1 /*
2 * Copyright (c) 2003, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "ci/ciUtilities.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/barrierSetNMethod.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "memory/universe.hpp"
34 #include "nativeInst_x86.hpp"
35 #include "oops/instanceOop.hpp"
36 #include "oops/method.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/methodHandles.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/handles.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubCodeGenerator.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/thread.inline.hpp"
46 #ifdef COMPILER2
47 #include "opto/runtime.hpp"
48 #endif
49 #if INCLUDE_ZGC
50 #include "gc/z/zThreadLocalData.hpp"
51 #endif
52
53 // Declaration and definition of StubGenerator (no .hpp file).
54 // For a more detailed description of the stub routine structure
55 // see the comment in stubRoutines.hpp
56
57 #define __ _masm->
58 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
59 #define a__ ((Assembler*)_masm)->
60
61 #ifdef PRODUCT
62 #define BLOCK_COMMENT(str) /* nothing */
63 #else
64 #define BLOCK_COMMENT(str) __ block_comment(str)
65 #endif
66
67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
68 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
69
70 // Stub Code definitions
71
72 class StubGenerator: public StubCodeGenerator {
73 private:
74
75 #ifdef PRODUCT
76 #define inc_counter_np(counter) ((void)0)
77 #else
78 void inc_counter_np_(int& counter) {
79 // This can destroy rscratch1 if counter is far from the code cache
80 __ incrementl(ExternalAddress((address)&counter));
81 }
82 #define inc_counter_np(counter) \
83 BLOCK_COMMENT("inc_counter " #counter); \
84 inc_counter_np_(counter);
85 #endif
86
87 // Call stubs are used to call Java from C
88 //
89 // Linux Arguments:
90 // c_rarg0: call wrapper address address
91 // c_rarg1: result address
92 // c_rarg2: result type BasicType
93 // c_rarg3: method Method*
94 // c_rarg4: (interpreter) entry point address
95 // c_rarg5: parameters intptr_t*
96 // 16(rbp): parameter size (in words) int
97 // 24(rbp): thread Thread*
98 //
99 // [ return_from_Java ] <--- rsp
100 // [ argument word n ]
101 // ...
102 // -12 [ argument word 1 ]
103 // -11 [ saved r15 ] <--- rsp_after_call
104 // -10 [ saved r14 ]
105 // -9 [ saved r13 ]
106 // -8 [ saved r12 ]
107 // -7 [ saved rbx ]
108 // -6 [ call wrapper ]
109 // -5 [ result ]
110 // -4 [ result type ]
111 // -3 [ method ]
112 // -2 [ entry point ]
113 // -1 [ parameters ]
114 // 0 [ saved rbp ] <--- rbp
115 // 1 [ return address ]
116 // 2 [ parameter size ]
117 // 3 [ thread ]
118 //
119 // Windows Arguments:
120 // c_rarg0: call wrapper address address
121 // c_rarg1: result address
122 // c_rarg2: result type BasicType
123 // c_rarg3: method Method*
124 // 48(rbp): (interpreter) entry point address
125 // 56(rbp): parameters intptr_t*
126 // 64(rbp): parameter size (in words) int
127 // 72(rbp): thread Thread*
128 //
129 // [ return_from_Java ] <--- rsp
130 // [ argument word n ]
131 // ...
132 // -60 [ argument word 1 ]
133 // -59 [ saved xmm31 ] <--- rsp after_call
134 // [ saved xmm16-xmm30 ] (EVEX enabled, else the space is blank)
135 // -27 [ saved xmm15 ]
136 // [ saved xmm7-xmm14 ]
137 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
138 // -7 [ saved r15 ]
139 // -6 [ saved r14 ]
140 // -5 [ saved r13 ]
141 // -4 [ saved r12 ]
142 // -3 [ saved rdi ]
143 // -2 [ saved rsi ]
144 // -1 [ saved rbx ]
145 // 0 [ saved rbp ] <--- rbp
146 // 1 [ return address ]
147 // 2 [ call wrapper ]
148 // 3 [ result ]
149 // 4 [ result type ]
150 // 5 [ method ]
151 // 6 [ entry point ]
152 // 7 [ parameters ]
153 // 8 [ parameter size ]
154 // 9 [ thread ]
155 //
156 // Windows reserves the callers stack space for arguments 1-4.
157 // We spill c_rarg0-c_rarg3 to this space.
158
159 // Call stub stack layout word offsets from rbp
160 enum call_stub_layout {
161 #ifdef _WIN64
162 xmm_save_first = 6, // save from xmm6
163 xmm_save_last = 31, // to xmm31
164 xmm_save_base = -9,
165 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
166 r15_off = -7,
167 r14_off = -6,
168 r13_off = -5,
169 r12_off = -4,
170 rdi_off = -3,
171 rsi_off = -2,
172 rbx_off = -1,
173 rbp_off = 0,
174 retaddr_off = 1,
175 call_wrapper_off = 2,
176 result_off = 3,
177 result_type_off = 4,
178 method_off = 5,
179 entry_point_off = 6,
180 parameters_off = 7,
181 parameter_size_off = 8,
182 thread_off = 9
183 #else
184 rsp_after_call_off = -12,
185 mxcsr_off = rsp_after_call_off,
186 r15_off = -11,
187 r14_off = -10,
188 r13_off = -9,
189 r12_off = -8,
190 rbx_off = -7,
191 call_wrapper_off = -6,
192 result_off = -5,
193 result_type_off = -4,
194 method_off = -3,
195 entry_point_off = -2,
196 parameters_off = -1,
197 rbp_off = 0,
198 retaddr_off = 1,
199 parameter_size_off = 2,
200 thread_off = 3
201 #endif
202 };
203
204 #ifdef _WIN64
205 Address xmm_save(int reg) {
206 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
207 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
208 }
209 #endif
210
211 address generate_call_stub(address& return_address) {
212 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
213 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
214 "adjust this code");
215 StubCodeMark mark(this, "StubRoutines", "call_stub");
216 address start = __ pc();
217
218 // same as in generate_catch_exception()!
219 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
220
221 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
222 const Address result (rbp, result_off * wordSize);
223 const Address result_type (rbp, result_type_off * wordSize);
224 const Address method (rbp, method_off * wordSize);
225 const Address entry_point (rbp, entry_point_off * wordSize);
226 const Address parameters (rbp, parameters_off * wordSize);
227 const Address parameter_size(rbp, parameter_size_off * wordSize);
228
229 // same as in generate_catch_exception()!
230 const Address thread (rbp, thread_off * wordSize);
231
232 const Address r15_save(rbp, r15_off * wordSize);
233 const Address r14_save(rbp, r14_off * wordSize);
234 const Address r13_save(rbp, r13_off * wordSize);
235 const Address r12_save(rbp, r12_off * wordSize);
236 const Address rbx_save(rbp, rbx_off * wordSize);
237
238 // stub code
239 __ enter();
240 __ subptr(rsp, -rsp_after_call_off * wordSize);
241
242 // save register parameters
243 #ifndef _WIN64
244 __ movptr(parameters, c_rarg5); // parameters
245 __ movptr(entry_point, c_rarg4); // entry_point
246 #endif
247
248 __ movptr(method, c_rarg3); // method
249 __ movl(result_type, c_rarg2); // result type
250 __ movptr(result, c_rarg1); // result
251 __ movptr(call_wrapper, c_rarg0); // call wrapper
252
253 // save regs belonging to calling function
254 __ movptr(rbx_save, rbx);
255 __ movptr(r12_save, r12);
256 __ movptr(r13_save, r13);
257 __ movptr(r14_save, r14);
258 __ movptr(r15_save, r15);
259
260 #ifdef _WIN64
261 int last_reg = 15;
262 if (UseAVX > 2) {
263 last_reg = 31;
264 }
265 if (VM_Version::supports_evex()) {
266 for (int i = xmm_save_first; i <= last_reg; i++) {
267 __ vextractf32x4(xmm_save(i), as_XMMRegister(i), 0);
268 }
269 } else {
270 for (int i = xmm_save_first; i <= last_reg; i++) {
271 __ movdqu(xmm_save(i), as_XMMRegister(i));
272 }
273 }
274
275 const Address rdi_save(rbp, rdi_off * wordSize);
276 const Address rsi_save(rbp, rsi_off * wordSize);
277
278 __ movptr(rsi_save, rsi);
279 __ movptr(rdi_save, rdi);
280 #else
281 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
282 {
283 Label skip_ldmx;
284 __ stmxcsr(mxcsr_save);
285 __ movl(rax, mxcsr_save);
286 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
287 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
288 __ cmp32(rax, mxcsr_std);
289 __ jcc(Assembler::equal, skip_ldmx);
290 __ ldmxcsr(mxcsr_std);
291 __ bind(skip_ldmx);
292 }
293 #endif
294
295 // Load up thread register
296 __ movptr(r15_thread, thread);
297 __ reinit_heapbase();
298
299 #ifdef ASSERT
300 // make sure we have no pending exceptions
301 {
302 Label L;
303 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
304 __ jcc(Assembler::equal, L);
305 __ stop("StubRoutines::call_stub: entered with pending exception");
306 __ bind(L);
307 }
308 #endif
309
310 // pass parameters if any
311 BLOCK_COMMENT("pass parameters if any");
312 Label parameters_done;
313 __ movl(c_rarg3, parameter_size);
314 __ testl(c_rarg3, c_rarg3);
315 __ jcc(Assembler::zero, parameters_done);
316
317 Label loop;
318 __ movptr(c_rarg2, parameters); // parameter pointer
319 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
320 __ BIND(loop);
321 __ movptr(rax, Address(c_rarg2, 0));// get parameter
322 __ addptr(c_rarg2, wordSize); // advance to next parameter
323 __ decrementl(c_rarg1); // decrement counter
324 __ push(rax); // pass parameter
325 __ jcc(Assembler::notZero, loop);
326
327 // call Java function
328 __ BIND(parameters_done);
329 __ movptr(rbx, method); // get Method*
330 __ movptr(c_rarg1, entry_point); // get entry_point
331 __ mov(r13, rsp); // set sender sp
332 BLOCK_COMMENT("call Java function");
333 __ call(c_rarg1);
334
335 BLOCK_COMMENT("call_stub_return_address:");
336 return_address = __ pc();
337
338 // store result depending on type (everything that is not
339 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
340 __ movptr(c_rarg0, result);
341 Label is_long, is_float, is_double, exit;
342 __ movl(c_rarg1, result_type);
343 __ cmpl(c_rarg1, T_OBJECT);
344 __ jcc(Assembler::equal, is_long);
345 __ cmpl(c_rarg1, T_LONG);
346 __ jcc(Assembler::equal, is_long);
347 __ cmpl(c_rarg1, T_FLOAT);
348 __ jcc(Assembler::equal, is_float);
349 __ cmpl(c_rarg1, T_DOUBLE);
350 __ jcc(Assembler::equal, is_double);
351
352 // handle T_INT case
353 __ movl(Address(c_rarg0, 0), rax);
354
355 __ BIND(exit);
356
357 // pop parameters
358 __ lea(rsp, rsp_after_call);
359
360 #ifdef ASSERT
361 // verify that threads correspond
362 {
363 Label L1, L2, L3;
364 __ cmpptr(r15_thread, thread);
365 __ jcc(Assembler::equal, L1);
366 __ stop("StubRoutines::call_stub: r15_thread is corrupted");
367 __ bind(L1);
368 __ get_thread(rbx);
369 __ cmpptr(r15_thread, thread);
370 __ jcc(Assembler::equal, L2);
371 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
372 __ bind(L2);
373 __ cmpptr(r15_thread, rbx);
374 __ jcc(Assembler::equal, L3);
375 __ stop("StubRoutines::call_stub: threads must correspond");
376 __ bind(L3);
377 }
378 #endif
379
380 // restore regs belonging to calling function
381 #ifdef _WIN64
382 // emit the restores for xmm regs
383 if (VM_Version::supports_evex()) {
384 for (int i = xmm_save_first; i <= last_reg; i++) {
385 __ vinsertf32x4(as_XMMRegister(i), as_XMMRegister(i), xmm_save(i), 0);
386 }
387 } else {
388 for (int i = xmm_save_first; i <= last_reg; i++) {
389 __ movdqu(as_XMMRegister(i), xmm_save(i));
390 }
391 }
392 #endif
393 __ movptr(r15, r15_save);
394 __ movptr(r14, r14_save);
395 __ movptr(r13, r13_save);
396 __ movptr(r12, r12_save);
397 __ movptr(rbx, rbx_save);
398
399 #ifdef _WIN64
400 __ movptr(rdi, rdi_save);
401 __ movptr(rsi, rsi_save);
402 #else
403 __ ldmxcsr(mxcsr_save);
404 #endif
405
406 // restore rsp
407 __ addptr(rsp, -rsp_after_call_off * wordSize);
408
409 // return
410 __ vzeroupper();
411 __ pop(rbp);
412 __ ret(0);
413
414 // handle return types different from T_INT
415 __ BIND(is_long);
416 __ movq(Address(c_rarg0, 0), rax);
417 __ jmp(exit);
418
419 __ BIND(is_float);
420 __ movflt(Address(c_rarg0, 0), xmm0);
421 __ jmp(exit);
422
423 __ BIND(is_double);
424 __ movdbl(Address(c_rarg0, 0), xmm0);
425 __ jmp(exit);
426
427 return start;
428 }
429
430 // Return point for a Java call if there's an exception thrown in
431 // Java code. The exception is caught and transformed into a
432 // pending exception stored in JavaThread that can be tested from
433 // within the VM.
434 //
435 // Note: Usually the parameters are removed by the callee. In case
436 // of an exception crossing an activation frame boundary, that is
437 // not the case if the callee is compiled code => need to setup the
438 // rsp.
439 //
440 // rax: exception oop
441
442 address generate_catch_exception() {
443 StubCodeMark mark(this, "StubRoutines", "catch_exception");
444 address start = __ pc();
445
446 // same as in generate_call_stub():
447 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
448 const Address thread (rbp, thread_off * wordSize);
449
450 #ifdef ASSERT
451 // verify that threads correspond
452 {
453 Label L1, L2, L3;
454 __ cmpptr(r15_thread, thread);
455 __ jcc(Assembler::equal, L1);
456 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
457 __ bind(L1);
458 __ get_thread(rbx);
459 __ cmpptr(r15_thread, thread);
460 __ jcc(Assembler::equal, L2);
461 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
462 __ bind(L2);
463 __ cmpptr(r15_thread, rbx);
464 __ jcc(Assembler::equal, L3);
465 __ stop("StubRoutines::catch_exception: threads must correspond");
466 __ bind(L3);
467 }
468 #endif
469
470 // set pending exception
471 __ verify_oop(rax);
472
473 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
474 __ lea(rscratch1, ExternalAddress((address)__FILE__));
475 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
476 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
477
478 // complete return to VM
479 assert(StubRoutines::_call_stub_return_address != NULL,
480 "_call_stub_return_address must have been generated before");
481 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
482
483 return start;
484 }
485
486 // Continuation point for runtime calls returning with a pending
487 // exception. The pending exception check happened in the runtime
488 // or native call stub. The pending exception in Thread is
489 // converted into a Java-level exception.
490 //
491 // Contract with Java-level exception handlers:
492 // rax: exception
493 // rdx: throwing pc
494 //
495 // NOTE: At entry of this stub, exception-pc must be on stack !!
496
497 address generate_forward_exception() {
498 StubCodeMark mark(this, "StubRoutines", "forward exception");
499 address start = __ pc();
500
501 // Upon entry, the sp points to the return address returning into
502 // Java (interpreted or compiled) code; i.e., the return address
503 // becomes the throwing pc.
504 //
505 // Arguments pushed before the runtime call are still on the stack
506 // but the exception handler will reset the stack pointer ->
507 // ignore them. A potential result in registers can be ignored as
508 // well.
509
510 #ifdef ASSERT
511 // make sure this code is only executed if there is a pending exception
512 {
513 Label L;
514 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
515 __ jcc(Assembler::notEqual, L);
516 __ stop("StubRoutines::forward exception: no pending exception (1)");
517 __ bind(L);
518 }
519 #endif
520
521 // compute exception handler into rbx
522 __ movptr(c_rarg0, Address(rsp, 0));
523 BLOCK_COMMENT("call exception_handler_for_return_address");
524 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
525 SharedRuntime::exception_handler_for_return_address),
526 r15_thread, c_rarg0);
527 __ mov(rbx, rax);
528
529 // setup rax & rdx, remove return address & clear pending exception
530 __ pop(rdx);
531 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
532 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
533
534 #ifdef ASSERT
535 // make sure exception is set
536 {
537 Label L;
538 __ testptr(rax, rax);
539 __ jcc(Assembler::notEqual, L);
540 __ stop("StubRoutines::forward exception: no pending exception (2)");
541 __ bind(L);
542 }
543 #endif
544
545 // continue at exception handler (return address removed)
546 // rax: exception
547 // rbx: exception handler
548 // rdx: throwing pc
549 __ verify_oop(rax);
550 __ jmp(rbx);
551
552 return start;
553 }
554
555 // Support for jint atomic::xchg(volatile jint* dest, jint exchange_value)
556 //
557 // Arguments :
558 // c_rarg0: exchange_value
559 // c_rarg0: dest
560 //
561 // Result:
562 // *dest <- ex, return (orig *dest)
563 address generate_atomic_xchg() {
564 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
565 address start = __ pc();
566
567 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
568 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
569 __ ret(0);
570
571 return start;
572 }
573
574 // Support for intptr_t atomic::xchg_long(volatile jlong* dest, jlong exchange_value)
575 //
576 // Arguments :
577 // c_rarg0: exchange_value
578 // c_rarg1: dest
579 //
580 // Result:
581 // *dest <- ex, return (orig *dest)
582 address generate_atomic_xchg_long() {
583 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_long");
584 address start = __ pc();
585
586 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
587 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
588 __ ret(0);
589
590 return start;
591 }
592
593 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
594 // jint compare_value)
595 //
596 // Arguments :
597 // c_rarg0: exchange_value
598 // c_rarg1: dest
599 // c_rarg2: compare_value
600 //
601 // Result:
602 // if ( compare_value == *dest ) {
603 // *dest = exchange_value
604 // return compare_value;
605 // else
606 // return *dest;
607 address generate_atomic_cmpxchg() {
608 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
609 address start = __ pc();
610
611 __ movl(rax, c_rarg2);
612 __ lock();
613 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
614 __ ret(0);
615
616 return start;
617 }
618
619 // Support for int8_t atomic::atomic_cmpxchg(int8_t exchange_value, volatile int8_t* dest,
620 // int8_t compare_value)
621 //
622 // Arguments :
623 // c_rarg0: exchange_value
624 // c_rarg1: dest
625 // c_rarg2: compare_value
626 //
627 // Result:
628 // if ( compare_value == *dest ) {
629 // *dest = exchange_value
630 // return compare_value;
631 // else
632 // return *dest;
633 address generate_atomic_cmpxchg_byte() {
634 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_byte");
635 address start = __ pc();
636
637 __ movsbq(rax, c_rarg2);
638 __ lock();
639 __ cmpxchgb(c_rarg0, Address(c_rarg1, 0));
640 __ ret(0);
641
642 return start;
643 }
644
645 // Support for int64_t atomic::atomic_cmpxchg(int64_t exchange_value,
646 // volatile int64_t* dest,
647 // int64_t compare_value)
648 // Arguments :
649 // c_rarg0: exchange_value
650 // c_rarg1: dest
651 // c_rarg2: compare_value
652 //
653 // Result:
654 // if ( compare_value == *dest ) {
655 // *dest = exchange_value
656 // return compare_value;
657 // else
658 // return *dest;
659 address generate_atomic_cmpxchg_long() {
660 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
661 address start = __ pc();
662
663 __ movq(rax, c_rarg2);
664 __ lock();
665 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
666 __ ret(0);
667
668 return start;
669 }
670
671 // Support for jint atomic::add(jint add_value, volatile jint* dest)
672 //
673 // Arguments :
674 // c_rarg0: add_value
675 // c_rarg1: dest
676 //
677 // Result:
678 // *dest += add_value
679 // return *dest;
680 address generate_atomic_add() {
681 StubCodeMark mark(this, "StubRoutines", "atomic_add");
682 address start = __ pc();
683
684 __ movl(rax, c_rarg0);
685 __ lock();
686 __ xaddl(Address(c_rarg1, 0), c_rarg0);
687 __ addl(rax, c_rarg0);
688 __ ret(0);
689
690 return start;
691 }
692
693 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
694 //
695 // Arguments :
696 // c_rarg0: add_value
697 // c_rarg1: dest
698 //
699 // Result:
700 // *dest += add_value
701 // return *dest;
702 address generate_atomic_add_long() {
703 StubCodeMark mark(this, "StubRoutines", "atomic_add_long");
704 address start = __ pc();
705
706 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
707 __ lock();
708 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
709 __ addptr(rax, c_rarg0);
710 __ ret(0);
711
712 return start;
713 }
714
715 // Support for intptr_t OrderAccess::fence()
716 //
717 // Arguments :
718 //
719 // Result:
720 address generate_orderaccess_fence() {
721 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
722 address start = __ pc();
723 __ membar(Assembler::StoreLoad);
724 __ ret(0);
725
726 return start;
727 }
728
729 // Support for intptr_t get_previous_fp()
730 //
731 // This routine is used to find the previous frame pointer for the
732 // caller (current_frame_guess). This is used as part of debugging
733 // ps() is seemingly lost trying to find frames.
734 // This code assumes that caller current_frame_guess) has a frame.
735 address generate_get_previous_fp() {
736 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
737 const Address old_fp(rbp, 0);
738 const Address older_fp(rax, 0);
739 address start = __ pc();
740
741 __ enter();
742 __ movptr(rax, old_fp); // callers fp
743 __ movptr(rax, older_fp); // the frame for ps()
744 __ pop(rbp);
745 __ ret(0);
746
747 return start;
748 }
749
750 // Support for intptr_t get_previous_sp()
751 //
752 // This routine is used to find the previous stack pointer for the
753 // caller.
754 address generate_get_previous_sp() {
755 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
756 address start = __ pc();
757
758 __ movptr(rax, rsp);
759 __ addptr(rax, 8); // return address is at the top of the stack.
760 __ ret(0);
761
762 return start;
763 }
764
765 //----------------------------------------------------------------------------------------------------
766 // Support for void verify_mxcsr()
767 //
768 // This routine is used with -Xcheck:jni to verify that native
769 // JNI code does not return to Java code without restoring the
770 // MXCSR register to our expected state.
771
772 address generate_verify_mxcsr() {
773 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
774 address start = __ pc();
775
776 const Address mxcsr_save(rsp, 0);
777
778 if (CheckJNICalls) {
779 Label ok_ret;
780 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
781 __ push(rax);
782 __ subptr(rsp, wordSize); // allocate a temp location
783 __ stmxcsr(mxcsr_save);
784 __ movl(rax, mxcsr_save);
785 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
786 __ cmp32(rax, mxcsr_std);
787 __ jcc(Assembler::equal, ok_ret);
788
789 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
790
791 __ ldmxcsr(mxcsr_std);
792
793 __ bind(ok_ret);
794 __ addptr(rsp, wordSize);
795 __ pop(rax);
796 }
797
798 __ ret(0);
799
800 return start;
801 }
802
803 address generate_f2i_fixup() {
804 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
805 Address inout(rsp, 5 * wordSize); // return address + 4 saves
806
807 address start = __ pc();
808
809 Label L;
810
811 __ push(rax);
812 __ push(c_rarg3);
813 __ push(c_rarg2);
814 __ push(c_rarg1);
815
816 __ movl(rax, 0x7f800000);
817 __ xorl(c_rarg3, c_rarg3);
818 __ movl(c_rarg2, inout);
819 __ movl(c_rarg1, c_rarg2);
820 __ andl(c_rarg1, 0x7fffffff);
821 __ cmpl(rax, c_rarg1); // NaN? -> 0
822 __ jcc(Assembler::negative, L);
823 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
824 __ movl(c_rarg3, 0x80000000);
825 __ movl(rax, 0x7fffffff);
826 __ cmovl(Assembler::positive, c_rarg3, rax);
827
828 __ bind(L);
829 __ movptr(inout, c_rarg3);
830
831 __ pop(c_rarg1);
832 __ pop(c_rarg2);
833 __ pop(c_rarg3);
834 __ pop(rax);
835
836 __ ret(0);
837
838 return start;
839 }
840
841 address generate_f2l_fixup() {
842 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
843 Address inout(rsp, 5 * wordSize); // return address + 4 saves
844 address start = __ pc();
845
846 Label L;
847
848 __ push(rax);
849 __ push(c_rarg3);
850 __ push(c_rarg2);
851 __ push(c_rarg1);
852
853 __ movl(rax, 0x7f800000);
854 __ xorl(c_rarg3, c_rarg3);
855 __ movl(c_rarg2, inout);
856 __ movl(c_rarg1, c_rarg2);
857 __ andl(c_rarg1, 0x7fffffff);
858 __ cmpl(rax, c_rarg1); // NaN? -> 0
859 __ jcc(Assembler::negative, L);
860 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
861 __ mov64(c_rarg3, 0x8000000000000000);
862 __ mov64(rax, 0x7fffffffffffffff);
863 __ cmov(Assembler::positive, c_rarg3, rax);
864
865 __ bind(L);
866 __ movptr(inout, c_rarg3);
867
868 __ pop(c_rarg1);
869 __ pop(c_rarg2);
870 __ pop(c_rarg3);
871 __ pop(rax);
872
873 __ ret(0);
874
875 return start;
876 }
877
878 address generate_d2i_fixup() {
879 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
880 Address inout(rsp, 6 * wordSize); // return address + 5 saves
881
882 address start = __ pc();
883
884 Label L;
885
886 __ push(rax);
887 __ push(c_rarg3);
888 __ push(c_rarg2);
889 __ push(c_rarg1);
890 __ push(c_rarg0);
891
892 __ movl(rax, 0x7ff00000);
893 __ movq(c_rarg2, inout);
894 __ movl(c_rarg3, c_rarg2);
895 __ mov(c_rarg1, c_rarg2);
896 __ mov(c_rarg0, c_rarg2);
897 __ negl(c_rarg3);
898 __ shrptr(c_rarg1, 0x20);
899 __ orl(c_rarg3, c_rarg2);
900 __ andl(c_rarg1, 0x7fffffff);
901 __ xorl(c_rarg2, c_rarg2);
902 __ shrl(c_rarg3, 0x1f);
903 __ orl(c_rarg1, c_rarg3);
904 __ cmpl(rax, c_rarg1);
905 __ jcc(Assembler::negative, L); // NaN -> 0
906 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
907 __ movl(c_rarg2, 0x80000000);
908 __ movl(rax, 0x7fffffff);
909 __ cmov(Assembler::positive, c_rarg2, rax);
910
911 __ bind(L);
912 __ movptr(inout, c_rarg2);
913
914 __ pop(c_rarg0);
915 __ pop(c_rarg1);
916 __ pop(c_rarg2);
917 __ pop(c_rarg3);
918 __ pop(rax);
919
920 __ ret(0);
921
922 return start;
923 }
924
925 address generate_d2l_fixup() {
926 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
927 Address inout(rsp, 6 * wordSize); // return address + 5 saves
928
929 address start = __ pc();
930
931 Label L;
932
933 __ push(rax);
934 __ push(c_rarg3);
935 __ push(c_rarg2);
936 __ push(c_rarg1);
937 __ push(c_rarg0);
938
939 __ movl(rax, 0x7ff00000);
940 __ movq(c_rarg2, inout);
941 __ movl(c_rarg3, c_rarg2);
942 __ mov(c_rarg1, c_rarg2);
943 __ mov(c_rarg0, c_rarg2);
944 __ negl(c_rarg3);
945 __ shrptr(c_rarg1, 0x20);
946 __ orl(c_rarg3, c_rarg2);
947 __ andl(c_rarg1, 0x7fffffff);
948 __ xorl(c_rarg2, c_rarg2);
949 __ shrl(c_rarg3, 0x1f);
950 __ orl(c_rarg1, c_rarg3);
951 __ cmpl(rax, c_rarg1);
952 __ jcc(Assembler::negative, L); // NaN -> 0
953 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
954 __ mov64(c_rarg2, 0x8000000000000000);
955 __ mov64(rax, 0x7fffffffffffffff);
956 __ cmovq(Assembler::positive, c_rarg2, rax);
957
958 __ bind(L);
959 __ movq(inout, c_rarg2);
960
961 __ pop(c_rarg0);
962 __ pop(c_rarg1);
963 __ pop(c_rarg2);
964 __ pop(c_rarg3);
965 __ pop(rax);
966
967 __ ret(0);
968
969 return start;
970 }
971
972 address generate_fp_mask(const char *stub_name, int64_t mask) {
973 __ align(CodeEntryAlignment);
974 StubCodeMark mark(this, "StubRoutines", stub_name);
975 address start = __ pc();
976
977 __ emit_data64( mask, relocInfo::none );
978 __ emit_data64( mask, relocInfo::none );
979
980 return start;
981 }
982
983 address generate_vector_mask(const char *stub_name, int64_t mask) {
984 __ align(CodeEntryAlignment);
985 StubCodeMark mark(this, "StubRoutines", stub_name);
986 address start = __ pc();
987
988 __ emit_data64(mask, relocInfo::none);
989 __ emit_data64(mask, relocInfo::none);
990 __ emit_data64(mask, relocInfo::none);
991 __ emit_data64(mask, relocInfo::none);
992 __ emit_data64(mask, relocInfo::none);
993 __ emit_data64(mask, relocInfo::none);
994 __ emit_data64(mask, relocInfo::none);
995 __ emit_data64(mask, relocInfo::none);
996
997 return start;
998 }
999
1000 address generate_vector_byte_perm_mask(const char *stub_name) {
1001 __ align(CodeEntryAlignment);
1002 StubCodeMark mark(this, "StubRoutines", stub_name);
1003 address start = __ pc();
1004
1005 __ emit_data64(0x0000000000000001, relocInfo::none);
1006 __ emit_data64(0x0000000000000003, relocInfo::none);
1007 __ emit_data64(0x0000000000000005, relocInfo::none);
1008 __ emit_data64(0x0000000000000007, relocInfo::none);
1009 __ emit_data64(0x0000000000000000, relocInfo::none);
1010 __ emit_data64(0x0000000000000002, relocInfo::none);
1011 __ emit_data64(0x0000000000000004, relocInfo::none);
1012 __ emit_data64(0x0000000000000006, relocInfo::none);
1013
1014 return start;
1015 }
1016
1017 // Non-destructive plausibility checks for oops
1018 //
1019 // Arguments:
1020 // all args on stack!
