228
2229 // Fall through on failure!
2230 __ BIND(L_miss);
2231 }
2232
2233 //
2234 // Generate checkcasting array copy stub
2235 //
2236 // Input:
2237 // c_rarg0 - source array address
2238 // c_rarg1 - destination array address
2239 // c_rarg2 - element count, treated as ssize_t, can be zero
2240 // c_rarg3 - size_t ckoff (super_check_offset)
2241 // not Win64
2242 // c_rarg4 - oop ckval (super_klass)
2243 // Win64
2244 // rsp+40 - oop ckval (super_klass)
2245 //
2246 // Output:
2247 // rax == 0 - success
2248 // rax == -1^K - failure, where K is partial transfer count
2249 //
2250 address generate_checkcast_copy(const char *name, address *entry,
2251 bool dest_uninitialized = false) {
2252
2253 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2254
2255 // Input registers (after setup_arg_regs)
2256 const Register from = rdi; // source array address
2257 const Register to = rsi; // destination array address
2258 const Register length = rdx; // elements count
2259 const Register ckoff = rcx; // super_check_offset
2260 const Register ckval = r8; // super_klass
2261
2262 // Registers used as temps (r13, r14 are save-on-entry)
2263 const Register end_from = from; // source array end address
2264 const Register end_to = r13; // destination array end address
2265 const Register count = rdx; // -(count_remaining)
2266 const Register r14_length = r14; // saved copy of length
2267 // End pointers are inclusive, and if length is not zero they point
2268 // to the last unit copied: end_to[0] := end_from[0]
2269
2270 const Register rax_oop = rax; // actual oop copied
2271 const Register r11_klass = r11; // oop._klass
2272
2273 //---------------------------------------------------------------
2274 // Assembler stub will be used for this call to arraycopy
2275 // if the two arrays are subtypes of Object[] but the
2276 // destination array type is not equal to or a supertype
2277 // of the source type. Each element must be separately
2278 // checked.
2279
2280 __ align(CodeEntryAlignment);
2281 StubCodeMark mark(this, "StubRoutines", name);
2282 address start = __ pc();
2283
2284 __ enter(); // required for proper stackwalking of RuntimeStub frame
2285
2286 #ifdef ASSERT
2287 // caller guarantees that the arrays really are different
2288 // otherwise, we would have to make conjoint checks
2289 { Label L;
2290 array_overlap_test(L, TIMES_OOP);
2291 __ stop("checkcast_copy within a single array");
2292 __ bind(L);
2293 }
2294 #endif //ASSERT
2295
2296 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2297 // ckoff => rcx, ckval => r8
2298 // r9 and r10 may be used to save non-volatile registers
2299 #ifdef _WIN64
2300 // last argument (#4) is on stack on Win64
2301 __ movptr(ckval, Address(rsp, 6 * wordSize));
2302 #endif
2303
2304 // Caller of this entry point must set up the argument registers.
2305 if (entry != NULL) {
2306 *entry = __ pc();
2307 BLOCK_COMMENT("Entry:");
2308 }
2309
2310 // allocate spill slots for r13, r14
2311 enum {
2312 saved_r13_offset,
2313 saved_r14_offset,
2314 saved_r10_offset,
2315 saved_rbp_offset
2316 };
2317 __ subptr(rsp, saved_rbp_offset * wordSize);
2318 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2319 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2320 __ movptr(Address(rsp, saved_r10_offset * wordSize), r10);
2321
2322 #ifdef ASSERT
2323 Label L2;
2324 __ get_thread(r14);
2325 __ cmpptr(r15_thread, r14);
2326 __ jcc(Assembler::equal, L2);
2327 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
2328 __ bind(L2);
2329 #endif // ASSERT
2330
2331 // check that int operands are properly extended to size_t
2332 assert_clean_int(length, rax);
2333 assert_clean_int(ckoff, rax);
2334
2335 #ifdef ASSERT
2336 BLOCK_COMMENT("assert consistent ckoff/ckval");
2337 // The ckoff and ckval must be mutually consistent,
2338 // even though caller generates both.
2339 { Label L;
2340 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2341 __ cmpl(ckoff, Address(ckval, sco_offset));
2342 __ jcc(Assembler::equal, L);
2343 __ stop("super_check_offset inconsistent");
2344 __ bind(L);
2345 }
2346 #endif //ASSERT
2347
2348 // Loop-invariant addresses. They are exclusive end pointers.
2349 Address end_from_addr(from, length, TIMES_OOP, 0);
2350 Address end_to_addr(to, length, TIMES_OOP, 0);
2351 // Loop-variant addresses. They assume post-incremented count < 0.
2352 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2353 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2354
2355 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST;
2356 if (dest_uninitialized) {
2357 decorators |= IS_DEST_UNINITIALIZED;
2358 }
2359
2360 BasicType type = T_OBJECT;
2361 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2362 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2363
2364 // Copy from low to high addresses, indexed from the end of each array.
2365 __ lea(end_from, end_from_addr);
2366 __ lea(end_to, end_to_addr);
2367 __ movptr(r14_length, length); // save a copy of the length
2368 assert(length == count, ""); // else fix next line:
2369 __ negptr(count); // negate and test the length
2370 __ jcc(Assembler::notZero, L_load_element);
2371
2372 // Empty array: Nothing to do.
2373 __ xorptr(rax, rax); // return 0 on (trivial) success
2374 __ jmp(L_done);
2375
2376 // ======== begin loop ========
2377 // (Loop is rotated; its entry is L_load_element.)
2378 // Loop control:
2379 // for (count = -count; count != 0; count++)
2380 // Base pointers src, dst are biased by 8*(count-1),to last element.
2381 __ align(OptoLoopAlignment);
2382
2383 __ BIND(L_store_element);
2384 __ store_heap_oop(to_element_addr, rax_oop, noreg, noreg, AS_RAW); // store the oop
2385 __ increment(count); // increment the count toward zero
2386 __ jcc(Assembler::zero, L_do_card_marks);
2387
2388 // ======== loop entry is here ========
2389 __ BIND(L_load_element);
2390 __ load_heap_oop(rax_oop, from_element_addr, noreg, noreg, AS_RAW); // load the oop
2391 __ testptr(rax_oop, rax_oop);
2392 __ jcc(Assembler::zero, L_store_element);
2393
2394 __ load_klass(r11_klass, rax_oop);// query the object klass
2395 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2396 // ======== end loop ========
2397
2398 // It was a real error; we must depend on the caller to finish the job.
2399 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2400 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2401 // and report their number to the caller.
2402 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2403 Label L_post_barrier;
2404 __ addptr(r14_length, count); // K = (original - remaining) oops
2405 __ movptr(rax, r14_length); // save the value
2406 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2407 __ jccb(Assembler::notZero, L_post_barrier);
2408 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2409
2410 // Come here on success only.
2411 __ BIND(L_do_card_marks);
2412 __ xorptr(rax, rax); // return 0 on success
2413
2414 __ BIND(L_post_barrier);
2415 bs->arraycopy_epilogue(_masm, decorators, type, from, to, r14_length);
2416
2417 // Common exit point (success or failure).
2418 __ BIND(L_done);
2419 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2420 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2421 __ movptr(r10, Address(rsp, saved_r10_offset * wordSize));
2422 restore_arg_regs();
2423 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2424 __ leave(); // required for proper stackwalking of RuntimeStub frame
2425 __ ret(0);
2426
2427 return start;
2428 }
2429
2430 //
2431 // Generate 'unsafe' array copy stub
2432 // Though just as safe as the other stubs, it takes an unscaled
2433 // size_t argument instead of an element count.
2434 //
2435 // Input:
2436 // c_rarg0 - source array address
2437 // c_rarg1 - destination array address
2438 // c_rarg2 - byte count, treated as ssize_t, can be zero
2439 //
2440 // Examines the alignment of the operands and dispatches
2441 // to a long, int, short, or byte copy loop.
2442 //
2443 address generate_unsafe_copy(const char *name,
2444 address byte_copy_entry, address short_copy_entry,
2445 address int_copy_entry, address long_copy_entry) {
2446
2447 Label L_long_aligned, L_int_aligned, L_short_aligned;
2448
2449 // Input registers (before setup_arg_regs)
2450 const Register from = c_rarg0; // source array address
2451 const Register to = c_rarg1; // destination array address
2452 const Register size = c_rarg2; // byte count (size_t)
2453
2454 // Register used as a temp
2455 const Register bits = rax; // test copy of low bits
2456
2457 __ align(CodeEntryAlignment);
2458 StubCodeMark mark(this, "StubRoutines", name);
2459 address start = __ pc();
2460
2461 __ enter(); // required for proper stackwalking of RuntimeStub frame
2462
2463 // bump this on entry, not on exit:
2464 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2465
2466 __ mov(bits, from);
2467 __ orptr(bits, to);
2468 __ orptr(bits, size);
2469
2470 __ testb(bits, BytesPerLong-1);
2471 __ jccb(Assembler::zero, L_long_aligned);
2472
2473 __ testb(bits, BytesPerInt-1);
2474 __ jccb(Assembler::zero, L_int_aligned);
2475
2476 __ testb(bits, BytesPerShort-1);
2477 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2478
2479 __ BIND(L_short_aligned);
2480 __ shrptr(size, LogBytesPerShort); // size => short_count
2481 __ jump(RuntimeAddress(short_copy_entry));
2482
2483 __ BIND(L_int_aligned);
2484 __ shrptr(size, LogBytesPerInt); // size => int_count
2485 __ jump(RuntimeAddress(int_copy_entry));
2486
2487 __ BIND(L_long_aligned);
2488 __ shrptr(size, LogBytesPerLong); // size => qword_count
2489 __ jump(RuntimeAddress(long_copy_entry));
2490
2491 return start;
2492 }
2493
2494 // Perform range checks on the proposed arraycopy.
2495 // Kills temp, but nothing else.
2496 // Also, clean the sign bits of src_pos and dst_pos.
2497 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2498 Register src_pos, // source position (c_rarg1)
2499 Register dst, // destination array oo (c_rarg2)
2500 Register dst_pos, // destination position (c_rarg3)
2501 Register length,
2502 Register temp,
2503 Label& L_failed) {
2504 BLOCK_COMMENT("arraycopy_range_checks:");
2505
2506 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2507 __ movl(temp, length);
2508 __ addl(temp, src_pos); // src_pos + length
2509 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2510 __ jcc(Assembler::above, L_failed);
2511
2512 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2513 __ movl(temp, length);
2514 __ addl(temp, dst_pos); // dst_pos + length
2515 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2516 __ jcc(Assembler::above, L_failed);
2517
2518 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2519 // Move with sign extension can be used since they are positive.
2520 __ movslq(src_pos, src_pos);
2521 __ movslq(dst_pos, dst_pos);
2522
2523 BLOCK_COMMENT("arraycopy_range_checks done");
2524 }
2525
2526 //
2527 // Generate generic array copy stubs
2528 //
2529 // Input:
2530 // c_rarg0 - src oop
2531 // c_rarg1 - src_pos (32-bits)
2532 // c_rarg2 - dst oop
2533 // c_rarg3 - dst_pos (32-bits)
2534 // not Win64
2535 // c_rarg4 - element count (32-bits)
2536 // Win64
2537 // rsp+40 - element count (32-bits)
2538 //
2539 // Output:
2540 // rax == 0 - success
2541 // rax == -1^K - failure, where K is partial transfer count
2542 //
2543 address generate_generic_copy(const char *name,
2544 address byte_copy_entry, address short_copy_entry,
2545 address int_copy_entry, address oop_copy_entry,
2546 address long_copy_entry, address checkcast_copy_entry) {
2547
2548 Label L_failed, L_failed_0, L_objArray;
2549 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2550
2551 // Input registers
2552 const Register src = c_rarg0; // source array oop
2553 const Register src_pos = c_rarg1; // source position
2554 const Register dst = c_rarg2; // destination array oop
2555 const Register dst_pos = c_rarg3; // destination position
2556 #ifndef _WIN64
2557 const Register length = c_rarg4;
2558 #else
2559 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2560 #endif
2561
2562 { int modulus = CodeEntryAlignment;
2563 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2564 int advance = target - (__ offset() % modulus);
2565 if (advance < 0) advance += modulus;
2566 if (advance > 0) __ nop(advance);
2567 }
2568 StubCodeMark mark(this, "StubRoutines", name);
2569
2570 // Short-hop target to L_failed. Makes for denser prologue code.
2571 __ BIND(L_failed_0);
2572 __ jmp(L_failed);
2573 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2574
2575 __ align(CodeEntryAlignment);
2576 address start = __ pc();
2577
2578 __ enter(); // required for proper stackwalking of RuntimeStub frame
2579
2580 // bump this on entry, not on exit:
2581 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2582
2583 //-----------------------------------------------------------------------
2584 // Assembler stub will be used for this call to arraycopy
2585 // if the following conditions are met:
2586 //
2587 // (1) src and dst must not be null.
2588 // (2) src_pos must not be negative.
2589 // (3) dst_pos must not be negative.
2590 // (4) length must not be negative.
2591 // (5) src klass and dst klass should be the same and not NULL.
2592 // (6) src and dst should be arrays.
2593 // (7) src_pos + length must not exceed length of src.
2594 // (8) dst_pos + length must not exceed length of dst.
2595 //
2596
2597 // if (src == NULL) return -1;
2598 __ testptr(src, src); // src oop
2599 size_t j1off = __ offset();
2600 __ jccb(Assembler::zero, L_failed_0);
2601
2602 // if (src_pos < 0) return -1;
2603 __ testl(src_pos, src_pos); // src_pos (32-bits)
2604 __ jccb(Assembler::negative, L_failed_0);
2605
2606 // if (dst == NULL) return -1;
2607 __ testptr(dst, dst); // dst oop
2608 __ jccb(Assembler::zero, L_failed_0);
2609
2610 // if (dst_pos < 0) return -1;
2611 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2612 size_t j4off = __ offset();
2613 __ jccb(Assembler::negative, L_failed_0);
2614
2615 // The first four tests are very dense code,
2616 // but not quite dense enough to put four
2617 // jumps in a 16-byte instruction fetch buffer.
2618 // That's good, because some branch predicters
2619 // do not like jumps so close together.
2620 // Make sure of this.
2621 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2622
2623 // registers used as temp
2624 const Register r11_length = r11; // elements count to copy
2625 const Register r10_src_klass = r10; // array klass
2626
2627 // if (length < 0) return -1;
2628 __ movl(r11_length, length); // length (elements count, 32-bits value)
2629 __ testl(r11_length, r11_length);
2630 __ jccb(Assembler::negative, L_failed_0);
2631
2632 __ load_klass(r10_src_klass, src);
2633 #ifdef ASSERT
2634 // assert(src->klass() != NULL);
2635 {
2636 BLOCK_COMMENT("assert klasses not null {");
2637 Label L1, L2;
2638 __ testptr(r10_src_klass, r10_src_klass);
2639 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2640 __ bind(L1);
2641 __ stop("broken null klass");
2642 __ bind(L2);
2643 __ load_klass(rax, dst);
2644 __ cmpq(rax, 0);
2645 __ jcc(Assembler::equal, L1); // this would be broken also
2646 BLOCK_COMMENT("} assert klasses not null done");
2647 }
2648 #endif
2649
2650 // Load layout helper (32-bits)
2651 //
2652 // |array_tag| | header_size | element_type | |log2_element_size|
2653 // 32 30 24 16 8 2 0
2654 //
2655 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2656 //
2657
2658 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2659
2660 // Handle objArrays completely differently...
2661 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2662 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2663 __ jcc(Assembler::equal, L_objArray);
2664
2665 // if (src->klass() != dst->klass()) return -1;
2666 __ load_klass(rax, dst);
2667 __ cmpq(r10_src_klass, rax);
2668 __ jcc(Assembler::notEqual, L_failed);
2669
2670 const Register rax_lh = rax; // layout helper
2671 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2672
2673 // if (!src->is_Array()) return -1;
2674 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2675 __ jcc(Assembler::greaterEqual, L_failed);
2676
2677 // At this point, it is known to be a typeArray (array_tag 0x3).
2678 #ifdef ASSERT
2679 {
2680 BLOCK_COMMENT("assert primitive array {");
2681 Label L;
2682 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2683 __ jcc(Assembler::greaterEqual, L);
2684 __ stop("must be a primitive array");
2685 __ bind(L);
2686 BLOCK_COMMENT("} assert primitive array done");
2687 }
2688 #endif
2689
2690 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2691 r10, L_failed);
2692
2693 // TypeArrayKlass
2694 //
2695 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2696 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2697 //
2698
2699 const Register r10_offset = r10; // array offset
2700 const Register rax_elsize = rax_lh; // element size
2701
2702 __ movl(r10_offset, rax_lh);
2703 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2704 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2705 __ addptr(src, r10_offset); // src array offset
2706 __ addptr(dst, r10_offset); // dst array offset
2707 BLOCK_COMMENT("choose copy loop based on element size");
2708 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2709
2710 // next registers should be set before the jump to corresponding stub
2711 const Register from = c_rarg0; // source array address
2712 const Register to = c_rarg1; // destination array address
2713 const Register count = c_rarg2; // elements count
2714
2715 // 'from', 'to', 'count' registers should be set in such order
2716 // since they are the same as 'src', 'src_pos', 'dst'.
2717
2718 __ BIND(L_copy_bytes);
2719 __ cmpl(rax_elsize, 0);
2720 __ jccb(Assembler::notEqual, L_copy_shorts);
2721 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2722 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2723 __ movl2ptr(count, r11_length); // length
2724 __ jump(RuntimeAddress(byte_copy_entry));
2725
2726 __ BIND(L_copy_shorts);
2727 __ cmpl(rax_elsize, LogBytesPerShort);
2728 __ jccb(Assembler::notEqual, L_copy_ints);
2729 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2730 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2731 __ movl2ptr(count, r11_length); // length
2732 __ jump(RuntimeAddress(short_copy_entry));
2733
2734 __ BIND(L_copy_ints);
2735 __ cmpl(rax_elsize, LogBytesPerInt);
2736 __ jccb(Assembler::notEqual, L_copy_longs);
2737 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2738 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2739 __ movl2ptr(count, r11_length); // length
2740 __ jump(RuntimeAddress(int_copy_entry));
2741
2742 __ BIND(L_copy_longs);
2743 #ifdef ASSERT
2744 {
2745 BLOCK_COMMENT("assert long copy {");
2746 Label L;
2747 __ cmpl(rax_elsize, LogBytesPerLong);
2748 __ jcc(Assembler::equal, L);
2749 __ stop("must be long copy, but elsize is wrong");
2750 __ bind(L);
2751 BLOCK_COMMENT("} assert long copy done");
2752 }
2753 #endif
2754 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2755 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2756 __ movl2ptr(count, r11_length); // length
2757 __ jump(RuntimeAddress(long_copy_entry));
2758
2759 // ObjArrayKlass
2760 __ BIND(L_objArray);
2761 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2762
2763 Label L_plain_copy, L_checkcast_copy;
2764 // test array classes for subtyping
2765 __ load_klass(rax, dst);
2766 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2767 __ jcc(Assembler::notEqual, L_checkcast_copy);
2768
2769 // Identically typed arrays can be copied without element-wise checks.