1021 //
1022 // Stack after saving c_rarg3:
1023 // [tos + 0]: saved c_rarg3
1024 // [tos + 1]: saved c_rarg2
1025 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1026 // [tos + 3]: saved flags
1027 // [tos + 4]: return address
1028 // * [tos + 5]: error message (char*)
1029 // * [tos + 6]: object to verify (oop)
1030 // * [tos + 7]: saved rax - saved by caller and bashed
1031 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1032 // * = popped on exit
1033 address generate_verify_oop() {
1034 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1035 address start = __ pc();
1036
1037 Label exit, error;
1038
1039 __ pushf();
1040 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1041
1042 __ push(r12);
1043
1044 // save c_rarg2 and c_rarg3
1045 __ push(c_rarg2);
1046 __ push(c_rarg3);
1047
1048 enum {
1049 // After previous pushes.
1050 oop_to_verify = 6 * wordSize,
1051 saved_rax = 7 * wordSize,
1052 saved_r10 = 8 * wordSize,
1053
1054 // Before the call to MacroAssembler::debug(), see below.
1055 return_addr = 16 * wordSize,
1056 error_msg = 17 * wordSize
1057 };
1058
1059 // get object
1060 __ movptr(rax, Address(rsp, oop_to_verify));
1061
1062 // make sure object is 'reasonable'
1063 __ testptr(rax, rax);
1064 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1065
1066 #if INCLUDE_ZGC
1067 if (UseZGC) {
1068 // Check if metadata bits indicate a bad oop
1069 __ testptr(rax, Address(r15_thread, ZThreadLocalData::address_bad_mask_offset()));
1070 __ jcc(Assembler::notZero, error);
1071 }
1072 #endif
1073
1074 // Check if the oop is in the right area of memory
1075 __ movptr(c_rarg2, rax);
1076 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1077 __ andptr(c_rarg2, c_rarg3);
1078 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1079 __ cmpptr(c_rarg2, c_rarg3);
1080 __ jcc(Assembler::notZero, error);
1081
1082 // set r12 to heapbase for load_klass()
1083 __ reinit_heapbase();
1084
1085 // make sure klass is 'reasonable', which is not zero.
1086 __ load_klass(rax, rax); // get klass
1087 __ testptr(rax, rax);
1088 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1089
1090 // return if everything seems ok
1091 __ bind(exit);
1092 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1093 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1094 __ pop(c_rarg3); // restore c_rarg3
1095 __ pop(c_rarg2); // restore c_rarg2
1096 __ pop(r12); // restore r12
1097 __ popf(); // restore flags
1098 __ ret(4 * wordSize); // pop caller saved stuff
1099
1100 // handle errors
1101 __ bind(error);
1102 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1103 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1104 __ pop(c_rarg3); // get saved c_rarg3 back
1105 __ pop(c_rarg2); // get saved c_rarg2 back
1106 __ pop(r12); // get saved r12 back
1107 __ popf(); // get saved flags off stack --
1108 // will be ignored
1109
1110 __ pusha(); // push registers
1111 // (rip is already
1112 // already pushed)
1113 // debug(char* msg, int64_t pc, int64_t regs[])
1114 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1115 // pushed all the registers, so now the stack looks like:
1116 // [tos + 0] 16 saved registers
1117 // [tos + 16] return address
1118 // * [tos + 17] error message (char*)
1119 // * [tos + 18] object to verify (oop)
1120 // * [tos + 19] saved rax - saved by caller and bashed
1121 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1122 // * = popped on exit
1123
1124 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1125 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1126 __ movq(c_rarg2, rsp); // pass address of regs on stack
1127 __ mov(r12, rsp); // remember rsp
1128 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1129 __ andptr(rsp, -16); // align stack as required by ABI
1130 BLOCK_COMMENT("call MacroAssembler::debug");
1131 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1132 __ hlt();
1133 return start;
1134 }
1135
1136 //
1137 // Verify that a register contains clean 32-bits positive value
1138 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1139 //
1140 // Input:
1141 // Rint - 32-bits value
1142 // Rtmp - scratch
1143 //
1144 void assert_clean_int(Register Rint, Register Rtmp) {
1145 #ifdef ASSERT
1146 Label L;
1147 assert_different_registers(Rtmp, Rint);
1148 __ movslq(Rtmp, Rint);
1149 __ cmpq(Rtmp, Rint);
1150 __ jcc(Assembler::equal, L);
1151 __ stop("high 32-bits of int value are not 0");
1152 __ bind(L);
1153 #endif
1154 }
1155
1156 // Generate overlap test for array copy stubs
1157 //
1158 // Input:
1159 // c_rarg0 - from
1160 // c_rarg1 - to
1161 // c_rarg2 - element count
1162 //
1163 // Output:
1164 // rax - &from[element count - 1]
1165 //
1166 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1167 assert(no_overlap_target != NULL, "must be generated");
1168 array_overlap_test(no_overlap_target, NULL, sf);
1169 }
1170 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1171 array_overlap_test(NULL, &L_no_overlap, sf);
1172 }
1173 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1174 const Register from = c_rarg0;
1175 const Register to = c_rarg1;
1176 const Register count = c_rarg2;
1177 const Register end_from = rax;
1178
1179 __ cmpptr(to, from);
1180 __ lea(end_from, Address(from, count, sf, 0));
1181 if (NOLp == NULL) {
1182 ExternalAddress no_overlap(no_overlap_target);
1183 __ jump_cc(Assembler::belowEqual, no_overlap);
1184 __ cmpptr(to, end_from);
1185 __ jump_cc(Assembler::aboveEqual, no_overlap);
1186 } else {
1187 __ jcc(Assembler::belowEqual, (*NOLp));
1188 __ cmpptr(to, end_from);
1189 __ jcc(Assembler::aboveEqual, (*NOLp));
1190 }
1191 }
1192
1193 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1194 //
1195 // Outputs:
1196 // rdi - rcx
1197 // rsi - rdx
1198 // rdx - r8
1199 // rcx - r9
1200 //
1201 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1202 // are non-volatile. r9 and r10 should not be used by the caller.
1203 //
1204 DEBUG_ONLY(bool regs_in_thread;)
1205
1206 void setup_arg_regs(int nargs = 3) {
1207 const Register saved_rdi = r9;
1208 const Register saved_rsi = r10;
1209 assert(nargs == 3 || nargs == 4, "else fix");
1210 #ifdef _WIN64
1211 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1212 "unexpected argument registers");
1213 if (nargs >= 4)
1214 __ mov(rax, r9); // r9 is also saved_rdi
1215 __ movptr(saved_rdi, rdi);
1216 __ movptr(saved_rsi, rsi);
1217 __ mov(rdi, rcx); // c_rarg0
1218 __ mov(rsi, rdx); // c_rarg1
1219 __ mov(rdx, r8); // c_rarg2
1220 if (nargs >= 4)
1221 __ mov(rcx, rax); // c_rarg3 (via rax)
1222 #else
1223 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1224 "unexpected argument registers");
1225 #endif
1226 DEBUG_ONLY(regs_in_thread = false;)
1227 }
1228
1229 void restore_arg_regs() {
1230 assert(!regs_in_thread, "wrong call to restore_arg_regs");
1231 const Register saved_rdi = r9;
1232 const Register saved_rsi = r10;
1233 #ifdef _WIN64
1234 __ movptr(rdi, saved_rdi);
1235 __ movptr(rsi, saved_rsi);
1236 #endif
1237 }
1238
1239 // This is used in places where r10 is a scratch register, and can
1240 // be adapted if r9 is needed also.
1241 void setup_arg_regs_using_thread() {
1242 const Register saved_r15 = r9;
1243 #ifdef _WIN64
1244 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1245 __ get_thread(r15_thread);
1246 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1247 "unexpected argument registers");
1248 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1249 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1250
1251 __ mov(rdi, rcx); // c_rarg0
1252 __ mov(rsi, rdx); // c_rarg1
1253 __ mov(rdx, r8); // c_rarg2
1254 #else
1255 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1256 "unexpected argument registers");
1257 #endif
1258 DEBUG_ONLY(regs_in_thread = true;)
1259 }
1260
1261 void restore_arg_regs_using_thread() {
1262 assert(regs_in_thread, "wrong call to restore_arg_regs");
1263 const Register saved_r15 = r9;
1264 #ifdef _WIN64
1265 __ get_thread(r15_thread);
1266 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1267 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1268 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1269 #endif
1270 }
1271
1272 // Copy big chunks forward
1273 //
1274 // Inputs:
1275 // end_from - source arrays end address
1276 // end_to - destination array end address
1277 // qword_count - 64-bits element count, negative
1278 // to - scratch
1279 // L_copy_bytes - entry label
1280 // L_copy_8_bytes - exit label
1281 //
1282 void copy_bytes_forward(Register end_from, Register end_to,
1283 Register qword_count, Register to,
1284 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1285 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1286 Label L_loop;
1287 __ align(OptoLoopAlignment);
1288 if (UseUnalignedLoadStores) {
1289 Label L_end;
1290 // Copy 64-bytes per iteration
1291 if (UseAVX > 2) {
1292 Label L_loop_avx512, L_loop_avx2, L_32_byte_head, L_above_threshold, L_below_threshold;
1293
1294 __ BIND(L_copy_bytes);
1295 __ cmpptr(qword_count, (-1 * AVX3Threshold / 8));
1296 __ jccb(Assembler::less, L_above_threshold);
1297 __ jmpb(L_below_threshold);
1298
1299 __ bind(L_loop_avx512);
1300 __ evmovdqul(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1301 __ evmovdqul(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1302 __ bind(L_above_threshold);
1303 __ addptr(qword_count, 8);
1304 __ jcc(Assembler::lessEqual, L_loop_avx512);
1305 __ jmpb(L_32_byte_head);
1306
1307 __ bind(L_loop_avx2);
1308 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1309 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1310 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1311 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1312 __ bind(L_below_threshold);
1313 __ addptr(qword_count, 8);
1314 __ jcc(Assembler::lessEqual, L_loop_avx2);
1315
1316 __ bind(L_32_byte_head);
1317 __ subptr(qword_count, 4); // sub(8) and add(4)
1318 __ jccb(Assembler::greater, L_end);
1319 } else {
1320 __ BIND(L_loop);
1321 if (UseAVX == 2) {
1322 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1323 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1324 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1325 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1326 } else {
1327 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1328 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1329 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1330 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1331 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1332 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1333 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1334 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1335 }
1336
1337 __ BIND(L_copy_bytes);
1338 __ addptr(qword_count, 8);
1339 __ jcc(Assembler::lessEqual, L_loop);
1340 __ subptr(qword_count, 4); // sub(8) and add(4)
1341 __ jccb(Assembler::greater, L_end);
1342 }
1343 // Copy trailing 32 bytes
1344 if (UseAVX >= 2) {
1345 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1346 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1347 } else {
1348 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1349 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1350 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1351 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1352 }
1353 __ addptr(qword_count, 4);
1354 __ BIND(L_end);
1355 if (UseAVX >= 2) {
1356 // clean upper bits of YMM registers
1357 __ vpxor(xmm0, xmm0);
1358 __ vpxor(xmm1, xmm1);
1359 }
1360 } else {
1361 // Copy 32-bytes per iteration
1362 __ BIND(L_loop);
1363 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1364 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1365 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1366 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1367 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1368 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1369 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1370 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1371
1372 __ BIND(L_copy_bytes);
1373 __ addptr(qword_count, 4);
1374 __ jcc(Assembler::lessEqual, L_loop);
1375 }
1376 __ subptr(qword_count, 4);
1377 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1378 }
1379
1380 // Copy big chunks backward
1381 //
1382 // Inputs:
1383 // from - source arrays address
1384 // dest - destination array address
1385 // qword_count - 64-bits element count
1386 // to - scratch
1387 // L_copy_bytes - entry label
1388 // L_copy_8_bytes - exit label
1389 //
1390 void copy_bytes_backward(Register from, Register dest,
1391 Register qword_count, Register to,
1392 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1393 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1394 Label L_loop;
1395 __ align(OptoLoopAlignment);
1396 if (UseUnalignedLoadStores) {
1397 Label L_end;
1398 // Copy 64-bytes per iteration
1399 if (UseAVX > 2) {
1400 Label L_loop_avx512, L_loop_avx2, L_32_byte_head, L_above_threshold, L_below_threshold;
1401
1402 __ BIND(L_copy_bytes);
1403 __ cmpptr(qword_count, (AVX3Threshold / 8));
1404 __ jccb(Assembler::greater, L_above_threshold);
1405 __ jmpb(L_below_threshold);
1406
1407 __ BIND(L_loop_avx512);
1408 __ evmovdqul(xmm0, Address(from, qword_count, Address::times_8, 0), Assembler::AVX_512bit);
1409 __ evmovdqul(Address(dest, qword_count, Address::times_8, 0), xmm0, Assembler::AVX_512bit);
1410 __ bind(L_above_threshold);
1411 __ subptr(qword_count, 8);
1412 __ jcc(Assembler::greaterEqual, L_loop_avx512);
1413 __ jmpb(L_32_byte_head);
1414
1415 __ bind(L_loop_avx2);
1416 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1417 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1418 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1419 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1420 __ bind(L_below_threshold);
1421 __ subptr(qword_count, 8);
1422 __ jcc(Assembler::greaterEqual, L_loop_avx2);
1423
1424 __ bind(L_32_byte_head);
1425 __ addptr(qword_count, 4); // add(8) and sub(4)
1426 __ jccb(Assembler::less, L_end);
1427 } else {
1428 __ BIND(L_loop);
1429 if (UseAVX == 2) {
1430 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1431 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1432 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1433 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1434 } else {
1435 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1436 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1437 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1438 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1439 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1440 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1441 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1442 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1443 }
1444
1445 __ BIND(L_copy_bytes);
1446 __ subptr(qword_count, 8);
1447 __ jcc(Assembler::greaterEqual, L_loop);
1448
1449 __ addptr(qword_count, 4); // add(8) and sub(4)
1450 __ jccb(Assembler::less, L_end);
1451 }
1452 // Copy trailing 32 bytes
1453 if (UseAVX >= 2) {
1454 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1455 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1456 } else {
1457 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1458 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1459 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1460 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1461 }
1462 __ subptr(qword_count, 4);
1463 __ BIND(L_end);
1464 if (UseAVX >= 2) {
1465 // clean upper bits of YMM registers
1466 __ vpxor(xmm0, xmm0);
1467 __ vpxor(xmm1, xmm1);
1468 }
1469 } else {
1470 // Copy 32-bytes per iteration
1471 __ BIND(L_loop);
1472 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1473 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1474 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1475 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1476 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1477 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1478 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1479 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1480
1481 __ BIND(L_copy_bytes);
1482 __ subptr(qword_count, 4);
1483 __ jcc(Assembler::greaterEqual, L_loop);
1484 }
1485 __ addptr(qword_count, 4);
1486 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1487 }
1488
1489 // Arguments:
1490 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1491 // ignored
1492 // name - stub name string
1493 //
1494 // Inputs:
1495 // c_rarg0 - source array address
1496 // c_rarg1 - destination array address
1497 // c_rarg2 - element count, treated as ssize_t, can be zero
1498 //
1499 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1500 // we let the hardware handle it. The one to eight bytes within words,
1501 // dwords or qwords that span cache line boundaries will still be loaded
1502 // and stored atomically.
1503 //
1504 // Side Effects:
1505 // disjoint_byte_copy_entry is set to the no-overlap entry point
1506 // used by generate_conjoint_byte_copy().
1507 //
1508 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1509 __ align(CodeEntryAlignment);
1510 StubCodeMark mark(this, "StubRoutines", name);
1511 address start = __ pc();
1512
1513 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1514 Label L_copy_byte, L_exit;
1515 const Register from = rdi; // source array address
1516 const Register to = rsi; // destination array address
1517 const Register count = rdx; // elements count
1518 const Register byte_count = rcx;
1519 const Register qword_count = count;
1520 const Register end_from = from; // source array end address
1521 const Register end_to = to; // destination array end address
1522 // End pointers are inclusive, and if count is not zero they point
1523 // to the last unit copied: end_to[0] := end_from[0]
1524
1525 __ enter(); // required for proper stackwalking of RuntimeStub frame
1526 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1527
1528 if (entry != NULL) {
1529 *entry = __ pc();
1530 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1531 BLOCK_COMMENT("Entry:");
1532 }
1533
1534 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1535 // r9 and r10 may be used to save non-volatile registers
1536
1537 {
1538 // UnsafeCopyMemory page error: continue after ucm
1539 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1540 // 'from', 'to' and 'count' are now valid
1541 __ movptr(byte_count, count);
1542 __ shrptr(count, 3); // count => qword_count
1543
1544 // Copy from low to high addresses. Use 'to' as scratch.
1545 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1546 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1547 __ negptr(qword_count); // make the count negative
1548 __ jmp(L_copy_bytes);
1549
1550 // Copy trailing qwords
1551 __ BIND(L_copy_8_bytes);
1552 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1553 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1554 __ increment(qword_count);
1555 __ jcc(Assembler::notZero, L_copy_8_bytes);
1556
1557 // Check for and copy trailing dword
1558 __ BIND(L_copy_4_bytes);
1559 __ testl(byte_count, 4);
1560 __ jccb(Assembler::zero, L_copy_2_bytes);
1561 __ movl(rax, Address(end_from, 8));
1562 __ movl(Address(end_to, 8), rax);
1563
1564 __ addptr(end_from, 4);
1565 __ addptr(end_to, 4);
1566
1567 // Check for and copy trailing word
1568 __ BIND(L_copy_2_bytes);
1569 __ testl(byte_count, 2);
1570 __ jccb(Assembler::zero, L_copy_byte);
1571 __ movw(rax, Address(end_from, 8));
1572 __ movw(Address(end_to, 8), rax);
1573
1574 __ addptr(end_from, 2);
1575 __ addptr(end_to, 2);
1576
1577 // Check for and copy trailing byte
1578 __ BIND(L_copy_byte);
1579 __ testl(byte_count, 1);
1580 __ jccb(Assembler::zero, L_exit);
1581 __ movb(rax, Address(end_from, 8));
1582 __ movb(Address(end_to, 8), rax);
1583 }
1584 __ BIND(L_exit);
1585 address ucme_exit_pc = __ pc();
1586 restore_arg_regs();
1587 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1588 __ xorptr(rax, rax); // return 0
1589 __ vzeroupper();
1590 __ leave(); // required for proper stackwalking of RuntimeStub frame
1591 __ ret(0);
1592
1593 {
1594 UnsafeCopyMemoryMark ucmm(this, !aligned, false, ucme_exit_pc);
1595 // Copy in multi-bytes chunks
1596 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1597 __ jmp(L_copy_4_bytes);
1598 }
1599 return start;
1600 }
1601
1602 // Arguments:
1603 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1604 // ignored
1605 // name - stub name string
1606 //
1607 // Inputs:
1608 // c_rarg0 - source array address
1609 // c_rarg1 - destination array address
1610 // c_rarg2 - element count, treated as ssize_t, can be zero
1611 //
1612 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1613 // we let the hardware handle it. The one to eight bytes within words,
1614 // dwords or qwords that span cache line boundaries will still be loaded
1615 // and stored atomically.
1616 //
1617 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1618 address* entry, const char *name) {
1619 __ align(CodeEntryAlignment);
1620 StubCodeMark mark(this, "StubRoutines", name);
1621 address start = __ pc();
1622
1623 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1624 const Register from = rdi; // source array address
1625 const Register to = rsi; // destination array address
1626 const Register count = rdx; // elements count
1627 const Register byte_count = rcx;
1628 const Register qword_count = count;
1629
1630 __ enter(); // required for proper stackwalking of RuntimeStub frame
1631 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1632
1633 if (entry != NULL) {
1634 *entry = __ pc();
1635 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1636 BLOCK_COMMENT("Entry:");
1637 }
1638
1639 array_overlap_test(nooverlap_target, Address::times_1);
1640 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1641 // r9 and r10 may be used to save non-volatile registers
1642
1643 {
1644 // UnsafeCopyMemory page error: continue after ucm
1645 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1646 // 'from', 'to' and 'count' are now valid
1647 __ movptr(byte_count, count);
1648 __ shrptr(count, 3); // count => qword_count
1649
1650 // Copy from high to low addresses.
1651
1652 // Check for and copy trailing byte
1653 __ testl(byte_count, 1);
1654 __ jcc(Assembler::zero, L_copy_2_bytes);
1655 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1656 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1657 __ decrement(byte_count); // Adjust for possible trailing word
1658
1659 // Check for and copy trailing word
1660 __ BIND(L_copy_2_bytes);
1661 __ testl(byte_count, 2);
1662 __ jcc(Assembler::zero, L_copy_4_bytes);
1663 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1664 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1665
1666 // Check for and copy trailing dword
1667 __ BIND(L_copy_4_bytes);
1668 __ testl(byte_count, 4);
1669 __ jcc(Assembler::zero, L_copy_bytes);
1670 __ movl(rax, Address(from, qword_count, Address::times_8));
1671 __ movl(Address(to, qword_count, Address::times_8), rax);
1672 __ jmp(L_copy_bytes);
1673
1674 // Copy trailing qwords
1675 __ BIND(L_copy_8_bytes);
1676 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1677 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1678 __ decrement(qword_count);
1679 __ jcc(Assembler::notZero, L_copy_8_bytes);
1680 }
1681 restore_arg_regs();
1682 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1683 __ xorptr(rax, rax); // return 0
1684 __ vzeroupper();
1685 __ leave(); // required for proper stackwalking of RuntimeStub frame
1686 __ ret(0);
1687
1688 {
1689 // UnsafeCopyMemory page error: continue after ucm
1690 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1691 // Copy in multi-bytes chunks
1692 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1693 }
1694 restore_arg_regs();
1695 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1696 __ xorptr(rax, rax); // return 0
1697 __ vzeroupper();
1698 __ leave(); // required for proper stackwalking of RuntimeStub frame
1699 __ ret(0);
1700
1701 return start;
1702 }
1703
1704 // Arguments:
1705 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1706 // ignored
1707 // name - stub name string
1708 //
1709 // Inputs:
1710 // c_rarg0 - source array address
1711 // c_rarg1 - destination array address
1712 // c_rarg2 - element count, treated as ssize_t, can be zero
1713 //
1714 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1715 // let the hardware handle it. The two or four words within dwords
1716 // or qwords that span cache line boundaries will still be loaded
1717 // and stored atomically.
1718 //
1719 // Side Effects:
1720 // disjoint_short_copy_entry is set to the no-overlap entry point
1721 // used by generate_conjoint_short_copy().
1722 //
1723 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1724 __ align(CodeEntryAlignment);
1725 StubCodeMark mark(this, "StubRoutines", name);
1726 address start = __ pc();
1727
1728 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1729 const Register from = rdi; // source array address
1730 const Register to = rsi; // destination array address
1731 const Register count = rdx; // elements count
1732 const Register word_count = rcx;
1733 const Register qword_count = count;
1734 const Register end_from = from; // source array end address
1735 const Register end_to = to; // destination array end address
1736 // End pointers are inclusive, and if count is not zero they point
1737 // to the last unit copied: end_to[0] := end_from[0]
1738
1739 __ enter(); // required for proper stackwalking of RuntimeStub frame
1740 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1741
1742 if (entry != NULL) {
1743 *entry = __ pc();
1744 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1745 BLOCK_COMMENT("Entry:");
1746 }
1747
1748 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1749 // r9 and r10 may be used to save non-volatile registers
1750
1751 {
1752 // UnsafeCopyMemory page error: continue after ucm
1753 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1754 // 'from', 'to' and 'count' are now valid
1755 __ movptr(word_count, count);
1756 __ shrptr(count, 2); // count => qword_count
1757
1758 // Copy from low to high addresses. Use 'to' as scratch.
1759 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1760 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1761 __ negptr(qword_count);
1762 __ jmp(L_copy_bytes);
1763
1764 // Copy trailing qwords
1765 __ BIND(L_copy_8_bytes);
1766 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1767 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1768 __ increment(qword_count);
1769 __ jcc(Assembler::notZero, L_copy_8_bytes);
1770
1771 // Original 'dest' is trashed, so we can't use it as a
1772 // base register for a possible trailing word copy
1773
1774 // Check for and copy trailing dword
1775 __ BIND(L_copy_4_bytes);
1776 __ testl(word_count, 2);
1777 __ jccb(Assembler::zero, L_copy_2_bytes);
1778 __ movl(rax, Address(end_from, 8));
1779 __ movl(Address(end_to, 8), rax);
1780
1781 __ addptr(end_from, 4);
1782 __ addptr(end_to, 4);
1783
1784 // Check for and copy trailing word
1785 __ BIND(L_copy_2_bytes);
1786 __ testl(word_count, 1);
1787 __ jccb(Assembler::zero, L_exit);
1788 __ movw(rax, Address(end_from, 8));
1789 __ movw(Address(end_to, 8), rax);
1790 }
1791 __ BIND(L_exit);
1792 address ucme_exit_pc = __ pc();
1793 restore_arg_regs();
1794 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1795 __ xorptr(rax, rax); // return 0
1796 __ vzeroupper();
1797 __ leave(); // required for proper stackwalking of RuntimeStub frame
1798 __ ret(0);
1799
1800 {
1801 UnsafeCopyMemoryMark ucmm(this, !aligned, false, ucme_exit_pc);
1802 // Copy in multi-bytes chunks
1803 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1804 __ jmp(L_copy_4_bytes);
1805 }
1806
1807 return start;
1808 }
1809
1810 address generate_fill(BasicType t, bool aligned, const char *name) {
1811 __ align(CodeEntryAlignment);
1812 StubCodeMark mark(this, "StubRoutines", name);
1813 address start = __ pc();
1814
1815 BLOCK_COMMENT("Entry:");
1816
1817 const Register to = c_rarg0; // source array address
1818 const Register value = c_rarg1; // value
1819 const Register count = c_rarg2; // elements count
1820
1821 __ enter(); // required for proper stackwalking of RuntimeStub frame
1822
1823 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1824
1825 __ vzeroupper();
1826 __ leave(); // required for proper stackwalking of RuntimeStub frame
1827 __ ret(0);
1828 return start;
1829 }
1830
1831 // Arguments:
1832 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1833 // ignored
1834 // name - stub name string
1835 //
1836 // Inputs:
1837 // c_rarg0 - source array address
1838 // c_rarg1 - destination array address
1839 // c_rarg2 - element count, treated as ssize_t, can be zero
1840 //
1841 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1842 // let the hardware handle it. The two or four words within dwords
1843 // or qwords that span cache line boundaries will still be loaded
1844 // and stored atomically.
1845 //
1846 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1847 address *entry, const char *name) {
1848 __ align(CodeEntryAlignment);
1849 StubCodeMark mark(this, "StubRoutines", name);
1850 address start = __ pc();
1851
1852 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1853 const Register from = rdi; // source array address
1854 const Register to = rsi; // destination array address
1855 const Register count = rdx; // elements count
1856 const Register word_count = rcx;
1857 const Register qword_count = count;
1858
1859 __ enter(); // required for proper stackwalking of RuntimeStub frame
1860 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1861
1862 if (entry != NULL) {
1863 *entry = __ pc();
1864 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1865 BLOCK_COMMENT("Entry:");
1866 }
1867
1868 array_overlap_test(nooverlap_target, Address::times_2);
1869 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1870 // r9 and r10 may be used to save non-volatile registers
1871
1872 {
1873 // UnsafeCopyMemory page error: continue after ucm
1874 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1875 // 'from', 'to' and 'count' are now valid
1876 __ movptr(word_count, count);
1877 __ shrptr(count, 2); // count => qword_count
1878
1879 // Copy from high to low addresses. Use 'to' as scratch.
1880
1881 // Check for and copy trailing word
1882 __ testl(word_count, 1);
1883 __ jccb(Assembler::zero, L_copy_4_bytes);
1884 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1885 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1886
1887 // Check for and copy trailing dword
1888 __ BIND(L_copy_4_bytes);
1889 __ testl(word_count, 2);
1890 __ jcc(Assembler::zero, L_copy_bytes);
1891 __ movl(rax, Address(from, qword_count, Address::times_8));
1892 __ movl(Address(to, qword_count, Address::times_8), rax);
1893 __ jmp(L_copy_bytes);
1894
1895 // Copy trailing qwords
1896 __ BIND(L_copy_8_bytes);
1897 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1898 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1899 __ decrement(qword_count);
1900 __ jcc(Assembler::notZero, L_copy_8_bytes);
1901 }
1902 restore_arg_regs();
1903 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1904 __ xorptr(rax, rax); // return 0
1905 __ vzeroupper();
1906 __ leave(); // required for proper stackwalking of RuntimeStub frame
1907 __ ret(0);
1908
1909 {
1910 // UnsafeCopyMemory page error: continue after ucm
1911 UnsafeCopyMemoryMark ucmm(this, !aligned, true);
1912 // Copy in multi-bytes chunks
1913 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1914 }
1915 restore_arg_regs();
1916 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1917 __ xorptr(rax, rax); // return 0
1918 __ vzeroupper();
1919 __ leave(); // required for proper stackwalking of RuntimeStub frame
1920 __ ret(0);
1921
1922 return start;
1923 }
1924
1925 // Arguments:
1926 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1927 // ignored
1928 // is_oop - true => oop array, so generate store check code
1929 // name - stub name string
1930 //
1931 // Inputs:
1932 // c_rarg0 - source array address
1933 // c_rarg1 - destination array address
1934 // c_rarg2 - element count, treated as ssize_t, can be zero
1935 //
1936 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1937 // the hardware handle it. The two dwords within qwords that span
1938 // cache line boundaries will still be loaded and stored atomicly.
1939 //
1940 // Side Effects:
1941 // disjoint_int_copy_entry is set to the no-overlap entry point
1942 // used by generate_conjoint_int_oop_copy().
1943 //
1944 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1945 const char *name, bool dest_uninitialized = false) {
1946 __ align(CodeEntryAlignment);
1947 StubCodeMark mark(this, "StubRoutines", name);
1948 address start = __ pc();
1949
1950 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1951 const Register from = rdi; // source array address
1952 const Register to = rsi; // destination array address
1953 const Register count = rdx; // elements count
1954 const Register dword_count = rcx;
1955 const Register qword_count = count;
1956 const Register end_from = from; // source array end address
1957 const Register end_to = to; // destination array end address
1958 // End pointers are inclusive, and if count is not zero they point
1959 // to the last unit copied: end_to[0] := end_from[0]
1960
1961 __ enter(); // required for proper stackwalking of RuntimeStub frame
1962 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1963
1964 if (entry != NULL) {
1965 *entry = __ pc();
1966 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1967 BLOCK_COMMENT("Entry:");
1968 }
1969
1970 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1971 // r9 is used to save r15_thread
1972
1973 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
1974 if (dest_uninitialized) {
1975 decorators |= IS_DEST_UNINITIALIZED;
1976 }
1977 if (aligned) {
1978 decorators |= ARRAYCOPY_ALIGNED;
1979 }
1980
1981 BasicType type = is_oop ? T_OBJECT : T_INT;
1982 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1983 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1984
1985 {
1986 // UnsafeCopyMemory page error: continue after ucm
1987 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
1988 // 'from', 'to' and 'count' are now valid
1989 __ movptr(dword_count, count);
1990 __ shrptr(count, 1); // count => qword_count
1991
1992 // Copy from low to high addresses. Use 'to' as scratch.
1993 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1994 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1995 __ negptr(qword_count);
1996 __ jmp(L_copy_bytes);
1997
1998 // Copy trailing qwords
1999 __ BIND(L_copy_8_bytes);
2000 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2001 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2002 __ increment(qword_count);
2003 __ jcc(Assembler::notZero, L_copy_8_bytes);
2004
2005 // Check for and copy trailing dword
2006 __ BIND(L_copy_4_bytes);
2007 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
2008 __ jccb(Assembler::zero, L_exit);
2009 __ movl(rax, Address(end_from, 8));
2010 __ movl(Address(end_to, 8), rax);
2011 }
2012 __ BIND(L_exit);
2013 address ucme_exit_pc = __ pc();
2014 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2015 restore_arg_regs_using_thread();
2016 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2017 __ vzeroupper();
2018 __ xorptr(rax, rax); // return 0
2019 __ leave(); // required for proper stackwalking of RuntimeStub frame
2020 __ ret(0);
2021
2022 {
2023 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, false, ucme_exit_pc);
2024 // Copy in multi-bytes chunks
2025 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2026 __ jmp(L_copy_4_bytes);
2027 }
2028
2029 return start;
2030 }
2031
2032 // Arguments:
2033 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
2034 // ignored
2035 // is_oop - true => oop array, so generate store check code
2036 // name - stub name string
2037 //
2038 // Inputs:
2039 // c_rarg0 - source array address
2040 // c_rarg1 - destination array address
2041 // c_rarg2 - element count, treated as ssize_t, can be zero
2042 //
2043 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
2044 // the hardware handle it. The two dwords within qwords that span
2045 // cache line boundaries will still be loaded and stored atomicly.
2046 //
2047 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
2048 address *entry, const char *name,
2049 bool dest_uninitialized = false) {
2050 __ align(CodeEntryAlignment);
2051 StubCodeMark mark(this, "StubRoutines", name);
2052 address start = __ pc();
2053
2054 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2055 const Register from = rdi; // source array address
2056 const Register to = rsi; // destination array address
2057 const Register count = rdx; // elements count
2058 const Register dword_count = rcx;
2059 const Register qword_count = count;
2060
2061 __ enter(); // required for proper stackwalking of RuntimeStub frame
2062 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2063
2064 if (entry != NULL) {
2065 *entry = __ pc();
2066 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2067 BLOCK_COMMENT("Entry:");
2068 }
2069
2070 array_overlap_test(nooverlap_target, Address::times_4);
2071 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2072 // r9 is used to save r15_thread
2073
2074 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2075 if (dest_uninitialized) {
2076 decorators |= IS_DEST_UNINITIALIZED;
2077 }
2078 if (aligned) {
2079 decorators |= ARRAYCOPY_ALIGNED;
2080 }
2081
2082 BasicType type = is_oop ? T_OBJECT : T_INT;
2083 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2084 // no registers are destroyed by this call
2085 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2086
2087 assert_clean_int(count, rax); // Make sure 'count' is clean int.
2088 {
2089 // UnsafeCopyMemory page error: continue after ucm
2090 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2091 // 'from', 'to' and 'count' are now valid
2092 __ movptr(dword_count, count);
2093 __ shrptr(count, 1); // count => qword_count
2094
2095 // Copy from high to low addresses. Use 'to' as scratch.
2096
2097 // Check for and copy trailing dword
2098 __ testl(dword_count, 1);
2099 __ jcc(Assembler::zero, L_copy_bytes);
2100 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
2101 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
2102 __ jmp(L_copy_bytes);
2103
2104 // Copy trailing qwords
2105 __ BIND(L_copy_8_bytes);
2106 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2107 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2108 __ decrement(qword_count);
2109 __ jcc(Assembler::notZero, L_copy_8_bytes);
2110 }
2111 if (is_oop) {
2112 __ jmp(L_exit);
2113 }
2114 restore_arg_regs_using_thread();
2115 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2116 __ xorptr(rax, rax); // return 0
2117 __ vzeroupper();
2118 __ leave(); // required for proper stackwalking of RuntimeStub frame
2119 __ ret(0);
2120
2121 {
2122 // UnsafeCopyMemory page error: continue after ucm
2123 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2124 // Copy in multi-bytes chunks
2125 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2126 }
2127
2128 __ BIND(L_exit);
2129 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2130 restore_arg_regs_using_thread();
2131 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2132 __ xorptr(rax, rax); // return 0
2133 __ vzeroupper();
2134 __ leave(); // required for proper stackwalking of RuntimeStub frame
2135 __ ret(0);
2136
2137 return start;
2138 }
2139
2140 // Arguments:
2141 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2142 // ignored
2143 // is_oop - true => oop array, so generate store check code
2144 // name - stub name string
2145 //
2146 // Inputs:
2147 // c_rarg0 - source array address
2148 // c_rarg1 - destination array address
2149 // c_rarg2 - element count, treated as ssize_t, can be zero
2150 //
2151 // Side Effects:
2152 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2153 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2154 //
2155 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2156 const char *name, bool dest_uninitialized = false) {
2157 __ align(CodeEntryAlignment);
2158 StubCodeMark mark(this, "StubRoutines", name);
2159 address start = __ pc();
2160
2161 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2162 const Register from = rdi; // source array address
2163 const Register to = rsi; // destination array address
2164 const Register qword_count = rdx; // elements count
2165 const Register end_from = from; // source array end address
2166 const Register end_to = rcx; // destination array end address
2167 const Register saved_count = r11;
2168 // End pointers are inclusive, and if count is not zero they point
2169 // to the last unit copied: end_to[0] := end_from[0]
2170
2171 __ enter(); // required for proper stackwalking of RuntimeStub frame
2172 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2173 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2174
2175 if (entry != NULL) {
2176 *entry = __ pc();
2177 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2178 BLOCK_COMMENT("Entry:");
2179 }
2180
2181 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2182 // r9 is used to save r15_thread
2183 // 'from', 'to' and 'qword_count' are now valid
2184
2185 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2186 if (dest_uninitialized) {
2187 decorators |= IS_DEST_UNINITIALIZED;
2188 }
2189 if (aligned) {
2190 decorators |= ARRAYCOPY_ALIGNED;
2191 }
2192
2193 BasicType type = is_oop ? T_OBJECT : T_LONG;
2194 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2195 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2196 {
2197 // UnsafeCopyMemory page error: continue after ucm
2198 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2199
2200 // Copy from low to high addresses. Use 'to' as scratch.
2201 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2202 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2203 __ negptr(qword_count);
2204 __ jmp(L_copy_bytes);
2205
2206 // Copy trailing qwords
2207 __ BIND(L_copy_8_bytes);
2208 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2209 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2210 __ increment(qword_count);
2211 __ jcc(Assembler::notZero, L_copy_8_bytes);
2212 }
2213 if (is_oop) {
2214 __ jmp(L_exit);
2215 } else {
2216 restore_arg_regs_using_thread();
2217 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2218 __ xorptr(rax, rax); // return 0
2219 __ vzeroupper();
2220 __ leave(); // required for proper stackwalking of RuntimeStub frame
2221 __ ret(0);
2222 }
2223
2224 {
2225 // UnsafeCopyMemory page error: continue after ucm
2226 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2227 // Copy in multi-bytes chunks
2228 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2229 }
2230
2231 __ BIND(L_exit);
2232 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2233 restore_arg_regs_using_thread();
2234 if (is_oop) {
2235 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2236 } else {
2237 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2238 }
2239 __ vzeroupper();
2240 __ xorptr(rax, rax); // return 0
2241 __ leave(); // required for proper stackwalking of RuntimeStub frame
2242 __ ret(0);
2243
2244 return start;
2245 }
2246
2247 // Arguments:
2248 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2249 // ignored
2250 // is_oop - true => oop array, so generate store check code
2251 // name - stub name string
2252 //
2253 // Inputs:
2254 // c_rarg0 - source array address
2255 // c_rarg1 - destination array address
2256 // c_rarg2 - element count, treated as ssize_t, can be zero
2257 //
2258 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2259 address nooverlap_target, address *entry,
2260 const char *name, bool dest_uninitialized = false) {
2261 __ align(CodeEntryAlignment);
2262 StubCodeMark mark(this, "StubRoutines", name);
2263 address start = __ pc();
2264
2265 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2266 const Register from = rdi; // source array address
2267 const Register to = rsi; // destination array address
2268 const Register qword_count = rdx; // elements count
2269 const Register saved_count = rcx;
2270
2271 __ enter(); // required for proper stackwalking of RuntimeStub frame
2272 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2273
2274 if (entry != NULL) {
2275 *entry = __ pc();
2276 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2277 BLOCK_COMMENT("Entry:");
2278 }
2279
2280 array_overlap_test(nooverlap_target, Address::times_8);
2281 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2282 // r9 is used to save r15_thread
2283 // 'from', 'to' and 'qword_count' are now valid
2284
2285 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2286 if (dest_uninitialized) {
2287 decorators |= IS_DEST_UNINITIALIZED;
2288 }
2289 if (aligned) {
2290 decorators |= ARRAYCOPY_ALIGNED;
2291 }
2292
2293 BasicType type = is_oop ? T_OBJECT : T_LONG;
2294 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2295 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2296 {
2297 // UnsafeCopyMemory page error: continue after ucm
2298 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2299
2300 __ jmp(L_copy_bytes);
2301
2302 // Copy trailing qwords
2303 __ BIND(L_copy_8_bytes);
2304 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2305 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2306 __ decrement(qword_count);
2307 __ jcc(Assembler::notZero, L_copy_8_bytes);
2308 }
2309 if (is_oop) {
2310 __ jmp(L_exit);
2311 } else {
2312 restore_arg_regs_using_thread();
2313 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2314 __ xorptr(rax, rax); // return 0
2315 __ vzeroupper();
2316 __ leave(); // required for proper stackwalking of RuntimeStub frame
2317 __ ret(0);
2318 }
2319 {
2320 // UnsafeCopyMemory page error: continue after ucm
2321 UnsafeCopyMemoryMark ucmm(this, !is_oop && !aligned, true);
2322
2323 // Copy in multi-bytes chunks
2324 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2325 }
2326 __ BIND(L_exit);
2327 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2328 restore_arg_regs_using_thread();
2329 if (is_oop) {
2330 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2331 } else {
2332 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2333 }
2334 __ vzeroupper();
2335 __ xorptr(rax, rax); // return 0
2336 __ leave(); // required for proper stackwalking of RuntimeStub frame
2337 __ ret(0);
2338
2339 return start;
2340 }
2341
2342
2343 // Helper for generating a dynamic type check.
2344 // Smashes no registers.
2345 void generate_type_check(Register sub_klass,
2346 Register super_check_offset,
2347 Register super_klass,
2348 Label& L_success) {
2349 assert_different_registers(sub_klass, super_check_offset, super_klass);
2350
2351 BLOCK_COMMENT("type_check:");
2352
2353 Label L_miss;
2354
2355 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2356 super_check_offset);
2357 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2358
2359 // Fall through on failure!
2360 __ BIND(L_miss);
2361 }
2362
2363 //
2364 // Generate checkcasting array copy stub
2365 //
2366 // Input:
2367 // c_rarg0 - source array address
2368 // c_rarg1 - destination array address
2369 // c_rarg2 - element count, treated as ssize_t, can be zero
2370 // c_rarg3 - size_t ckoff (super_check_offset)
2371 // not Win64
2372 // c_rarg4 - oop ckval (super_klass)
2373 // Win64
2374 // rsp+40 - oop ckval (super_klass)
2375 //
2376 // Output:
2377 // rax == 0 - success
2378 // rax == -1^K - failure, where K is partial transfer count
2379 //
2380 address generate_checkcast_copy(const char *name, address *entry,
2381 bool dest_uninitialized = false) {
2382
2383 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2384
2385 // Input registers (after setup_arg_regs)
2386 const Register from = rdi; // source array address
2387 const Register to = rsi; // destination array address
2388 const Register length = rdx; // elements count
2389 const Register ckoff = rcx; // super_check_offset
2390 const Register ckval = r8; // super_klass
2391
2392 // Registers used as temps (r13, r14 are save-on-entry)
2393 const Register end_from = from; // source array end address
2394 const Register end_to = r13; // destination array end address
2395 const Register count = rdx; // -(count_remaining)
2396 const Register r14_length = r14; // saved copy of length
2397 // End pointers are inclusive, and if length is not zero they point
2398 // to the last unit copied: end_to[0] := end_from[0]
2399
2400 const Register rax_oop = rax; // actual oop copied
2401 const Register r11_klass = r11; // oop._klass
2402
2403 //---------------------------------------------------------------
2404 // Assembler stub will be used for this call to arraycopy
2405 // if the two arrays are subtypes of Object[] but the
2406 // destination array type is not equal to or a supertype
2407 // of the source type. Each element must be separately
2408 // checked.
2409
2410 __ align(CodeEntryAlignment);
2411 StubCodeMark mark(this, "StubRoutines", name);
2412 address start = __ pc();
2413
2414 __ enter(); // required for proper stackwalking of RuntimeStub frame
2415
2416 #ifdef ASSERT
2417 // caller guarantees that the arrays really are different
2418 // otherwise, we would have to make conjoint checks
2419 { Label L;
2420 array_overlap_test(L, TIMES_OOP);
2421 __ stop("checkcast_copy within a single array");
2422 __ bind(L);
2423 }
2424 #endif //ASSERT
2425
2426 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2427 // ckoff => rcx, ckval => r8
2428 // r9 and r10 may be used to save non-volatile registers
2429 #ifdef _WIN64
2430 // last argument (#4) is on stack on Win64
2431 __ movptr(ckval, Address(rsp, 6 * wordSize));
2432 #endif
2433
2434 // Caller of this entry point must set up the argument registers.
2435 if (entry != NULL) {
2436 *entry = __ pc();
2437 BLOCK_COMMENT("Entry:");
2438 }
2439
2440 // allocate spill slots for r13, r14
2441 enum {
2442 saved_r13_offset,
2443 saved_r14_offset,
2444 saved_r10_offset,
2445 saved_rbp_offset
2446 };
2447 __ subptr(rsp, saved_rbp_offset * wordSize);
2448 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2449 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2450 __ movptr(Address(rsp, saved_r10_offset * wordSize), r10);
2451
2452 #ifdef ASSERT
2453 Label L2;
2454 __ get_thread(r14);
2455 __ cmpptr(r15_thread, r14);
2456 __ jcc(Assembler::equal, L2);
2457 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
2458 __ bind(L2);
2459 #endif // ASSERT
2460
2461 // check that int operands are properly extended to size_t
2462 assert_clean_int(length, rax);
2463 assert_clean_int(ckoff, rax);
2464
2465 #ifdef ASSERT
2466 BLOCK_COMMENT("assert consistent ckoff/ckval");
2467 // The ckoff and ckval must be mutually consistent,
2468 // even though caller generates both.
2469 { Label L;
2470 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2471 __ cmpl(ckoff, Address(ckval, sco_offset));
2472 __ jcc(Assembler::equal, L);
2473 __ stop("super_check_offset inconsistent");
2474 __ bind(L);
2475 }
2476 #endif //ASSERT
2477
2478 // Loop-invariant addresses. They are exclusive end pointers.
2479 Address end_from_addr(from, length, TIMES_OOP, 0);
2480 Address end_to_addr(to, length, TIMES_OOP, 0);
2481 // Loop-variant addresses. They assume post-incremented count < 0.
2482 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2483 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2484
2485 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST | ARRAYCOPY_DISJOINT;
2486 if (dest_uninitialized) {
2487 decorators |= IS_DEST_UNINITIALIZED;
2488 }
2489
2490 BasicType type = T_OBJECT;
2491 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2492 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2493
2494 // Copy from low to high addresses, indexed from the end of each array.
2495 __ lea(end_from, end_from_addr);
2496 __ lea(end_to, end_to_addr);
2497 __ movptr(r14_length, length); // save a copy of the length
2498 assert(length == count, ""); // else fix next line:
2499 __ negptr(count); // negate and test the length
2500 __ jcc(Assembler::notZero, L_load_element);
2501
2502 // Empty array: Nothing to do.
2503 __ xorptr(rax, rax); // return 0 on (trivial) success
2504 __ jmp(L_done);
2505
2506 // ======== begin loop ========
2507 // (Loop is rotated; its entry is L_load_element.)
2508 // Loop control:
2509 // for (count = -count; count != 0; count++)
2510 // Base pointers src, dst are biased by 8*(count-1),to last element.
2511 __ align(OptoLoopAlignment);
2512
2513 __ BIND(L_store_element);
2514 __ store_heap_oop(to_element_addr, rax_oop, noreg, noreg, AS_RAW); // store the oop
2515 __ increment(count); // increment the count toward zero
2516 __ jcc(Assembler::zero, L_do_card_marks);
2517
2518 // ======== loop entry is here ========
2519 __ BIND(L_load_element);
2520 __ load_heap_oop(rax_oop, from_element_addr, noreg, noreg, AS_RAW); // load the oop
2521 __ testptr(rax_oop, rax_oop);
2522 __ jcc(Assembler::zero, L_store_element);
2523
2524 __ load_klass(r11_klass, rax_oop);// query the object klass
2525 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2526 // ======== end loop ========
2527
2528 // It was a real error; we must depend on the caller to finish the job.
2529 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2530 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2531 // and report their number to the caller.
2532 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2533 Label L_post_barrier;
2534 __ addptr(r14_length, count); // K = (original - remaining) oops
2535 __ movptr(rax, r14_length); // save the value
2536 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2537 __ jccb(Assembler::notZero, L_post_barrier);
2538 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2539
2540 // Come here on success only.
2541 __ BIND(L_do_card_marks);
2542 __ xorptr(rax, rax); // return 0 on success
2543
2544 __ BIND(L_post_barrier);
2545 bs->arraycopy_epilogue(_masm, decorators, type, from, to, r14_length);
2546
2547 // Common exit point (success or failure).
2548 __ BIND(L_done);
2549 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2550 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2551 __ movptr(r10, Address(rsp, saved_r10_offset * wordSize));
2552 restore_arg_regs();
2553 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2554 __ leave(); // required for proper stackwalking of RuntimeStub frame
2555 __ ret(0);
2556
2557 return start;
2558 }
2559
2560 //
2561 // Generate 'unsafe' array copy stub
2562 // Though just as safe as the other stubs, it takes an unscaled
2563 // size_t argument instead of an element count.
2564 //
2565 // Input:
2566 // c_rarg0 - source array address
2567 // c_rarg1 - destination array address
2568 // c_rarg2 - byte count, treated as ssize_t, can be zero
2569 //
2570 // Examines the alignment of the operands and dispatches
2571 // to a long, int, short, or byte copy loop.