2770 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2771 r10, L_failed);
2772
2773 __ lea(from, Address(src, src_pos, TIMES_OOP,
2774 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2775 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2776 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2777 __ movl2ptr(count, r11_length); // length
2778 __ BIND(L_plain_copy);
2779 __ jump(RuntimeAddress(oop_copy_entry));
2780
2781 __ BIND(L_checkcast_copy);
2782 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2783 {
2784 // Before looking at dst.length, make sure dst is also an objArray.
2785 __ cmpl(Address(rax, lh_offset), objArray_lh);
2786 __ jcc(Assembler::notEqual, L_failed);
2787
2788 // It is safe to examine both src.length and dst.length.
2789 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2790 rax, L_failed);
2791
2792 const Register r11_dst_klass = r11;
2793 __ load_klass(r11_dst_klass, dst); // reload
2794
2795 // Marshal the base address arguments now, freeing registers.
2796 __ lea(from, Address(src, src_pos, TIMES_OOP,
2797 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2798 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2799 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2800 __ movl(count, length); // length (reloaded)
2801 Register sco_temp = c_rarg3; // this register is free now
2802 assert_different_registers(from, to, count, sco_temp,
2803 r11_dst_klass, r10_src_klass);
2804 assert_clean_int(count, sco_temp);
2805
2806 // Generate the type check.
2807 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2808 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2809 assert_clean_int(sco_temp, rax);
2810 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2811
2812 // Fetch destination element klass from the ObjArrayKlass header.
2813 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2814 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2815 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2816 assert_clean_int(sco_temp, rax);
2817
2818 // the checkcast_copy loop needs two extra arguments:
2819 assert(c_rarg3 == sco_temp, "#3 already in place");
2820 // Set up arguments for checkcast_copy_entry.
2821 setup_arg_regs(4);
2822 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2823 __ jump(RuntimeAddress(checkcast_copy_entry));
2824 }
2825
2826 __ BIND(L_failed);
2827 __ xorptr(rax, rax);
2828 __ notptr(rax); // return -1
2829 __ leave(); // required for proper stackwalking of RuntimeStub frame
2830 __ ret(0);
2831
2832 return start;
2833 }
2834
2835 void generate_arraycopy_stubs() {
2836 address entry;
2837 address entry_jbyte_arraycopy;
2838 address entry_jshort_arraycopy;
2839 address entry_jint_arraycopy;
2840 address entry_oop_arraycopy;
2841 address entry_jlong_arraycopy;
2842 address entry_checkcast_arraycopy;
2843
2844 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2845 "jbyte_disjoint_arraycopy");
2846 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2847 "jbyte_arraycopy");
2848
2849 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2850 "jshort_disjoint_arraycopy");
2851 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2852 "jshort_arraycopy");
2853
2854 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2855 "jint_disjoint_arraycopy");
2856 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2857 &entry_jint_arraycopy, "jint_arraycopy");
2858
2859 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2860 "jlong_disjoint_arraycopy");
2861 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2862 &entry_jlong_arraycopy, "jlong_arraycopy");
2863
2864
2865 if (UseCompressedOops) {
2866 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2867 "oop_disjoint_arraycopy");
2868 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2869 &entry_oop_arraycopy, "oop_arraycopy");
2870 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2871 "oop_disjoint_arraycopy_uninit",
2872 /*dest_uninitialized*/true);
2873 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2874 NULL, "oop_arraycopy_uninit",
2875 /*dest_uninitialized*/true);
2876 } else {
2877 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2878 "oop_disjoint_arraycopy");
2879 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2880 &entry_oop_arraycopy, "oop_arraycopy");
2881 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2882 "oop_disjoint_arraycopy_uninit",
2883 /*dest_uninitialized*/true);
2884 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2885 NULL, "oop_arraycopy_uninit",
2886 /*dest_uninitialized*/true);
2887 }
2888
2889 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2890 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2891 /*dest_uninitialized*/true);
2892
2893 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2894 entry_jbyte_arraycopy,
2895 entry_jshort_arraycopy,
2896 entry_jint_arraycopy,
2897 entry_jlong_arraycopy);
2898 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2899 entry_jbyte_arraycopy,
2900 entry_jshort_arraycopy,
2901 entry_jint_arraycopy,
2902 entry_oop_arraycopy,
2903 entry_jlong_arraycopy,
2904 entry_checkcast_arraycopy);
2905
2906 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2907 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2908 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2909 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2910 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2911 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2912
2913 // We don't generate specialized code for HeapWord-aligned source
2914 // arrays, so just use the code we've already generated
2915 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2916 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2917
2918 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2919 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2920
2921 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2922 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2923
2924 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2925 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2926
2927 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2928 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2929
2930 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2931 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2932 }
2933
2934 // AES intrinsic stubs
2935 enum {AESBlockSize = 16};
2936
2937 address generate_key_shuffle_mask() {
2938 __ align(16);
2939 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2940 address start = __ pc();
2941 __ emit_data64( 0x0405060700010203, relocInfo::none );
2942 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
2943 return start;
2944 }
2945
2946 address generate_counter_shuffle_mask() {
2947 __ align(16);
2948 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2949 address start = __ pc();
2950 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
2951 __ emit_data64(0x0001020304050607, relocInfo::none);
2952 return start;
2953 }
2954
2955 // Utility routine for loading a 128-bit key word in little endian format
2956 // can optionally specify that the shuffle mask is already in an xmmregister
2957 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2958 __ movdqu(xmmdst, Address(key, offset));
2959 if (xmm_shuf_mask != NULL) {
2960 __ pshufb(xmmdst, xmm_shuf_mask);
2961 } else {
2962 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2963 }
2964 }
2965
2966 // Utility routine for increase 128bit counter (iv in CTR mode)
2967 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
2968 __ pextrq(reg, xmmdst, 0x0);
2969 __ addq(reg, inc_delta);
2970 __ pinsrq(xmmdst, reg, 0x0);
2971 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2972 __ pextrq(reg, xmmdst, 0x01); // Carry
2973 __ addq(reg, 0x01);
2974 __ pinsrq(xmmdst, reg, 0x01); //Carry end
2975 __ BIND(next_block); // next instruction
2976 }
2977
2978 // Arguments:
2979 //
2980 // Inputs:
2981 // c_rarg0 - source byte array address
2982 // c_rarg1 - destination byte array address
2983 // c_rarg2 - K (key) in little endian int array
2984 //
2985 address generate_aescrypt_encryptBlock() {
2986 assert(UseAES, "need AES instructions and misaligned SSE support");
2987 __ align(CodeEntryAlignment);
2988 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2989 Label L_doLast;
2990 address start = __ pc();
2991
2992 const Register from = c_rarg0; // source array address
2993 const Register to = c_rarg1; // destination array address
2994 const Register key = c_rarg2; // key array address
2995 const Register keylen = rax;
2996
2997 const XMMRegister xmm_result = xmm0;
2998 const XMMRegister xmm_key_shuf_mask = xmm1;
2999 // On win64 xmm6-xmm15 must be preserved so don't use them.
3000 const XMMRegister xmm_temp1 = xmm2;
3001 const XMMRegister xmm_temp2 = xmm3;
3002 const XMMRegister xmm_temp3 = xmm4;
3003 const XMMRegister xmm_temp4 = xmm5;
3004
3005 __ enter(); // required for proper stackwalking of RuntimeStub frame
3006
3007 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3008 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3009
3010 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3011 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
3012
3013 // For encryption, the java expanded key ordering is just what we need
3014 // we don't know if the key is aligned, hence not using load-execute form
3015
3016 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3017 __ pxor(xmm_result, xmm_temp1);
3018
3019 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3020 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3021 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3022 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3023
3024 __ aesenc(xmm_result, xmm_temp1);
3025 __ aesenc(xmm_result, xmm_temp2);
3026 __ aesenc(xmm_result, xmm_temp3);
3027 __ aesenc(xmm_result, xmm_temp4);
3028
3029 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3030 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3031 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3032 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3033
3034 __ aesenc(xmm_result, xmm_temp1);
3035 __ aesenc(xmm_result, xmm_temp2);
3036 __ aesenc(xmm_result, xmm_temp3);
3037 __ aesenc(xmm_result, xmm_temp4);
3038
3039 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3040 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3041
3042 __ cmpl(keylen, 44);
3043 __ jccb(Assembler::equal, L_doLast);
3044
3045 __ aesenc(xmm_result, xmm_temp1);
3046 __ aesenc(xmm_result, xmm_temp2);
3047
3048 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3049 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3050
3051 __ cmpl(keylen, 52);
3052 __ jccb(Assembler::equal, L_doLast);
3053
3054 __ aesenc(xmm_result, xmm_temp1);
3055 __ aesenc(xmm_result, xmm_temp2);
3056
3057 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3058 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3059
3060 __ BIND(L_doLast);
3061 __ aesenc(xmm_result, xmm_temp1);
3062 __ aesenclast(xmm_result, xmm_temp2);
3063 __ movdqu(Address(to, 0), xmm_result); // store the result
3064 __ xorptr(rax, rax); // return 0
3065 __ leave(); // required for proper stackwalking of RuntimeStub frame
3066 __ ret(0);
3067
3068 return start;
3069 }
3070
3071
3072 // Arguments:
3073 //
3074 // Inputs:
3075 // c_rarg0 - source byte array address
3076 // c_rarg1 - destination byte array address
3077 // c_rarg2 - K (key) in little endian int array
3078 //
3079 address generate_aescrypt_decryptBlock() {
3080 assert(UseAES, "need AES instructions and misaligned SSE support");
3081 __ align(CodeEntryAlignment);
3082 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3083 Label L_doLast;
3084 address start = __ pc();
3085
3086 const Register from = c_rarg0; // source array address
3087 const Register to = c_rarg1; // destination array address
3088 const Register key = c_rarg2; // key array address
3089 const Register keylen = rax;
3090
3091 const XMMRegister xmm_result = xmm0;
3092 const XMMRegister xmm_key_shuf_mask = xmm1;
3093 // On win64 xmm6-xmm15 must be preserved so don't use them.
3094 const XMMRegister xmm_temp1 = xmm2;
3095 const XMMRegister xmm_temp2 = xmm3;
3096 const XMMRegister xmm_temp3 = xmm4;
3097 const XMMRegister xmm_temp4 = xmm5;
3098
3099 __ enter(); // required for proper stackwalking of RuntimeStub frame
3100
3101 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3102 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3103
3104 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3105 __ movdqu(xmm_result, Address(from, 0));
3106
3107 // for decryption java expanded key ordering is rotated one position from what we want
3108 // so we start from 0x10 here and hit 0x00 last
3109 // we don't know if the key is aligned, hence not using load-execute form
3110 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3111 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3112 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3113 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3114
3115 __ pxor (xmm_result, xmm_temp1);
3116 __ aesdec(xmm_result, xmm_temp2);
3117 __ aesdec(xmm_result, xmm_temp3);
3118 __ aesdec(xmm_result, xmm_temp4);
3119
3120 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3121 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3122 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3123 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3124
3125 __ aesdec(xmm_result, xmm_temp1);
3126 __ aesdec(xmm_result, xmm_temp2);
3127 __ aesdec(xmm_result, xmm_temp3);
3128 __ aesdec(xmm_result, xmm_temp4);
3129
3130 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3131 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3132 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3133
3134 __ cmpl(keylen, 44);
3135 __ jccb(Assembler::equal, L_doLast);
3136
3137 __ aesdec(xmm_result, xmm_temp1);
3138 __ aesdec(xmm_result, xmm_temp2);
3139
3140 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3141 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3142
3143 __ cmpl(keylen, 52);
3144 __ jccb(Assembler::equal, L_doLast);
3145
3146 __ aesdec(xmm_result, xmm_temp1);
3147 __ aesdec(xmm_result, xmm_temp2);
3148
3149 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3150 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3151
3152 __ BIND(L_doLast);
3153 __ aesdec(xmm_result, xmm_temp1);
3154 __ aesdec(xmm_result, xmm_temp2);
3155
3156 // for decryption the aesdeclast operation is always on key+0x00
3157 __ aesdeclast(xmm_result, xmm_temp3);
3158 __ movdqu(Address(to, 0), xmm_result); // store the result
3159 __ xorptr(rax, rax); // return 0
3160 __ leave(); // required for proper stackwalking of RuntimeStub frame
3161 __ ret(0);
3162
3163 return start;
3164 }
3165
3166
3167 // Arguments:
3168 //
3169 // Inputs:
3170 // c_rarg0 - source byte array address
3171 // c_rarg1 - destination byte array address
3172 // c_rarg2 - K (key) in little endian int array
3173 // c_rarg3 - r vector byte array address
3174 // c_rarg4 - input length
3175 //
3176 // Output:
3177 // rax - input length
3178 //
3179 address generate_cipherBlockChaining_encryptAESCrypt() {
3180 assert(UseAES, "need AES instructions and misaligned SSE support");
3181 __ align(CodeEntryAlignment);
3182 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3183 address start = __ pc();
3184
3185 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3186 const Register from = c_rarg0; // source array address
3187 const Register to = c_rarg1; // destination array address
3188 const Register key = c_rarg2; // key array address
3189 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3190 // and left with the results of the last encryption block
3191 #ifndef _WIN64
3192 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3193 #else
3194 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3195 const Register len_reg = r11; // pick the volatile windows register
3196 #endif
3197 const Register pos = rax;
3198
3199 // xmm register assignments for the loops below
3200 const XMMRegister xmm_result = xmm0;
3201 const XMMRegister xmm_temp = xmm1;
3202 // keys 0-10 preloaded into xmm2-xmm12
3203 const int XMM_REG_NUM_KEY_FIRST = 2;
3204 const int XMM_REG_NUM_KEY_LAST = 15;
3205 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3206 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3207 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3208 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3209 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3210
3211 __ enter(); // required for proper stackwalking of RuntimeStub frame
3212
3213 #ifdef _WIN64
3214 // on win64, fill len_reg from stack position
3215 __ movl(len_reg, len_mem);
3216 #else
3217 __ push(len_reg); // Save
3218 #endif
3219
3220 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3221 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3222 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3223 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3224 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3225 offset += 0x10;
3226 }
3227 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3228
3229 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3230 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3231 __ cmpl(rax, 44);
3232 __ jcc(Assembler::notEqual, L_key_192_256);
3233
3234 // 128 bit code follows here
3235 __ movptr(pos, 0);
3236 __ align(OptoLoopAlignment);
3237
3238 __ BIND(L_loopTop_128);
3239 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3240 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3241 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3242 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3243 __ aesenc(xmm_result, as_XMMRegister(rnum));
3244 }
3245 __ aesenclast(xmm_result, xmm_key10);
3246 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3247 // no need to store r to memory until we exit
3248 __ addptr(pos, AESBlockSize);
3249 __ subptr(len_reg, AESBlockSize);
3250 __ jcc(Assembler::notEqual, L_loopTop_128);
3251
3252 __ BIND(L_exit);
3253 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3254
3255 #ifdef _WIN64
3256 __ movl(rax, len_mem);
3257 #else
3258 __ pop(rax); // return length
3259 #endif
3260 __ leave(); // required for proper stackwalking of RuntimeStub frame
3261 __ ret(0);
3262
3263 __ BIND(L_key_192_256);
3264 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3265 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3266 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3267 __ cmpl(rax, 52);
3268 __ jcc(Assembler::notEqual, L_key_256);
3269
3270 // 192-bit code follows here (could be changed to use more xmm registers)
3271 __ movptr(pos, 0);
3272 __ align(OptoLoopAlignment);
3273
3274 __ BIND(L_loopTop_192);
3275 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3276 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3277 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3278 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3279 __ aesenc(xmm_result, as_XMMRegister(rnum));
3280 }
3281 __ aesenclast(xmm_result, xmm_key12);
3282 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3283 // no need to store r to memory until we exit
3284 __ addptr(pos, AESBlockSize);
3285 __ subptr(len_reg, AESBlockSize);
3286 __ jcc(Assembler::notEqual, L_loopTop_192);
3287 __ jmp(L_exit);
3288
3289 __ BIND(L_key_256);
3290 // 256-bit code follows here (could be changed to use more xmm registers)
3291 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3292 __ movptr(pos, 0);
3293 __ align(OptoLoopAlignment);
3294
3295 __ BIND(L_loopTop_256);
3296 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3297 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3298 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3299 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3300 __ aesenc(xmm_result, as_XMMRegister(rnum));
3301 }
3302 load_key(xmm_temp, key, 0xe0);
3303 __ aesenclast(xmm_result, xmm_temp);
3304 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3305 // no need to store r to memory until we exit
3306 __ addptr(pos, AESBlockSize);
3307 __ subptr(len_reg, AESBlockSize);
3308 __ jcc(Assembler::notEqual, L_loopTop_256);
3309 __ jmp(L_exit);
3310
3311 return start;
3312 }
3313
3314 // Safefetch stubs.
3315 void generate_safefetch(const char* name, int size, address* entry,
3316 address* fault_pc, address* continuation_pc) {
3317 // safefetch signatures:
3318 // int SafeFetch32(int* adr, int errValue);
3319 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3320 //
3321 // arguments:
3322 // c_rarg0 = adr
3323 // c_rarg1 = errValue
3324 //
3325 // result:
3326 // PPC_RET = *adr or errValue
3327
3328 StubCodeMark mark(this, "StubRoutines", name);
3329
3330 // Entry point, pc or function descriptor.
3331 *entry = __ pc();
3332
3333 // Load *adr into c_rarg1, may fault.