2572 //
2573 address generate_unsafe_copy(const char *name,
2574 address byte_copy_entry, address short_copy_entry,
2575 address int_copy_entry, address long_copy_entry) {
2576
2577 Label L_long_aligned, L_int_aligned, L_short_aligned;
2578
2579 // Input registers (before setup_arg_regs)
2580 const Register from = c_rarg0; // source array address
2581 const Register to = c_rarg1; // destination array address
2582 const Register size = c_rarg2; // byte count (size_t)
2583
2584 // Register used as a temp
2585 const Register bits = rax; // test copy of low bits
2586
2587 __ align(CodeEntryAlignment);
2588 StubCodeMark mark(this, "StubRoutines", name);
2589 address start = __ pc();
2590
2591 __ enter(); // required for proper stackwalking of RuntimeStub frame
2592
2593 // bump this on entry, not on exit:
2594 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2595
2596 __ mov(bits, from);
2597 __ orptr(bits, to);
2598 __ orptr(bits, size);
2599
2600 __ testb(bits, BytesPerLong-1);
2601 __ jccb(Assembler::zero, L_long_aligned);
2602
2603 __ testb(bits, BytesPerInt-1);
2604 __ jccb(Assembler::zero, L_int_aligned);
2605
2606 __ testb(bits, BytesPerShort-1);
2607 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2608
2609 __ BIND(L_short_aligned);
2610 __ shrptr(size, LogBytesPerShort); // size => short_count
2611 __ jump(RuntimeAddress(short_copy_entry));
2612
2613 __ BIND(L_int_aligned);
2614 __ shrptr(size, LogBytesPerInt); // size => int_count
2615 __ jump(RuntimeAddress(int_copy_entry));
2616
2617 __ BIND(L_long_aligned);
2618 __ shrptr(size, LogBytesPerLong); // size => qword_count
2619 __ jump(RuntimeAddress(long_copy_entry));
2620
2621 return start;
2622 }
2623
2624 // Perform range checks on the proposed arraycopy.
2625 // Kills temp, but nothing else.
2626 // Also, clean the sign bits of src_pos and dst_pos.
2627 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2628 Register src_pos, // source position (c_rarg1)
2629 Register dst, // destination array oo (c_rarg2)
2630 Register dst_pos, // destination position (c_rarg3)
2631 Register length,
2632 Register temp,
2633 Label& L_failed) {
2634 BLOCK_COMMENT("arraycopy_range_checks:");
2635
2636 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2637 __ movl(temp, length);
2638 __ addl(temp, src_pos); // src_pos + length
2639 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2640 __ jcc(Assembler::above, L_failed);
2641
2642 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2643 __ movl(temp, length);
2644 __ addl(temp, dst_pos); // dst_pos + length
2645 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2646 __ jcc(Assembler::above, L_failed);
2647
2648 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2649 // Move with sign extension can be used since they are positive.
2650 __ movslq(src_pos, src_pos);
2651 __ movslq(dst_pos, dst_pos);
2652
2653 BLOCK_COMMENT("arraycopy_range_checks done");
2654 }
2655
2656 //
2657 // Generate generic array copy stubs
2658 //
2659 // Input:
2660 // c_rarg0 - src oop
2661 // c_rarg1 - src_pos (32-bits)
2662 // c_rarg2 - dst oop
2663 // c_rarg3 - dst_pos (32-bits)
2664 // not Win64
2665 // c_rarg4 - element count (32-bits)
2666 // Win64
2667 // rsp+40 - element count (32-bits)
2668 //
2669 // Output:
2670 // rax == 0 - success
2671 // rax == -1^K - failure, where K is partial transfer count
2672 //
2673 address generate_generic_copy(const char *name,
2674 address byte_copy_entry, address short_copy_entry,
2675 address int_copy_entry, address oop_copy_entry,
2676 address long_copy_entry, address checkcast_copy_entry) {
2677
2678 Label L_failed, L_failed_0, L_objArray;
2679 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2680
2681 // Input registers
2682 const Register src = c_rarg0; // source array oop
2683 const Register src_pos = c_rarg1; // source position
2684 const Register dst = c_rarg2; // destination array oop
2685 const Register dst_pos = c_rarg3; // destination position
2686 #ifndef _WIN64
2687 const Register length = c_rarg4;
2688 #else
2689 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2690 #endif
2691
2692 { int modulus = CodeEntryAlignment;
2693 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2694 int advance = target - (__ offset() % modulus);
2695 if (advance < 0) advance += modulus;
2696 if (advance > 0) __ nop(advance);
2697 }
2698 StubCodeMark mark(this, "StubRoutines", name);
2699
2700 // Short-hop target to L_failed. Makes for denser prologue code.
2701 __ BIND(L_failed_0);
2702 __ jmp(L_failed);
2703 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2704
2705 __ align(CodeEntryAlignment);
2706 address start = __ pc();
2707
2708 __ enter(); // required for proper stackwalking of RuntimeStub frame
2709
2710 // bump this on entry, not on exit:
2711 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2712
2713 //-----------------------------------------------------------------------
2714 // Assembler stub will be used for this call to arraycopy
2715 // if the following conditions are met:
2716 //
2717 // (1) src and dst must not be null.
2718 // (2) src_pos must not be negative.
2719 // (3) dst_pos must not be negative.
2720 // (4) length must not be negative.
2721 // (5) src klass and dst klass should be the same and not NULL.
2722 // (6) src and dst should be arrays.
2723 // (7) src_pos + length must not exceed length of src.
2724 // (8) dst_pos + length must not exceed length of dst.
2725 //
2726
2727 // if (src == NULL) return -1;
2728 __ testptr(src, src); // src oop
2729 size_t j1off = __ offset();
2730 __ jccb(Assembler::zero, L_failed_0);
2731
2732 // if (src_pos < 0) return -1;
2733 __ testl(src_pos, src_pos); // src_pos (32-bits)
2734 __ jccb(Assembler::negative, L_failed_0);
2735
2736 // if (dst == NULL) return -1;
2737 __ testptr(dst, dst); // dst oop
2738 __ jccb(Assembler::zero, L_failed_0);
2739
2740 // if (dst_pos < 0) return -1;
2741 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2742 size_t j4off = __ offset();
2743 __ jccb(Assembler::negative, L_failed_0);
2744
2745 // The first four tests are very dense code,
2746 // but not quite dense enough to put four
2747 // jumps in a 16-byte instruction fetch buffer.
2748 // That's good, because some branch predicters
2749 // do not like jumps so close together.
2750 // Make sure of this.
2751 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2752
2753 // registers used as temp
2754 const Register r11_length = r11; // elements count to copy
2755 const Register r10_src_klass = r10; // array klass
2756
2757 // if (length < 0) return -1;
2758 __ movl(r11_length, length); // length (elements count, 32-bits value)
2759 __ testl(r11_length, r11_length);
2760 __ jccb(Assembler::negative, L_failed_0);
2761
2762 __ load_klass(r10_src_klass, src);
2763 #ifdef ASSERT
2764 // assert(src->klass() != NULL);
2765 {
2766 BLOCK_COMMENT("assert klasses not null {");
2767 Label L1, L2;
2768 __ testptr(r10_src_klass, r10_src_klass);
2769 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2770 __ bind(L1);
2771 __ stop("broken null klass");
2772 __ bind(L2);
2773 __ load_klass(rax, dst);
2774 __ cmpq(rax, 0);
2775 __ jcc(Assembler::equal, L1); // this would be broken also
2776 BLOCK_COMMENT("} assert klasses not null done");
2777 }
2778 #endif
2779
2780 // Load layout helper (32-bits)
2781 //
2782 // |array_tag| | header_size | element_type | |log2_element_size|
2783 // 32 30 24 16 8 2 0
2784 //
2785 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2786 //
2787
2788 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2789
2790 // Handle objArrays completely differently...
2791 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2792 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2793 __ jcc(Assembler::equal, L_objArray);
2794
2795 // if (src->klass() != dst->klass()) return -1;
2796 __ load_klass(rax, dst);
2797 __ cmpq(r10_src_klass, rax);
2798 __ jcc(Assembler::notEqual, L_failed);
2799
2800 const Register rax_lh = rax; // layout helper
2801 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2802
2803 // if (!src->is_Array()) return -1;
2804 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2805 __ jcc(Assembler::greaterEqual, L_failed);
2806
2807 // At this point, it is known to be a typeArray (array_tag 0x3).
2808 #ifdef ASSERT
2809 {
2810 BLOCK_COMMENT("assert primitive array {");
2811 Label L;
2812 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2813 __ jcc(Assembler::greaterEqual, L);
2814 __ stop("must be a primitive array");
2815 __ bind(L);
2816 BLOCK_COMMENT("} assert primitive array done");
2817 }
2818 #endif
2819
2820 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2821 r10, L_failed);
2822
2823 // TypeArrayKlass
2824 //
2825 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2826 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2827 //
2828
2829 const Register r10_offset = r10; // array offset
2830 const Register rax_elsize = rax_lh; // element size
2831
2832 __ movl(r10_offset, rax_lh);
2833 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2834 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2835 __ addptr(src, r10_offset); // src array offset
2836 __ addptr(dst, r10_offset); // dst array offset
2837 BLOCK_COMMENT("choose copy loop based on element size");
2838 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2839
2840 // next registers should be set before the jump to corresponding stub
2841 const Register from = c_rarg0; // source array address
2842 const Register to = c_rarg1; // destination array address
2843 const Register count = c_rarg2; // elements count
2844
2845 // 'from', 'to', 'count' registers should be set in such order
2846 // since they are the same as 'src', 'src_pos', 'dst'.
2847
2848 __ BIND(L_copy_bytes);
2849 __ cmpl(rax_elsize, 0);
2850 __ jccb(Assembler::notEqual, L_copy_shorts);
2851 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2852 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2853 __ movl2ptr(count, r11_length); // length
2854 __ jump(RuntimeAddress(byte_copy_entry));
2855
2856 __ BIND(L_copy_shorts);
2857 __ cmpl(rax_elsize, LogBytesPerShort);
2858 __ jccb(Assembler::notEqual, L_copy_ints);
2859 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2860 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2861 __ movl2ptr(count, r11_length); // length
2862 __ jump(RuntimeAddress(short_copy_entry));
2863
2864 __ BIND(L_copy_ints);
2865 __ cmpl(rax_elsize, LogBytesPerInt);
2866 __ jccb(Assembler::notEqual, L_copy_longs);
2867 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2868 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2869 __ movl2ptr(count, r11_length); // length
2870 __ jump(RuntimeAddress(int_copy_entry));
2871
2872 __ BIND(L_copy_longs);
2873 #ifdef ASSERT
2874 {
2875 BLOCK_COMMENT("assert long copy {");
2876 Label L;
2877 __ cmpl(rax_elsize, LogBytesPerLong);
2878 __ jcc(Assembler::equal, L);
2879 __ stop("must be long copy, but elsize is wrong");
2880 __ bind(L);
2881 BLOCK_COMMENT("} assert long copy done");
2882 }
2883 #endif
2884 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2885 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2886 __ movl2ptr(count, r11_length); // length
2887 __ jump(RuntimeAddress(long_copy_entry));
2888
2889 // ObjArrayKlass
2890 __ BIND(L_objArray);
2891 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2892
2893 Label L_plain_copy, L_checkcast_copy;
2894 // test array classes for subtyping
2895 __ load_klass(rax, dst);
2896 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2897 __ jcc(Assembler::notEqual, L_checkcast_copy);
2898
2899 // Identically typed arrays can be copied without element-wise checks.
2900 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2901 r10, L_failed);
2902
2903 __ lea(from, Address(src, src_pos, TIMES_OOP,
2904 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2905 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2906 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2907 __ movl2ptr(count, r11_length); // length
2908 __ BIND(L_plain_copy);
2909 __ jump(RuntimeAddress(oop_copy_entry));
2910
2911 __ BIND(L_checkcast_copy);
2912 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2913 {
2914 // Before looking at dst.length, make sure dst is also an objArray.
2915 __ cmpl(Address(rax, lh_offset), objArray_lh);
2916 __ jcc(Assembler::notEqual, L_failed);
2917
2918 // It is safe to examine both src.length and dst.length.
2919 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2920 rax, L_failed);
2921
2922 const Register r11_dst_klass = r11;
2923 __ load_klass(r11_dst_klass, dst); // reload
2924
2925 // Marshal the base address arguments now, freeing registers.
2926 __ lea(from, Address(src, src_pos, TIMES_OOP,
2927 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2928 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2929 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2930 __ movl(count, length); // length (reloaded)
2931 Register sco_temp = c_rarg3; // this register is free now
2932 assert_different_registers(from, to, count, sco_temp,
2933 r11_dst_klass, r10_src_klass);
2934 assert_clean_int(count, sco_temp);
2935
2936 // Generate the type check.
2937 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2938 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2939 assert_clean_int(sco_temp, rax);
2940 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2941
2942 // Fetch destination element klass from the ObjArrayKlass header.
2943 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2944 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2945 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2946 assert_clean_int(sco_temp, rax);
2947
2948 // the checkcast_copy loop needs two extra arguments:
2949 assert(c_rarg3 == sco_temp, "#3 already in place");
2950 // Set up arguments for checkcast_copy_entry.
2951 setup_arg_regs(4);
2952 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2953 __ jump(RuntimeAddress(checkcast_copy_entry));
2954 }
2955
2956 __ BIND(L_failed);
2957 __ xorptr(rax, rax);
2958 __ notptr(rax); // return -1
2959 __ leave(); // required for proper stackwalking of RuntimeStub frame
2960 __ ret(0);
2961
2962 return start;
2963 }
2964
2965 address generate_data_cache_writeback() {
2966 const Register src = c_rarg0; // source address
2967
2968 __ align(CodeEntryAlignment);
2969
2970 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback");
2971
2972 address start = __ pc();
2973 __ enter();
2974 __ cache_wb(Address(src, 0));
2975 __ leave();
2976 __ ret(0);
2977
2978 return start;
2979 }
2980
2981 address generate_data_cache_writeback_sync() {
2982 const Register is_pre = c_rarg0; // pre or post sync
2983
2984 __ align(CodeEntryAlignment);
2985
2986 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback_sync");
2987
2988 // pre wbsync is a no-op
2989 // post wbsync translates to an sfence
2990
2991 Label skip;
2992 address start = __ pc();
2993 __ enter();
2994 __ cmpl(is_pre, 0);
2995 __ jcc(Assembler::notEqual, skip);
2996 __ cache_wbsync(false);
2997 __ bind(skip);
2998 __ leave();
2999 __ ret(0);
3000
3001 return start;
3002 }
3003
3004 void generate_arraycopy_stubs() {
3005 address entry;
3006 address entry_jbyte_arraycopy;
3007 address entry_jshort_arraycopy;
3008 address entry_jint_arraycopy;
3009 address entry_oop_arraycopy;
3010 address entry_jlong_arraycopy;
3011 address entry_checkcast_arraycopy;
3012
3013 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
3014 "jbyte_disjoint_arraycopy");
3015 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
3016 "jbyte_arraycopy");
3017
3018 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
3019 "jshort_disjoint_arraycopy");
3020 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
3021 "jshort_arraycopy");
3022
3023 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
3024 "jint_disjoint_arraycopy");
3025 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
3026 &entry_jint_arraycopy, "jint_arraycopy");
3027
3028 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
3029 "jlong_disjoint_arraycopy");
3030 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
3031 &entry_jlong_arraycopy, "jlong_arraycopy");
3032
3033
3034 if (UseCompressedOops) {
3035 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
3036 "oop_disjoint_arraycopy");
3037 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
3038 &entry_oop_arraycopy, "oop_arraycopy");
3039 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
3040 "oop_disjoint_arraycopy_uninit",
3041 /*dest_uninitialized*/true);
3042 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
3043 NULL, "oop_arraycopy_uninit",
3044 /*dest_uninitialized*/true);
3045 } else {
3046 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
3047 "oop_disjoint_arraycopy");
3048 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
3049 &entry_oop_arraycopy, "oop_arraycopy");
3050 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
3051 "oop_disjoint_arraycopy_uninit",
3052 /*dest_uninitialized*/true);
3053 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
3054 NULL, "oop_arraycopy_uninit",
3055 /*dest_uninitialized*/true);
3056 }
3057
3058 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
3059 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
3060 /*dest_uninitialized*/true);
3061
3062 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
3063 entry_jbyte_arraycopy,
3064 entry_jshort_arraycopy,
3065 entry_jint_arraycopy,
3066 entry_jlong_arraycopy);
3067 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
3068 entry_jbyte_arraycopy,
3069 entry_jshort_arraycopy,
3070 entry_jint_arraycopy,
3071 entry_oop_arraycopy,
3072 entry_jlong_arraycopy,
3073 entry_checkcast_arraycopy);
3074
3075 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
3076 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
3077 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
3078 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
3079 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
3080 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
3081
3082 // We don't generate specialized code for HeapWord-aligned source
3083 // arrays, so just use the code we've already generated
3084 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
3085 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
3086
3087 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
3088 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
3089
3090 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
3091 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
3092
3093 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
3094 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
3095
3096 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
3097 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
3098
3099 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
3100 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
3101 }
3102
3103 // AES intrinsic stubs
3104 enum {AESBlockSize = 16};
3105
3106 address generate_key_shuffle_mask() {
3107 __ align(16);
3108 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
3109 address start = __ pc();
3110 __ emit_data64( 0x0405060700010203, relocInfo::none );
3111 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
3112 return start;
3113 }
3114
3115 address generate_counter_shuffle_mask() {
3116 __ align(16);
3117 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
3118 address start = __ pc();
3119 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3120 __ emit_data64(0x0001020304050607, relocInfo::none);
3121 return start;
3122 }
3123
3124 // Utility routine for loading a 128-bit key word in little endian format
3125 // can optionally specify that the shuffle mask is already in an xmmregister
3126 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
3127 __ movdqu(xmmdst, Address(key, offset));
3128 if (xmm_shuf_mask != NULL) {
3129 __ pshufb(xmmdst, xmm_shuf_mask);
3130 } else {
3131 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3132 }
3133 }
3134
3135 // Utility routine for increase 128bit counter (iv in CTR mode)
3136 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
3137 __ pextrq(reg, xmmdst, 0x0);
3138 __ addq(reg, inc_delta);
3139 __ pinsrq(xmmdst, reg, 0x0);
3140 __ jcc(Assembler::carryClear, next_block); // jump if no carry
3141 __ pextrq(reg, xmmdst, 0x01); // Carry
3142 __ addq(reg, 0x01);
3143 __ pinsrq(xmmdst, reg, 0x01); //Carry end
3144 __ BIND(next_block); // next instruction
3145 }
3146
3147 // Arguments:
3148 //
3149 // Inputs:
3150 // c_rarg0 - source byte array address
3151 // c_rarg1 - destination byte array address
3152 // c_rarg2 - K (key) in little endian int array
3153 //
3154 address generate_aescrypt_encryptBlock() {
3155 assert(UseAES, "need AES instructions and misaligned SSE support");
3156 __ align(CodeEntryAlignment);
3157 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
3158 Label L_doLast;
3159 address start = __ pc();
3160
3161 const Register from = c_rarg0; // source array address
3162 const Register to = c_rarg1; // destination array address
3163 const Register key = c_rarg2; // key array address
3164 const Register keylen = rax;
3165
3166 const XMMRegister xmm_result = xmm0;
3167 const XMMRegister xmm_key_shuf_mask = xmm1;
3168 // On win64 xmm6-xmm15 must be preserved so don't use them.
3169 const XMMRegister xmm_temp1 = xmm2;
3170 const XMMRegister xmm_temp2 = xmm3;
3171 const XMMRegister xmm_temp3 = xmm4;
3172 const XMMRegister xmm_temp4 = xmm5;
3173
3174 __ enter(); // required for proper stackwalking of RuntimeStub frame
3175
3176 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3177 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3178
3179 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3180 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
3181
3182 // For encryption, the java expanded key ordering is just what we need
3183 // we don't know if the key is aligned, hence not using load-execute form
3184
3185 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3186 __ pxor(xmm_result, xmm_temp1);
3187
3188 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3189 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3190 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3191 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3192
3193 __ aesenc(xmm_result, xmm_temp1);
3194 __ aesenc(xmm_result, xmm_temp2);
3195 __ aesenc(xmm_result, xmm_temp3);
3196 __ aesenc(xmm_result, xmm_temp4);
3197
3198 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3199 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3200 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3201 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3202
3203 __ aesenc(xmm_result, xmm_temp1);
3204 __ aesenc(xmm_result, xmm_temp2);
3205 __ aesenc(xmm_result, xmm_temp3);
3206 __ aesenc(xmm_result, xmm_temp4);
3207
3208 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3209 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3210
3211 __ cmpl(keylen, 44);
3212 __ jccb(Assembler::equal, L_doLast);
3213
3214 __ aesenc(xmm_result, xmm_temp1);
3215 __ aesenc(xmm_result, xmm_temp2);
3216
3217 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3218 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3219
3220 __ cmpl(keylen, 52);
3221 __ jccb(Assembler::equal, L_doLast);
3222
3223 __ aesenc(xmm_result, xmm_temp1);
3224 __ aesenc(xmm_result, xmm_temp2);
3225
3226 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3227 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3228
3229 __ BIND(L_doLast);
3230 __ aesenc(xmm_result, xmm_temp1);
3231 __ aesenclast(xmm_result, xmm_temp2);
3232 __ movdqu(Address(to, 0), xmm_result); // store the result
3233 __ xorptr(rax, rax); // return 0
3234 __ leave(); // required for proper stackwalking of RuntimeStub frame
3235 __ ret(0);
3236
3237 return start;
3238 }
3239
3240
3241 // Arguments:
3242 //
3243 // Inputs:
3244 // c_rarg0 - source byte array address
3245 // c_rarg1 - destination byte array address
3246 // c_rarg2 - K (key) in little endian int array
3247 //
3248 address generate_aescrypt_decryptBlock() {
3249 assert(UseAES, "need AES instructions and misaligned SSE support");
3250 __ align(CodeEntryAlignment);
3251 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3252 Label L_doLast;
3253 address start = __ pc();
3254
3255 const Register from = c_rarg0; // source array address
3256 const Register to = c_rarg1; // destination array address
3257 const Register key = c_rarg2; // key array address
3258 const Register keylen = rax;
3259
3260 const XMMRegister xmm_result = xmm0;
3261 const XMMRegister xmm_key_shuf_mask = xmm1;
3262 // On win64 xmm6-xmm15 must be preserved so don't use them.
3263 const XMMRegister xmm_temp1 = xmm2;
3264 const XMMRegister xmm_temp2 = xmm3;
3265 const XMMRegister xmm_temp3 = xmm4;
3266 const XMMRegister xmm_temp4 = xmm5;
3267
3268 __ enter(); // required for proper stackwalking of RuntimeStub frame
3269
3270 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3271 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3272
3273 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3274 __ movdqu(xmm_result, Address(from, 0));
3275
3276 // for decryption java expanded key ordering is rotated one position from what we want
3277 // so we start from 0x10 here and hit 0x00 last
3278 // we don't know if the key is aligned, hence not using load-execute form
3279 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3280 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3281 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3282 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3283
3284 __ pxor (xmm_result, xmm_temp1);
3285 __ aesdec(xmm_result, xmm_temp2);
3286 __ aesdec(xmm_result, xmm_temp3);
3287 __ aesdec(xmm_result, xmm_temp4);
3288
3289 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3290 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3291 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3292 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3293
3294 __ aesdec(xmm_result, xmm_temp1);
3295 __ aesdec(xmm_result, xmm_temp2);
3296 __ aesdec(xmm_result, xmm_temp3);
3297 __ aesdec(xmm_result, xmm_temp4);
3298
3299 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3300 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3301 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3302
3303 __ cmpl(keylen, 44);
3304 __ jccb(Assembler::equal, L_doLast);
3305
3306 __ aesdec(xmm_result, xmm_temp1);
3307 __ aesdec(xmm_result, xmm_temp2);
3308
3309 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3310 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3311
3312 __ cmpl(keylen, 52);
3313 __ jccb(Assembler::equal, L_doLast);
3314
3315 __ aesdec(xmm_result, xmm_temp1);
3316 __ aesdec(xmm_result, xmm_temp2);
3317
3318 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3319 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3320
3321 __ BIND(L_doLast);
3322 __ aesdec(xmm_result, xmm_temp1);
3323 __ aesdec(xmm_result, xmm_temp2);
3324
3325 // for decryption the aesdeclast operation is always on key+0x00
3326 __ aesdeclast(xmm_result, xmm_temp3);
3327 __ movdqu(Address(to, 0), xmm_result); // store the result
3328 __ xorptr(rax, rax); // return 0
3329 __ leave(); // required for proper stackwalking of RuntimeStub frame
3330 __ ret(0);
3331
3332 return start;
3333 }
3334
3335
3336 // Arguments:
3337 //
3338 // Inputs:
3339 // c_rarg0 - source byte array address
3340 // c_rarg1 - destination byte array address
3341 // c_rarg2 - K (key) in little endian int array
3342 // c_rarg3 - r vector byte array address
3343 // c_rarg4 - input length
3344 //
3345 // Output:
3346 // rax - input length
3347 //
3348 address generate_cipherBlockChaining_encryptAESCrypt() {
3349 assert(UseAES, "need AES instructions and misaligned SSE support");
3350 __ align(CodeEntryAlignment);
3351 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3352 address start = __ pc();
3353
3354 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3355 const Register from = c_rarg0; // source array address
3356 const Register to = c_rarg1; // destination array address
3357 const Register key = c_rarg2; // key array address
3358 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3359 // and left with the results of the last encryption block
3360 #ifndef _WIN64
3361 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3362 #else
3363 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3364 const Register len_reg = r11; // pick the volatile windows register
3365 #endif
3366 const Register pos = rax;
3367
3368 // xmm register assignments for the loops below
3369 const XMMRegister xmm_result = xmm0;
3370 const XMMRegister xmm_temp = xmm1;
3371 // keys 0-10 preloaded into xmm2-xmm12
3372 const int XMM_REG_NUM_KEY_FIRST = 2;
3373 const int XMM_REG_NUM_KEY_LAST = 15;
3374 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3375 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3376 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3377 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3378 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3379
3380 __ enter(); // required for proper stackwalking of RuntimeStub frame
3381
3382 #ifdef _WIN64
3383 // on win64, fill len_reg from stack position
3384 __ movl(len_reg, len_mem);
3385 #else
3386 __ push(len_reg); // Save
3387 #endif
3388
3389 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3390 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3391 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3392 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3393 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3394 offset += 0x10;
3395 }
3396 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3397
3398 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3399 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3400 __ cmpl(rax, 44);
3401 __ jcc(Assembler::notEqual, L_key_192_256);
3402
3403 // 128 bit code follows here
3404 __ movptr(pos, 0);
3405 __ align(OptoLoopAlignment);
3406
3407 __ BIND(L_loopTop_128);
3408 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3409 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3410 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3411 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3412 __ aesenc(xmm_result, as_XMMRegister(rnum));
3413 }
3414 __ aesenclast(xmm_result, xmm_key10);
3415 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3416 // no need to store r to memory until we exit
3417 __ addptr(pos, AESBlockSize);
3418 __ subptr(len_reg, AESBlockSize);
3419 __ jcc(Assembler::notEqual, L_loopTop_128);
3420
3421 __ BIND(L_exit);
3422 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3423
3424 #ifdef _WIN64
3425 __ movl(rax, len_mem);
3426 #else
3427 __ pop(rax); // return length
3428 #endif
3429 __ leave(); // required for proper stackwalking of RuntimeStub frame
3430 __ ret(0);
3431
3432 __ BIND(L_key_192_256);
3433 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3434 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3435 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3436 __ cmpl(rax, 52);
3437 __ jcc(Assembler::notEqual, L_key_256);
3438
3439 // 192-bit code follows here (could be changed to use more xmm registers)
3440 __ movptr(pos, 0);
3441 __ align(OptoLoopAlignment);
3442
3443 __ BIND(L_loopTop_192);
3444 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3445 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3446 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3447 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3448 __ aesenc(xmm_result, as_XMMRegister(rnum));
3449 }
3450 __ aesenclast(xmm_result, xmm_key12);
3451 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3452 // no need to store r to memory until we exit
3453 __ addptr(pos, AESBlockSize);
3454 __ subptr(len_reg, AESBlockSize);
3455 __ jcc(Assembler::notEqual, L_loopTop_192);
3456 __ jmp(L_exit);
3457
3458 __ BIND(L_key_256);
3459 // 256-bit code follows here (could be changed to use more xmm registers)
3460 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3461 __ movptr(pos, 0);
3462 __ align(OptoLoopAlignment);
3463
3464 __ BIND(L_loopTop_256);
3465 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3466 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3467 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3468 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3469 __ aesenc(xmm_result, as_XMMRegister(rnum));
3470 }
3471 load_key(xmm_temp, key, 0xe0);
3472 __ aesenclast(xmm_result, xmm_temp);
3473 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3474 // no need to store r to memory until we exit
3475 __ addptr(pos, AESBlockSize);
3476 __ subptr(len_reg, AESBlockSize);
3477 __ jcc(Assembler::notEqual, L_loopTop_256);
3478 __ jmp(L_exit);
3479
3480 return start;
3481 }
3482
3483 // Safefetch stubs.
3484 void generate_safefetch(const char* name, int size, address* entry,
3485 address* fault_pc, address* continuation_pc) {
3486 // safefetch signatures:
3487 // int SafeFetch32(int* adr, int errValue);
3488 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3489 //
3490 // arguments:
3491 // c_rarg0 = adr
3492 // c_rarg1 = errValue
3493 //
3494 // result:
3495 // PPC_RET = *adr or errValue
3496
3497 StubCodeMark mark(this, "StubRoutines", name);
3498
3499 // Entry point, pc or function descriptor.
3500 *entry = __ pc();
3501
3502 // Load *adr into c_rarg1, may fault.
3503 *fault_pc = __ pc();
3504 switch (size) {
3505 case 4:
3506 // int32_t
3507 __ movl(c_rarg1, Address(c_rarg0, 0));
3508 break;
3509 case 8:
3510 // int64_t
3511 __ movq(c_rarg1, Address(c_rarg0, 0));
3512 break;
3513 default:
3514 ShouldNotReachHere();
3515 }
3516
3517 // return errValue or *adr
3518 *continuation_pc = __ pc();
3519 __ movq(rax, c_rarg1);
3520 __ ret(0);
3521 }
3522
3523 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3524 // to hide instruction latency
3525 //
3526 // Arguments:
3527 //
3528 // Inputs:
3529 // c_rarg0 - source byte array address
3530 // c_rarg1 - destination byte array address
3531 // c_rarg2 - K (key) in little endian int array
3532 // c_rarg3 - r vector byte array address
3533 // c_rarg4 - input length
3534 //
3535 // Output:
3536 // rax - input length
3537 //
3538 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3539 assert(UseAES, "need AES instructions and misaligned SSE support");
3540 __ align(CodeEntryAlignment);
3541 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3542 address start = __ pc();
3543
3544 const Register from = c_rarg0; // source array address
3545 const Register to = c_rarg1; // destination array address
3546 const Register key = c_rarg2; // key array address
3547 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3548 // and left with the results of the last encryption block
3549 #ifndef _WIN64
3550 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3551 #else
3552 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3553 const Register len_reg = r11; // pick the volatile windows register
3554 #endif
3555 const Register pos = rax;
3556
3557 const int PARALLEL_FACTOR = 4;
3558 const int ROUNDS[3] = { 10, 12, 14 }; // aes rounds for key128, key192, key256
3559
3560 Label L_exit;
3561 Label L_singleBlock_loopTopHead[3]; // 128, 192, 256
3562 Label L_singleBlock_loopTopHead2[3]; // 128, 192, 256
3563 Label L_singleBlock_loopTop[3]; // 128, 192, 256
3564 Label L_multiBlock_loopTopHead[3]; // 128, 192, 256
3565 Label L_multiBlock_loopTop[3]; // 128, 192, 256
3566
3567 // keys 0-10 preloaded into xmm5-xmm15
3568 const int XMM_REG_NUM_KEY_FIRST = 5;
3569 const int XMM_REG_NUM_KEY_LAST = 15;
3570 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3571 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3572
3573 __ enter(); // required for proper stackwalking of RuntimeStub frame
3574
3575 #ifdef _WIN64
3576 // on win64, fill len_reg from stack position
3577 __ movl(len_reg, len_mem);
3578 #else
3579 __ push(len_reg); // Save
3580 #endif
3581 __ push(rbx);
3582 // the java expanded key ordering is rotated one position from what we want
3583 // so we start from 0x10 here and hit 0x00 last
3584 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3585 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3586 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3587 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3588 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3589 offset += 0x10;
3590 }
3591 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3592
3593 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3594
3595 // registers holding the four results in the parallelized loop
3596 const XMMRegister xmm_result0 = xmm0;
3597 const XMMRegister xmm_result1 = xmm2;
3598 const XMMRegister xmm_result2 = xmm3;
3599 const XMMRegister xmm_result3 = xmm4;
3600
3601 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3602
3603 __ xorptr(pos, pos);
3604
3605 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3606 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3607 __ cmpl(rbx, 52);
3608 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[1]);
3609 __ cmpl(rbx, 60);
3610 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[2]);
3611
3612 #define DoFour(opc, src_reg) \
3613 __ opc(xmm_result0, src_reg); \
3614 __ opc(xmm_result1, src_reg); \
3615 __ opc(xmm_result2, src_reg); \
3616 __ opc(xmm_result3, src_reg); \
3617
3618 for (int k = 0; k < 3; ++k) {
3619 __ BIND(L_multiBlock_loopTopHead[k]);
3620 if (k != 0) {
3621 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3622 __ jcc(Assembler::less, L_singleBlock_loopTopHead2[k]);
3623 }
3624 if (k == 1) {
3625 __ subptr(rsp, 6 * wordSize);
3626 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3627 load_key(xmm15, key, 0xb0); // 0xb0; 192-bit key goes up to 0xc0
3628 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3629 load_key(xmm1, key, 0xc0); // 0xc0;
3630 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3631 } else if (k == 2) {
3632 __ subptr(rsp, 10 * wordSize);
3633 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3634 load_key(xmm15, key, 0xd0); // 0xd0; 256-bit key goes upto 0xe0
3635 __ movdqu(Address(rsp, 6 * wordSize), xmm15);
3636 load_key(xmm1, key, 0xe0); // 0xe0;
3637 __ movdqu(Address(rsp, 8 * wordSize), xmm1);
3638 load_key(xmm15, key, 0xb0); // 0xb0;
3639 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3640 load_key(xmm1, key, 0xc0); // 0xc0;
3641 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3642 }
3643 __ align(OptoLoopAlignment);
3644 __ BIND(L_multiBlock_loopTop[k]);
3645 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3646 __ jcc(Assembler::less, L_singleBlock_loopTopHead[k]);
3647
3648 if (k != 0) {
3649 __ movdqu(xmm15, Address(rsp, 2 * wordSize));
3650 __ movdqu(xmm1, Address(rsp, 4 * wordSize));
3651 }
3652
3653 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
3654 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3655 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3656 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3657
3658 DoFour(pxor, xmm_key_first);
3659 if (k == 0) {
3660 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
3661 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3662 }
3663 DoFour(aesdeclast, xmm_key_last);
3664 } else if (k == 1) {
3665 for (int rnum = 1; rnum <= ROUNDS[k]-2; rnum++) {
3666 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3667 }
3668 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3669 DoFour(aesdec, xmm1); // key : 0xc0
3670 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3671 DoFour(aesdeclast, xmm_key_last);
3672 } else if (k == 2) {
3673 for (int rnum = 1; rnum <= ROUNDS[k] - 4; rnum++) {
3674 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3675 }
3676 DoFour(aesdec, xmm1); // key : 0xc0
3677 __ movdqu(xmm15, Address(rsp, 6 * wordSize));
3678 __ movdqu(xmm1, Address(rsp, 8 * wordSize));
3679 DoFour(aesdec, xmm15); // key : 0xd0
3680 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3681 DoFour(aesdec, xmm1); // key : 0xe0
3682 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3683 DoFour(aesdeclast, xmm_key_last);
3684 }
3685
3686 // for each result, xor with the r vector of previous cipher block
3687 __ pxor(xmm_result0, xmm_prev_block_cipher);
3688 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3689 __ pxor(xmm_result1, xmm_prev_block_cipher);
3690 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3691 __ pxor(xmm_result2, xmm_prev_block_cipher);
3692 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3693 __ pxor(xmm_result3, xmm_prev_block_cipher);
3694 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
3695 if (k != 0) {
3696 __ movdqu(Address(rvec, 0x00), xmm_prev_block_cipher);
3697 }
3698
3699 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3700 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3701 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3702 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3703
3704 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize);
3705 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize);
3706 __ jmp(L_multiBlock_loopTop[k]);
3707
3708 // registers used in the non-parallelized loops
3709 // xmm register assignments for the loops below
3710 const XMMRegister xmm_result = xmm0;
3711 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3712 const XMMRegister xmm_key11 = xmm3;
3713 const XMMRegister xmm_key12 = xmm4;
3714 const XMMRegister key_tmp = xmm4;
3715
3716 __ BIND(L_singleBlock_loopTopHead[k]);
3717 if (k == 1) {
3718 __ addptr(rsp, 6 * wordSize);
3719 } else if (k == 2) {
3720 __ addptr(rsp, 10 * wordSize);
3721 }
3722 __ cmpptr(len_reg, 0); // any blocks left??
3723 __ jcc(Assembler::equal, L_exit);
3724 __ BIND(L_singleBlock_loopTopHead2[k]);
3725 if (k == 1) {
3726 load_key(xmm_key11, key, 0xb0); // 0xb0; 192-bit key goes upto 0xc0
3727 load_key(xmm_key12, key, 0xc0); // 0xc0; 192-bit key goes upto 0xc0
3728 }
3729 if (k == 2) {
3730 load_key(xmm_key11, key, 0xb0); // 0xb0; 256-bit key goes upto 0xe0
3731 }
3732 __ align(OptoLoopAlignment);
3733 __ BIND(L_singleBlock_loopTop[k]);
3734 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3735 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3736 __ pxor(xmm_result, xmm_key_first); // do the aes dec rounds
3737 for (int rnum = 1; rnum <= 9 ; rnum++) {
3738 __ aesdec(xmm_result, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3739 }
3740 if (k == 1) {
3741 __ aesdec(xmm_result, xmm_key11);
3742 __ aesdec(xmm_result, xmm_key12);
3743 }
3744 if (k == 2) {
3745 __ aesdec(xmm_result, xmm_key11);
3746 load_key(key_tmp, key, 0xc0);
3747 __ aesdec(xmm_result, key_tmp);
3748 load_key(key_tmp, key, 0xd0);
3749 __ aesdec(xmm_result, key_tmp);
3750 load_key(key_tmp, key, 0xe0);
3751 __ aesdec(xmm_result, key_tmp);
3752 }
3753
3754 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3755 __ pxor(xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3756 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3757 // no need to store r to memory until we exit
3758 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3759 __ addptr(pos, AESBlockSize);
3760 __ subptr(len_reg, AESBlockSize);
3761 __ jcc(Assembler::notEqual, L_singleBlock_loopTop[k]);
3762 if (k != 2) {
3763 __ jmp(L_exit);
3764 }
3765 } //for 128/192/256
3766
3767 __ BIND(L_exit);
3768 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3769 __ pop(rbx);
3770 #ifdef _WIN64
3771 __ movl(rax, len_mem);
3772 #else
3773 __ pop(rax); // return length
3774 #endif
3775 __ leave(); // required for proper stackwalking of RuntimeStub frame
3776 __ ret(0);
3777 return start;
3778 }
3779
3780 address generate_electronicCodeBook_encryptAESCrypt() {
3781 __ align(CodeEntryAlignment);
3782 StubCodeMark mark(this, "StubRoutines", "electronicCodeBook_encryptAESCrypt");
3783 address start = __ pc();
3784 const Register from = c_rarg0; // source array address
3785 const Register to = c_rarg1; // destination array address
3786 const Register key = c_rarg2; // key array address
3787 const Register len = c_rarg3; // src len (must be multiple of blocksize 16)
3788 __ enter(); // required for proper stackwalking of RuntimeStub frame
3789 __ aesecb_encrypt(from, to, key, len);
3790 __ leave(); // required for proper stackwalking of RuntimeStub frame
3791 __ ret(0);
3792 return start;
3793 }
3794
3795 address generate_electronicCodeBook_decryptAESCrypt() {
3796 __ align(CodeEntryAlignment);
3797 StubCodeMark mark(this, "StubRoutines", "electronicCodeBook_decryptAESCrypt");
3798 address start = __ pc();
3799 const Register from = c_rarg0; // source array address
3800 const Register to = c_rarg1; // destination array address
3801 const Register key = c_rarg2; // key array address
3802 const Register len = c_rarg3; // src len (must be multiple of blocksize 16)
3803 __ enter(); // required for proper stackwalking of RuntimeStub frame
3804 __ aesecb_decrypt(from, to, key, len);
3805 __ leave(); // required for proper stackwalking of RuntimeStub frame
3806 __ ret(0);
3807 return start;
3808 }
3809
3810 address generate_upper_word_mask() {
3811 __ align(64);
3812 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3813 address start = __ pc();
3814 __ emit_data64(0x0000000000000000, relocInfo::none);
3815 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
3816 return start;
3817 }
3818
3819 address generate_shuffle_byte_flip_mask() {
3820 __ align(64);
3821 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3822 address start = __ pc();
3823 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3824 __ emit_data64(0x0001020304050607, relocInfo::none);
3825 return start;
3826 }
3827
3828 // ofs and limit are use for multi-block byte array.
3829 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3830 address generate_sha1_implCompress(bool multi_block, const char *name) {
3831 __ align(CodeEntryAlignment);
3832 StubCodeMark mark(this, "StubRoutines", name);
3833 address start = __ pc();
3834
3835 Register buf = c_rarg0;
3836 Register state = c_rarg1;
3837 Register ofs = c_rarg2;
3838 Register limit = c_rarg3;
3839
3840 const XMMRegister abcd = xmm0;
3841 const XMMRegister e0 = xmm1;
3842 const XMMRegister e1 = xmm2;
3843 const XMMRegister msg0 = xmm3;
3844
3845 const XMMRegister msg1 = xmm4;
3846 const XMMRegister msg2 = xmm5;
3847 const XMMRegister msg3 = xmm6;
3848 const XMMRegister shuf_mask = xmm7;
3849
3850 __ enter();
3851
3852 __ subptr(rsp, 4 * wordSize);
3853
3854 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3855 buf, state, ofs, limit, rsp, multi_block);
3856
3857 __ addptr(rsp, 4 * wordSize);
3858
3859 __ leave();
3860 __ ret(0);
3861 return start;
3862 }
3863
3864 address generate_pshuffle_byte_flip_mask() {
3865 __ align(64);
3866 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3867 address start = __ pc();
3868 __ emit_data64(0x0405060700010203, relocInfo::none);
3869 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3870
3871 if (VM_Version::supports_avx2()) {
3872 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
3873 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3874 // _SHUF_00BA
3875 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3876 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3877 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3878 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3879 // _SHUF_DC00
3880 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3881 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3882 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3883 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3884 }
3885
3886 return start;
3887 }
3888
3889 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
3890 address generate_pshuffle_byte_flip_mask_sha512() {
3891 __ align(32);
3892 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
3893 address start = __ pc();
3894 if (VM_Version::supports_avx2()) {
3895 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
3896 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3897 __ emit_data64(0x1011121314151617, relocInfo::none);
3898 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
3899 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
3900 __ emit_data64(0x0000000000000000, relocInfo::none);
3901 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3902 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3903 }
3904
3905 return start;
3906 }
3907
3908 // ofs and limit are use for multi-block byte array.
3909 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3910 address generate_sha256_implCompress(bool multi_block, const char *name) {
3911 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
3912 __ align(CodeEntryAlignment);
3913 StubCodeMark mark(this, "StubRoutines", name);
3914 address start = __ pc();
3915
3916 Register buf = c_rarg0;
3917 Register state = c_rarg1;
3918 Register ofs = c_rarg2;
3919 Register limit = c_rarg3;
3920
3921 const XMMRegister msg = xmm0;
3922 const XMMRegister state0 = xmm1;
3923 const XMMRegister state1 = xmm2;
3924 const XMMRegister msgtmp0 = xmm3;
3925
3926 const XMMRegister msgtmp1 = xmm4;
3927 const XMMRegister msgtmp2 = xmm5;
3928 const XMMRegister msgtmp3 = xmm6;
3929 const XMMRegister msgtmp4 = xmm7;
3930
3931 const XMMRegister shuf_mask = xmm8;
3932
3933 __ enter();
3934
3935 __ subptr(rsp, 4 * wordSize);
3936
3937 if (VM_Version::supports_sha()) {
3938 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3939 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3940 } else if (VM_Version::supports_avx2()) {
3941 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3942 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3943 }
3944 __ addptr(rsp, 4 * wordSize);
3945 __ vzeroupper();
3946 __ leave();
3947 __ ret(0);
3948 return start;
3949 }
3950
3951 address generate_sha512_implCompress(bool multi_block, const char *name) {
3952 assert(VM_Version::supports_avx2(), "");
3953 assert(VM_Version::supports_bmi2(), "");
3954 __ align(CodeEntryAlignment);
3955 StubCodeMark mark(this, "StubRoutines", name);
3956 address start = __ pc();
3957
3958 Register buf = c_rarg0;
3959 Register state = c_rarg1;
3960 Register ofs = c_rarg2;
3961 Register limit = c_rarg3;
3962
3963 const XMMRegister msg = xmm0;
3964 const XMMRegister state0 = xmm1;
3965 const XMMRegister state1 = xmm2;
3966 const XMMRegister msgtmp0 = xmm3;
3967 const XMMRegister msgtmp1 = xmm4;
3968 const XMMRegister msgtmp2 = xmm5;
3969 const XMMRegister msgtmp3 = xmm6;
3970 const XMMRegister msgtmp4 = xmm7;
3971
3972 const XMMRegister shuf_mask = xmm8;
3973
3974 __ enter();
3975
3976 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3977 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3978
3979 __ vzeroupper();
3980 __ leave();
3981 __ ret(0);
3982 return start;
3983 }
3984
3985 // This mask is used for incrementing counter value(linc0, linc4, etc.)
3986 address counter_mask_addr() {
3987 __ align(64);
3988 StubCodeMark mark(this, "StubRoutines", "counter_mask_addr");
3989 address start = __ pc();
3990 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);//lbswapmask
3991 __ emit_data64(0x0001020304050607, relocInfo::none);
3992 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3993 __ emit_data64(0x0001020304050607, relocInfo::none);
3994 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3995 __ emit_data64(0x0001020304050607, relocInfo::none);
3996 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3997 __ emit_data64(0x0001020304050607, relocInfo::none);
3998 __ emit_data64(0x0000000000000000, relocInfo::none);//linc0 = counter_mask_addr+64
3999 __ emit_data64(0x0000000000000000, relocInfo::none);
4000 __ emit_data64(0x0000000000000001, relocInfo::none);//counter_mask_addr() + 80
4001 __ emit_data64(0x0000000000000000, relocInfo::none);
4002 __ emit_data64(0x0000000000000002, relocInfo::none);
4003 __ emit_data64(0x0000000000000000, relocInfo::none);
4004 __ emit_data64(0x0000000000000003, relocInfo::none);
4005 __ emit_data64(0x0000000000000000, relocInfo::none);
4006 __ emit_data64(0x0000000000000004, relocInfo::none);//linc4 = counter_mask_addr() + 128
4007 __ emit_data64(0x0000000000000000, relocInfo::none);
4008 __ emit_data64(0x0000000000000004, relocInfo::none);
4009 __ emit_data64(0x0000000000000000, relocInfo::none);
4010 __ emit_data64(0x0000000000000004, relocInfo::none);
4011 __ emit_data64(0x0000000000000000, relocInfo::none);
4012 __ emit_data64(0x0000000000000004, relocInfo::none);
4013 __ emit_data64(0x0000000000000000, relocInfo::none);
4014 __ emit_data64(0x0000000000000008, relocInfo::none);//linc8 = counter_mask_addr() + 192
4015 __ emit_data64(0x0000000000000000, relocInfo::none);
4016 __ emit_data64(0x0000000000000008, relocInfo::none);
4017 __ emit_data64(0x0000000000000000, relocInfo::none);
4018 __ emit_data64(0x0000000000000008, relocInfo::none);
4019 __ emit_data64(0x0000000000000000, relocInfo::none);
4020 __ emit_data64(0x0000000000000008, relocInfo::none);
4021 __ emit_data64(0x0000000000000000, relocInfo::none);
4022 __ emit_data64(0x0000000000000020, relocInfo::none);//linc32 = counter_mask_addr() + 256
4023 __ emit_data64(0x0000000000000000, relocInfo::none);
4024 __ emit_data64(0x0000000000000020, relocInfo::none);
4025 __ emit_data64(0x0000000000000000, relocInfo::none);
4026 __ emit_data64(0x0000000000000020, relocInfo::none);
4027 __ emit_data64(0x0000000000000000, relocInfo::none);
4028 __ emit_data64(0x0000000000000020, relocInfo::none);
4029 __ emit_data64(0x0000000000000000, relocInfo::none);
4030 __ emit_data64(0x0000000000000010, relocInfo::none);//linc16 = counter_mask_addr() + 320
4031 __ emit_data64(0x0000000000000000, relocInfo::none);
4032 __ emit_data64(0x0000000000000010, relocInfo::none);
4033 __ emit_data64(0x0000000000000000, relocInfo::none);
4034 __ emit_data64(0x0000000000000010, relocInfo::none);
4035 __ emit_data64(0x0000000000000000, relocInfo::none);
4036 __ emit_data64(0x0000000000000010, relocInfo::none);
4037 __ emit_data64(0x0000000000000000, relocInfo::none);
4038 return start;
4039 }
4040
4041 // Vector AES Counter implementation
4042 address generate_counterMode_VectorAESCrypt() {
4043 __ align(CodeEntryAlignment);
4044 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
4045 address start = __ pc();
4046 const Register from = c_rarg0; // source array address
4047 const Register to = c_rarg1; // destination array address
4048 const Register key = c_rarg2; // key array address r8
4049 const Register counter = c_rarg3; // counter byte array initialized from counter array address
4050 // and updated with the incremented counter in the end
4051 #ifndef _WIN64
4052 const Register len_reg = c_rarg4;
4053 const Register saved_encCounter_start = c_rarg5;
4054 const Register used_addr = r10;
4055 const Address used_mem(rbp, 2 * wordSize);
4056 const Register used = r11;
4057 #else
4058 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4059 const Address saved_encCounter_mem(rbp, 7 * wordSize); // saved encrypted counter is on stack on Win64
4060 const Address used_mem(rbp, 8 * wordSize); // used length is on stack on Win64
4061 const Register len_reg = r10; // pick the first volatile windows register
4062 const Register saved_encCounter_start = r11;
4063 const Register used_addr = r13;
4064 const Register used = r14;
4065 #endif
4066 __ enter();
4067 // Save state before entering routine
4068 __ push(r12);
4069 __ push(r13);
4070 __ push(r14);
4071 __ push(r15);
4072 #ifdef _WIN64
4073 // on win64, fill len_reg from stack position
4074 __ movl(len_reg, len_mem);
4075 __ movptr(saved_encCounter_start, saved_encCounter_mem);
4076 __ movptr(used_addr, used_mem);
4077 __ movl(used, Address(used_addr, 0));
4078 #else
4079 __ push(len_reg); // Save
4080 __ movptr(used_addr, used_mem);
4081 __ movl(used, Address(used_addr, 0));
4082 #endif
4083 __ push(rbx);
4084 __ aesctr_encrypt(from, to, key, counter, len_reg, used, used_addr, saved_encCounter_start);
4085 // Restore state before leaving routine
4086 __ pop(rbx);
4087 #ifdef _WIN64
4088 __ movl(rax, len_mem); // return length
4089 #else
4090 __ pop(rax); // return length
4091 #endif
4092 __ pop(r15);
4093 __ pop(r14);
4094 __ pop(r13);
4095 __ pop(r12);
4096
4097 __ leave(); // required for proper stackwalking of RuntimeStub frame
4098 __ ret(0);
4099 return start;
4100 }
4101
4102 // This is a version of CTR/AES crypt which does 6 blocks in a loop at a time
4103 // to hide instruction latency
4104 //
4105 // Arguments:
4106 //
4107 // Inputs:
4108 // c_rarg0 - source byte array address
4109 // c_rarg1 - destination byte array address
4110 // c_rarg2 - K (key) in little endian int array
4111 // c_rarg3 - counter vector byte array address
4112 // Linux
4113 // c_rarg4 - input length
4114 // c_rarg5 - saved encryptedCounter start
4115 // rbp + 6 * wordSize - saved used length
4116 // Windows
4117 // rbp + 6 * wordSize - input length
4118 // rbp + 7 * wordSize - saved encryptedCounter start
4119 // rbp + 8 * wordSize - saved used length
4120 //
4121 // Output:
4122 // rax - input length
4123 //
4124 address generate_counterMode_AESCrypt_Parallel() {
4125 assert(UseAES, "need AES instructions and misaligned SSE support");
4126 __ align(CodeEntryAlignment);
4127 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
4128 address start = __ pc();
4129 const Register from = c_rarg0; // source array address
4130 const Register to = c_rarg1; // destination array address
4131 const Register key = c_rarg2; // key array address
4132 const Register counter = c_rarg3; // counter byte array initialized from counter array address
4133 // and updated with the incremented counter in the end
4134 #ifndef _WIN64
4135 const Register len_reg = c_rarg4;
4136 const Register saved_encCounter_start = c_rarg5;
4137 const Register used_addr = r10;
4138 const Address used_mem(rbp, 2 * wordSize);
4139 const Register used = r11;
4140 #else
4141 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4142 const Address saved_encCounter_mem(rbp, 7 * wordSize); // length is on stack on Win64
4143 const Address used_mem(rbp, 8 * wordSize); // length is on stack on Win64
4144 const Register len_reg = r10; // pick the first volatile windows register
4145 const Register saved_encCounter_start = r11;
4146 const Register used_addr = r13;
4147 const Register used = r14;
4148 #endif
4149 const Register pos = rax;
4150
4151 const int PARALLEL_FACTOR = 6;
4152 const XMMRegister xmm_counter_shuf_mask = xmm0;
4153 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
4154 const XMMRegister xmm_curr_counter = xmm2;
4155
4156 const XMMRegister xmm_key_tmp0 = xmm3;
4157 const XMMRegister xmm_key_tmp1 = xmm4;
4158
4159 // registers holding the four results in the parallelized loop
4160 const XMMRegister xmm_result0 = xmm5;
4161 const XMMRegister xmm_result1 = xmm6;
4162 const XMMRegister xmm_result2 = xmm7;
4163 const XMMRegister xmm_result3 = xmm8;
4164 const XMMRegister xmm_result4 = xmm9;
4165 const XMMRegister xmm_result5 = xmm10;
4166
4167 const XMMRegister xmm_from0 = xmm11;
4168 const XMMRegister xmm_from1 = xmm12;
4169 const XMMRegister xmm_from2 = xmm13;
4170 const XMMRegister xmm_from3 = xmm14; //the last one is xmm14. we have to preserve it on WIN64.