3334 *fault_pc = __ pc();
3335 switch (size) {
3336 case 4:
3337 // int32_t
3338 __ movl(c_rarg1, Address(c_rarg0, 0));
3339 break;
3340 case 8:
3341 // int64_t
3342 __ movq(c_rarg1, Address(c_rarg0, 0));
3343 break;
3344 default:
3345 ShouldNotReachHere();
3346 }
3347
3348 // return errValue or *adr
3349 *continuation_pc = __ pc();
3350 __ movq(rax, c_rarg1);
3351 __ ret(0);
3352 }
3353
3354 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3355 // to hide instruction latency
3356 //
3357 // Arguments:
3358 //
3359 // Inputs:
3360 // c_rarg0 - source byte array address
3361 // c_rarg1 - destination byte array address
3362 // c_rarg2 - K (key) in little endian int array
3363 // c_rarg3 - r vector byte array address
3364 // c_rarg4 - input length
3365 //
3366 // Output:
3367 // rax - input length
3368 //
3369 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3370 assert(UseAES, "need AES instructions and misaligned SSE support");
3371 __ align(CodeEntryAlignment);
3372 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3373 address start = __ pc();
3374
3375 const Register from = c_rarg0; // source array address
3376 const Register to = c_rarg1; // destination array address
3377 const Register key = c_rarg2; // key array address
3378 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3379 // and left with the results of the last encryption block
3380 #ifndef _WIN64
3381 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3382 #else
3383 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3384 const Register len_reg = r11; // pick the volatile windows register
3385 #endif
3386 const Register pos = rax;
3387
3388 const int PARALLEL_FACTOR = 4;
3389 const int ROUNDS[3] = { 10, 12, 14 }; // aes rounds for key128, key192, key256
3390
3391 Label L_exit;
3392 Label L_singleBlock_loopTopHead[3]; // 128, 192, 256
3393 Label L_singleBlock_loopTopHead2[3]; // 128, 192, 256
3394 Label L_singleBlock_loopTop[3]; // 128, 192, 256
3395 Label L_multiBlock_loopTopHead[3]; // 128, 192, 256
3396 Label L_multiBlock_loopTop[3]; // 128, 192, 256
3397
3398 // keys 0-10 preloaded into xmm5-xmm15
3399 const int XMM_REG_NUM_KEY_FIRST = 5;
3400 const int XMM_REG_NUM_KEY_LAST = 15;
3401 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3402 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3403
3404 __ enter(); // required for proper stackwalking of RuntimeStub frame
3405
3406 #ifdef _WIN64
3407 // on win64, fill len_reg from stack position
3408 __ movl(len_reg, len_mem);
3409 #else
3410 __ push(len_reg); // Save
3411 #endif
3412 __ push(rbx);
3413 // the java expanded key ordering is rotated one position from what we want
3414 // so we start from 0x10 here and hit 0x00 last
3415 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3416 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3417 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3418 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3419 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3420 offset += 0x10;
3421 }
3422 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3423
3424 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3425
3426 // registers holding the four results in the parallelized loop
3427 const XMMRegister xmm_result0 = xmm0;
3428 const XMMRegister xmm_result1 = xmm2;
3429 const XMMRegister xmm_result2 = xmm3;
3430 const XMMRegister xmm_result3 = xmm4;
3431
3432 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3433
3434 __ xorptr(pos, pos);
3435
3436 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3437 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3438 __ cmpl(rbx, 52);
3439 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[1]);
3440 __ cmpl(rbx, 60);
3441 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[2]);
3442
3443 #define DoFour(opc, src_reg) \
3444 __ opc(xmm_result0, src_reg); \
3445 __ opc(xmm_result1, src_reg); \
3446 __ opc(xmm_result2, src_reg); \
3447 __ opc(xmm_result3, src_reg); \
3448
3449 for (int k = 0; k < 3; ++k) {
3450 __ BIND(L_multiBlock_loopTopHead[k]);
3451 if (k != 0) {
3452 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3453 __ jcc(Assembler::less, L_singleBlock_loopTopHead2[k]);
3454 }
3455 if (k == 1) {
3456 __ subptr(rsp, 6 * wordSize);
3457 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3458 load_key(xmm15, key, 0xb0); // 0xb0; 192-bit key goes up to 0xc0
3459 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3460 load_key(xmm1, key, 0xc0); // 0xc0;
3461 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3462 } else if (k == 2) {
3463 __ subptr(rsp, 10 * wordSize);
3464 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3465 load_key(xmm15, key, 0xd0); // 0xd0; 256-bit key goes upto 0xe0
3466 __ movdqu(Address(rsp, 6 * wordSize), xmm15);
3467 load_key(xmm1, key, 0xe0); // 0xe0;
3468 __ movdqu(Address(rsp, 8 * wordSize), xmm1);
3469 load_key(xmm15, key, 0xb0); // 0xb0;
3470 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3471 load_key(xmm1, key, 0xc0); // 0xc0;
3472 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3473 }
3474 __ align(OptoLoopAlignment);
3475 __ BIND(L_multiBlock_loopTop[k]);
3476 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3477 __ jcc(Assembler::less, L_singleBlock_loopTopHead[k]);
3478
3479 if (k != 0) {
3480 __ movdqu(xmm15, Address(rsp, 2 * wordSize));
3481 __ movdqu(xmm1, Address(rsp, 4 * wordSize));
3482 }
3483
3484 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
3485 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3486 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3487 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3488
3489 DoFour(pxor, xmm_key_first);
3490 if (k == 0) {
3491 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
3492 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3493 }
3494 DoFour(aesdeclast, xmm_key_last);
3495 } else if (k == 1) {
3496 for (int rnum = 1; rnum <= ROUNDS[k]-2; rnum++) {
3497 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3498 }
3499 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3500 DoFour(aesdec, xmm1); // key : 0xc0
3501 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3502 DoFour(aesdeclast, xmm_key_last);
3503 } else if (k == 2) {
3504 for (int rnum = 1; rnum <= ROUNDS[k] - 4; rnum++) {
3505 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3506 }
3507 DoFour(aesdec, xmm1); // key : 0xc0
3508 __ movdqu(xmm15, Address(rsp, 6 * wordSize));
3509 __ movdqu(xmm1, Address(rsp, 8 * wordSize));
3510 DoFour(aesdec, xmm15); // key : 0xd0
3511 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3512 DoFour(aesdec, xmm1); // key : 0xe0
3513 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3514 DoFour(aesdeclast, xmm_key_last);
3515 }
3516
3517 // for each result, xor with the r vector of previous cipher block
3518 __ pxor(xmm_result0, xmm_prev_block_cipher);
3519 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3520 __ pxor(xmm_result1, xmm_prev_block_cipher);
3521 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3522 __ pxor(xmm_result2, xmm_prev_block_cipher);
3523 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3524 __ pxor(xmm_result3, xmm_prev_block_cipher);
3525 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
3526 if (k != 0) {
3527 __ movdqu(Address(rvec, 0x00), xmm_prev_block_cipher);
3528 }
3529
3530 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3531 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3532 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3533 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3534
3535 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize);
3536 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize);
3537 __ jmp(L_multiBlock_loopTop[k]);
3538
3539 // registers used in the non-parallelized loops
3540 // xmm register assignments for the loops below
3541 const XMMRegister xmm_result = xmm0;
3542 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3543 const XMMRegister xmm_key11 = xmm3;
3544 const XMMRegister xmm_key12 = xmm4;
3545 const XMMRegister key_tmp = xmm4;
3546
3547 __ BIND(L_singleBlock_loopTopHead[k]);
3548 if (k == 1) {
3549 __ addptr(rsp, 6 * wordSize);
3550 } else if (k == 2) {
3551 __ addptr(rsp, 10 * wordSize);
3552 }
3553 __ cmpptr(len_reg, 0); // any blocks left??
3554 __ jcc(Assembler::equal, L_exit);
3555 __ BIND(L_singleBlock_loopTopHead2[k]);
3556 if (k == 1) {
3557 load_key(xmm_key11, key, 0xb0); // 0xb0; 192-bit key goes upto 0xc0
3558 load_key(xmm_key12, key, 0xc0); // 0xc0; 192-bit key goes upto 0xc0
3559 }
3560 if (k == 2) {
3561 load_key(xmm_key11, key, 0xb0); // 0xb0; 256-bit key goes upto 0xe0
3562 }
3563 __ align(OptoLoopAlignment);
3564 __ BIND(L_singleBlock_loopTop[k]);
3565 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3566 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3567 __ pxor(xmm_result, xmm_key_first); // do the aes dec rounds
3568 for (int rnum = 1; rnum <= 9 ; rnum++) {
3569 __ aesdec(xmm_result, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3570 }
3571 if (k == 1) {
3572 __ aesdec(xmm_result, xmm_key11);
3573 __ aesdec(xmm_result, xmm_key12);
3574 }
3575 if (k == 2) {
3576 __ aesdec(xmm_result, xmm_key11);
3577 load_key(key_tmp, key, 0xc0);
3578 __ aesdec(xmm_result, key_tmp);
3579 load_key(key_tmp, key, 0xd0);
3580 __ aesdec(xmm_result, key_tmp);
3581 load_key(key_tmp, key, 0xe0);
3582 __ aesdec(xmm_result, key_tmp);
3583 }
3584
3585 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3586 __ pxor(xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3587 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3588 // no need to store r to memory until we exit
3589 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3590 __ addptr(pos, AESBlockSize);
3591 __ subptr(len_reg, AESBlockSize);
3592 __ jcc(Assembler::notEqual, L_singleBlock_loopTop[k]);
3593 if (k != 2) {
3594 __ jmp(L_exit);
3595 }
3596 } //for 128/192/256
3597
3598 __ BIND(L_exit);
3599 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3600 __ pop(rbx);
3601 #ifdef _WIN64
3602 __ movl(rax, len_mem);
3603 #else
3604 __ pop(rax); // return length
3605 #endif
3606 __ leave(); // required for proper stackwalking of RuntimeStub frame
3607 __ ret(0);
3608 return start;
3609 }
3610
3611 address generate_upper_word_mask() {
3612 __ align(64);
3613 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3614 address start = __ pc();
3615 __ emit_data64(0x0000000000000000, relocInfo::none);
3616 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
3617 return start;
3618 }
3619
3620 address generate_shuffle_byte_flip_mask() {
3621 __ align(64);
3622 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3623 address start = __ pc();
3624 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3625 __ emit_data64(0x0001020304050607, relocInfo::none);
3626 return start;
3627 }
3628
3629 // ofs and limit are use for multi-block byte array.
3630 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3631 address generate_sha1_implCompress(bool multi_block, const char *name) {
3632 __ align(CodeEntryAlignment);
3633 StubCodeMark mark(this, "StubRoutines", name);
3634 address start = __ pc();
3635
3636 Register buf = c_rarg0;
3637 Register state = c_rarg1;
3638 Register ofs = c_rarg2;
3639 Register limit = c_rarg3;
3640
3641 const XMMRegister abcd = xmm0;
3642 const XMMRegister e0 = xmm1;
3643 const XMMRegister e1 = xmm2;
3644 const XMMRegister msg0 = xmm3;
3645
3646 const XMMRegister msg1 = xmm4;
3647 const XMMRegister msg2 = xmm5;
3648 const XMMRegister msg3 = xmm6;
3649 const XMMRegister shuf_mask = xmm7;
3650
3651 __ enter();
3652
3653 __ subptr(rsp, 4 * wordSize);
3654
3655 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3656 buf, state, ofs, limit, rsp, multi_block);
3657
3658 __ addptr(rsp, 4 * wordSize);
3659
3660 __ leave();
3661 __ ret(0);
3662 return start;
3663 }
3664
3665 address generate_pshuffle_byte_flip_mask() {
3666 __ align(64);
3667 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3668 address start = __ pc();
3669 __ emit_data64(0x0405060700010203, relocInfo::none);
3670 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3671
3672 if (VM_Version::supports_avx2()) {
3673 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
3674 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3675 // _SHUF_00BA
3676 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3677 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3678 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3679 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3680 // _SHUF_DC00
3681 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3682 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3683 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3684 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3685 }
3686
3687 return start;
3688 }
3689
3690 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
3691 address generate_pshuffle_byte_flip_mask_sha512() {
3692 __ align(32);
3693 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
3694 address start = __ pc();
3695 if (VM_Version::supports_avx2()) {
3696 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
3697 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3698 __ emit_data64(0x1011121314151617, relocInfo::none);
3699 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
3700 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
3701 __ emit_data64(0x0000000000000000, relocInfo::none);
3702 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3703 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3704 }
3705
3706 return start;
3707 }
3708
3709 // ofs and limit are use for multi-block byte array.
3710 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3711 address generate_sha256_implCompress(bool multi_block, const char *name) {
3712 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
3713 __ align(CodeEntryAlignment);
3714 StubCodeMark mark(this, "StubRoutines", name);
3715 address start = __ pc();
3716
3717 Register buf = c_rarg0;
3718 Register state = c_rarg1;
3719 Register ofs = c_rarg2;
3720 Register limit = c_rarg3;
3721
3722 const XMMRegister msg = xmm0;
3723 const XMMRegister state0 = xmm1;
3724 const XMMRegister state1 = xmm2;
3725 const XMMRegister msgtmp0 = xmm3;
3726
3727 const XMMRegister msgtmp1 = xmm4;
3728 const XMMRegister msgtmp2 = xmm5;
3729 const XMMRegister msgtmp3 = xmm6;
3730 const XMMRegister msgtmp4 = xmm7;
3731
3732 const XMMRegister shuf_mask = xmm8;
3733
3734 __ enter();
3735
3736 __ subptr(rsp, 4 * wordSize);
3737
3738 if (VM_Version::supports_sha()) {
3739 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3740 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3741 } else if (VM_Version::supports_avx2()) {
3742 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3743 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3744 }
3745 __ addptr(rsp, 4 * wordSize);
3746 __ vzeroupper();
3747 __ leave();
3748 __ ret(0);
3749 return start;
3750 }
3751
3752 address generate_sha512_implCompress(bool multi_block, const char *name) {
3753 assert(VM_Version::supports_avx2(), "");
3754 assert(VM_Version::supports_bmi2(), "");
3755 __ align(CodeEntryAlignment);
3756 StubCodeMark mark(this, "StubRoutines", name);
3757 address start = __ pc();
3758
3759 Register buf = c_rarg0;
3760 Register state = c_rarg1;
3761 Register ofs = c_rarg2;
3762 Register limit = c_rarg3;
3763
3764 const XMMRegister msg = xmm0;
3765 const XMMRegister state0 = xmm1;
3766 const XMMRegister state1 = xmm2;
3767 const XMMRegister msgtmp0 = xmm3;
3768 const XMMRegister msgtmp1 = xmm4;
3769 const XMMRegister msgtmp2 = xmm5;
3770 const XMMRegister msgtmp3 = xmm6;
3771 const XMMRegister msgtmp4 = xmm7;
3772
3773 const XMMRegister shuf_mask = xmm8;
3774
3775 __ enter();
3776
3777 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3778 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3779
3780 __ vzeroupper();
3781 __ leave();
3782 __ ret(0);
3783 return start;
3784 }
3785
3786 // This is a version of CTR/AES crypt which does 6 blocks in a loop at a time
3787 // to hide instruction latency
3788 //
3789 // Arguments:
3790 //
3791 // Inputs:
3792 // c_rarg0 - source byte array address
3793 // c_rarg1 - destination byte array address
3794 // c_rarg2 - K (key) in little endian int array
3795 // c_rarg3 - counter vector byte array address
3796 // Linux
3797 // c_rarg4 - input length
3798 // c_rarg5 - saved encryptedCounter start
3799 // rbp + 6 * wordSize - saved used length
3800 // Windows
3801 // rbp + 6 * wordSize - input length
3802 // rbp + 7 * wordSize - saved encryptedCounter start
3803 // rbp + 8 * wordSize - saved used length
3804 //
3805 // Output:
3806 // rax - input length
3807 //
3808 address generate_counterMode_AESCrypt_Parallel() {
3809 assert(UseAES, "need AES instructions and misaligned SSE support");
3810 __ align(CodeEntryAlignment);
3811 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
3812 address start = __ pc();
3813 const Register from = c_rarg0; // source array address
3814 const Register to = c_rarg1; // destination array address
3815 const Register key = c_rarg2; // key array address
3816 const Register counter = c_rarg3; // counter byte array initialized from counter array address
3817 // and updated with the incremented counter in the end
3818 #ifndef _WIN64
3819 const Register len_reg = c_rarg4;
3820 const Register saved_encCounter_start = c_rarg5;
3821 const Register used_addr = r10;
3822 const Address used_mem(rbp, 2 * wordSize);
3823 const Register used = r11;
3824 #else
3825 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3826 const Address saved_encCounter_mem(rbp, 7 * wordSize); // length is on stack on Win64
3827 const Address used_mem(rbp, 8 * wordSize); // length is on stack on Win64
3828 const Register len_reg = r10; // pick the first volatile windows register
3829 const Register saved_encCounter_start = r11;
3830 const Register used_addr = r13;
3831 const Register used = r14;
3832 #endif
3833 const Register pos = rax;
3834
3835 const int PARALLEL_FACTOR = 6;
3836 const XMMRegister xmm_counter_shuf_mask = xmm0;
3837 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3838 const XMMRegister xmm_curr_counter = xmm2;
3839
3840 const XMMRegister xmm_key_tmp0 = xmm3;
3841 const XMMRegister xmm_key_tmp1 = xmm4;
3842
3843 // registers holding the four results in the parallelized loop
3844 const XMMRegister xmm_result0 = xmm5;
3845 const XMMRegister xmm_result1 = xmm6;
3846 const XMMRegister xmm_result2 = xmm7;
3847 const XMMRegister xmm_result3 = xmm8;
3848 const XMMRegister xmm_result4 = xmm9;
3849 const XMMRegister xmm_result5 = xmm10;
3850
3851 const XMMRegister xmm_from0 = xmm11;
3852 const XMMRegister xmm_from1 = xmm12;
3853 const XMMRegister xmm_from2 = xmm13;
3854 const XMMRegister xmm_from3 = xmm14; //the last one is xmm14. we have to preserve it on WIN64.
3855 const XMMRegister xmm_from4 = xmm3; //reuse xmm3~4. Because xmm_key_tmp0~1 are useless when loading input text
3856 const XMMRegister xmm_from5 = xmm4;
3857
3858 //for key_128, key_192, key_256
3859 const int rounds[3] = {10, 12, 14};
3860 Label L_exit_preLoop, L_preLoop_start;
3861 Label L_multiBlock_loopTop[3];
3862 Label L_singleBlockLoopTop[3];
3863 Label L__incCounter[3][6]; //for 6 blocks
3864 Label L__incCounter_single[3]; //for single block, key128, key192, key256
3865 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];
3866 Label L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
3867
3868 Label L_exit;
3869
3870 __ enter(); // required for proper stackwalking of RuntimeStub frame
3871
3872 #ifdef _WIN64
3873 // allocate spill slots for r13, r14
3874 enum {
3875 saved_r13_offset,
3876 saved_r14_offset
3877 };
3878 __ subptr(rsp, 2 * wordSize);
3879 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
3880 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
3881
3882 // on win64, fill len_reg from stack position
3883 __ movl(len_reg, len_mem);
3884 __ movptr(saved_encCounter_start, saved_encCounter_mem);
3885 __ movptr(used_addr, used_mem);
3886 __ movl(used, Address(used_addr, 0));
3887 #else
3888 __ push(len_reg); // Save
3889 __ movptr(used_addr, used_mem);
3890 __ movl(used, Address(used_addr, 0));
3891 #endif
3892
3893 __ push(rbx); // Save RBX
3894 __ movdqu(xmm_curr_counter, Address(counter, 0x00)); // initialize counter with initial counter
3895 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()), pos); // pos as scratch
3896 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled
3897 __ movptr(pos, 0);
3898
3899 // Use the partially used encrpyted counter from last invocation
3900 __ BIND(L_preLoop_start);
3901 __ cmpptr(used, 16);
3902 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
3903 __ cmpptr(len_reg, 0);
3904 __ jcc(Assembler::lessEqual, L_exit_preLoop);
3905 __ movb(rbx, Address(saved_encCounter_start, used));
3906 __ xorb(rbx, Address(from, pos));
3907 __ movb(Address(to, pos), rbx);
3908 __ addptr(pos, 1);
3909 __ addptr(used, 1);
3910 __ subptr(len_reg, 1);
3911
3912 __ jmp(L_preLoop_start);
3913
3914 __ BIND(L_exit_preLoop);
3915 __ movl(Address(used_addr, 0), used);
3916
3917 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
3918 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()), rbx); // rbx as scratch
3919 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3920 __ cmpl(rbx, 52);
3921 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
3922 __ cmpl(rbx, 60);
3923 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
3924
3925 #define CTR_DoSix(opc, src_reg) \
3926 __ opc(xmm_result0, src_reg); \
3927 __ opc(xmm_result1, src_reg); \
3928 __ opc(xmm_result2, src_reg); \
3929 __ opc(xmm_result3, src_reg); \
3930 __ opc(xmm_result4, src_reg); \
3931 __ opc(xmm_result5, src_reg);
3932
3933 // k == 0 : generate code for key_128
3934 // k == 1 : generate code for key_192
3935 // k == 2 : generate code for key_256
3936 for (int k = 0; k < 3; ++k) {
3937 //multi blocks starts here
3938 __ align(OptoLoopAlignment);
3939 __ BIND(L_multiBlock_loopTop[k]);
3940 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
3941 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
3942 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
3943
3944 //load, then increase counters
3945 CTR_DoSix(movdqa, xmm_curr_counter);
3946 inc_counter(rbx, xmm_result1, 0x01, L__incCounter[k][0]);
3947 inc_counter(rbx, xmm_result2, 0x02, L__incCounter[k][1]);
3948 inc_counter(rbx, xmm_result3, 0x03, L__incCounter[k][2]);
3949 inc_counter(rbx, xmm_result4, 0x04, L__incCounter[k][3]);
3950 inc_counter(rbx, xmm_result5, 0x05, L__incCounter[k][4]);
3951 inc_counter(rbx, xmm_curr_counter, 0x06, L__incCounter[k][5]);
3952 CTR_DoSix(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
3953 CTR_DoSix(pxor, xmm_key_tmp0); //PXOR with Round 0 key
3954
3955 //load two ROUND_KEYs at a time
3956 for (int i = 1; i < rounds[k]; ) {
3957 load_key(xmm_key_tmp1, key, (0x10 * i), xmm_key_shuf_mask);
3958 load_key(xmm_key_tmp0, key, (0x10 * (i+1)), xmm_key_shuf_mask);
3959 CTR_DoSix(aesenc, xmm_key_tmp1);
3960 i++;
3961 if (i != rounds[k]) {
3962 CTR_DoSix(aesenc, xmm_key_tmp0);
3963 } else {
3964 CTR_DoSix(aesenclast, xmm_key_tmp0);
3965 }
3966 i++;
3967 }
3968
3969 // get next PARALLEL_FACTOR blocks into xmm_result registers
3970 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3971 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3972 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3973 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3974 __ movdqu(xmm_from4, Address(from, pos, Address::times_1, 4 * AESBlockSize));
3975 __ movdqu(xmm_from5, Address(from, pos, Address::times_1, 5 * AESBlockSize));
3976
3977 __ pxor(xmm_result0, xmm_from0);
3978 __ pxor(xmm_result1, xmm_from1);
3979 __ pxor(xmm_result2, xmm_from2);
3980 __ pxor(xmm_result3, xmm_from3);
3981 __ pxor(xmm_result4, xmm_from4);
3982 __ pxor(xmm_result5, xmm_from5);
3983
3984 // store 6 results into the next 64 bytes of output
3985 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
3986 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3987 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3988 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3989 __ movdqu(Address(to, pos, Address::times_1, 4 * AESBlockSize), xmm_result4);
3990 __ movdqu(Address(to, pos, Address::times_1, 5 * AESBlockSize), xmm_result5);
3991
3992 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
3993 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
3994 __ jmp(L_multiBlock_loopTop[k]);
3995
3996 // singleBlock starts here
3997 __ align(OptoLoopAlignment);
3998 __ BIND(L_singleBlockLoopTop[k]);
3999 __ cmpptr(len_reg, 0);
4000 __ jcc(Assembler::lessEqual, L_exit);
4001 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4002 __ movdqa(xmm_result0, xmm_curr_counter);
4003 inc_counter(rbx, xmm_curr_counter, 0x01, L__incCounter_single[k]);
4004 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
4005 __ pxor(xmm_result0, xmm_key_tmp0);
4006 for (int i = 1; i < rounds[k]; i++) {
4007 load_key(xmm_key_tmp0, key, (0x10 * i), xmm_key_shuf_mask);
4008 __ aesenc(xmm_result0, xmm_key_tmp0);
4009 }
4010 load_key(xmm_key_tmp0, key, (rounds[k] * 0x10), xmm_key_shuf_mask);
4011 __ aesenclast(xmm_result0, xmm_key_tmp0);
4012 __ cmpptr(len_reg, AESBlockSize);
4013 __ jcc(Assembler::less, L_processTail_insr[k]);
4014 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4015 __ pxor(xmm_result0, xmm_from0);
4016 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4017 __ addptr(pos, AESBlockSize);
4018 __ subptr(len_reg, AESBlockSize);
4019 __ jmp(L_singleBlockLoopTop[k]);
4020 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
4021 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
4022 __ testptr(len_reg, 8);
4023 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
4024 __ subptr(pos,8);
4025 __ pinsrq(xmm_from0, Address(from, pos), 0);
4026 __ BIND(L_processTail_4_insr[k]);
4027 __ testptr(len_reg, 4);
4028 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
4029 __ subptr(pos,4);
4030 __ pslldq(xmm_from0, 4);
4031 __ pinsrd(xmm_from0, Address(from, pos), 0);
4032 __ BIND(L_processTail_2_insr[k]);
4033 __ testptr(len_reg, 2);
4034 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
4035 __ subptr(pos, 2);
4036 __ pslldq(xmm_from0, 2);
4037 __ pinsrw(xmm_from0, Address(from, pos), 0);
4038 __ BIND(L_processTail_1_insr[k]);
4039 __ testptr(len_reg, 1);
4040 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
4041 __ subptr(pos, 1);
4042 __ pslldq(xmm_from0, 1);
4043 __ pinsrb(xmm_from0, Address(from, pos), 0);
4044 __ BIND(L_processTail_exit_insr[k]);
4045
4046 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
4047 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
4048
4049 __ testptr(len_reg, 8);
4050 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
4051 __ pextrq(Address(to, pos), xmm_result0, 0);
4052 __ psrldq(xmm_result0, 8);
4053 __ addptr(pos, 8);
4054 __ BIND(L_processTail_4_extr[k]);
4055 __ testptr(len_reg, 4);
4056 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
4057 __ pextrd(Address(to, pos), xmm_result0, 0);
4058 __ psrldq(xmm_result0, 4);
4059 __ addptr(pos, 4);
4060 __ BIND(L_processTail_2_extr[k]);
4061 __ testptr(len_reg, 2);
4062 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
4063 __ pextrw(Address(to, pos), xmm_result0, 0);
4064 __ psrldq(xmm_result0, 2);
4065 __ addptr(pos, 2);
4066 __ BIND(L_processTail_1_extr[k]);
4067 __ testptr(len_reg, 1);
4068 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
4069 __ pextrb(Address(to, pos), xmm_result0, 0);
4070
4071 __ BIND(L_processTail_exit_extr[k]);
4072 __ movl(Address(used_addr, 0), len_reg);
4073 __ jmp(L_exit);
4074
4075 }
4076
4077 __ BIND(L_exit);
4078 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
4079 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
4080 __ pop(rbx); // pop the saved RBX.
4081 #ifdef _WIN64
4082 __ movl(rax, len_mem);
4083 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
4084 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
4085 __ addptr(rsp, 2 * wordSize);
4086 #else
4087 __ pop(rax); // return 'len'
4088 #endif
4089 __ leave(); // required for proper stackwalking of RuntimeStub frame
4090 __ ret(0);
4091 return start;
4092 }
4093
4094 void roundDec(XMMRegister xmm_reg) {
4095 __ vaesdec(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4096 __ vaesdec(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4097 __ vaesdec(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4098 __ vaesdec(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4099 __ vaesdec(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4100 __ vaesdec(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4101 __ vaesdec(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4102 __ vaesdec(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4103 }
4104
4105 void roundDeclast(XMMRegister xmm_reg) {
4106 __ vaesdeclast(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4107 __ vaesdeclast(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4108 __ vaesdeclast(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4109 __ vaesdeclast(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4110 __ vaesdeclast(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4111 __ vaesdeclast(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4112 __ vaesdeclast(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4113 __ vaesdeclast(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4114 }
4115
4116 void ev_load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask = NULL) {
4117 __ movdqu(xmmdst, Address(key, offset));
4118 if (xmm_shuf_mask != NULL) {
4119 __ pshufb(xmmdst, xmm_shuf_mask);
4120 } else {
4121 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4122 }
4123 __ evshufi64x2(xmmdst, xmmdst, xmmdst, 0x0, Assembler::AVX_512bit);
4124
4125 }
4126
4127 address generate_cipherBlockChaining_decryptVectorAESCrypt() {
4128 assert(VM_Version::supports_vaes(), "need AES instructions and misaligned SSE support");
4129 __ align(CodeEntryAlignment);
4130 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
4131 address start = __ pc();
4132
4133 const Register from = c_rarg0; // source array address
4134 const Register to = c_rarg1; // destination array address
4135 const Register key = c_rarg2; // key array address
4136 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
4137 // and left with the results of the last encryption block
4138 #ifndef _WIN64
4139 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
4140 #else
4141 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4142 const Register len_reg = r11; // pick the volatile windows register
4143 #endif
4144
4145 Label Loop, Loop1, L_128, L_256, L_192, KEY_192, KEY_256, Loop2, Lcbc_dec_rem_loop,
4146 Lcbc_dec_rem_last, Lcbc_dec_ret, Lcbc_dec_rem, Lcbc_exit;
4147
4148 __ enter();
4149
4150 #ifdef _WIN64
4151 // on win64, fill len_reg from stack position
4152 __ movl(len_reg, len_mem);
4153 #else
4154 __ push(len_reg); // Save
4155 #endif
4156 __ push(rbx);
4157 __ vzeroupper();
4158
4159 // Temporary variable declaration for swapping key bytes
4160 const XMMRegister xmm_key_shuf_mask = xmm1;
4161 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4162
4163 // Calculate number of rounds from key size: 44 for 10-rounds, 52 for 12-rounds, 60 for 14-rounds
4164 const Register rounds = rbx;
4165 __ movl(rounds, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4166
4167 const XMMRegister IV = xmm0;
4168 // Load IV and broadcast value to 512-bits
4169 __ evbroadcasti64x2(IV, Address(rvec, 0), Assembler::AVX_512bit);
4170
4171 // Temporary variables for storing round keys
4172 const XMMRegister RK0 = xmm30;
4173 const XMMRegister RK1 = xmm9;
4174 const XMMRegister RK2 = xmm18;
4175 const XMMRegister RK3 = xmm19;
4176 const XMMRegister RK4 = xmm20;
4177 const XMMRegister RK5 = xmm21;
4178 const XMMRegister RK6 = xmm22;
4179 const XMMRegister RK7 = xmm23;
4180 const XMMRegister RK8 = xmm24;
4181 const XMMRegister RK9 = xmm25;
4182 const XMMRegister RK10 = xmm26;
4183
4184 // Load and shuffle key
4185 // the java expanded key ordering is rotated one position from what we want
4186 // so we start from 1*16 here and hit 0*16 last
4187 ev_load_key(RK1, key, 1 * 16, xmm_key_shuf_mask);
4188 ev_load_key(RK2, key, 2 * 16, xmm_key_shuf_mask);
4189 ev_load_key(RK3, key, 3 * 16, xmm_key_shuf_mask);
4190 ev_load_key(RK4, key, 4 * 16, xmm_key_shuf_mask);
4191 ev_load_key(RK5, key, 5 * 16, xmm_key_shuf_mask);
4192 ev_load_key(RK6, key, 6 * 16, xmm_key_shuf_mask);
4193 ev_load_key(RK7, key, 7 * 16, xmm_key_shuf_mask);
4194 ev_load_key(RK8, key, 8 * 16, xmm_key_shuf_mask);
4195 ev_load_key(RK9, key, 9 * 16, xmm_key_shuf_mask);
4196 ev_load_key(RK10, key, 10 * 16, xmm_key_shuf_mask);
4197 ev_load_key(RK0, key, 0*16, xmm_key_shuf_mask);
4198
4199 // Variables for storing source cipher text
4200 const XMMRegister S0 = xmm10;
4201 const XMMRegister S1 = xmm11;
4202 const XMMRegister S2 = xmm12;
4203 const XMMRegister S3 = xmm13;
4204 const XMMRegister S4 = xmm14;
4205 const XMMRegister S5 = xmm15;
4206 const XMMRegister S6 = xmm16;
4207 const XMMRegister S7 = xmm17;
4208
4209 // Variables for storing decrypted text
4210 const XMMRegister B0 = xmm1;
4211 const XMMRegister B1 = xmm2;
4212 const XMMRegister B2 = xmm3;
4213 const XMMRegister B3 = xmm4;
4214 const XMMRegister B4 = xmm5;
4215 const XMMRegister B5 = xmm6;
4216 const XMMRegister B6 = xmm7;
4217 const XMMRegister B7 = xmm8;
4218
4219 __ cmpl(rounds, 44);
4220 __ jcc(Assembler::greater, KEY_192);
4221 __ jmp(Loop);
4222
4223 __ BIND(KEY_192);
4224 const XMMRegister RK11 = xmm27;
4225 const XMMRegister RK12 = xmm28;
4226 ev_load_key(RK11, key, 11*16, xmm_key_shuf_mask);
4227 ev_load_key(RK12, key, 12*16, xmm_key_shuf_mask);
4228
4229 __ cmpl(rounds, 52);
4230 __ jcc(Assembler::greater, KEY_256);
4231 __ jmp(Loop);
4232
4233 __ BIND(KEY_256);
4234 const XMMRegister RK13 = xmm29;
4235 const XMMRegister RK14 = xmm31;
4236 ev_load_key(RK13, key, 13*16, xmm_key_shuf_mask);
4237 ev_load_key(RK14, key, 14*16, xmm_key_shuf_mask);
4238
4239 __ BIND(Loop);
4240 __ cmpl(len_reg, 512);
4241 __ jcc(Assembler::below, Lcbc_dec_rem);
4242 __ BIND(Loop1);
4243 __ subl(len_reg, 512);
4244 __ evmovdquq(S0, Address(from, 0 * 64), Assembler::AVX_512bit);
4245 __ evmovdquq(S1, Address(from, 1 * 64), Assembler::AVX_512bit);
4246 __ evmovdquq(S2, Address(from, 2 * 64), Assembler::AVX_512bit);
4247 __ evmovdquq(S3, Address(from, 3 * 64), Assembler::AVX_512bit);
4248 __ evmovdquq(S4, Address(from, 4 * 64), Assembler::AVX_512bit);
4249 __ evmovdquq(S5, Address(from, 5 * 64), Assembler::AVX_512bit);
4250 __ evmovdquq(S6, Address(from, 6 * 64), Assembler::AVX_512bit);
4251 __ evmovdquq(S7, Address(from, 7 * 64), Assembler::AVX_512bit);
4252 __ leaq(from, Address(from, 8 * 64));
4253
4254 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4255 __ evpxorq(B1, S1, RK1, Assembler::AVX_512bit);
4256 __ evpxorq(B2, S2, RK1, Assembler::AVX_512bit);
4257 __ evpxorq(B3, S3, RK1, Assembler::AVX_512bit);
4258 __ evpxorq(B4, S4, RK1, Assembler::AVX_512bit);
4259 __ evpxorq(B5, S5, RK1, Assembler::AVX_512bit);
4260 __ evpxorq(B6, S6, RK1, Assembler::AVX_512bit);
4261 __ evpxorq(B7, S7, RK1, Assembler::AVX_512bit);
4262
4263 __ evalignq(IV, S0, IV, 0x06);
4264 __ evalignq(S0, S1, S0, 0x06);
4265 __ evalignq(S1, S2, S1, 0x06);
4266 __ evalignq(S2, S3, S2, 0x06);
4267 __ evalignq(S3, S4, S3, 0x06);
4268 __ evalignq(S4, S5, S4, 0x06);
4269 __ evalignq(S5, S6, S5, 0x06);
4270 __ evalignq(S6, S7, S6, 0x06);
4271
4272 roundDec(RK2);
4273 roundDec(RK3);
4274 roundDec(RK4);
4275 roundDec(RK5);
4276 roundDec(RK6);
4277 roundDec(RK7);
4278 roundDec(RK8);
4279 roundDec(RK9);
4280 roundDec(RK10);
4281
4282 __ cmpl(rounds, 44);
4283 __ jcc(Assembler::belowEqual, L_128);
4284 roundDec(RK11);
4285 roundDec(RK12);
4286
4287 __ cmpl(rounds, 52);
4288 __ jcc(Assembler::belowEqual, L_192);
4289 roundDec(RK13);
4290 roundDec(RK14);
4291
4292 __ BIND(L_256);
4293 roundDeclast(RK0);
4294 __ jmp(Loop2);
4295
4296 __ BIND(L_128);
4297 roundDeclast(RK0);
4298 __ jmp(Loop2);
4299
4300 __ BIND(L_192);
4301 roundDeclast(RK0);
4302
4303 __ BIND(Loop2);
4304 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4305 __ evpxorq(B1, B1, S0, Assembler::AVX_512bit);
4306 __ evpxorq(B2, B2, S1, Assembler::AVX_512bit);