4171 const XMMRegister xmm_from4 = xmm3; //reuse xmm3~4. Because xmm_key_tmp0~1 are useless when loading input text
4172 const XMMRegister xmm_from5 = xmm4;
4173
4174 //for key_128, key_192, key_256
4175 const int rounds[3] = {10, 12, 14};
4176 Label L_exit_preLoop, L_preLoop_start;
4177 Label L_multiBlock_loopTop[3];
4178 Label L_singleBlockLoopTop[3];
4179 Label L__incCounter[3][6]; //for 6 blocks
4180 Label L__incCounter_single[3]; //for single block, key128, key192, key256
4181 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];
4182 Label L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
4183
4184 Label L_exit;
4185
4186 __ enter(); // required for proper stackwalking of RuntimeStub frame
4187
4188 #ifdef _WIN64
4189 // allocate spill slots for r13, r14
4190 enum {
4191 saved_r13_offset,
4192 saved_r14_offset
4193 };
4194 __ subptr(rsp, 2 * wordSize);
4195 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
4196 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
4197
4198 // on win64, fill len_reg from stack position
4199 __ movl(len_reg, len_mem);
4200 __ movptr(saved_encCounter_start, saved_encCounter_mem);
4201 __ movptr(used_addr, used_mem);
4202 __ movl(used, Address(used_addr, 0));
4203 #else
4204 __ push(len_reg); // Save
4205 __ movptr(used_addr, used_mem);
4206 __ movl(used, Address(used_addr, 0));
4207 #endif
4208
4209 __ push(rbx); // Save RBX
4210 __ movdqu(xmm_curr_counter, Address(counter, 0x00)); // initialize counter with initial counter
4211 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()), pos); // pos as scratch
4212 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled
4213 __ movptr(pos, 0);
4214
4215 // Use the partially used encrpyted counter from last invocation
4216 __ BIND(L_preLoop_start);
4217 __ cmpptr(used, 16);
4218 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
4219 __ cmpptr(len_reg, 0);
4220 __ jcc(Assembler::lessEqual, L_exit_preLoop);
4221 __ movb(rbx, Address(saved_encCounter_start, used));
4222 __ xorb(rbx, Address(from, pos));
4223 __ movb(Address(to, pos), rbx);
4224 __ addptr(pos, 1);
4225 __ addptr(used, 1);
4226 __ subptr(len_reg, 1);
4227
4228 __ jmp(L_preLoop_start);
4229
4230 __ BIND(L_exit_preLoop);
4231 __ movl(Address(used_addr, 0), used);
4232
4233 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
4234 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()), rbx); // rbx as scratch
4235 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4236 __ cmpl(rbx, 52);
4237 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
4238 __ cmpl(rbx, 60);
4239 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
4240
4241 #define CTR_DoSix(opc, src_reg) \
4242 __ opc(xmm_result0, src_reg); \
4243 __ opc(xmm_result1, src_reg); \
4244 __ opc(xmm_result2, src_reg); \
4245 __ opc(xmm_result3, src_reg); \
4246 __ opc(xmm_result4, src_reg); \
4247 __ opc(xmm_result5, src_reg);
4248
4249 // k == 0 : generate code for key_128
4250 // k == 1 : generate code for key_192
4251 // k == 2 : generate code for key_256
4252 for (int k = 0; k < 3; ++k) {
4253 //multi blocks starts here
4254 __ align(OptoLoopAlignment);
4255 __ BIND(L_multiBlock_loopTop[k]);
4256 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
4257 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
4258 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4259
4260 //load, then increase counters
4261 CTR_DoSix(movdqa, xmm_curr_counter);
4262 inc_counter(rbx, xmm_result1, 0x01, L__incCounter[k][0]);
4263 inc_counter(rbx, xmm_result2, 0x02, L__incCounter[k][1]);
4264 inc_counter(rbx, xmm_result3, 0x03, L__incCounter[k][2]);
4265 inc_counter(rbx, xmm_result4, 0x04, L__incCounter[k][3]);
4266 inc_counter(rbx, xmm_result5, 0x05, L__incCounter[k][4]);
4267 inc_counter(rbx, xmm_curr_counter, 0x06, L__incCounter[k][5]);
4268 CTR_DoSix(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
4269 CTR_DoSix(pxor, xmm_key_tmp0); //PXOR with Round 0 key
4270
4271 //load two ROUND_KEYs at a time
4272 for (int i = 1; i < rounds[k]; ) {
4273 load_key(xmm_key_tmp1, key, (0x10 * i), xmm_key_shuf_mask);
4274 load_key(xmm_key_tmp0, key, (0x10 * (i+1)), xmm_key_shuf_mask);
4275 CTR_DoSix(aesenc, xmm_key_tmp1);
4276 i++;
4277 if (i != rounds[k]) {
4278 CTR_DoSix(aesenc, xmm_key_tmp0);
4279 } else {
4280 CTR_DoSix(aesenclast, xmm_key_tmp0);
4281 }
4282 i++;
4283 }
4284
4285 // get next PARALLEL_FACTOR blocks into xmm_result registers
4286 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4287 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
4288 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
4289 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
4290 __ movdqu(xmm_from4, Address(from, pos, Address::times_1, 4 * AESBlockSize));
4291 __ movdqu(xmm_from5, Address(from, pos, Address::times_1, 5 * AESBlockSize));
4292
4293 __ pxor(xmm_result0, xmm_from0);
4294 __ pxor(xmm_result1, xmm_from1);
4295 __ pxor(xmm_result2, xmm_from2);
4296 __ pxor(xmm_result3, xmm_from3);
4297 __ pxor(xmm_result4, xmm_from4);
4298 __ pxor(xmm_result5, xmm_from5);
4299
4300 // store 6 results into the next 64 bytes of output
4301 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4302 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
4303 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
4304 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
4305 __ movdqu(Address(to, pos, Address::times_1, 4 * AESBlockSize), xmm_result4);
4306 __ movdqu(Address(to, pos, Address::times_1, 5 * AESBlockSize), xmm_result5);
4307
4308 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
4309 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
4310 __ jmp(L_multiBlock_loopTop[k]);
4311
4312 // singleBlock starts here
4313 __ align(OptoLoopAlignment);
4314 __ BIND(L_singleBlockLoopTop[k]);
4315 __ cmpptr(len_reg, 0);
4316 __ jcc(Assembler::lessEqual, L_exit);
4317 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4318 __ movdqa(xmm_result0, xmm_curr_counter);
4319 inc_counter(rbx, xmm_curr_counter, 0x01, L__incCounter_single[k]);
4320 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
4321 __ pxor(xmm_result0, xmm_key_tmp0);
4322 for (int i = 1; i < rounds[k]; i++) {
4323 load_key(xmm_key_tmp0, key, (0x10 * i), xmm_key_shuf_mask);
4324 __ aesenc(xmm_result0, xmm_key_tmp0);
4325 }
4326 load_key(xmm_key_tmp0, key, (rounds[k] * 0x10), xmm_key_shuf_mask);
4327 __ aesenclast(xmm_result0, xmm_key_tmp0);
4328 __ cmpptr(len_reg, AESBlockSize);
4329 __ jcc(Assembler::less, L_processTail_insr[k]);
4330 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4331 __ pxor(xmm_result0, xmm_from0);
4332 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4333 __ addptr(pos, AESBlockSize);
4334 __ subptr(len_reg, AESBlockSize);
4335 __ jmp(L_singleBlockLoopTop[k]);
4336 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
4337 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
4338 __ testptr(len_reg, 8);
4339 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
4340 __ subptr(pos,8);
4341 __ pinsrq(xmm_from0, Address(from, pos), 0);
4342 __ BIND(L_processTail_4_insr[k]);
4343 __ testptr(len_reg, 4);
4344 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
4345 __ subptr(pos,4);
4346 __ pslldq(xmm_from0, 4);
4347 __ pinsrd(xmm_from0, Address(from, pos), 0);
4348 __ BIND(L_processTail_2_insr[k]);
4349 __ testptr(len_reg, 2);
4350 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
4351 __ subptr(pos, 2);
4352 __ pslldq(xmm_from0, 2);
4353 __ pinsrw(xmm_from0, Address(from, pos), 0);
4354 __ BIND(L_processTail_1_insr[k]);
4355 __ testptr(len_reg, 1);
4356 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
4357 __ subptr(pos, 1);
4358 __ pslldq(xmm_from0, 1);
4359 __ pinsrb(xmm_from0, Address(from, pos), 0);
4360 __ BIND(L_processTail_exit_insr[k]);
4361
4362 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
4363 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
4364
4365 __ testptr(len_reg, 8);
4366 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
4367 __ pextrq(Address(to, pos), xmm_result0, 0);
4368 __ psrldq(xmm_result0, 8);
4369 __ addptr(pos, 8);
4370 __ BIND(L_processTail_4_extr[k]);
4371 __ testptr(len_reg, 4);
4372 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
4373 __ pextrd(Address(to, pos), xmm_result0, 0);
4374 __ psrldq(xmm_result0, 4);
4375 __ addptr(pos, 4);
4376 __ BIND(L_processTail_2_extr[k]);
4377 __ testptr(len_reg, 2);
4378 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
4379 __ pextrw(Address(to, pos), xmm_result0, 0);
4380 __ psrldq(xmm_result0, 2);
4381 __ addptr(pos, 2);
4382 __ BIND(L_processTail_1_extr[k]);
4383 __ testptr(len_reg, 1);
4384 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
4385 __ pextrb(Address(to, pos), xmm_result0, 0);
4386
4387 __ BIND(L_processTail_exit_extr[k]);
4388 __ movl(Address(used_addr, 0), len_reg);
4389 __ jmp(L_exit);
4390
4391 }
4392
4393 __ BIND(L_exit);
4394 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
4395 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
4396 __ pop(rbx); // pop the saved RBX.
4397 #ifdef _WIN64
4398 __ movl(rax, len_mem);
4399 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
4400 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
4401 __ addptr(rsp, 2 * wordSize);
4402 #else
4403 __ pop(rax); // return 'len'
4404 #endif
4405 __ leave(); // required for proper stackwalking of RuntimeStub frame
4406 __ ret(0);
4407 return start;
4408 }
4409
4410 void roundDec(XMMRegister xmm_reg) {
4411 __ vaesdec(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4412 __ vaesdec(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4413 __ vaesdec(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4414 __ vaesdec(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4415 __ vaesdec(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4416 __ vaesdec(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4417 __ vaesdec(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4418 __ vaesdec(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4419 }
4420
4421 void roundDeclast(XMMRegister xmm_reg) {
4422 __ vaesdeclast(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4423 __ vaesdeclast(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4424 __ vaesdeclast(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4425 __ vaesdeclast(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4426 __ vaesdeclast(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4427 __ vaesdeclast(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4428 __ vaesdeclast(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4429 __ vaesdeclast(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4430 }
4431
4432 void ev_load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask = NULL) {
4433 __ movdqu(xmmdst, Address(key, offset));
4434 if (xmm_shuf_mask != NULL) {
4435 __ pshufb(xmmdst, xmm_shuf_mask);
4436 } else {
4437 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4438 }
4439 __ evshufi64x2(xmmdst, xmmdst, xmmdst, 0x0, Assembler::AVX_512bit);
4440
4441 }
4442
4443 address generate_cipherBlockChaining_decryptVectorAESCrypt() {
4444 assert(VM_Version::supports_vaes(), "need AES instructions and misaligned SSE support");
4445 __ align(CodeEntryAlignment);
4446 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
4447 address start = __ pc();
4448
4449 const Register from = c_rarg0; // source array address
4450 const Register to = c_rarg1; // destination array address
4451 const Register key = c_rarg2; // key array address
4452 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
4453 // and left with the results of the last encryption block
4454 #ifndef _WIN64
4455 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
4456 #else
4457 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4458 const Register len_reg = r11; // pick the volatile windows register
4459 #endif
4460
4461 Label Loop, Loop1, L_128, L_256, L_192, KEY_192, KEY_256, Loop2, Lcbc_dec_rem_loop,
4462 Lcbc_dec_rem_last, Lcbc_dec_ret, Lcbc_dec_rem, Lcbc_exit;
4463
4464 __ enter();
4465
4466 #ifdef _WIN64
4467 // on win64, fill len_reg from stack position
4468 __ movl(len_reg, len_mem);
4469 #else
4470 __ push(len_reg); // Save
4471 #endif
4472 __ push(rbx);
4473 __ vzeroupper();
4474
4475 // Temporary variable declaration for swapping key bytes
4476 const XMMRegister xmm_key_shuf_mask = xmm1;
4477 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4478
4479 // Calculate number of rounds from key size: 44 for 10-rounds, 52 for 12-rounds, 60 for 14-rounds
4480 const Register rounds = rbx;
4481 __ movl(rounds, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4482
4483 const XMMRegister IV = xmm0;
4484 // Load IV and broadcast value to 512-bits
4485 __ evbroadcasti64x2(IV, Address(rvec, 0), Assembler::AVX_512bit);
4486
4487 // Temporary variables for storing round keys
4488 const XMMRegister RK0 = xmm30;
4489 const XMMRegister RK1 = xmm9;
4490 const XMMRegister RK2 = xmm18;
4491 const XMMRegister RK3 = xmm19;
4492 const XMMRegister RK4 = xmm20;
4493 const XMMRegister RK5 = xmm21;
4494 const XMMRegister RK6 = xmm22;
4495 const XMMRegister RK7 = xmm23;
4496 const XMMRegister RK8 = xmm24;
4497 const XMMRegister RK9 = xmm25;
4498 const XMMRegister RK10 = xmm26;
4499
4500 // Load and shuffle key
4501 // the java expanded key ordering is rotated one position from what we want
4502 // so we start from 1*16 here and hit 0*16 last
4503 ev_load_key(RK1, key, 1 * 16, xmm_key_shuf_mask);
4504 ev_load_key(RK2, key, 2 * 16, xmm_key_shuf_mask);
4505 ev_load_key(RK3, key, 3 * 16, xmm_key_shuf_mask);
4506 ev_load_key(RK4, key, 4 * 16, xmm_key_shuf_mask);
4507 ev_load_key(RK5, key, 5 * 16, xmm_key_shuf_mask);
4508 ev_load_key(RK6, key, 6 * 16, xmm_key_shuf_mask);
4509 ev_load_key(RK7, key, 7 * 16, xmm_key_shuf_mask);
4510 ev_load_key(RK8, key, 8 * 16, xmm_key_shuf_mask);
4511 ev_load_key(RK9, key, 9 * 16, xmm_key_shuf_mask);
4512 ev_load_key(RK10, key, 10 * 16, xmm_key_shuf_mask);
4513 ev_load_key(RK0, key, 0*16, xmm_key_shuf_mask);
4514
4515 // Variables for storing source cipher text
4516 const XMMRegister S0 = xmm10;
4517 const XMMRegister S1 = xmm11;
4518 const XMMRegister S2 = xmm12;
4519 const XMMRegister S3 = xmm13;
4520 const XMMRegister S4 = xmm14;
4521 const XMMRegister S5 = xmm15;
4522 const XMMRegister S6 = xmm16;
4523 const XMMRegister S7 = xmm17;
4524
4525 // Variables for storing decrypted text
4526 const XMMRegister B0 = xmm1;
4527 const XMMRegister B1 = xmm2;
4528 const XMMRegister B2 = xmm3;
4529 const XMMRegister B3 = xmm4;
4530 const XMMRegister B4 = xmm5;
4531 const XMMRegister B5 = xmm6;
4532 const XMMRegister B6 = xmm7;
4533 const XMMRegister B7 = xmm8;
4534
4535 __ cmpl(rounds, 44);
4536 __ jcc(Assembler::greater, KEY_192);
4537 __ jmp(Loop);
4538
4539 __ BIND(KEY_192);
4540 const XMMRegister RK11 = xmm27;
4541 const XMMRegister RK12 = xmm28;
4542 ev_load_key(RK11, key, 11*16, xmm_key_shuf_mask);
4543 ev_load_key(RK12, key, 12*16, xmm_key_shuf_mask);
4544
4545 __ cmpl(rounds, 52);
4546 __ jcc(Assembler::greater, KEY_256);
4547 __ jmp(Loop);
4548
4549 __ BIND(KEY_256);
4550 const XMMRegister RK13 = xmm29;
4551 const XMMRegister RK14 = xmm31;
4552 ev_load_key(RK13, key, 13*16, xmm_key_shuf_mask);
4553 ev_load_key(RK14, key, 14*16, xmm_key_shuf_mask);
4554
4555 __ BIND(Loop);
4556 __ cmpl(len_reg, 512);
4557 __ jcc(Assembler::below, Lcbc_dec_rem);
4558 __ BIND(Loop1);
4559 __ subl(len_reg, 512);
4560 __ evmovdquq(S0, Address(from, 0 * 64), Assembler::AVX_512bit);
4561 __ evmovdquq(S1, Address(from, 1 * 64), Assembler::AVX_512bit);
4562 __ evmovdquq(S2, Address(from, 2 * 64), Assembler::AVX_512bit);
4563 __ evmovdquq(S3, Address(from, 3 * 64), Assembler::AVX_512bit);
4564 __ evmovdquq(S4, Address(from, 4 * 64), Assembler::AVX_512bit);
4565 __ evmovdquq(S5, Address(from, 5 * 64), Assembler::AVX_512bit);
4566 __ evmovdquq(S6, Address(from, 6 * 64), Assembler::AVX_512bit);
4567 __ evmovdquq(S7, Address(from, 7 * 64), Assembler::AVX_512bit);
4568 __ leaq(from, Address(from, 8 * 64));
4569
4570 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4571 __ evpxorq(B1, S1, RK1, Assembler::AVX_512bit);
4572 __ evpxorq(B2, S2, RK1, Assembler::AVX_512bit);
4573 __ evpxorq(B3, S3, RK1, Assembler::AVX_512bit);
4574 __ evpxorq(B4, S4, RK1, Assembler::AVX_512bit);
4575 __ evpxorq(B5, S5, RK1, Assembler::AVX_512bit);
4576 __ evpxorq(B6, S6, RK1, Assembler::AVX_512bit);
4577 __ evpxorq(B7, S7, RK1, Assembler::AVX_512bit);
4578
4579 __ evalignq(IV, S0, IV, 0x06);
4580 __ evalignq(S0, S1, S0, 0x06);
4581 __ evalignq(S1, S2, S1, 0x06);
4582 __ evalignq(S2, S3, S2, 0x06);
4583 __ evalignq(S3, S4, S3, 0x06);
4584 __ evalignq(S4, S5, S4, 0x06);
4585 __ evalignq(S5, S6, S5, 0x06);
4586 __ evalignq(S6, S7, S6, 0x06);
4587
4588 roundDec(RK2);
4589 roundDec(RK3);
4590 roundDec(RK4);
4591 roundDec(RK5);
4592 roundDec(RK6);
4593 roundDec(RK7);
4594 roundDec(RK8);
4595 roundDec(RK9);
4596 roundDec(RK10);
4597
4598 __ cmpl(rounds, 44);
4599 __ jcc(Assembler::belowEqual, L_128);
4600 roundDec(RK11);
4601 roundDec(RK12);
4602
4603 __ cmpl(rounds, 52);
4604 __ jcc(Assembler::belowEqual, L_192);
4605 roundDec(RK13);
4606 roundDec(RK14);
4607
4608 __ BIND(L_256);
4609 roundDeclast(RK0);
4610 __ jmp(Loop2);
4611
4612 __ BIND(L_128);
4613 roundDeclast(RK0);
4614 __ jmp(Loop2);
4615
4616 __ BIND(L_192);
4617 roundDeclast(RK0);
4618
4619 __ BIND(Loop2);
4620 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4621 __ evpxorq(B1, B1, S0, Assembler::AVX_512bit);
4622 __ evpxorq(B2, B2, S1, Assembler::AVX_512bit);
4623 __ evpxorq(B3, B3, S2, Assembler::AVX_512bit);
4624 __ evpxorq(B4, B4, S3, Assembler::AVX_512bit);
4625 __ evpxorq(B5, B5, S4, Assembler::AVX_512bit);
4626 __ evpxorq(B6, B6, S5, Assembler::AVX_512bit);
4627 __ evpxorq(B7, B7, S6, Assembler::AVX_512bit);
4628 __ evmovdquq(IV, S7, Assembler::AVX_512bit);
4629
4630 __ evmovdquq(Address(to, 0 * 64), B0, Assembler::AVX_512bit);
4631 __ evmovdquq(Address(to, 1 * 64), B1, Assembler::AVX_512bit);
4632 __ evmovdquq(Address(to, 2 * 64), B2, Assembler::AVX_512bit);
4633 __ evmovdquq(Address(to, 3 * 64), B3, Assembler::AVX_512bit);
4634 __ evmovdquq(Address(to, 4 * 64), B4, Assembler::AVX_512bit);
4635 __ evmovdquq(Address(to, 5 * 64), B5, Assembler::AVX_512bit);
4636 __ evmovdquq(Address(to, 6 * 64), B6, Assembler::AVX_512bit);
4637 __ evmovdquq(Address(to, 7 * 64), B7, Assembler::AVX_512bit);
4638 __ leaq(to, Address(to, 8 * 64));
4639 __ jmp(Loop);
4640
4641 __ BIND(Lcbc_dec_rem);
4642 __ evshufi64x2(IV, IV, IV, 0x03, Assembler::AVX_512bit);
4643
4644 __ BIND(Lcbc_dec_rem_loop);
4645 __ subl(len_reg, 16);
4646 __ jcc(Assembler::carrySet, Lcbc_dec_ret);
4647
4648 __ movdqu(S0, Address(from, 0));
4649 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4650 __ vaesdec(B0, B0, RK2, Assembler::AVX_512bit);
4651 __ vaesdec(B0, B0, RK3, Assembler::AVX_512bit);
4652 __ vaesdec(B0, B0, RK4, Assembler::AVX_512bit);
4653 __ vaesdec(B0, B0, RK5, Assembler::AVX_512bit);
4654 __ vaesdec(B0, B0, RK6, Assembler::AVX_512bit);
4655 __ vaesdec(B0, B0, RK7, Assembler::AVX_512bit);
4656 __ vaesdec(B0, B0, RK8, Assembler::AVX_512bit);
4657 __ vaesdec(B0, B0, RK9, Assembler::AVX_512bit);
4658 __ vaesdec(B0, B0, RK10, Assembler::AVX_512bit);
4659 __ cmpl(rounds, 44);
4660 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4661
4662 __ vaesdec(B0, B0, RK11, Assembler::AVX_512bit);
4663 __ vaesdec(B0, B0, RK12, Assembler::AVX_512bit);
4664 __ cmpl(rounds, 52);
4665 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4666
4667 __ vaesdec(B0, B0, RK13, Assembler::AVX_512bit);
4668 __ vaesdec(B0, B0, RK14, Assembler::AVX_512bit);
4669
4670 __ BIND(Lcbc_dec_rem_last);
4671 __ vaesdeclast(B0, B0, RK0, Assembler::AVX_512bit);
4672
4673 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4674 __ evmovdquq(IV, S0, Assembler::AVX_512bit);
4675 __ movdqu(Address(to, 0), B0);
4676 __ leaq(from, Address(from, 16));
4677 __ leaq(to, Address(to, 16));
4678 __ jmp(Lcbc_dec_rem_loop);
4679
4680 __ BIND(Lcbc_dec_ret);
4681 __ movdqu(Address(rvec, 0), IV);
4682
4683 // Zero out the round keys
4684 __ evpxorq(RK0, RK0, RK0, Assembler::AVX_512bit);
4685 __ evpxorq(RK1, RK1, RK1, Assembler::AVX_512bit);
4686 __ evpxorq(RK2, RK2, RK2, Assembler::AVX_512bit);
4687 __ evpxorq(RK3, RK3, RK3, Assembler::AVX_512bit);
4688 __ evpxorq(RK4, RK4, RK4, Assembler::AVX_512bit);
4689 __ evpxorq(RK5, RK5, RK5, Assembler::AVX_512bit);
4690 __ evpxorq(RK6, RK6, RK6, Assembler::AVX_512bit);
4691 __ evpxorq(RK7, RK7, RK7, Assembler::AVX_512bit);
4692 __ evpxorq(RK8, RK8, RK8, Assembler::AVX_512bit);
4693 __ evpxorq(RK9, RK9, RK9, Assembler::AVX_512bit);
4694 __ evpxorq(RK10, RK10, RK10, Assembler::AVX_512bit);
4695 __ cmpl(rounds, 44);
4696 __ jcc(Assembler::belowEqual, Lcbc_exit);
4697 __ evpxorq(RK11, RK11, RK11, Assembler::AVX_512bit);
4698 __ evpxorq(RK12, RK12, RK12, Assembler::AVX_512bit);
4699 __ cmpl(rounds, 52);
4700 __ jcc(Assembler::belowEqual, Lcbc_exit);
4701 __ evpxorq(RK13, RK13, RK13, Assembler::AVX_512bit);
4702 __ evpxorq(RK14, RK14, RK14, Assembler::AVX_512bit);
4703
4704 __ BIND(Lcbc_exit);
4705 __ pop(rbx);
4706 #ifdef _WIN64
4707 __ movl(rax, len_mem);
4708 #else
4709 __ pop(rax); // return length
4710 #endif
4711 __ leave(); // required for proper stackwalking of RuntimeStub frame
4712 __ ret(0);
4713 return start;
4714 }
4715
4716 // Polynomial x^128+x^127+x^126+x^121+1
4717 address ghash_polynomial_addr() {
4718 __ align(CodeEntryAlignment);
4719 StubCodeMark mark(this, "StubRoutines", "_ghash_poly_addr");
4720 address start = __ pc();
4721 __ emit_data64(0x0000000000000001, relocInfo::none);
4722 __ emit_data64(0xc200000000000000, relocInfo::none);
4723 return start;
4724 }
4725
4726 address ghash_shufflemask_addr() {
4727 __ align(CodeEntryAlignment);
4728 StubCodeMark mark(this, "StubRoutines", "_ghash_shuffmask_addr");
4729 address start = __ pc();
4730 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4731 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4732 return start;
4733 }
4734
4735 // Ghash single and multi block operations using AVX instructions
4736 address generate_avx_ghash_processBlocks() {
4737 __ align(CodeEntryAlignment);
4738
4739 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4740 address start = __ pc();
4741
4742 // arguments
4743 const Register state = c_rarg0;
4744 const Register htbl = c_rarg1;
4745 const Register data = c_rarg2;
4746 const Register blocks = c_rarg3;
4747 __ enter();
4748 // Save state before entering routine
4749 __ avx_ghash(state, htbl, data, blocks);
4750 __ leave(); // required for proper stackwalking of RuntimeStub frame
4751 __ ret(0);
4752 return start;
4753 }
4754
4755 // byte swap x86 long
4756 address generate_ghash_long_swap_mask() {
4757 __ align(CodeEntryAlignment);
4758 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
4759 address start = __ pc();
4760 __ emit_data64(0x0f0e0d0c0b0a0908, relocInfo::none );
4761 __ emit_data64(0x0706050403020100, relocInfo::none );
4762 return start;
4763 }
4764
4765 // byte swap x86 byte array
4766 address generate_ghash_byte_swap_mask() {
4767 __ align(CodeEntryAlignment);
4768 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
4769 address start = __ pc();
4770 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none );
4771 __ emit_data64(0x0001020304050607, relocInfo::none );
4772 return start;
4773 }
4774
4775 /* Single and multi-block ghash operations */
4776 address generate_ghash_processBlocks() {
4777 __ align(CodeEntryAlignment);
4778 Label L_ghash_loop, L_exit;
4779 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4780 address start = __ pc();
4781
4782 const Register state = c_rarg0;
4783 const Register subkeyH = c_rarg1;
4784 const Register data = c_rarg2;
4785 const Register blocks = c_rarg3;
4786
4787 const XMMRegister xmm_temp0 = xmm0;
4788 const XMMRegister xmm_temp1 = xmm1;
4789 const XMMRegister xmm_temp2 = xmm2;
4790 const XMMRegister xmm_temp3 = xmm3;
4791 const XMMRegister xmm_temp4 = xmm4;
4792 const XMMRegister xmm_temp5 = xmm5;
4793 const XMMRegister xmm_temp6 = xmm6;
4794 const XMMRegister xmm_temp7 = xmm7;
4795 const XMMRegister xmm_temp8 = xmm8;
4796 const XMMRegister xmm_temp9 = xmm9;
4797 const XMMRegister xmm_temp10 = xmm10;
4798
4799 __ enter();
4800
4801 __ movdqu(xmm_temp10, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
4802
4803 __ movdqu(xmm_temp0, Address(state, 0));
4804 __ pshufb(xmm_temp0, xmm_temp10);
4805
4806
4807 __ BIND(L_ghash_loop);
4808 __ movdqu(xmm_temp2, Address(data, 0));
4809 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
4810
4811 __ movdqu(xmm_temp1, Address(subkeyH, 0));
4812 __ pshufb(xmm_temp1, xmm_temp10);
4813
4814 __ pxor(xmm_temp0, xmm_temp2);
4815
4816 //
4817 // Multiply with the hash key
4818 //
4819 __ movdqu(xmm_temp3, xmm_temp0);
4820 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
4821 __ movdqu(xmm_temp4, xmm_temp0);
4822 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
4823
4824 __ movdqu(xmm_temp5, xmm_temp0);
4825 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
4826 __ movdqu(xmm_temp6, xmm_temp0);
4827 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
4828
4829 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
4830
4831 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
4832 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
4833 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
4834 __ pxor(xmm_temp3, xmm_temp5);
4835 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
4836 // of the carry-less multiplication of
4837 // xmm0 by xmm1.