4307 __ evpxorq(B3, B3, S2, Assembler::AVX_512bit);
4308 __ evpxorq(B4, B4, S3, Assembler::AVX_512bit);
4309 __ evpxorq(B5, B5, S4, Assembler::AVX_512bit);
4310 __ evpxorq(B6, B6, S5, Assembler::AVX_512bit);
4311 __ evpxorq(B7, B7, S6, Assembler::AVX_512bit);
4312 __ evmovdquq(IV, S7, Assembler::AVX_512bit);
4313
4314 __ evmovdquq(Address(to, 0 * 64), B0, Assembler::AVX_512bit);
4315 __ evmovdquq(Address(to, 1 * 64), B1, Assembler::AVX_512bit);
4316 __ evmovdquq(Address(to, 2 * 64), B2, Assembler::AVX_512bit);
4317 __ evmovdquq(Address(to, 3 * 64), B3, Assembler::AVX_512bit);
4318 __ evmovdquq(Address(to, 4 * 64), B4, Assembler::AVX_512bit);
4319 __ evmovdquq(Address(to, 5 * 64), B5, Assembler::AVX_512bit);
4320 __ evmovdquq(Address(to, 6 * 64), B6, Assembler::AVX_512bit);
4321 __ evmovdquq(Address(to, 7 * 64), B7, Assembler::AVX_512bit);
4322 __ leaq(to, Address(to, 8 * 64));
4323 __ jmp(Loop);
4324
4325 __ BIND(Lcbc_dec_rem);
4326 __ evshufi64x2(IV, IV, IV, 0x03, Assembler::AVX_512bit);
4327
4328 __ BIND(Lcbc_dec_rem_loop);
4329 __ subl(len_reg, 16);
4330 __ jcc(Assembler::carrySet, Lcbc_dec_ret);
4331
4332 __ movdqu(S0, Address(from, 0));
4333 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4334 __ vaesdec(B0, B0, RK2, Assembler::AVX_512bit);
4335 __ vaesdec(B0, B0, RK3, Assembler::AVX_512bit);
4336 __ vaesdec(B0, B0, RK4, Assembler::AVX_512bit);
4337 __ vaesdec(B0, B0, RK5, Assembler::AVX_512bit);
4338 __ vaesdec(B0, B0, RK6, Assembler::AVX_512bit);
4339 __ vaesdec(B0, B0, RK7, Assembler::AVX_512bit);
4340 __ vaesdec(B0, B0, RK8, Assembler::AVX_512bit);
4341 __ vaesdec(B0, B0, RK9, Assembler::AVX_512bit);
4342 __ vaesdec(B0, B0, RK10, Assembler::AVX_512bit);
4343 __ cmpl(rounds, 44);
4344 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4345
4346 __ vaesdec(B0, B0, RK11, Assembler::AVX_512bit);
4347 __ vaesdec(B0, B0, RK12, Assembler::AVX_512bit);
4348 __ cmpl(rounds, 52);
4349 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4350
4351 __ vaesdec(B0, B0, RK13, Assembler::AVX_512bit);
4352 __ vaesdec(B0, B0, RK14, Assembler::AVX_512bit);
4353
4354 __ BIND(Lcbc_dec_rem_last);
4355 __ vaesdeclast(B0, B0, RK0, Assembler::AVX_512bit);
4356
4357 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4358 __ evmovdquq(IV, S0, Assembler::AVX_512bit);
4359 __ movdqu(Address(to, 0), B0);
4360 __ leaq(from, Address(from, 16));
4361 __ leaq(to, Address(to, 16));
4362 __ jmp(Lcbc_dec_rem_loop);
4363
4364 __ BIND(Lcbc_dec_ret);
4365 __ movdqu(Address(rvec, 0), IV);
4366
4367 // Zero out the round keys
4368 __ evpxorq(RK0, RK0, RK0, Assembler::AVX_512bit);
4369 __ evpxorq(RK1, RK1, RK1, Assembler::AVX_512bit);
4370 __ evpxorq(RK2, RK2, RK2, Assembler::AVX_512bit);
4371 __ evpxorq(RK3, RK3, RK3, Assembler::AVX_512bit);
4372 __ evpxorq(RK4, RK4, RK4, Assembler::AVX_512bit);
4373 __ evpxorq(RK5, RK5, RK5, Assembler::AVX_512bit);
4374 __ evpxorq(RK6, RK6, RK6, Assembler::AVX_512bit);
4375 __ evpxorq(RK7, RK7, RK7, Assembler::AVX_512bit);
4376 __ evpxorq(RK8, RK8, RK8, Assembler::AVX_512bit);
4377 __ evpxorq(RK9, RK9, RK9, Assembler::AVX_512bit);
4378 __ evpxorq(RK10, RK10, RK10, Assembler::AVX_512bit);
4379 __ cmpl(rounds, 44);
4380 __ jcc(Assembler::belowEqual, Lcbc_exit);
4381 __ evpxorq(RK11, RK11, RK11, Assembler::AVX_512bit);
4382 __ evpxorq(RK12, RK12, RK12, Assembler::AVX_512bit);
4383 __ cmpl(rounds, 52);
4384 __ jcc(Assembler::belowEqual, Lcbc_exit);
4385 __ evpxorq(RK13, RK13, RK13, Assembler::AVX_512bit);
4386 __ evpxorq(RK14, RK14, RK14, Assembler::AVX_512bit);
4387
4388 __ BIND(Lcbc_exit);
4389 __ pop(rbx);
4390 #ifdef _WIN64
4391 __ movl(rax, len_mem);
4392 #else
4393 __ pop(rax); // return length
4394 #endif
4395 __ leave(); // required for proper stackwalking of RuntimeStub frame
4396 __ ret(0);
4397 return start;
4398 }
4399
4400 // Polynomial x^128+x^127+x^126+x^121+1
4401 address ghash_polynomial_addr() {
4402 __ align(CodeEntryAlignment);
4403 StubCodeMark mark(this, "StubRoutines", "_ghash_poly_addr");
4404 address start = __ pc();
4405 __ emit_data64(0x0000000000000001, relocInfo::none);
4406 __ emit_data64(0xc200000000000000, relocInfo::none);
4407 return start;
4408 }
4409
4410 address ghash_shufflemask_addr() {
4411 __ align(CodeEntryAlignment);
4412 StubCodeMark mark(this, "StubRoutines", "_ghash_shuffmask_addr");
4413 address start = __ pc();
4414 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4415 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4416 return start;
4417 }
4418
4419 // Ghash single and multi block operations using AVX instructions
4420 address generate_avx_ghash_processBlocks() {
4421 __ align(CodeEntryAlignment);
4422
4423 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4424 address start = __ pc();
4425
4426 // arguments
4427 const Register state = c_rarg0;
4428 const Register htbl = c_rarg1;
4429 const Register data = c_rarg2;
4430 const Register blocks = c_rarg3;
4431 __ enter();
4432 // Save state before entering routine
4433 __ avx_ghash(state, htbl, data, blocks);
4434 __ leave(); // required for proper stackwalking of RuntimeStub frame
4435 __ ret(0);
4436 return start;
4437 }
4438
4439 // byte swap x86 long
4440 address generate_ghash_long_swap_mask() {
4441 __ align(CodeEntryAlignment);
4442 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
4443 address start = __ pc();
4444 __ emit_data64(0x0f0e0d0c0b0a0908, relocInfo::none );
4445 __ emit_data64(0x0706050403020100, relocInfo::none );
4446 return start;
4447 }
4448
4449 // byte swap x86 byte array
4450 address generate_ghash_byte_swap_mask() {
4451 __ align(CodeEntryAlignment);
4452 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
4453 address start = __ pc();
4454 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none );
4455 __ emit_data64(0x0001020304050607, relocInfo::none );
4456 return start;
4457 }
4458
4459 /* Single and multi-block ghash operations */
4460 address generate_ghash_processBlocks() {
4461 __ align(CodeEntryAlignment);
4462 Label L_ghash_loop, L_exit;
4463 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4464 address start = __ pc();
4465
4466 const Register state = c_rarg0;
4467 const Register subkeyH = c_rarg1;
4468 const Register data = c_rarg2;
4469 const Register blocks = c_rarg3;
4470
4471 const XMMRegister xmm_temp0 = xmm0;
4472 const XMMRegister xmm_temp1 = xmm1;
4473 const XMMRegister xmm_temp2 = xmm2;
4474 const XMMRegister xmm_temp3 = xmm3;
4475 const XMMRegister xmm_temp4 = xmm4;
4476 const XMMRegister xmm_temp5 = xmm5;
4477 const XMMRegister xmm_temp6 = xmm6;
4478 const XMMRegister xmm_temp7 = xmm7;
4479 const XMMRegister xmm_temp8 = xmm8;
4480 const XMMRegister xmm_temp9 = xmm9;
4481 const XMMRegister xmm_temp10 = xmm10;
4482
4483 __ enter();
4484
4485 __ movdqu(xmm_temp10, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
4486
4487 __ movdqu(xmm_temp0, Address(state, 0));
4488 __ pshufb(xmm_temp0, xmm_temp10);
4489
4490
4491 __ BIND(L_ghash_loop);
4492 __ movdqu(xmm_temp2, Address(data, 0));
4493 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
4494
4495 __ movdqu(xmm_temp1, Address(subkeyH, 0));
4496 __ pshufb(xmm_temp1, xmm_temp10);
4497
4498 __ pxor(xmm_temp0, xmm_temp2);
4499
4500 //
4501 // Multiply with the hash key
4502 //
4503 __ movdqu(xmm_temp3, xmm_temp0);
4504 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
4505 __ movdqu(xmm_temp4, xmm_temp0);
4506 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
4507
4508 __ movdqu(xmm_temp5, xmm_temp0);
4509 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
4510 __ movdqu(xmm_temp6, xmm_temp0);
4511 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
4512
4513 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
4514
4515 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
4516 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
4517 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
4518 __ pxor(xmm_temp3, xmm_temp5);
4519 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
4520 // of the carry-less multiplication of
4521 // xmm0 by xmm1.
4522
4523 // We shift the result of the multiplication by one bit position
4524 // to the left to cope for the fact that the bits are reversed.
4525 __ movdqu(xmm_temp7, xmm_temp3);
4526 __ movdqu(xmm_temp8, xmm_temp6);
4527 __ pslld(xmm_temp3, 1);
4528 __ pslld(xmm_temp6, 1);
4529 __ psrld(xmm_temp7, 31);
4530 __ psrld(xmm_temp8, 31);
4531 __ movdqu(xmm_temp9, xmm_temp7);
4532 __ pslldq(xmm_temp8, 4);
4533 __ pslldq(xmm_temp7, 4);
4534 __ psrldq(xmm_temp9, 12);
4535 __ por(xmm_temp3, xmm_temp7);
4536 __ por(xmm_temp6, xmm_temp8);
4537 __ por(xmm_temp6, xmm_temp9);
4538
4539 //
4540 // First phase of the reduction
4541 //
4542 // Move xmm3 into xmm7, xmm8, xmm9 in order to perform the shifts
4543 // independently.
4544 __ movdqu(xmm_temp7, xmm_temp3);
4545 __ movdqu(xmm_temp8, xmm_temp3);
4546 __ movdqu(xmm_temp9, xmm_temp3);
4547 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
4548 __ pslld(xmm_temp8, 30); // packed right shift shifting << 30
4549 __ pslld(xmm_temp9, 25); // packed right shift shifting << 25
4550 __ pxor(xmm_temp7, xmm_temp8); // xor the shifted versions
4551 __ pxor(xmm_temp7, xmm_temp9);
4552 __ movdqu(xmm_temp8, xmm_temp7);
4553 __ pslldq(xmm_temp7, 12);
4554 __ psrldq(xmm_temp8, 4);
4555 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
4556
4557 //
4558 // Second phase of the reduction
4559 //
4560 // Make 3 copies of xmm3 in xmm2, xmm4, xmm5 for doing these
4561 // shift operations.
4562 __ movdqu(xmm_temp2, xmm_temp3);
4563 __ movdqu(xmm_temp4, xmm_temp3);
4564 __ movdqu(xmm_temp5, xmm_temp3);
4565 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
4566 __ psrld(xmm_temp4, 2); // packed left shifting >> 2
4567 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
4568 __ pxor(xmm_temp2, xmm_temp4); // xor the shifted versions
4569 __ pxor(xmm_temp2, xmm_temp5);
4570 __ pxor(xmm_temp2, xmm_temp8);
4571 __ pxor(xmm_temp3, xmm_temp2);
4572 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
4573
4574 __ decrement(blocks);
4575 __ jcc(Assembler::zero, L_exit);
4576 __ movdqu(xmm_temp0, xmm_temp6);
4577 __ addptr(data, 16);
4578 __ jmp(L_ghash_loop);
4579
4580 __ BIND(L_exit);
4581 __ pshufb(xmm_temp6, xmm_temp10); // Byte swap 16-byte result
4582 __ movdqu(Address(state, 0), xmm_temp6); // store the result
4583 __ leave();
4584 __ ret(0);
4585 return start;
4586 }
4587
4588 //base64 character set
4589 address base64_charset_addr() {
4590 __ align(CodeEntryAlignment);
4591 StubCodeMark mark(this, "StubRoutines", "base64_charset");
4592 address start = __ pc();
4593 __ emit_data64(0x0000004200000041, relocInfo::none);
4594 __ emit_data64(0x0000004400000043, relocInfo::none);
4595 __ emit_data64(0x0000004600000045, relocInfo::none);
4596 __ emit_data64(0x0000004800000047, relocInfo::none);
4597 __ emit_data64(0x0000004a00000049, relocInfo::none);
4598 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4599 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4600 __ emit_data64(0x000000500000004f, relocInfo::none);
4601 __ emit_data64(0x0000005200000051, relocInfo::none);
4602 __ emit_data64(0x0000005400000053, relocInfo::none);
4603 __ emit_data64(0x0000005600000055, relocInfo::none);
4604 __ emit_data64(0x0000005800000057, relocInfo::none);
4605 __ emit_data64(0x0000005a00000059, relocInfo::none);
4606 __ emit_data64(0x0000006200000061, relocInfo::none);
4607 __ emit_data64(0x0000006400000063, relocInfo::none);
4608 __ emit_data64(0x0000006600000065, relocInfo::none);
4609 __ emit_data64(0x0000006800000067, relocInfo::none);
4610 __ emit_data64(0x0000006a00000069, relocInfo::none);
4611 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4612 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4613 __ emit_data64(0x000000700000006f, relocInfo::none);
4614 __ emit_data64(0x0000007200000071, relocInfo::none);
4615 __ emit_data64(0x0000007400000073, relocInfo::none);
4616 __ emit_data64(0x0000007600000075, relocInfo::none);
4617 __ emit_data64(0x0000007800000077, relocInfo::none);
4618 __ emit_data64(0x0000007a00000079, relocInfo::none);
4619 __ emit_data64(0x0000003100000030, relocInfo::none);
4620 __ emit_data64(0x0000003300000032, relocInfo::none);
4621 __ emit_data64(0x0000003500000034, relocInfo::none);
4622 __ emit_data64(0x0000003700000036, relocInfo::none);
4623 __ emit_data64(0x0000003900000038, relocInfo::none);
4624 __ emit_data64(0x0000002f0000002b, relocInfo::none);
4625 return start;
4626 }
4627
4628 //base64 url character set
4629 address base64url_charset_addr() {
4630 __ align(CodeEntryAlignment);
4631 StubCodeMark mark(this, "StubRoutines", "base64url_charset");
4632 address start = __ pc();
4633 __ emit_data64(0x0000004200000041, relocInfo::none);
4634 __ emit_data64(0x0000004400000043, relocInfo::none);
4635 __ emit_data64(0x0000004600000045, relocInfo::none);
4636 __ emit_data64(0x0000004800000047, relocInfo::none);
4637 __ emit_data64(0x0000004a00000049, relocInfo::none);
4638 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4639 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4640 __ emit_data64(0x000000500000004f, relocInfo::none);
4641 __ emit_data64(0x0000005200000051, relocInfo::none);
4642 __ emit_data64(0x0000005400000053, relocInfo::none);
4643 __ emit_data64(0x0000005600000055, relocInfo::none);
4644 __ emit_data64(0x0000005800000057, relocInfo::none);
4645 __ emit_data64(0x0000005a00000059, relocInfo::none);
4646 __ emit_data64(0x0000006200000061, relocInfo::none);
4647 __ emit_data64(0x0000006400000063, relocInfo::none);
4648 __ emit_data64(0x0000006600000065, relocInfo::none);
4649 __ emit_data64(0x0000006800000067, relocInfo::none);
4650 __ emit_data64(0x0000006a00000069, relocInfo::none);
4651 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4652 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4653 __ emit_data64(0x000000700000006f, relocInfo::none);
4654 __ emit_data64(0x0000007200000071, relocInfo::none);
4655 __ emit_data64(0x0000007400000073, relocInfo::none);
4656 __ emit_data64(0x0000007600000075, relocInfo::none);
4657 __ emit_data64(0x0000007800000077, relocInfo::none);
4658 __ emit_data64(0x0000007a00000079, relocInfo::none);
4659 __ emit_data64(0x0000003100000030, relocInfo::none);
4660 __ emit_data64(0x0000003300000032, relocInfo::none);
4661 __ emit_data64(0x0000003500000034, relocInfo::none);
4662 __ emit_data64(0x0000003700000036, relocInfo::none);
4663 __ emit_data64(0x0000003900000038, relocInfo::none);
4664 __ emit_data64(0x0000005f0000002d, relocInfo::none);
4665
4666 return start;
4667 }
4668
4669 address base64_bswap_mask_addr() {
4670 __ align(CodeEntryAlignment);
4671 StubCodeMark mark(this, "StubRoutines", "bswap_mask_base64");
4672 address start = __ pc();
4673 __ emit_data64(0x0504038002010080, relocInfo::none);
4674 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4675 __ emit_data64(0x0908078006050480, relocInfo::none);
4676 __ emit_data64(0x0f0e0d800c0b0a80, relocInfo::none);
4677 __ emit_data64(0x0605048003020180, relocInfo::none);
4678 __ emit_data64(0x0c0b0a8009080780, relocInfo::none);
4679 __ emit_data64(0x0504038002010080, relocInfo::none);
4680 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4681
4682 return start;
4683 }
4684
4685 address base64_right_shift_mask_addr() {
4686 __ align(CodeEntryAlignment);
4687 StubCodeMark mark(this, "StubRoutines", "right_shift_mask");
4688 address start = __ pc();
4689 __ emit_data64(0x0006000400020000, relocInfo::none);
4690 __ emit_data64(0x0006000400020000, relocInfo::none);
4691 __ emit_data64(0x0006000400020000, relocInfo::none);
4692 __ emit_data64(0x0006000400020000, relocInfo::none);
4693 __ emit_data64(0x0006000400020000, relocInfo::none);
4694 __ emit_data64(0x0006000400020000, relocInfo::none);
4695 __ emit_data64(0x0006000400020000, relocInfo::none);
4696 __ emit_data64(0x0006000400020000, relocInfo::none);
4697
4698 return start;
4699 }
4700
4701 address base64_left_shift_mask_addr() {
4702 __ align(CodeEntryAlignment);
4703 StubCodeMark mark(this, "StubRoutines", "left_shift_mask");
4704 address start = __ pc();
4705 __ emit_data64(0x0000000200040000, relocInfo::none);
4706 __ emit_data64(0x0000000200040000, relocInfo::none);
4707 __ emit_data64(0x0000000200040000, relocInfo::none);
4708 __ emit_data64(0x0000000200040000, relocInfo::none);
4709 __ emit_data64(0x0000000200040000, relocInfo::none);
4710 __ emit_data64(0x0000000200040000, relocInfo::none);
4711 __ emit_data64(0x0000000200040000, relocInfo::none);
4712 __ emit_data64(0x0000000200040000, relocInfo::none);
4713
4714 return start;
4715 }
4716
4717 address base64_and_mask_addr() {
4718 __ align(CodeEntryAlignment);
4719 StubCodeMark mark(this, "StubRoutines", "and_mask");
4720 address start = __ pc();
4721 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4722 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4723 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4724 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4725 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4726 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4727 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4728 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4729 return start;
4730 }
4731
4732 address base64_gather_mask_addr() {
4733 __ align(CodeEntryAlignment);
4734 StubCodeMark mark(this, "StubRoutines", "gather_mask");
4735 address start = __ pc();
4736 __ emit_data64(0xffffffffffffffff, relocInfo::none);
4737 return start;
4738 }
4739
4740 // Code for generating Base64 encoding.