4838
4839 // We shift the result of the multiplication by one bit position
4840 // to the left to cope for the fact that the bits are reversed.
4841 __ movdqu(xmm_temp7, xmm_temp3);
4842 __ movdqu(xmm_temp8, xmm_temp6);
4843 __ pslld(xmm_temp3, 1);
4844 __ pslld(xmm_temp6, 1);
4845 __ psrld(xmm_temp7, 31);
4846 __ psrld(xmm_temp8, 31);
4847 __ movdqu(xmm_temp9, xmm_temp7);
4848 __ pslldq(xmm_temp8, 4);
4849 __ pslldq(xmm_temp7, 4);
4850 __ psrldq(xmm_temp9, 12);
4851 __ por(xmm_temp3, xmm_temp7);
4852 __ por(xmm_temp6, xmm_temp8);
4853 __ por(xmm_temp6, xmm_temp9);
4854
4855 //
4856 // First phase of the reduction
4857 //
4858 // Move xmm3 into xmm7, xmm8, xmm9 in order to perform the shifts
4859 // independently.
4860 __ movdqu(xmm_temp7, xmm_temp3);
4861 __ movdqu(xmm_temp8, xmm_temp3);
4862 __ movdqu(xmm_temp9, xmm_temp3);
4863 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
4864 __ pslld(xmm_temp8, 30); // packed right shift shifting << 30
4865 __ pslld(xmm_temp9, 25); // packed right shift shifting << 25
4866 __ pxor(xmm_temp7, xmm_temp8); // xor the shifted versions
4867 __ pxor(xmm_temp7, xmm_temp9);
4868 __ movdqu(xmm_temp8, xmm_temp7);
4869 __ pslldq(xmm_temp7, 12);
4870 __ psrldq(xmm_temp8, 4);
4871 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
4872
4873 //
4874 // Second phase of the reduction
4875 //
4876 // Make 3 copies of xmm3 in xmm2, xmm4, xmm5 for doing these
4877 // shift operations.
4878 __ movdqu(xmm_temp2, xmm_temp3);
4879 __ movdqu(xmm_temp4, xmm_temp3);
4880 __ movdqu(xmm_temp5, xmm_temp3);
4881 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
4882 __ psrld(xmm_temp4, 2); // packed left shifting >> 2
4883 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
4884 __ pxor(xmm_temp2, xmm_temp4); // xor the shifted versions
4885 __ pxor(xmm_temp2, xmm_temp5);
4886 __ pxor(xmm_temp2, xmm_temp8);
4887 __ pxor(xmm_temp3, xmm_temp2);
4888 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
4889
4890 __ decrement(blocks);
4891 __ jcc(Assembler::zero, L_exit);
4892 __ movdqu(xmm_temp0, xmm_temp6);
4893 __ addptr(data, 16);
4894 __ jmp(L_ghash_loop);
4895
4896 __ BIND(L_exit);
4897 __ pshufb(xmm_temp6, xmm_temp10); // Byte swap 16-byte result
4898 __ movdqu(Address(state, 0), xmm_temp6); // store the result
4899 __ leave();
4900 __ ret(0);
4901 return start;
4902 }
4903
4904 //base64 character set
4905 address base64_charset_addr() {
4906 __ align(CodeEntryAlignment);
4907 StubCodeMark mark(this, "StubRoutines", "base64_charset");
4908 address start = __ pc();
4909 __ emit_data64(0x0000004200000041, relocInfo::none);
4910 __ emit_data64(0x0000004400000043, relocInfo::none);
4911 __ emit_data64(0x0000004600000045, relocInfo::none);
4912 __ emit_data64(0x0000004800000047, relocInfo::none);
4913 __ emit_data64(0x0000004a00000049, relocInfo::none);
4914 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4915 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4916 __ emit_data64(0x000000500000004f, relocInfo::none);
4917 __ emit_data64(0x0000005200000051, relocInfo::none);
4918 __ emit_data64(0x0000005400000053, relocInfo::none);
4919 __ emit_data64(0x0000005600000055, relocInfo::none);
4920 __ emit_data64(0x0000005800000057, relocInfo::none);
4921 __ emit_data64(0x0000005a00000059, relocInfo::none);
4922 __ emit_data64(0x0000006200000061, relocInfo::none);
4923 __ emit_data64(0x0000006400000063, relocInfo::none);
4924 __ emit_data64(0x0000006600000065, relocInfo::none);
4925 __ emit_data64(0x0000006800000067, relocInfo::none);
4926 __ emit_data64(0x0000006a00000069, relocInfo::none);
4927 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4928 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4929 __ emit_data64(0x000000700000006f, relocInfo::none);
4930 __ emit_data64(0x0000007200000071, relocInfo::none);
4931 __ emit_data64(0x0000007400000073, relocInfo::none);
4932 __ emit_data64(0x0000007600000075, relocInfo::none);
4933 __ emit_data64(0x0000007800000077, relocInfo::none);
4934 __ emit_data64(0x0000007a00000079, relocInfo::none);
4935 __ emit_data64(0x0000003100000030, relocInfo::none);
4936 __ emit_data64(0x0000003300000032, relocInfo::none);
4937 __ emit_data64(0x0000003500000034, relocInfo::none);
4938 __ emit_data64(0x0000003700000036, relocInfo::none);
4939 __ emit_data64(0x0000003900000038, relocInfo::none);
4940 __ emit_data64(0x0000002f0000002b, relocInfo::none);
4941 return start;
4942 }
4943
4944 //base64 url character set
4945 address base64url_charset_addr() {
4946 __ align(CodeEntryAlignment);
4947 StubCodeMark mark(this, "StubRoutines", "base64url_charset");
4948 address start = __ pc();
4949 __ emit_data64(0x0000004200000041, relocInfo::none);
4950 __ emit_data64(0x0000004400000043, relocInfo::none);
4951 __ emit_data64(0x0000004600000045, relocInfo::none);
4952 __ emit_data64(0x0000004800000047, relocInfo::none);
4953 __ emit_data64(0x0000004a00000049, relocInfo::none);
4954 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4955 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4956 __ emit_data64(0x000000500000004f, relocInfo::none);
4957 __ emit_data64(0x0000005200000051, relocInfo::none);
4958 __ emit_data64(0x0000005400000053, relocInfo::none);
4959 __ emit_data64(0x0000005600000055, relocInfo::none);
4960 __ emit_data64(0x0000005800000057, relocInfo::none);
4961 __ emit_data64(0x0000005a00000059, relocInfo::none);
4962 __ emit_data64(0x0000006200000061, relocInfo::none);
4963 __ emit_data64(0x0000006400000063, relocInfo::none);
4964 __ emit_data64(0x0000006600000065, relocInfo::none);
4965 __ emit_data64(0x0000006800000067, relocInfo::none);
4966 __ emit_data64(0x0000006a00000069, relocInfo::none);
4967 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4968 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4969 __ emit_data64(0x000000700000006f, relocInfo::none);
4970 __ emit_data64(0x0000007200000071, relocInfo::none);
4971 __ emit_data64(0x0000007400000073, relocInfo::none);
4972 __ emit_data64(0x0000007600000075, relocInfo::none);
4973 __ emit_data64(0x0000007800000077, relocInfo::none);
4974 __ emit_data64(0x0000007a00000079, relocInfo::none);
4975 __ emit_data64(0x0000003100000030, relocInfo::none);
4976 __ emit_data64(0x0000003300000032, relocInfo::none);
4977 __ emit_data64(0x0000003500000034, relocInfo::none);
4978 __ emit_data64(0x0000003700000036, relocInfo::none);
4979 __ emit_data64(0x0000003900000038, relocInfo::none);
4980 __ emit_data64(0x0000005f0000002d, relocInfo::none);
4981
4982 return start;
4983 }
4984
4985 address base64_bswap_mask_addr() {
4986 __ align(CodeEntryAlignment);
4987 StubCodeMark mark(this, "StubRoutines", "bswap_mask_base64");
4988 address start = __ pc();
4989 __ emit_data64(0x0504038002010080, relocInfo::none);
4990 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4991 __ emit_data64(0x0908078006050480, relocInfo::none);
4992 __ emit_data64(0x0f0e0d800c0b0a80, relocInfo::none);
4993 __ emit_data64(0x0605048003020180, relocInfo::none);
4994 __ emit_data64(0x0c0b0a8009080780, relocInfo::none);
4995 __ emit_data64(0x0504038002010080, relocInfo::none);
4996 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4997
4998 return start;
4999 }
5000
5001 address base64_right_shift_mask_addr() {
5002 __ align(CodeEntryAlignment);
5003 StubCodeMark mark(this, "StubRoutines", "right_shift_mask");
5004 address start = __ pc();
5005 __ emit_data64(0x0006000400020000, relocInfo::none);
5006 __ emit_data64(0x0006000400020000, relocInfo::none);
5007 __ emit_data64(0x0006000400020000, relocInfo::none);
5008 __ emit_data64(0x0006000400020000, relocInfo::none);
5009 __ emit_data64(0x0006000400020000, relocInfo::none);
5010 __ emit_data64(0x0006000400020000, relocInfo::none);
5011 __ emit_data64(0x0006000400020000, relocInfo::none);
5012 __ emit_data64(0x0006000400020000, relocInfo::none);
5013
5014 return start;
5015 }
5016
5017 address base64_left_shift_mask_addr() {
5018 __ align(CodeEntryAlignment);
5019 StubCodeMark mark(this, "StubRoutines", "left_shift_mask");
5020 address start = __ pc();
5021 __ emit_data64(0x0000000200040000, relocInfo::none);
5022 __ emit_data64(0x0000000200040000, relocInfo::none);
5023 __ emit_data64(0x0000000200040000, relocInfo::none);
5024 __ emit_data64(0x0000000200040000, relocInfo::none);
5025 __ emit_data64(0x0000000200040000, relocInfo::none);
5026 __ emit_data64(0x0000000200040000, relocInfo::none);
5027 __ emit_data64(0x0000000200040000, relocInfo::none);
5028 __ emit_data64(0x0000000200040000, relocInfo::none);
5029
5030 return start;
5031 }
5032
5033 address base64_and_mask_addr() {
5034 __ align(CodeEntryAlignment);
5035 StubCodeMark mark(this, "StubRoutines", "and_mask");
5036 address start = __ pc();
5037 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5038 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5039 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5040 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5041 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5042 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5043 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5044 __ emit_data64(0x3f003f003f000000, relocInfo::none);
5045 return start;
5046 }
5047
5048 address base64_gather_mask_addr() {
5049 __ align(CodeEntryAlignment);
5050 StubCodeMark mark(this, "StubRoutines", "gather_mask");
5051 address start = __ pc();
5052 __ emit_data64(0xffffffffffffffff, relocInfo::none);
5053 return start;
5054 }
5055
5056 // Code for generating Base64 encoding.
5057 // Intrinsic function prototype in Base64.java:
5058 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL) {
5059 address generate_base64_encodeBlock() {
5060 __ align(CodeEntryAlignment);
5061 StubCodeMark mark(this, "StubRoutines", "implEncode");
5062 address start = __ pc();
5063 __ enter();
5064
5065 // Save callee-saved registers before using them
5066 __ push(r12);
5067 __ push(r13);
5068 __ push(r14);
5069 __ push(r15);
5070
5071 // arguments
5072 const Register source = c_rarg0; // Source Array
5073 const Register start_offset = c_rarg1; // start offset
5074 const Register end_offset = c_rarg2; // end offset
5075 const Register dest = c_rarg3; // destination array
5076
5077 #ifndef _WIN64
5078 const Register dp = c_rarg4; // Position for writing to dest array
5079 const Register isURL = c_rarg5;// Base64 or URL character set
5080 #else
5081 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
5082 const Address isURL_mem(rbp, 7 * wordSize);
5083 const Register isURL = r10; // pick the volatile windows register
5084 const Register dp = r12;
5085 __ movl(dp, dp_mem);
5086 __ movl(isURL, isURL_mem);
5087 #endif
5088
5089 const Register length = r14;
5090 Label L_process80, L_process32, L_process3, L_exit, L_processdata;
5091
5092 // calculate length from offsets
5093 __ movl(length, end_offset);
5094 __ subl(length, start_offset);
5095 __ cmpl(length, 0);
5096 __ jcc(Assembler::lessEqual, L_exit);
5097
5098 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_charset_addr()));
5099 // check if base64 charset(isURL=0) or base64 url charset(isURL=1) needs to be loaded
5100 __ cmpl(isURL, 0);
5101 __ jcc(Assembler::equal, L_processdata);
5102 __ lea(r11, ExternalAddress(StubRoutines::x86::base64url_charset_addr()));
5103
5104 // load masks required for encoding data
5105 __ BIND(L_processdata);
5106 __ movdqu(xmm16, ExternalAddress(StubRoutines::x86::base64_gather_mask_addr()));
5107 // Set 64 bits of K register.
5108 __ evpcmpeqb(k3, xmm16, xmm16, Assembler::AVX_512bit);
5109 __ evmovdquq(xmm12, ExternalAddress(StubRoutines::x86::base64_bswap_mask_addr()), Assembler::AVX_256bit, r13);
5110 __ evmovdquq(xmm13, ExternalAddress(StubRoutines::x86::base64_right_shift_mask_addr()), Assembler::AVX_512bit, r13);
5111 __ evmovdquq(xmm14, ExternalAddress(StubRoutines::x86::base64_left_shift_mask_addr()), Assembler::AVX_512bit, r13);
5112 __ evmovdquq(xmm15, ExternalAddress(StubRoutines::x86::base64_and_mask_addr()), Assembler::AVX_512bit, r13);
5113
5114 // Vector Base64 implementation, producing 96 bytes of encoded data
5115 __ BIND(L_process80);
5116 __ cmpl(length, 80);
5117 __ jcc(Assembler::below, L_process32);
5118 __ evmovdquq(xmm0, Address(source, start_offset, Address::times_1, 0), Assembler::AVX_256bit);
5119 __ evmovdquq(xmm1, Address(source, start_offset, Address::times_1, 24), Assembler::AVX_256bit);
5120 __ evmovdquq(xmm2, Address(source, start_offset, Address::times_1, 48), Assembler::AVX_256bit);
5121
5122 //permute the input data in such a manner that we have continuity of the source
5123 __ vpermq(xmm3, xmm0, 148, Assembler::AVX_256bit);
5124 __ vpermq(xmm4, xmm1, 148, Assembler::AVX_256bit);
5125 __ vpermq(xmm5, xmm2, 148, Assembler::AVX_256bit);
5126
5127 //shuffle input and group 3 bytes of data and to it add 0 as the 4th byte.
5128 //we can deal with 12 bytes at a time in a 128 bit register
5129 __ vpshufb(xmm3, xmm3, xmm12, Assembler::AVX_256bit);
5130 __ vpshufb(xmm4, xmm4, xmm12, Assembler::AVX_256bit);
5131 __ vpshufb(xmm5, xmm5, xmm12, Assembler::AVX_256bit);
5132
5133 //convert byte to word. Each 128 bit register will have 6 bytes for processing
5134 __ vpmovzxbw(xmm3, xmm3, Assembler::AVX_512bit);
5135 __ vpmovzxbw(xmm4, xmm4, Assembler::AVX_512bit);
5136 __ vpmovzxbw(xmm5, xmm5, Assembler::AVX_512bit);
5137
5138 // Extract bits in the following pattern 6, 4+2, 2+4, 6 to convert 3, 8 bit numbers to 4, 6 bit numbers
5139 __ evpsrlvw(xmm0, xmm3, xmm13, Assembler::AVX_512bit);
5140 __ evpsrlvw(xmm1, xmm4, xmm13, Assembler::AVX_512bit);
5141 __ evpsrlvw(xmm2, xmm5, xmm13, Assembler::AVX_512bit);
5142
5143 __ evpsllvw(xmm3, xmm3, xmm14, Assembler::AVX_512bit);
5144 __ evpsllvw(xmm4, xmm4, xmm14, Assembler::AVX_512bit);
5145 __ evpsllvw(xmm5, xmm5, xmm14, Assembler::AVX_512bit);
5146
5147 __ vpsrlq(xmm0, xmm0, 8, Assembler::AVX_512bit);
5148 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
5149 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
5150
5151 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5152 __ vpsllq(xmm4, xmm4, 8, Assembler::AVX_512bit);
5153 __ vpsllq(xmm5, xmm5, 8, Assembler::AVX_512bit);
5154
5155 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
5156 __ vpandq(xmm4, xmm4, xmm15, Assembler::AVX_512bit);
5157 __ vpandq(xmm5, xmm5, xmm15, Assembler::AVX_512bit);
5158
5159 // Get the final 4*6 bits base64 encoding
5160 __ vporq(xmm3, xmm3, xmm0, Assembler::AVX_512bit);
5161 __ vporq(xmm4, xmm4, xmm1, Assembler::AVX_512bit);
5162 __ vporq(xmm5, xmm5, xmm2, Assembler::AVX_512bit);
5163
5164 // Shift
5165 __ vpsrlq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5166 __ vpsrlq(xmm4, xmm4, 8, Assembler::AVX_512bit);
5167 __ vpsrlq(xmm5, xmm5, 8, Assembler::AVX_512bit);
5168
5169 // look up 6 bits in the base64 character set to fetch the encoding
5170 // we are converting word to dword as gather instructions need dword indices for looking up encoding
5171 __ vextracti64x4(xmm6, xmm3, 0);
5172 __ vpmovzxwd(xmm0, xmm6, Assembler::AVX_512bit);
5173 __ vextracti64x4(xmm6, xmm3, 1);
5174 __ vpmovzxwd(xmm1, xmm6, Assembler::AVX_512bit);
5175
5176 __ vextracti64x4(xmm6, xmm4, 0);
5177 __ vpmovzxwd(xmm2, xmm6, Assembler::AVX_512bit);
5178 __ vextracti64x4(xmm6, xmm4, 1);
5179 __ vpmovzxwd(xmm3, xmm6, Assembler::AVX_512bit);
5180
5181 __ vextracti64x4(xmm4, xmm5, 0);
5182 __ vpmovzxwd(xmm6, xmm4, Assembler::AVX_512bit);
5183
5184 __ vextracti64x4(xmm4, xmm5, 1);
5185 __ vpmovzxwd(xmm7, xmm4, Assembler::AVX_512bit);
5186
5187 __ kmovql(k2, k3);
5188 __ evpgatherdd(xmm4, k2, Address(r11, xmm0, Address::times_4, 0), Assembler::AVX_512bit);
5189 __ kmovql(k2, k3);
5190 __ evpgatherdd(xmm5, k2, Address(r11, xmm1, Address::times_4, 0), Assembler::AVX_512bit);
5191 __ kmovql(k2, k3);
5192 __ evpgatherdd(xmm8, k2, Address(r11, xmm2, Address::times_4, 0), Assembler::AVX_512bit);
5193 __ kmovql(k2, k3);
5194 __ evpgatherdd(xmm9, k2, Address(r11, xmm3, Address::times_4, 0), Assembler::AVX_512bit);
5195 __ kmovql(k2, k3);
5196 __ evpgatherdd(xmm10, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
5197 __ kmovql(k2, k3);
5198 __ evpgatherdd(xmm11, k2, Address(r11, xmm7, Address::times_4, 0), Assembler::AVX_512bit);
5199
5200 //Down convert dword to byte. Final output is 16*6 = 96 bytes long
5201 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm4, Assembler::AVX_512bit);
5202 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm5, Assembler::AVX_512bit);
5203 __ evpmovdb(Address(dest, dp, Address::times_1, 32), xmm8, Assembler::AVX_512bit);
5204 __ evpmovdb(Address(dest, dp, Address::times_1, 48), xmm9, Assembler::AVX_512bit);
5205 __ evpmovdb(Address(dest, dp, Address::times_1, 64), xmm10, Assembler::AVX_512bit);
5206 __ evpmovdb(Address(dest, dp, Address::times_1, 80), xmm11, Assembler::AVX_512bit);
5207
5208 __ addq(dest, 96);
5209 __ addq(source, 72);
5210 __ subq(length, 72);
5211 __ jmp(L_process80);
5212
5213 // Vector Base64 implementation generating 32 bytes of encoded data
5214 __ BIND(L_process32);
5215 __ cmpl(length, 32);
5216 __ jcc(Assembler::below, L_process3);
5217 __ evmovdquq(xmm0, Address(source, start_offset), Assembler::AVX_256bit);
5218 __ vpermq(xmm0, xmm0, 148, Assembler::AVX_256bit);
5219 __ vpshufb(xmm6, xmm0, xmm12, Assembler::AVX_256bit);
5220 __ vpmovzxbw(xmm6, xmm6, Assembler::AVX_512bit);
5221 __ evpsrlvw(xmm2, xmm6, xmm13, Assembler::AVX_512bit);
5222 __ evpsllvw(xmm3, xmm6, xmm14, Assembler::AVX_512bit);
5223
5224 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
5225 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
5226 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
5227 __ vporq(xmm1, xmm2, xmm3, Assembler::AVX_512bit);
5228 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
5229 __ vextracti64x4(xmm9, xmm1, 0);
5230 __ vpmovzxwd(xmm6, xmm9, Assembler::AVX_512bit);
5231 __ vextracti64x4(xmm9, xmm1, 1);
5232 __ vpmovzxwd(xmm5, xmm9, Assembler::AVX_512bit);
5233 __ kmovql(k2, k3);
5234 __ evpgatherdd(xmm8, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
5235 __ kmovql(k2, k3);
5236 __ evpgatherdd(xmm10, k2, Address(r11, xmm5, Address::times_4, 0), Assembler::AVX_512bit);
5237 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm8, Assembler::AVX_512bit);
5238 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm10, Assembler::AVX_512bit);
5239 __ subq(length, 24);
5240 __ addq(dest, 32);
5241 __ addq(source, 24);
5242 __ jmp(L_process32);
5243
5244 // Scalar data processing takes 3 bytes at a time and produces 4 bytes of encoded data
5245 /* This code corresponds to the scalar version of the following snippet in Base64.java
5246 ** int bits = (src[sp0++] & 0xff) << 16 |(src[sp0++] & 0xff) << 8 |(src[sp0++] & 0xff);
5247 ** dst[dp0++] = (byte)base64[(bits >> > 18) & 0x3f];
5248 ** dst[dp0++] = (byte)base64[(bits >> > 12) & 0x3f];
5249 ** dst[dp0++] = (byte)base64[(bits >> > 6) & 0x3f];
5250 ** dst[dp0++] = (byte)base64[bits & 0x3f];*/
5251 __ BIND(L_process3);
5252 __ cmpl(length, 3);
5253 __ jcc(Assembler::below, L_exit);
5254 // Read 1 byte at a time
5255 __ movzbl(rax, Address(source, start_offset));
5256 __ shll(rax, 0x10);
5257 __ movl(r15, rax);
5258 __ movzbl(rax, Address(source, start_offset, Address::times_1, 1));
5259 __ shll(rax, 0x8);
5260 __ movzwl(rax, rax);
5261 __ orl(r15, rax);
5262 __ movzbl(rax, Address(source, start_offset, Address::times_1, 2));
5263 __ orl(rax, r15);
5264 // Save 3 bytes read in r15
5265 __ movl(r15, rax);
5266 __ shrl(rax, 0x12);
5267 __ andl(rax, 0x3f);
5268 // rax contains the index, r11 contains base64 lookup table
5269 __ movb(rax, Address(r11, rax, Address::times_4));
5270 // Write the encoded byte to destination
5271 __ movb(Address(dest, dp, Address::times_1, 0), rax);
5272 __ movl(rax, r15);
5273 __ shrl(rax, 0xc);
5274 __ andl(rax, 0x3f);
5275 __ movb(rax, Address(r11, rax, Address::times_4));
5276 __ movb(Address(dest, dp, Address::times_1, 1), rax);
5277 __ movl(rax, r15);
5278 __ shrl(rax, 0x6);
5279 __ andl(rax, 0x3f);
5280 __ movb(rax, Address(r11, rax, Address::times_4));
5281 __ movb(Address(dest, dp, Address::times_1, 2), rax);
5282 __ movl(rax, r15);
5283 __ andl(rax, 0x3f);
5284 __ movb(rax, Address(r11, rax, Address::times_4));
5285 __ movb(Address(dest, dp, Address::times_1, 3), rax);
5286 __ subl(length, 3);
5287 __ addq(dest, 4);
5288 __ addq(source, 3);
5289 __ jmp(L_process3);
5290 __ BIND(L_exit);
5291 __ pop(r15);
5292 __ pop(r14);
5293 __ pop(r13);
5294 __ pop(r12);
5295 __ leave();
5296 __ ret(0);
5297 return start;
5298 }
5299
5300 /**
5301 * Arguments:
5302 *
5303 * Inputs:
5304 * c_rarg0 - int crc
5305 * c_rarg1 - byte* buf
5306 * c_rarg2 - int length
5307 *
5308 * Ouput:
5309 * rax - int crc result
5310 */
5311 address generate_updateBytesCRC32() {
5312 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
5313
5314 __ align(CodeEntryAlignment);
5315 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
5316
5317 address start = __ pc();
5318 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5319 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5320 // rscratch1: r10
5321 const Register crc = c_rarg0; // crc
5322 const Register buf = c_rarg1; // source java byte array address
5323 const Register len = c_rarg2; // length
5324 const Register table = c_rarg3; // crc_table address (reuse register)
5325 const Register tmp = r11;
5326 assert_different_registers(crc, buf, len, table, tmp, rax);
5327
5328 BLOCK_COMMENT("Entry:");
5329 __ enter(); // required for proper stackwalking of RuntimeStub frame
5330
5331 __ kernel_crc32(crc, buf, len, table, tmp);
5332
5333 __ movl(rax, crc);
5334 __ vzeroupper();
5335 __ leave(); // required for proper stackwalking of RuntimeStub frame
5336 __ ret(0);
5337
5338 return start;
5339 }
5340
5341 /**
5342 * Arguments:
5343 *
5344 * Inputs:
5345 * c_rarg0 - int crc
5346 * c_rarg1 - byte* buf
5347 * c_rarg2 - long length
5348 * c_rarg3 - table_start - optional (present only when doing a library_call,
5349 * not used by x86 algorithm)
5350 *
5351 * Ouput:
5352 * rax - int crc result
5353 */
5354 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
5355 assert(UseCRC32CIntrinsics, "need SSE4_2");
5356 __ align(CodeEntryAlignment);
5357 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
5358 address start = __ pc();
5359 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
5360 //Windows RCX RDX R8 R9 none none XMM0..XMM3
5361 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
5362 const Register crc = c_rarg0; // crc
5363 const Register buf = c_rarg1; // source java byte array address
5364 const Register len = c_rarg2; // length
5365 const Register a = rax;
5366 const Register j = r9;
5367 const Register k = r10;
5368 const Register l = r11;
5369 #ifdef _WIN64
5370 const Register y = rdi;
5371 const Register z = rsi;
5372 #else
5373 const Register y = rcx;
5374 const Register z = r8;
5375 #endif
5376 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
5377
5378 BLOCK_COMMENT("Entry:");
5379 __ enter(); // required for proper stackwalking of RuntimeStub frame
5380 #ifdef _WIN64
5381 __ push(y);
5382 __ push(z);
5383 #endif
5384 __ crc32c_ipl_alg2_alt2(crc, buf, len,
5385 a, j, k,
5386 l, y, z,
5387 c_farg0, c_farg1, c_farg2,
5388 is_pclmulqdq_supported);
5389 __ movl(rax, crc);
5390 #ifdef _WIN64
5391 __ pop(z);
5392 __ pop(y);
5393 #endif
5394 __ vzeroupper();
5395 __ leave(); // required for proper stackwalking of RuntimeStub frame
5396 __ ret(0);
5397
5398 return start;
5399 }
5400
5401 /**
5402 * Arguments:
5403 *
5404 * Input:
5405 * c_rarg0 - x address
5406 * c_rarg1 - x length
5407 * c_rarg2 - y address
5408 * c_rarg3 - y length
5409 * not Win64
5410 * c_rarg4 - z address
5411 * c_rarg5 - z length
5412 * Win64
5413 * rsp+40 - z address
5414 * rsp+48 - z length
5415 */
5416 address generate_multiplyToLen() {
5417 __ align(CodeEntryAlignment);
5418 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
5419
5420 address start = __ pc();
5421 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5422 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5423 const Register x = rdi;
5424 const Register xlen = rax;
5425 const Register y = rsi;
5426 const Register ylen = rcx;
5427 const Register z = r8;
5428 const Register zlen = r11;
5429
5430 // Next registers will be saved on stack in multiply_to_len().