4741 // Intrinsic function prototype in Base64.java:
4742 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL) {
4743 address generate_base64_encodeBlock() {
4744 __ align(CodeEntryAlignment);
4745 StubCodeMark mark(this, "StubRoutines", "implEncode");
4746 address start = __ pc();
4747 __ enter();
4748
4749 // Save callee-saved registers before using them
4750 __ push(r12);
4751 __ push(r13);
4752 __ push(r14);
4753 __ push(r15);
4754
4755 // arguments
4756 const Register source = c_rarg0; // Source Array
4757 const Register start_offset = c_rarg1; // start offset
4758 const Register end_offset = c_rarg2; // end offset
4759 const Register dest = c_rarg3; // destination array
4760
4761 #ifndef _WIN64
4762 const Register dp = c_rarg4; // Position for writing to dest array
4763 const Register isURL = c_rarg5;// Base64 or URL character set
4764 #else
4765 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
4766 const Address isURL_mem(rbp, 7 * wordSize);
4767 const Register isURL = r10; // pick the volatile windows register
4768 const Register dp = r12;
4769 __ movl(dp, dp_mem);
4770 __ movl(isURL, isURL_mem);
4771 #endif
4772
4773 const Register length = r14;
4774 Label L_process80, L_process32, L_process3, L_exit, L_processdata;
4775
4776 // calculate length from offsets
4777 __ movl(length, end_offset);
4778 __ subl(length, start_offset);
4779 __ cmpl(length, 0);
4780 __ jcc(Assembler::lessEqual, L_exit);
4781
4782 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_charset_addr()));
4783 // check if base64 charset(isURL=0) or base64 url charset(isURL=1) needs to be loaded
4784 __ cmpl(isURL, 0);
4785 __ jcc(Assembler::equal, L_processdata);
4786 __ lea(r11, ExternalAddress(StubRoutines::x86::base64url_charset_addr()));
4787
4788 // load masks required for encoding data
4789 __ BIND(L_processdata);
4790 __ movdqu(xmm16, ExternalAddress(StubRoutines::x86::base64_gather_mask_addr()));
4791 // Set 64 bits of K register.
4792 __ evpcmpeqb(k3, xmm16, xmm16, Assembler::AVX_512bit);
4793 __ evmovdquq(xmm12, ExternalAddress(StubRoutines::x86::base64_bswap_mask_addr()), Assembler::AVX_256bit, r13);
4794 __ evmovdquq(xmm13, ExternalAddress(StubRoutines::x86::base64_right_shift_mask_addr()), Assembler::AVX_512bit, r13);
4795 __ evmovdquq(xmm14, ExternalAddress(StubRoutines::x86::base64_left_shift_mask_addr()), Assembler::AVX_512bit, r13);
4796 __ evmovdquq(xmm15, ExternalAddress(StubRoutines::x86::base64_and_mask_addr()), Assembler::AVX_512bit, r13);
4797
4798 // Vector Base64 implementation, producing 96 bytes of encoded data
4799 __ BIND(L_process80);
4800 __ cmpl(length, 80);
4801 __ jcc(Assembler::below, L_process32);
4802 __ evmovdquq(xmm0, Address(source, start_offset, Address::times_1, 0), Assembler::AVX_256bit);
4803 __ evmovdquq(xmm1, Address(source, start_offset, Address::times_1, 24), Assembler::AVX_256bit);
4804 __ evmovdquq(xmm2, Address(source, start_offset, Address::times_1, 48), Assembler::AVX_256bit);
4805
4806 //permute the input data in such a manner that we have continuity of the source
4807 __ vpermq(xmm3, xmm0, 148, Assembler::AVX_256bit);
4808 __ vpermq(xmm4, xmm1, 148, Assembler::AVX_256bit);
4809 __ vpermq(xmm5, xmm2, 148, Assembler::AVX_256bit);
4810
4811 //shuffle input and group 3 bytes of data and to it add 0 as the 4th byte.
4812 //we can deal with 12 bytes at a time in a 128 bit register
4813 __ vpshufb(xmm3, xmm3, xmm12, Assembler::AVX_256bit);
4814 __ vpshufb(xmm4, xmm4, xmm12, Assembler::AVX_256bit);
4815 __ vpshufb(xmm5, xmm5, xmm12, Assembler::AVX_256bit);
4816
4817 //convert byte to word. Each 128 bit register will have 6 bytes for processing
4818 __ vpmovzxbw(xmm3, xmm3, Assembler::AVX_512bit);
4819 __ vpmovzxbw(xmm4, xmm4, Assembler::AVX_512bit);
4820 __ vpmovzxbw(xmm5, xmm5, Assembler::AVX_512bit);
4821
4822 // Extract bits in the following pattern 6, 4+2, 2+4, 6 to convert 3, 8 bit numbers to 4, 6 bit numbers
4823 __ evpsrlvw(xmm0, xmm3, xmm13, Assembler::AVX_512bit);
4824 __ evpsrlvw(xmm1, xmm4, xmm13, Assembler::AVX_512bit);
4825 __ evpsrlvw(xmm2, xmm5, xmm13, Assembler::AVX_512bit);
4826
4827 __ evpsllvw(xmm3, xmm3, xmm14, Assembler::AVX_512bit);
4828 __ evpsllvw(xmm4, xmm4, xmm14, Assembler::AVX_512bit);
4829 __ evpsllvw(xmm5, xmm5, xmm14, Assembler::AVX_512bit);
4830
4831 __ vpsrlq(xmm0, xmm0, 8, Assembler::AVX_512bit);
4832 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
4833 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
4834
4835 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4836 __ vpsllq(xmm4, xmm4, 8, Assembler::AVX_512bit);
4837 __ vpsllq(xmm5, xmm5, 8, Assembler::AVX_512bit);
4838
4839 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
4840 __ vpandq(xmm4, xmm4, xmm15, Assembler::AVX_512bit);
4841 __ vpandq(xmm5, xmm5, xmm15, Assembler::AVX_512bit);
4842
4843 // Get the final 4*6 bits base64 encoding
4844 __ vporq(xmm3, xmm3, xmm0, Assembler::AVX_512bit);
4845 __ vporq(xmm4, xmm4, xmm1, Assembler::AVX_512bit);
4846 __ vporq(xmm5, xmm5, xmm2, Assembler::AVX_512bit);
4847
4848 // Shift
4849 __ vpsrlq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4850 __ vpsrlq(xmm4, xmm4, 8, Assembler::AVX_512bit);
4851 __ vpsrlq(xmm5, xmm5, 8, Assembler::AVX_512bit);
4852
4853 // look up 6 bits in the base64 character set to fetch the encoding
4854 // we are converting word to dword as gather instructions need dword indices for looking up encoding
4855 __ vextracti64x4(xmm6, xmm3, 0);
4856 __ vpmovzxwd(xmm0, xmm6, Assembler::AVX_512bit);
4857 __ vextracti64x4(xmm6, xmm3, 1);
4858 __ vpmovzxwd(xmm1, xmm6, Assembler::AVX_512bit);
4859
4860 __ vextracti64x4(xmm6, xmm4, 0);
4861 __ vpmovzxwd(xmm2, xmm6, Assembler::AVX_512bit);
4862 __ vextracti64x4(xmm6, xmm4, 1);
4863 __ vpmovzxwd(xmm3, xmm6, Assembler::AVX_512bit);
4864
4865 __ vextracti64x4(xmm4, xmm5, 0);
4866 __ vpmovzxwd(xmm6, xmm4, Assembler::AVX_512bit);
4867
4868 __ vextracti64x4(xmm4, xmm5, 1);
4869 __ vpmovzxwd(xmm7, xmm4, Assembler::AVX_512bit);
4870
4871 __ kmovql(k2, k3);
4872 __ evpgatherdd(xmm4, k2, Address(r11, xmm0, Address::times_4, 0), Assembler::AVX_512bit);
4873 __ kmovql(k2, k3);
4874 __ evpgatherdd(xmm5, k2, Address(r11, xmm1, Address::times_4, 0), Assembler::AVX_512bit);
4875 __ kmovql(k2, k3);
4876 __ evpgatherdd(xmm8, k2, Address(r11, xmm2, Address::times_4, 0), Assembler::AVX_512bit);
4877 __ kmovql(k2, k3);
4878 __ evpgatherdd(xmm9, k2, Address(r11, xmm3, Address::times_4, 0), Assembler::AVX_512bit);
4879 __ kmovql(k2, k3);
4880 __ evpgatherdd(xmm10, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
4881 __ kmovql(k2, k3);
4882 __ evpgatherdd(xmm11, k2, Address(r11, xmm7, Address::times_4, 0), Assembler::AVX_512bit);
4883
4884 //Down convert dword to byte. Final output is 16*6 = 96 bytes long
4885 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm4, Assembler::AVX_512bit);
4886 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm5, Assembler::AVX_512bit);
4887 __ evpmovdb(Address(dest, dp, Address::times_1, 32), xmm8, Assembler::AVX_512bit);
4888 __ evpmovdb(Address(dest, dp, Address::times_1, 48), xmm9, Assembler::AVX_512bit);
4889 __ evpmovdb(Address(dest, dp, Address::times_1, 64), xmm10, Assembler::AVX_512bit);
4890 __ evpmovdb(Address(dest, dp, Address::times_1, 80), xmm11, Assembler::AVX_512bit);
4891
4892 __ addq(dest, 96);
4893 __ addq(source, 72);
4894 __ subq(length, 72);
4895 __ jmp(L_process80);
4896
4897 // Vector Base64 implementation generating 32 bytes of encoded data
4898 __ BIND(L_process32);
4899 __ cmpl(length, 32);
4900 __ jcc(Assembler::below, L_process3);
4901 __ evmovdquq(xmm0, Address(source, start_offset), Assembler::AVX_256bit);
4902 __ vpermq(xmm0, xmm0, 148, Assembler::AVX_256bit);
4903 __ vpshufb(xmm6, xmm0, xmm12, Assembler::AVX_256bit);
4904 __ vpmovzxbw(xmm6, xmm6, Assembler::AVX_512bit);
4905 __ evpsrlvw(xmm2, xmm6, xmm13, Assembler::AVX_512bit);
4906 __ evpsllvw(xmm3, xmm6, xmm14, Assembler::AVX_512bit);
4907
4908 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
4909 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4910 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
4911 __ vporq(xmm1, xmm2, xmm3, Assembler::AVX_512bit);
4912 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
4913 __ vextracti64x4(xmm9, xmm1, 0);
4914 __ vpmovzxwd(xmm6, xmm9, Assembler::AVX_512bit);
4915 __ vextracti64x4(xmm9, xmm1, 1);
4916 __ vpmovzxwd(xmm5, xmm9, Assembler::AVX_512bit);
4917 __ kmovql(k2, k3);
4918 __ evpgatherdd(xmm8, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
4919 __ kmovql(k2, k3);
4920 __ evpgatherdd(xmm10, k2, Address(r11, xmm5, Address::times_4, 0), Assembler::AVX_512bit);
4921 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm8, Assembler::AVX_512bit);
4922 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm10, Assembler::AVX_512bit);
4923 __ subq(length, 24);
4924 __ addq(dest, 32);
4925 __ addq(source, 24);
4926 __ jmp(L_process32);
4927
4928 // Scalar data processing takes 3 bytes at a time and produces 4 bytes of encoded data
4929 /* This code corresponds to the scalar version of the following snippet in Base64.java
4930 ** int bits = (src[sp0++] & 0xff) << 16 |(src[sp0++] & 0xff) << 8 |(src[sp0++] & 0xff);
4931 ** dst[dp0++] = (byte)base64[(bits >> > 18) & 0x3f];
4932 ** dst[dp0++] = (byte)base64[(bits >> > 12) & 0x3f];
4933 ** dst[dp0++] = (byte)base64[(bits >> > 6) & 0x3f];
4934 ** dst[dp0++] = (byte)base64[bits & 0x3f];*/
4935 __ BIND(L_process3);
4936 __ cmpl(length, 3);
4937 __ jcc(Assembler::below, L_exit);
4938 // Read 1 byte at a time
4939 __ movzbl(rax, Address(source, start_offset));
4940 __ shll(rax, 0x10);
4941 __ movl(r15, rax);
4942 __ movzbl(rax, Address(source, start_offset, Address::times_1, 1));
4943 __ shll(rax, 0x8);
4944 __ movzwl(rax, rax);
4945 __ orl(r15, rax);
4946 __ movzbl(rax, Address(source, start_offset, Address::times_1, 2));
4947 __ orl(rax, r15);
4948 // Save 3 bytes read in r15
4949 __ movl(r15, rax);
4950 __ shrl(rax, 0x12);
4951 __ andl(rax, 0x3f);
4952 // rax contains the index, r11 contains base64 lookup table
4953 __ movb(rax, Address(r11, rax, Address::times_4));
4954 // Write the encoded byte to destination
4955 __ movb(Address(dest, dp, Address::times_1, 0), rax);
4956 __ movl(rax, r15);
4957 __ shrl(rax, 0xc);
4958 __ andl(rax, 0x3f);
4959 __ movb(rax, Address(r11, rax, Address::times_4));
4960 __ movb(Address(dest, dp, Address::times_1, 1), rax);
4961 __ movl(rax, r15);
4962 __ shrl(rax, 0x6);
4963 __ andl(rax, 0x3f);
4964 __ movb(rax, Address(r11, rax, Address::times_4));
4965 __ movb(Address(dest, dp, Address::times_1, 2), rax);
4966 __ movl(rax, r15);
4967 __ andl(rax, 0x3f);
4968 __ movb(rax, Address(r11, rax, Address::times_4));
4969 __ movb(Address(dest, dp, Address::times_1, 3), rax);
4970 __ subl(length, 3);
4971 __ addq(dest, 4);
4972 __ addq(source, 3);
4973 __ jmp(L_process3);
4974 __ BIND(L_exit);
4975 __ pop(r15);
4976 __ pop(r14);
4977 __ pop(r13);
4978 __ pop(r12);
4979 __ leave();
4980 __ ret(0);
4981 return start;
4982 }
4983
4984 /**
4985 * Arguments:
4986 *
4987 * Inputs:
4988 * c_rarg0 - int crc
4989 * c_rarg1 - byte* buf
4990 * c_rarg2 - int length
4991 *
4992 * Ouput:
4993 * rax - int crc result
4994 */
4995 address generate_updateBytesCRC32() {
4996 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
4997
4998 __ align(CodeEntryAlignment);
4999 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
5000
5001 address start = __ pc();
5002 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5003 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5004 // rscratch1: r10
5005 const Register crc = c_rarg0; // crc
5006 const Register buf = c_rarg1; // source java byte array address
5007 const Register len = c_rarg2; // length
5008 const Register table = c_rarg3; // crc_table address (reuse register)
5009 const Register tmp = r11;
5010 assert_different_registers(crc, buf, len, table, tmp, rax);
5011
5012 BLOCK_COMMENT("Entry:");
5013 __ enter(); // required for proper stackwalking of RuntimeStub frame
5014
5015 __ kernel_crc32(crc, buf, len, table, tmp);
5016
5017 __ movl(rax, crc);
5018 __ vzeroupper();
5019 __ leave(); // required for proper stackwalking of RuntimeStub frame
5020 __ ret(0);
5021
5022 return start;
5023 }
5024
5025 /**
5026 * Arguments:
5027 *
5028 * Inputs:
5029 * c_rarg0 - int crc
5030 * c_rarg1 - byte* buf
5031 * c_rarg2 - long length
5032 * c_rarg3 - table_start - optional (present only when doing a library_call,
5033 * not used by x86 algorithm)
5034 *
5035 * Ouput:
5036 * rax - int crc result
5037 */
5038 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
5039 assert(UseCRC32CIntrinsics, "need SSE4_2");
5040 __ align(CodeEntryAlignment);
5041 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
5042 address start = __ pc();
5043 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
5044 //Windows RCX RDX R8 R9 none none XMM0..XMM3
5045 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
5046 const Register crc = c_rarg0; // crc
5047 const Register buf = c_rarg1; // source java byte array address
5048 const Register len = c_rarg2; // length
5049 const Register a = rax;
5050 const Register j = r9;
5051 const Register k = r10;
5052 const Register l = r11;
5053 #ifdef _WIN64
5054 const Register y = rdi;
5055 const Register z = rsi;
5056 #else
5057 const Register y = rcx;
5058 const Register z = r8;
5059 #endif
5060 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
5061
5062 BLOCK_COMMENT("Entry:");
5063 __ enter(); // required for proper stackwalking of RuntimeStub frame
5064 #ifdef _WIN64
5065 __ push(y);
5066 __ push(z);
5067 #endif
5068 __ crc32c_ipl_alg2_alt2(crc, buf, len,
5069 a, j, k,
5070 l, y, z,
5071 c_farg0, c_farg1, c_farg2,
5072 is_pclmulqdq_supported);
5073 __ movl(rax, crc);
5074 #ifdef _WIN64
5075 __ pop(z);
5076 __ pop(y);
5077 #endif
5078 __ vzeroupper();
5079 __ leave(); // required for proper stackwalking of RuntimeStub frame
5080 __ ret(0);
5081
5082 return start;
5083 }
5084
5085 /**
5086 * Arguments:
5087 *
5088 * Input:
5089 * c_rarg0 - x address
5090 * c_rarg1 - x length
5091 * c_rarg2 - y address
5092 * c_rarg3 - y length
5093 * not Win64
5094 * c_rarg4 - z address
5095 * c_rarg5 - z length
5096 * Win64
5097 * rsp+40 - z address
5098 * rsp+48 - z length
5099 */
5100 address generate_multiplyToLen() {
5101 __ align(CodeEntryAlignment);
5102 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
5103
5104 address start = __ pc();
5105 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5106 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5107 const Register x = rdi;
5108 const Register xlen = rax;
5109 const Register y = rsi;
5110 const Register ylen = rcx;
5111 const Register z = r8;
5112 const Register zlen = r11;
5113
5114 // Next registers will be saved on stack in multiply_to_len().
5115 const Register tmp1 = r12;
5116 const Register tmp2 = r13;
5117 const Register tmp3 = r14;
5118 const Register tmp4 = r15;
5119 const Register tmp5 = rbx;
5120
5121 BLOCK_COMMENT("Entry:");
5122 __ enter(); // required for proper stackwalking of RuntimeStub frame
5123
5124 #ifndef _WIN64
5125 __ movptr(zlen, r9); // Save r9 in r11 - zlen
5126 #endif
5127 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
5128 // ylen => rcx, z => r8, zlen => r11
5129 // r9 and r10 may be used to save non-volatile registers
5130 #ifdef _WIN64
5131 // last 2 arguments (#4, #5) are on stack on Win64
5132 __ movptr(z, Address(rsp, 6 * wordSize));
5133 __ movptr(zlen, Address(rsp, 7 * wordSize));
5134 #endif
5135
5136 __ movptr(xlen, rsi);
5137 __ movptr(y, rdx);
5138 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
5139
5140 restore_arg_regs();
5141
5142 __ leave(); // required for proper stackwalking of RuntimeStub frame
5143 __ ret(0);
5144
5145 return start;
5146 }
5147
5148 /**
5149 * Arguments:
5150 *
5151 * Input:
5152 * c_rarg0 - obja address
5153 * c_rarg1 - objb address
5154 * c_rarg3 - length length
5155 * c_rarg4 - scale log2_array_indxscale
5156 *
5157 * Output:
5158 * rax - int >= mismatched index, < 0 bitwise complement of tail
5159 */
5160 address generate_vectorizedMismatch() {
5161 __ align(CodeEntryAlignment);
5162 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
5163 address start = __ pc();
5164
5165 BLOCK_COMMENT("Entry:");
5166 __ enter();
5167
5168 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5169 const Register scale = c_rarg0; //rcx, will exchange with r9
5170 const Register objb = c_rarg1; //rdx
5171 const Register length = c_rarg2; //r8
5172 const Register obja = c_rarg3; //r9
5173 __ xchgq(obja, scale); //now obja and scale contains the correct contents
5174
5175 const Register tmp1 = r10;
5176 const Register tmp2 = r11;
5177 #endif
5178 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5179 const Register obja = c_rarg0; //U:rdi
5180 const Register objb = c_rarg1; //U:rsi
5181 const Register length = c_rarg2; //U:rdx
5182 const Register scale = c_rarg3; //U:rcx
5183 const Register tmp1 = r8;
5184 const Register tmp2 = r9;
5185 #endif
5186 const Register result = rax; //return value
5187 const XMMRegister vec0 = xmm0;
5188 const XMMRegister vec1 = xmm1;
5189 const XMMRegister vec2 = xmm2;
5190
5191 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
5192
5193 __ vzeroupper();
5194 __ leave();
5195 __ ret(0);
5196
5197 return start;
5198 }
5199
5200 /**
5201 * Arguments:
5202 *
5203 // Input:
5204 // c_rarg0 - x address
5205 // c_rarg1 - x length
5206 // c_rarg2 - z address
5207 // c_rarg3 - z lenth
5208 *
5209 */
5210 address generate_squareToLen() {
5211
5212 __ align(CodeEntryAlignment);
5213 StubCodeMark mark(this, "StubRoutines", "squareToLen");
5214
5215 address start = __ pc();
5216 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5217 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
5218 const Register x = rdi;
5219 const Register len = rsi;
5220 const Register z = r8;
5221 const Register zlen = rcx;
5222
5223 const Register tmp1 = r12;
5224 const Register tmp2 = r13;
5225 const Register tmp3 = r14;
5226 const Register tmp4 = r15;
5227 const Register tmp5 = rbx;
5228
5229 BLOCK_COMMENT("Entry:");
5230 __ enter(); // required for proper stackwalking of RuntimeStub frame
5231
5232 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
5233 // zlen => rcx
5234 // r9 and r10 may be used to save non-volatile registers
5235 __ movptr(r8, rdx);
5236 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5237
5238 restore_arg_regs();
5239
5240 __ leave(); // required for proper stackwalking of RuntimeStub frame
5241 __ ret(0);
5242
5243 return start;
5244 }
5245
5246 address generate_method_entry_barrier() {
5247 __ align(CodeEntryAlignment);
5248 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
5249
5250 Label deoptimize_label;
5251
5252 address start = __ pc();
5253
5254 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
5255
5256 BLOCK_COMMENT("Entry:");
5257 __ enter(); // save rbp
5258
5259 // save c_rarg0, because we want to use that value.
5260 // We could do without it but then we depend on the number of slots used by pusha
5261 __ push(c_rarg0);
5262
5263 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
5264
5265 __ pusha();
5266
5267 // The method may have floats as arguments, and we must spill them before calling
5268 // the VM runtime.
5269 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
5270 const int xmm_size = wordSize * 2;
5271 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
5272 __ subptr(rsp, xmm_spill_size);
5273 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
5274 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
5275 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
5276 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
5277 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
5278 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
5279 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
5280 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
5281
5282 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
5283
5284 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
5285 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
5286 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
5287 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
5288 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
5289 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
5290 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
5291 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
5292 __ addptr(rsp, xmm_spill_size);
5293
5294 __ cmpl(rax, 1); // 1 means deoptimize
5295 __ jcc(Assembler::equal, deoptimize_label);
5296
5297 __ popa();
5298 __ pop(c_rarg0);
5299
5300 __ leave();
5301
5302 __ addptr(rsp, 1 * wordSize); // cookie
5303 __ ret(0);
5304
5305
5306 __ BIND(deoptimize_label);
5307
5308 __ popa();
5309 __ pop(c_rarg0);
5310
5311 __ leave();
5312
5313 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
5314 // here while still having a correct stack is valuable
5315 __ testptr(rsp, Address(rsp, 0));
5316
5317 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
5318 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
5319
5320 return start;
5321 }
5322
5323 /**
5324 * Arguments:
5325 *
5326 * Input:
5327 * c_rarg0 - out address
5328 * c_rarg1 - in address
5329 * c_rarg2 - offset
5330 * c_rarg3 - len
5331 * not Win64
5332 * c_rarg4 - k
5333 * Win64
5334 * rsp+40 - k
5335 */
5336 address generate_mulAdd() {
5337 __ align(CodeEntryAlignment);
5338 StubCodeMark mark(this, "StubRoutines", "mulAdd");
5339
5340 address start = __ pc();
5341 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5342 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5343 const Register out = rdi;
5344 const Register in = rsi;
5345 const Register offset = r11;
5346 const Register len = rcx;
5347 const Register k = r8;
5348
5349 // Next registers will be saved on stack in mul_add().
5350 const Register tmp1 = r12;
5351 const Register tmp2 = r13;
5352 const Register tmp3 = r14;
5353 const Register tmp4 = r15;
5354 const Register tmp5 = rbx;
5355
5356 BLOCK_COMMENT("Entry:");
5357 __ enter(); // required for proper stackwalking of RuntimeStub frame
5358
5359 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
5360 // len => rcx, k => r8
5361 // r9 and r10 may be used to save non-volatile registers
5362 #ifdef _WIN64
5363 // last argument is on stack on Win64
5364 __ movl(k, Address(rsp, 6 * wordSize));
5365 #endif
5366 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
5367 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5368
5369 restore_arg_regs();
5370
5371 __ leave(); // required for proper stackwalking of RuntimeStub frame
5372 __ ret(0);
5373
5374 return start;
5375 }
5376
5377 address generate_libmExp() {
5378 StubCodeMark mark(this, "StubRoutines", "libmExp");
5379
5380 address start = __ pc();
5381
5382 const XMMRegister x0 = xmm0;
5383 const XMMRegister x1 = xmm1;
5384 const XMMRegister x2 = xmm2;
5385 const XMMRegister x3 = xmm3;
5386
5387 const XMMRegister x4 = xmm4;
5388 const XMMRegister x5 = xmm5;
5389 const XMMRegister x6 = xmm6;
5390 const XMMRegister x7 = xmm7;
5391
5392 const Register tmp = r11;
5393
5394 BLOCK_COMMENT("Entry:");
5395 __ enter(); // required for proper stackwalking of RuntimeStub frame
5396
5397 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5398
5399 __ leave(); // required for proper stackwalking of RuntimeStub frame
5400 __ ret(0);
5401
5402 return start;
5403
5404 }
5405
5406 address generate_libmLog() {
5407 StubCodeMark mark(this, "StubRoutines", "libmLog");
5408
5409 address start = __ pc();
5410
5411 const XMMRegister x0 = xmm0;
5412 const XMMRegister x1 = xmm1;
5413 const XMMRegister x2 = xmm2;
5414 const XMMRegister x3 = xmm3;
5415
5416 const XMMRegister x4 = xmm4;
5417 const XMMRegister x5 = xmm5;
5418 const XMMRegister x6 = xmm6;
5419 const XMMRegister x7 = xmm7;
5420
5421 const Register tmp1 = r11;
5422 const Register tmp2 = r8;
5423
5424 BLOCK_COMMENT("Entry:");
5425 __ enter(); // required for proper stackwalking of RuntimeStub frame
5426
5427 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2);
5428
5429 __ leave(); // required for proper stackwalking of RuntimeStub frame
5430 __ ret(0);
5431
5432 return start;
5433
5434 }
5435
5436 address generate_libmLog10() {
5437 StubCodeMark mark(this, "StubRoutines", "libmLog10");
5438
5439 address start = __ pc();
5440
5441 const XMMRegister x0 = xmm0;
5442 const XMMRegister x1 = xmm1;
5443 const XMMRegister x2 = xmm2;
5444 const XMMRegister x3 = xmm3;
5445
5446 const XMMRegister x4 = xmm4;
5447 const XMMRegister x5 = xmm5;
5448 const XMMRegister x6 = xmm6;
5449 const XMMRegister x7 = xmm7;
5450
5451 const Register tmp = r11;
5452
5453 BLOCK_COMMENT("Entry:");
5454 __ enter(); // required for proper stackwalking of RuntimeStub frame
5455
5456 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5457
5458 __ leave(); // required for proper stackwalking of RuntimeStub frame
5459 __ ret(0);
5460
5461 return start;
5462
5463 }
5464
5465 address generate_libmPow() {
5466 StubCodeMark mark(this, "StubRoutines", "libmPow");
5467
5468 address start = __ pc();
5469
5470 const XMMRegister x0 = xmm0;
5471 const XMMRegister x1 = xmm1;
5472 const XMMRegister x2 = xmm2;
5473 const XMMRegister x3 = xmm3;
5474
5475 const XMMRegister x4 = xmm4;
5476 const XMMRegister x5 = xmm5;
5477 const XMMRegister x6 = xmm6;
5478 const XMMRegister x7 = xmm7;
5479
5480 const Register tmp1 = r8;
5481 const Register tmp2 = r9;
5482 const Register tmp3 = r10;
5483 const Register tmp4 = r11;
5484
5485 BLOCK_COMMENT("Entry:");
5486 __ enter(); // required for proper stackwalking of RuntimeStub frame
5487
5488 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5489
5490 __ leave(); // required for proper stackwalking of RuntimeStub frame
5491 __ ret(0);
5492
5493 return start;
5494
5495 }
5496
5497 address generate_libmSin() {
5498 StubCodeMark mark(this, "StubRoutines", "libmSin");
5499
5500 address start = __ pc();
5501
5502 const XMMRegister x0 = xmm0;
5503 const XMMRegister x1 = xmm1;
5504 const XMMRegister x2 = xmm2;
5505 const XMMRegister x3 = xmm3;
5506
5507 const XMMRegister x4 = xmm4;
5508 const XMMRegister x5 = xmm5;
5509 const XMMRegister x6 = xmm6;
5510 const XMMRegister x7 = xmm7;
5511
5512 const Register tmp1 = r8;
5513 const Register tmp2 = r9;
5514 const Register tmp3 = r10;
5515 const Register tmp4 = r11;
5516
5517 BLOCK_COMMENT("Entry:");
5518 __ enter(); // required for proper stackwalking of RuntimeStub frame
5519
5520 #ifdef _WIN64
5521 __ push(rsi);
5522 __ push(rdi);
5523 #endif
5524 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5525
5526 #ifdef _WIN64
5527 __ pop(rdi);
5528 __ pop(rsi);
5529 #endif
5530
5531 __ leave(); // required for proper stackwalking of RuntimeStub frame
5532 __ ret(0);
5533
5534 return start;
5535
5536 }
5537
5538 address generate_libmCos() {
5539 StubCodeMark mark(this, "StubRoutines", "libmCos");
5540
5541 address start = __ pc();
5542
5543 const XMMRegister x0 = xmm0;
5544 const XMMRegister x1 = xmm1;
5545 const XMMRegister x2 = xmm2;
5546 const XMMRegister x3 = xmm3;
5547
5548 const XMMRegister x4 = xmm4;
5549 const XMMRegister x5 = xmm5;
5550 const XMMRegister x6 = xmm6;
5551 const XMMRegister x7 = xmm7;
5552
5553 const Register tmp1 = r8;
5554 const Register tmp2 = r9;
5555 const Register tmp3 = r10;
5556 const Register tmp4 = r11;
5557
5558 BLOCK_COMMENT("Entry:");
5559 __ enter(); // required for proper stackwalking of RuntimeStub frame
5560
5561 #ifdef _WIN64
5562 __ push(rsi);
5563 __ push(rdi);
5564 #endif
5565 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5566
5567 #ifdef _WIN64
5568 __ pop(rdi);
5569 __ pop(rsi);
5570 #endif
5571
5572 __ leave(); // required for proper stackwalking of RuntimeStub frame
5573 __ ret(0);
5574
5575 return start;
5576
5577 }
5578
5579 address generate_libmTan() {
5580 StubCodeMark mark(this, "StubRoutines", "libmTan");
5581
5582 address start = __ pc();
5583
5584 const XMMRegister x0 = xmm0;
5585 const XMMRegister x1 = xmm1;
5586 const XMMRegister x2 = xmm2;
5587 const XMMRegister x3 = xmm3;
5588
5589 const XMMRegister x4 = xmm4;
5590 const XMMRegister x5 = xmm5;
5591 const XMMRegister x6 = xmm6;
5592 const XMMRegister x7 = xmm7;
5593
5594 const Register tmp1 = r8;
5595 const Register tmp2 = r9;
5596 const Register tmp3 = r10;
5597 const Register tmp4 = r11;
5598
5599 BLOCK_COMMENT("Entry:");
5600 __ enter(); // required for proper stackwalking of RuntimeStub frame
5601
5602 #ifdef _WIN64
5603 __ push(rsi);
5604 __ push(rdi);
5605 #endif
5606 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5607
5608 #ifdef _WIN64
5609 __ pop(rdi);
5610 __ pop(rsi);
5611 #endif
5612
5613 __ leave(); // required for proper stackwalking of RuntimeStub frame
5614 __ ret(0);
5615
5616 return start;
5617
5618 }
5619
5620 #undef __
5621 #define __ masm->
5622
5623 // Continuation point for throwing of implicit exceptions that are
5624 // not handled in the current activation. Fabricates an exception
5625 // oop and initiates normal exception dispatching in this
5626 // frame. Since we need to preserve callee-saved values (currently
5627 // only for C2, but done for C1 as well) we need a callee-saved oop
5628 // map and therefore have to make these stubs into RuntimeStubs
5629 // rather than BufferBlobs. If the compiler needs all registers to
5630 // be preserved between the fault point and the exception handler
5631 // then it must assume responsibility for that in
5632 // AbstractCompiler::continuation_for_implicit_null_exception or
5633 // continuation_for_implicit_division_by_zero_exception. All other
5634 // implicit exceptions (e.g., NullPointerException or
5635 // AbstractMethodError on entry) are either at call sites or
5636 // otherwise assume that stack unwinding will be initiated, so
5637 // caller saved registers were assumed volatile in the compiler.
5638 address generate_throw_exception(const char* name,
5639 address runtime_entry,
5640 Register arg1 = noreg,
5641 Register arg2 = noreg) {
5642 // Information about frame layout at time of blocking runtime call.
5643 // Note that we only have to preserve callee-saved registers since
5644 // the compilers are responsible for supplying a continuation point
5645 // if they expect all registers to be preserved.
5646 enum layout {
5647 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
5648 rbp_off2,
5649 return_off,
5650 return_off2,
5651 framesize // inclusive of return address
5652 };
5653
5654 int insts_size = 512;
5655 int locs_size = 64;
5656
5657 CodeBuffer code(name, insts_size, locs_size);
5658 OopMapSet* oop_maps = new OopMapSet();
5659 MacroAssembler* masm = new MacroAssembler(&code);
5660
5661 address start = __ pc();
5662
5663 // This is an inlined and slightly modified version of call_VM
5664 // which has the ability to fetch the return PC out of
5665 // thread-local storage and also sets up last_Java_sp slightly
5666 // differently than the real call_VM
5667
5668 __ enter(); // required for proper stackwalking of RuntimeStub frame
5669
5670 assert(is_even(framesize/2), "sp not 16-byte aligned");
5671
5672 // return address and rbp are already in place
5673 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
5674
5675 int frame_complete = __ pc() - start;
5676
5677 // Set up last_Java_sp and last_Java_fp
5678 address the_pc = __ pc();
5679 __ set_last_Java_frame(rsp, rbp, the_pc);
5680 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
5681
5682 // Call runtime
5683 if (arg1 != noreg) {
5684 assert(arg2 != c_rarg1, "clobbered");
5685 __ movptr(c_rarg1, arg1);
5686 }
5687 if (arg2 != noreg) {
5688 __ movptr(c_rarg2, arg2);
5689 }
5690 __ movptr(c_rarg0, r15_thread);
5691 BLOCK_COMMENT("call runtime_entry");
5692 __ call(RuntimeAddress(runtime_entry));
5693
5694 // Generate oop map
5695 OopMap* map = new OopMap(framesize, 0);
5696
5697 oop_maps->add_gc_map(the_pc - start, map);
5698
5699 __ reset_last_Java_frame(true);
5700
5701 __ leave(); // required for proper stackwalking of RuntimeStub frame
5702
5703 // check for pending exceptions
5704 #ifdef ASSERT
5705 Label L;
5706 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
5707 (int32_t) NULL_WORD);
5708 __ jcc(Assembler::notEqual, L);
5709 __ should_not_reach_here();
5710 __ bind(L);
5711 #endif // ASSERT
5712 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5713
5714
5715 // codeBlob framesize is in words (not VMRegImpl::slot_size)
5716 RuntimeStub* stub =
5717 RuntimeStub::new_runtime_stub(name,
5718 &code,
5719 frame_complete,
5720 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
5721 oop_maps, false);
5722 return stub->entry_point();
5723 }
5724
5725 void create_control_words() {
5726 // Round to nearest, 53-bit mode, exceptions masked
5727 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
5728 // Round to zero, 53-bit mode, exception mased
5729 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
5730 // Round to nearest, 24-bit mode, exceptions masked
5731 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
5732 // Round to nearest, 64-bit mode, exceptions masked
5733 StubRoutines::_mxcsr_std = 0x1F80;
5734 // Note: the following two constants are 80-bit values
5735 // layout is critical for correct loading by FPU.
5736 // Bias for strict fp multiply/divide
5737 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
5738 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
5739 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
5740 // Un-Bias for strict fp multiply/divide
5741 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
5742 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
5743 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
5744 }
5745
5746 // Call here from the interpreter or compiled code to either load
5747 // multiple returned values from the value type instance being
5748 // returned to registers or to store returned values to a newly
5749 // allocated value type instance.
5750 address generate_return_value_stub(address destination, const char* name, bool has_res) {
5751 // We need to save all registers the calling convention may use so
5752 // the runtime calls read or update those registers. This needs to
5753 // be in sync with SharedRuntime::java_return_convention().
5754 enum layout {
5755 pad_off = frame::arg_reg_save_area_bytes/BytesPerInt, pad_off_2,
5756 rax_off, rax_off_2,
5757 j_rarg5_off, j_rarg5_2,
5758 j_rarg4_off, j_rarg4_2,
5759 j_rarg3_off, j_rarg3_2,
5760 j_rarg2_off, j_rarg2_2,
5761 j_rarg1_off, j_rarg1_2,
5762 j_rarg0_off, j_rarg0_2,
5763 j_farg0_off, j_farg0_2,
5764 j_farg1_off, j_farg1_2,
5765 j_farg2_off, j_farg2_2,
5766 j_farg3_off, j_farg3_2,
5767 j_farg4_off, j_farg4_2,
5768 j_farg5_off, j_farg5_2,
5769 j_farg6_off, j_farg6_2,
5770 j_farg7_off, j_farg7_2,
5771 rbp_off, rbp_off_2,
5772 return_off, return_off_2,
5773
5774 framesize
5775 };
5776
5777 CodeBuffer buffer(name, 1000, 512);
5778 MacroAssembler* masm = new MacroAssembler(&buffer);
5779
5780 int frame_size_in_bytes = align_up(framesize*BytesPerInt, 16);
5781 assert(frame_size_in_bytes == framesize*BytesPerInt, "misaligned");
5782 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
5783 int frame_size_in_words = frame_size_in_bytes / wordSize;
5784
5785 OopMapSet *oop_maps = new OopMapSet();
5786 OopMap* map = new OopMap(frame_size_in_slots, 0);
5787
5788 map->set_callee_saved(VMRegImpl::stack2reg(rax_off), rax->as_VMReg());
5789 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg5_off), j_rarg5->as_VMReg());
5790 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg4_off), j_rarg4->as_VMReg());
5791 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg3_off), j_rarg3->as_VMReg());
5792 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg2_off), j_rarg2->as_VMReg());
5793 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg1_off), j_rarg1->as_VMReg());
5794 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg0_off), j_rarg0->as_VMReg());
5795 map->set_callee_saved(VMRegImpl::stack2reg(j_farg0_off), j_farg0->as_VMReg());
5796 map->set_callee_saved(VMRegImpl::stack2reg(j_farg1_off), j_farg1->as_VMReg());
5797 map->set_callee_saved(VMRegImpl::stack2reg(j_farg2_off), j_farg2->as_VMReg());
5798 map->set_callee_saved(VMRegImpl::stack2reg(j_farg3_off), j_farg3->as_VMReg());
5799 map->set_callee_saved(VMRegImpl::stack2reg(j_farg4_off), j_farg4->as_VMReg());
5800 map->set_callee_saved(VMRegImpl::stack2reg(j_farg5_off), j_farg5->as_VMReg());
5801 map->set_callee_saved(VMRegImpl::stack2reg(j_farg6_off), j_farg6->as_VMReg());
5802 map->set_callee_saved(VMRegImpl::stack2reg(j_farg7_off), j_farg7->as_VMReg());
5803
5804 int start = __ offset();
5805
5806 __ subptr(rsp, frame_size_in_bytes - 8 /* return address*/);
5807
5808 __ movptr(Address(rsp, rbp_off * BytesPerInt), rbp);
5809 __ movdbl(Address(rsp, j_farg7_off * BytesPerInt), j_farg7);
5810 __ movdbl(Address(rsp, j_farg6_off * BytesPerInt), j_farg6);
5811 __ movdbl(Address(rsp, j_farg5_off * BytesPerInt), j_farg5);
5812 __ movdbl(Address(rsp, j_farg4_off * BytesPerInt), j_farg4);
5813 __ movdbl(Address(rsp, j_farg3_off * BytesPerInt), j_farg3);
5814 __ movdbl(Address(rsp, j_farg2_off * BytesPerInt), j_farg2);
5815 __ movdbl(Address(rsp, j_farg1_off * BytesPerInt), j_farg1);
5816 __ movdbl(Address(rsp, j_farg0_off * BytesPerInt), j_farg0);
5817
5818 __ movptr(Address(rsp, j_rarg0_off * BytesPerInt), j_rarg0);
5819 __ movptr(Address(rsp, j_rarg1_off * BytesPerInt), j_rarg1);
5820 __ movptr(Address(rsp, j_rarg2_off * BytesPerInt), j_rarg2);
5821 __ movptr(Address(rsp, j_rarg3_off * BytesPerInt), j_rarg3);
5822 __ movptr(Address(rsp, j_rarg4_off * BytesPerInt), j_rarg4);
5823 __ movptr(Address(rsp, j_rarg5_off * BytesPerInt), j_rarg5);
5824 __ movptr(Address(rsp, rax_off * BytesPerInt), rax);
5825
5826 int frame_complete = __ offset();
5827
5828 __ set_last_Java_frame(noreg, noreg, NULL);
5829
5830 __ mov(c_rarg0, r15_thread);
5831 __ mov(c_rarg1, rax);
5832
5833 __ call(RuntimeAddress(destination));
5834
5835 // Set an oopmap for the call site.
5836
5837 oop_maps->add_gc_map( __ offset() - start, map);
5838
5839 // clear last_Java_sp
5840 __ reset_last_Java_frame(false);
5841
5842 __ movptr(rbp, Address(rsp, rbp_off * BytesPerInt));
5843 __ movdbl(j_farg7, Address(rsp, j_farg7_off * BytesPerInt));
5844 __ movdbl(j_farg6, Address(rsp, j_farg6_off * BytesPerInt));
5845 __ movdbl(j_farg5, Address(rsp, j_farg5_off * BytesPerInt));
5846 __ movdbl(j_farg4, Address(rsp, j_farg4_off * BytesPerInt));
5847 __ movdbl(j_farg3, Address(rsp, j_farg3_off * BytesPerInt));
5848 __ movdbl(j_farg2, Address(rsp, j_farg2_off * BytesPerInt));
5849 __ movdbl(j_farg1, Address(rsp, j_farg1_off * BytesPerInt));
5850 __ movdbl(j_farg0, Address(rsp, j_farg0_off * BytesPerInt));
5851
5852 __ movptr(j_rarg0, Address(rsp, j_rarg0_off * BytesPerInt));
5853 __ movptr(j_rarg1, Address(rsp, j_rarg1_off * BytesPerInt));
5854 __ movptr(j_rarg2, Address(rsp, j_rarg2_off * BytesPerInt));
5855 __ movptr(j_rarg3, Address(rsp, j_rarg3_off * BytesPerInt));
5856 __ movptr(j_rarg4, Address(rsp, j_rarg4_off * BytesPerInt));
5857 __ movptr(j_rarg5, Address(rsp, j_rarg5_off * BytesPerInt));
5858 __ movptr(rax, Address(rsp, rax_off * BytesPerInt));
5859
5860 __ addptr(rsp, frame_size_in_bytes-8);
5861
5862 // check for pending exceptions
5863 Label pending;
5864 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
5865 __ jcc(Assembler::notEqual, pending);
5866
5867 if (has_res) {
5868 __ get_vm_result(rax, r15_thread);
5869 }
5870
5871 __ ret(0);
5872
5873 __ bind(pending);
5874
5875 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
5876 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5877
5878 // -------------
5879 // make sure all code is generated
5880 masm->flush();
5881
5882 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, false);
5883 return stub->entry_point();
5884 }
5885
5886 // Initialization
5887 void generate_initial() {
5888 // Generates all stubs and initializes the entry points
5889
5890 // This platform-specific settings are needed by generate_call_stub()
5891 create_control_words();
5892
5893 // entry points that exist in all platforms Note: This is code
5894 // that could be shared among different platforms - however the
5895 // benefit seems to be smaller than the disadvantage of having a
5896 // much more complicated generator structure. See also comment in
5897 // stubRoutines.hpp.
5898
5899 StubRoutines::_forward_exception_entry = generate_forward_exception();
5900
5901 StubRoutines::_call_stub_entry =
5902 generate_call_stub(StubRoutines::_call_stub_return_address);
5903
5904 // is referenced by megamorphic call
5905 StubRoutines::_catch_exception_entry = generate_catch_exception();
5906
5907 // atomic calls
5908 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
5909 StubRoutines::_atomic_xchg_long_entry = generate_atomic_xchg_long();
5910 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
5911 StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
5912 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
5913 StubRoutines::_atomic_add_entry = generate_atomic_add();
5914 StubRoutines::_atomic_add_long_entry = generate_atomic_add_long();
5915 StubRoutines::_fence_entry = generate_orderaccess_fence();
5916
5917 // platform dependent
5918 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
5919 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
5920
5921 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
5922
5923 // Build this early so it's available for the interpreter.
5924 StubRoutines::_throw_StackOverflowError_entry =
5925 generate_throw_exception("StackOverflowError throw_exception",
5926 CAST_FROM_FN_PTR(address,
5927 SharedRuntime::
5928 throw_StackOverflowError));
5929 StubRoutines::_throw_delayed_StackOverflowError_entry =
5930 generate_throw_exception("delayed StackOverflowError throw_exception",
5931 CAST_FROM_FN_PTR(address,
5932 SharedRuntime::
5933 throw_delayed_StackOverflowError));
5934 if (UseCRC32Intrinsics) {
5935 // set table address before stub generation which use it
5936 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
5937 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
5938 }
5939
5940 if (UseCRC32CIntrinsics) {
5941 bool supports_clmul = VM_Version::supports_clmul();
5942 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
5943 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
5944 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
5945 }
5946 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
5947 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
5948 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
5949 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
5950 StubRoutines::x86::_ONEHALF_adr = (address)StubRoutines::x86::_ONEHALF;
5951 StubRoutines::x86::_P_2_adr = (address)StubRoutines::x86::_P_2;
5952 StubRoutines::x86::_SC_4_adr = (address)StubRoutines::x86::_SC_4;
5953 StubRoutines::x86::_Ctable_adr = (address)StubRoutines::x86::_Ctable;
5954 StubRoutines::x86::_SC_2_adr = (address)StubRoutines::x86::_SC_2;
5955 StubRoutines::x86::_SC_3_adr = (address)StubRoutines::x86::_SC_3;
5956 StubRoutines::x86::_SC_1_adr = (address)StubRoutines::x86::_SC_1;
5957 StubRoutines::x86::_PI_INV_TABLE_adr = (address)StubRoutines::x86::_PI_INV_TABLE;
5958 StubRoutines::x86::_PI_4_adr = (address)StubRoutines::x86::_PI_4;
5959 StubRoutines::x86::_PI32INV_adr = (address)StubRoutines::x86::_PI32INV;
5960 StubRoutines::x86::_SIGN_MASK_adr = (address)StubRoutines::x86::_SIGN_MASK;
5961 StubRoutines::x86::_P_1_adr = (address)StubRoutines::x86::_P_1;
5962 StubRoutines::x86::_P_3_adr = (address)StubRoutines::x86::_P_3;
5963 StubRoutines::x86::_NEG_ZERO_adr = (address)StubRoutines::x86::_NEG_ZERO;
5964 }
5965 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
5966 StubRoutines::_dexp = generate_libmExp();
5967 }
5968 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
5969 StubRoutines::_dlog = generate_libmLog();
5970 }
5971 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
5972 StubRoutines::_dlog10 = generate_libmLog10();
5973 }
5974 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
5975 StubRoutines::_dpow = generate_libmPow();
5976 }
5977 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
5978 StubRoutines::_dsin = generate_libmSin();
5979 }
5980 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
5981 StubRoutines::_dcos = generate_libmCos();
5982 }
5983 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
5984 StubRoutines::_dtan = generate_libmTan();
5985 }
5986 }
5987
5988 StubRoutines::_load_value_type_fields_in_regs = generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::load_value_type_fields_in_regs), "load_value_type_fields_in_regs", false);
5989 StubRoutines::_store_value_type_fields_to_buf = generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::store_value_type_fields_to_buf), "store_value_type_fields_to_buf", true);
5990 }
5991
5992 void generate_all() {
5993 // Generates all stubs and initializes the entry points
5994
5995 // These entry points require SharedInfo::stack0 to be set up in
5996 // non-core builds and need to be relocatable, so they each
5997 // fabricate a RuntimeStub internally.
5998 StubRoutines::_throw_AbstractMethodError_entry =
5999 generate_throw_exception("AbstractMethodError throw_exception",
6000 CAST_FROM_FN_PTR(address,
6001 SharedRuntime::
6002 throw_AbstractMethodError));
6003
6004 StubRoutines::_throw_IncompatibleClassChangeError_entry =
6005 generate_throw_exception("IncompatibleClassChangeError throw_exception",
6006 CAST_FROM_FN_PTR(address,
6007 SharedRuntime::
6008 throw_IncompatibleClassChangeError));
6009
6010 StubRoutines::_throw_NullPointerException_at_call_entry =
6011 generate_throw_exception("NullPointerException at call throw_exception",
6012 CAST_FROM_FN_PTR(address,
6013 SharedRuntime::
6014 throw_NullPointerException_at_call));
6015
6016 // entry points that are platform specific
6017 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
6018 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
6019 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
6020 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
6021
6022 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6023 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
6024 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6025 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
6026
6027 // support for verify_oop (must happen after universe_init)
6028 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
6029
6030 // arraycopy stubs used by compilers
6031 generate_arraycopy_stubs();
6032
6033 // don't bother generating these AES intrinsic stubs unless global flag is set
6034 if (UseAESIntrinsics) {
6035 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
6036 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
6037 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
6038 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
6039 if (VM_Version::supports_vaes() && VM_Version::supports_avx512vl() && VM_Version::supports_avx512dq() ) {
6040 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptVectorAESCrypt();
6041 } else {
6042 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
6043 }
6044 }
6045 if (UseAESCTRIntrinsics){
6046 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
6047 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
6048 }
6049
6050 if (UseSHA1Intrinsics) {
6051 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
6052 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
6053 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
6054 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
6055 }
6056 if (UseSHA256Intrinsics) {
6057 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
6058 char* dst = (char*)StubRoutines::x86::_k256_W;
6059 char* src = (char*)StubRoutines::x86::_k256;
6060 for (int ii = 0; ii < 16; ++ii) {
6061 memcpy(dst + 32 * ii, src + 16 * ii, 16);
6062 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
6063 }
6064 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
6065 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
6066 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
6067 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
6068 }
6069 if (UseSHA512Intrinsics) {
6070 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
6071 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
6072 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
6073 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
6074 }
6075
6076 // Generate GHASH intrinsics code
6077 if (UseGHASHIntrinsics) {
6078 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
6079 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
6080 if (VM_Version::supports_avx()) {
6081 StubRoutines::x86::_ghash_shuffmask_addr = ghash_shufflemask_addr();
6082 StubRoutines::x86::_ghash_poly_addr = ghash_polynomial_addr();
6083 StubRoutines::_ghash_processBlocks = generate_avx_ghash_processBlocks();
6084 } else {
6085 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
6086 }
6087 }
6088
6089 if (UseBASE64Intrinsics) {
6090 StubRoutines::x86::_and_mask = base64_and_mask_addr();
6091 StubRoutines::x86::_bswap_mask = base64_bswap_mask_addr();
6092 StubRoutines::x86::_base64_charset = base64_charset_addr();
6093 StubRoutines::x86::_url_charset = base64url_charset_addr();
6094 StubRoutines::x86::_gather_mask = base64_gather_mask_addr();
6095 StubRoutines::x86::_left_shift_mask = base64_left_shift_mask_addr();
6096 StubRoutines::x86::_right_shift_mask = base64_right_shift_mask_addr();
6097 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
6098 }
6099
6100 // Safefetch stubs.
6101 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
6102 &StubRoutines::_safefetch32_fault_pc,
6103 &StubRoutines::_safefetch32_continuation_pc);
6104 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
6105 &StubRoutines::_safefetchN_fault_pc,
6106 &StubRoutines::_safefetchN_continuation_pc);
6107
6108 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
6109 if (bs_nm != NULL) {
6110 StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
6111 }
6112 #ifdef COMPILER2
6113 if (UseMultiplyToLenIntrinsic) {
6114 StubRoutines::_multiplyToLen = generate_multiplyToLen();
6115 }
6116 if (UseSquareToLenIntrinsic) {
6117 StubRoutines::_squareToLen = generate_squareToLen();
6118 }
6119 if (UseMulAddIntrinsic) {
6120 StubRoutines::_mulAdd = generate_mulAdd();
6121 }
6122 #ifndef _WINDOWS
6123 if (UseMontgomeryMultiplyIntrinsic) {
6124 StubRoutines::_montgomeryMultiply
6125 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
6126 }
6127 if (UseMontgomerySquareIntrinsic) {
6128 StubRoutines::_montgomerySquare
6129 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
6130 }
6131 #endif // WINDOWS
6132 #endif // COMPILER2
6133
6134 if (UseVectorizedMismatchIntrinsic) {
6135 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
6136 }
6137 }
6138
6139 public:
6140 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
6141 if (all) {
6142 generate_all();
6143 } else {
6144 generate_initial();
6145 }
6146 }
6147 }; // end class declaration
6148
6149 void StubGenerator_generate(CodeBuffer* code, bool all) {
6150 StubGenerator g(code, all);
6151 }