5431 const Register tmp1 = r12;
5432 const Register tmp2 = r13;
5433 const Register tmp3 = r14;
5434 const Register tmp4 = r15;
5435 const Register tmp5 = rbx;
5436
5437 BLOCK_COMMENT("Entry:");
5438 __ enter(); // required for proper stackwalking of RuntimeStub frame
5439
5440 #ifndef _WIN64
5441 __ movptr(zlen, r9); // Save r9 in r11 - zlen
5442 #endif
5443 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
5444 // ylen => rcx, z => r8, zlen => r11
5445 // r9 and r10 may be used to save non-volatile registers
5446 #ifdef _WIN64
5447 // last 2 arguments (#4, #5) are on stack on Win64
5448 __ movptr(z, Address(rsp, 6 * wordSize));
5449 __ movptr(zlen, Address(rsp, 7 * wordSize));
5450 #endif
5451
5452 __ movptr(xlen, rsi);
5453 __ movptr(y, rdx);
5454 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
5455
5456 restore_arg_regs();
5457
5458 __ leave(); // required for proper stackwalking of RuntimeStub frame
5459 __ ret(0);
5460
5461 return start;
5462 }
5463
5464 /**
5465 * Arguments:
5466 *
5467 * Input:
5468 * c_rarg0 - obja address
5469 * c_rarg1 - objb address
5470 * c_rarg3 - length length
5471 * c_rarg4 - scale log2_array_indxscale
5472 *
5473 * Output:
5474 * rax - int >= mismatched index, < 0 bitwise complement of tail
5475 */
5476 address generate_vectorizedMismatch() {
5477 __ align(CodeEntryAlignment);
5478 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
5479 address start = __ pc();
5480
5481 BLOCK_COMMENT("Entry:");
5482 __ enter();
5483
5484 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5485 const Register scale = c_rarg0; //rcx, will exchange with r9
5486 const Register objb = c_rarg1; //rdx
5487 const Register length = c_rarg2; //r8
5488 const Register obja = c_rarg3; //r9
5489 __ xchgq(obja, scale); //now obja and scale contains the correct contents
5490
5491 const Register tmp1 = r10;
5492 const Register tmp2 = r11;
5493 #endif
5494 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5495 const Register obja = c_rarg0; //U:rdi
5496 const Register objb = c_rarg1; //U:rsi
5497 const Register length = c_rarg2; //U:rdx
5498 const Register scale = c_rarg3; //U:rcx
5499 const Register tmp1 = r8;
5500 const Register tmp2 = r9;
5501 #endif
5502 const Register result = rax; //return value
5503 const XMMRegister vec0 = xmm0;
5504 const XMMRegister vec1 = xmm1;
5505 const XMMRegister vec2 = xmm2;
5506
5507 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
5508
5509 __ vzeroupper();
5510 __ leave();
5511 __ ret(0);
5512
5513 return start;
5514 }
5515
5516 /**
5517 * Arguments:
5518 *
5519 // Input:
5520 // c_rarg0 - x address
5521 // c_rarg1 - x length
5522 // c_rarg2 - z address
5523 // c_rarg3 - z lenth
5524 *
5525 */
5526 address generate_squareToLen() {
5527
5528 __ align(CodeEntryAlignment);
5529 StubCodeMark mark(this, "StubRoutines", "squareToLen");
5530
5531 address start = __ pc();
5532 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5533 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
5534 const Register x = rdi;
5535 const Register len = rsi;
5536 const Register z = r8;
5537 const Register zlen = rcx;
5538
5539 const Register tmp1 = r12;
5540 const Register tmp2 = r13;
5541 const Register tmp3 = r14;
5542 const Register tmp4 = r15;
5543 const Register tmp5 = rbx;
5544
5545 BLOCK_COMMENT("Entry:");
5546 __ enter(); // required for proper stackwalking of RuntimeStub frame
5547
5548 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
5549 // zlen => rcx
5550 // r9 and r10 may be used to save non-volatile registers
5551 __ movptr(r8, rdx);
5552 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5553
5554 restore_arg_regs();
5555
5556 __ leave(); // required for proper stackwalking of RuntimeStub frame
5557 __ ret(0);
5558
5559 return start;
5560 }
5561
5562 address generate_method_entry_barrier() {
5563 __ align(CodeEntryAlignment);
5564 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
5565
5566 Label deoptimize_label;
5567
5568 address start = __ pc();
5569
5570 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
5571
5572 BLOCK_COMMENT("Entry:");
5573 __ enter(); // save rbp
5574
5575 // save c_rarg0, because we want to use that value.
5576 // We could do without it but then we depend on the number of slots used by pusha
5577 __ push(c_rarg0);
5578
5579 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
5580
5581 __ pusha();
5582
5583 // The method may have floats as arguments, and we must spill them before calling
5584 // the VM runtime.
5585 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
5586 const int xmm_size = wordSize * 2;
5587 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
5588 __ subptr(rsp, xmm_spill_size);
5589 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
5590 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
5591 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
5592 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
5593 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
5594 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
5595 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
5596 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
5597
5598 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
5599
5600 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
5601 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
5602 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
5603 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
5604 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
5605 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
5606 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
5607 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
5608 __ addptr(rsp, xmm_spill_size);
5609
5610 __ cmpl(rax, 1); // 1 means deoptimize
5611 __ jcc(Assembler::equal, deoptimize_label);
5612
5613 __ popa();
5614 __ pop(c_rarg0);
5615
5616 __ leave();
5617
5618 __ addptr(rsp, 1 * wordSize); // cookie
5619 __ ret(0);
5620
5621
5622 __ BIND(deoptimize_label);
5623
5624 __ popa();
5625 __ pop(c_rarg0);
5626
5627 __ leave();
5628
5629 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
5630 // here while still having a correct stack is valuable
5631 __ testptr(rsp, Address(rsp, 0));
5632
5633 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
5634 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
5635
5636 return start;
5637 }
5638
5639 /**
5640 * Arguments:
5641 *
5642 * Input:
5643 * c_rarg0 - out address
5644 * c_rarg1 - in address
5645 * c_rarg2 - offset
5646 * c_rarg3 - len
5647 * not Win64
5648 * c_rarg4 - k
5649 * Win64
5650 * rsp+40 - k
5651 */
5652 address generate_mulAdd() {
5653 __ align(CodeEntryAlignment);
5654 StubCodeMark mark(this, "StubRoutines", "mulAdd");
5655
5656 address start = __ pc();
5657 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5658 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5659 const Register out = rdi;
5660 const Register in = rsi;
5661 const Register offset = r11;
5662 const Register len = rcx;
5663 const Register k = r8;
5664
5665 // Next registers will be saved on stack in mul_add().
5666 const Register tmp1 = r12;
5667 const Register tmp2 = r13;
5668 const Register tmp3 = r14;
5669 const Register tmp4 = r15;
5670 const Register tmp5 = rbx;
5671
5672 BLOCK_COMMENT("Entry:");
5673 __ enter(); // required for proper stackwalking of RuntimeStub frame
5674
5675 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
5676 // len => rcx, k => r8
5677 // r9 and r10 may be used to save non-volatile registers
5678 #ifdef _WIN64
5679 // last argument is on stack on Win64
5680 __ movl(k, Address(rsp, 6 * wordSize));
5681 #endif
5682 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
5683 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5684
5685 restore_arg_regs();
5686
5687 __ leave(); // required for proper stackwalking of RuntimeStub frame
5688 __ ret(0);
5689
5690 return start;
5691 }
5692
5693 address generate_libmExp() {
5694 StubCodeMark mark(this, "StubRoutines", "libmExp");
5695
5696 address start = __ pc();
5697
5698 const XMMRegister x0 = xmm0;
5699 const XMMRegister x1 = xmm1;
5700 const XMMRegister x2 = xmm2;
5701 const XMMRegister x3 = xmm3;
5702
5703 const XMMRegister x4 = xmm4;
5704 const XMMRegister x5 = xmm5;
5705 const XMMRegister x6 = xmm6;
5706 const XMMRegister x7 = xmm7;
5707
5708 const Register tmp = r11;
5709
5710 BLOCK_COMMENT("Entry:");
5711 __ enter(); // required for proper stackwalking of RuntimeStub frame
5712
5713 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5714
5715 __ leave(); // required for proper stackwalking of RuntimeStub frame
5716 __ ret(0);
5717
5718 return start;
5719
5720 }
5721
5722 address generate_libmLog() {
5723 StubCodeMark mark(this, "StubRoutines", "libmLog");
5724
5725 address start = __ pc();
5726
5727 const XMMRegister x0 = xmm0;
5728 const XMMRegister x1 = xmm1;
5729 const XMMRegister x2 = xmm2;
5730 const XMMRegister x3 = xmm3;
5731
5732 const XMMRegister x4 = xmm4;
5733 const XMMRegister x5 = xmm5;
5734 const XMMRegister x6 = xmm6;
5735 const XMMRegister x7 = xmm7;
5736
5737 const Register tmp1 = r11;
5738 const Register tmp2 = r8;
5739
5740 BLOCK_COMMENT("Entry:");
5741 __ enter(); // required for proper stackwalking of RuntimeStub frame
5742
5743 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2);
5744
5745 __ leave(); // required for proper stackwalking of RuntimeStub frame
5746 __ ret(0);
5747
5748 return start;
5749
5750 }
5751
5752 address generate_libmLog10() {
5753 StubCodeMark mark(this, "StubRoutines", "libmLog10");
5754
5755 address start = __ pc();
5756
5757 const XMMRegister x0 = xmm0;
5758 const XMMRegister x1 = xmm1;
5759 const XMMRegister x2 = xmm2;
5760 const XMMRegister x3 = xmm3;
5761
5762 const XMMRegister x4 = xmm4;
5763 const XMMRegister x5 = xmm5;
5764 const XMMRegister x6 = xmm6;
5765 const XMMRegister x7 = xmm7;
5766
5767 const Register tmp = r11;
5768
5769 BLOCK_COMMENT("Entry:");
5770 __ enter(); // required for proper stackwalking of RuntimeStub frame
5771
5772 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5773
5774 __ leave(); // required for proper stackwalking of RuntimeStub frame
5775 __ ret(0);
5776
5777 return start;
5778
5779 }
5780
5781 address generate_libmPow() {
5782 StubCodeMark mark(this, "StubRoutines", "libmPow");
5783
5784 address start = __ pc();
5785
5786 const XMMRegister x0 = xmm0;
5787 const XMMRegister x1 = xmm1;
5788 const XMMRegister x2 = xmm2;
5789 const XMMRegister x3 = xmm3;
5790
5791 const XMMRegister x4 = xmm4;
5792 const XMMRegister x5 = xmm5;
5793 const XMMRegister x6 = xmm6;
5794 const XMMRegister x7 = xmm7;
5795
5796 const Register tmp1 = r8;
5797 const Register tmp2 = r9;
5798 const Register tmp3 = r10;
5799 const Register tmp4 = r11;
5800
5801 BLOCK_COMMENT("Entry:");
5802 __ enter(); // required for proper stackwalking of RuntimeStub frame
5803
5804 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5805
5806 __ leave(); // required for proper stackwalking of RuntimeStub frame
5807 __ ret(0);
5808
5809 return start;
5810
5811 }
5812
5813 address generate_libmSin() {
5814 StubCodeMark mark(this, "StubRoutines", "libmSin");
5815
5816 address start = __ pc();
5817
5818 const XMMRegister x0 = xmm0;
5819 const XMMRegister x1 = xmm1;
5820 const XMMRegister x2 = xmm2;
5821 const XMMRegister x3 = xmm3;
5822
5823 const XMMRegister x4 = xmm4;
5824 const XMMRegister x5 = xmm5;
5825 const XMMRegister x6 = xmm6;
5826 const XMMRegister x7 = xmm7;
5827
5828 const Register tmp1 = r8;
5829 const Register tmp2 = r9;
5830 const Register tmp3 = r10;
5831 const Register tmp4 = r11;
5832
5833 BLOCK_COMMENT("Entry:");
5834 __ enter(); // required for proper stackwalking of RuntimeStub frame
5835
5836 #ifdef _WIN64
5837 __ push(rsi);
5838 __ push(rdi);
5839 #endif
5840 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5841
5842 #ifdef _WIN64
5843 __ pop(rdi);
5844 __ pop(rsi);
5845 #endif
5846
5847 __ leave(); // required for proper stackwalking of RuntimeStub frame
5848 __ ret(0);
5849
5850 return start;
5851
5852 }
5853
5854 address generate_libmCos() {
5855 StubCodeMark mark(this, "StubRoutines", "libmCos");
5856
5857 address start = __ pc();
5858
5859 const XMMRegister x0 = xmm0;
5860 const XMMRegister x1 = xmm1;
5861 const XMMRegister x2 = xmm2;
5862 const XMMRegister x3 = xmm3;
5863
5864 const XMMRegister x4 = xmm4;
5865 const XMMRegister x5 = xmm5;
5866 const XMMRegister x6 = xmm6;
5867 const XMMRegister x7 = xmm7;
5868
5869 const Register tmp1 = r8;
5870 const Register tmp2 = r9;
5871 const Register tmp3 = r10;
5872 const Register tmp4 = r11;
5873
5874 BLOCK_COMMENT("Entry:");
5875 __ enter(); // required for proper stackwalking of RuntimeStub frame
5876
5877 #ifdef _WIN64
5878 __ push(rsi);
5879 __ push(rdi);
5880 #endif
5881 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5882
5883 #ifdef _WIN64
5884 __ pop(rdi);
5885 __ pop(rsi);
5886 #endif
5887
5888 __ leave(); // required for proper stackwalking of RuntimeStub frame
5889 __ ret(0);
5890
5891 return start;
5892
5893 }
5894
5895 address generate_libmTan() {
5896 StubCodeMark mark(this, "StubRoutines", "libmTan");
5897
5898 address start = __ pc();
5899
5900 const XMMRegister x0 = xmm0;
5901 const XMMRegister x1 = xmm1;
5902 const XMMRegister x2 = xmm2;
5903 const XMMRegister x3 = xmm3;
5904
5905 const XMMRegister x4 = xmm4;
5906 const XMMRegister x5 = xmm5;
5907 const XMMRegister x6 = xmm6;
5908 const XMMRegister x7 = xmm7;
5909
5910 const Register tmp1 = r8;
5911 const Register tmp2 = r9;
5912 const Register tmp3 = r10;
5913 const Register tmp4 = r11;
5914
5915 BLOCK_COMMENT("Entry:");
5916 __ enter(); // required for proper stackwalking of RuntimeStub frame
5917
5918 #ifdef _WIN64
5919 __ push(rsi);
5920 __ push(rdi);
5921 #endif
5922 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5923
5924 #ifdef _WIN64
5925 __ pop(rdi);
5926 __ pop(rsi);
5927 #endif
5928
5929 __ leave(); // required for proper stackwalking of RuntimeStub frame
5930 __ ret(0);
5931
5932 return start;
5933
5934 }
5935
5936 #undef __
5937 #define __ masm->
5938
5939 // Continuation point for throwing of implicit exceptions that are
5940 // not handled in the current activation. Fabricates an exception
5941 // oop and initiates normal exception dispatching in this
5942 // frame. Since we need to preserve callee-saved values (currently
5943 // only for C2, but done for C1 as well) we need a callee-saved oop
5944 // map and therefore have to make these stubs into RuntimeStubs
5945 // rather than BufferBlobs. If the compiler needs all registers to
5946 // be preserved between the fault point and the exception handler
5947 // then it must assume responsibility for that in
5948 // AbstractCompiler::continuation_for_implicit_null_exception or
5949 // continuation_for_implicit_division_by_zero_exception. All other
5950 // implicit exceptions (e.g., NullPointerException or
5951 // AbstractMethodError on entry) are either at call sites or
5952 // otherwise assume that stack unwinding will be initiated, so
5953 // caller saved registers were assumed volatile in the compiler.
5954 address generate_throw_exception(const char* name,
5955 address runtime_entry,
5956 Register arg1 = noreg,
5957 Register arg2 = noreg) {
5958 // Information about frame layout at time of blocking runtime call.
5959 // Note that we only have to preserve callee-saved registers since
5960 // the compilers are responsible for supplying a continuation point
5961 // if they expect all registers to be preserved.
5962 enum layout {
5963 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
5964 rbp_off2,
5965 return_off,
5966 return_off2,
5967 framesize // inclusive of return address
5968 };
5969
5970 int insts_size = 512;
5971 int locs_size = 64;
5972
5973 CodeBuffer code(name, insts_size, locs_size);
5974 OopMapSet* oop_maps = new OopMapSet();
5975 MacroAssembler* masm = new MacroAssembler(&code);
5976
5977 address start = __ pc();
5978
5979 // This is an inlined and slightly modified version of call_VM
5980 // which has the ability to fetch the return PC out of
5981 // thread-local storage and also sets up last_Java_sp slightly
5982 // differently than the real call_VM
5983
5984 __ enter(); // required for proper stackwalking of RuntimeStub frame
5985
5986 assert(is_even(framesize/2), "sp not 16-byte aligned");
5987
5988 // return address and rbp are already in place
5989 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
5990
5991 int frame_complete = __ pc() - start;
5992
5993 // Set up last_Java_sp and last_Java_fp
5994 address the_pc = __ pc();
5995 __ set_last_Java_frame(rsp, rbp, the_pc);
5996 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
5997
5998 // Call runtime
5999 if (arg1 != noreg) {
6000 assert(arg2 != c_rarg1, "clobbered");
6001 __ movptr(c_rarg1, arg1);
6002 }
6003 if (arg2 != noreg) {
6004 __ movptr(c_rarg2, arg2);
6005 }
6006 __ movptr(c_rarg0, r15_thread);
6007 BLOCK_COMMENT("call runtime_entry");
6008 __ call(RuntimeAddress(runtime_entry));
6009
6010 // Generate oop map
6011 OopMap* map = new OopMap(framesize, 0);
6012
6013 oop_maps->add_gc_map(the_pc - start, map);
6014
6015 __ reset_last_Java_frame(true);
6016
6017 __ leave(); // required for proper stackwalking of RuntimeStub frame
6018
6019 // check for pending exceptions
6020 #ifdef ASSERT
6021 Label L;
6022 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
6023 (int32_t) NULL_WORD);
6024 __ jcc(Assembler::notEqual, L);
6025 __ should_not_reach_here();
6026 __ bind(L);
6027 #endif // ASSERT
6028 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
6029
6030
6031 // codeBlob framesize is in words (not VMRegImpl::slot_size)
6032 RuntimeStub* stub =
6033 RuntimeStub::new_runtime_stub(name,
6034 &code,
6035 frame_complete,
6036 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
6037 oop_maps, false);
6038 return stub->entry_point();
6039 }
6040
6041 void create_control_words() {
6042 // Round to nearest, 53-bit mode, exceptions masked
6043 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
6044 // Round to zero, 53-bit mode, exception mased
6045 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
6046 // Round to nearest, 24-bit mode, exceptions masked
6047 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
6048 // Round to nearest, 64-bit mode, exceptions masked
6049 StubRoutines::_mxcsr_std = 0x1F80;
6050 // Note: the following two constants are 80-bit values
6051 // layout is critical for correct loading by FPU.
6052 // Bias for strict fp multiply/divide
6053 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
6054 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
6055 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
6056 // Un-Bias for strict fp multiply/divide
6057 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
6058 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
6059 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
6060 }
6061
6062 // Initialization
6063 void generate_initial() {
6064 // Generates all stubs and initializes the entry points
6065
6066 // This platform-specific settings are needed by generate_call_stub()
6067 create_control_words();
6068
6069 // entry points that exist in all platforms Note: This is code
6070 // that could be shared among different platforms - however the
6071 // benefit seems to be smaller than the disadvantage of having a
6072 // much more complicated generator structure. See also comment in
6073 // stubRoutines.hpp.
6074
6075 StubRoutines::_forward_exception_entry = generate_forward_exception();
6076
6077 StubRoutines::_call_stub_entry =
6078 generate_call_stub(StubRoutines::_call_stub_return_address);
6079
6080 // is referenced by megamorphic call
6081 StubRoutines::_catch_exception_entry = generate_catch_exception();
6082
6083 // atomic calls
6084 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
6085 StubRoutines::_atomic_xchg_long_entry = generate_atomic_xchg_long();
6086 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
6087 StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
6088 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
6089 StubRoutines::_atomic_add_entry = generate_atomic_add();
6090 StubRoutines::_atomic_add_long_entry = generate_atomic_add_long();
6091 StubRoutines::_fence_entry = generate_orderaccess_fence();
6092
6093 // platform dependent
6094 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
6095 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
6096
6097 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
6098
6099 // Build this early so it's available for the interpreter.
6100 StubRoutines::_throw_StackOverflowError_entry =
6101 generate_throw_exception("StackOverflowError throw_exception",
6102 CAST_FROM_FN_PTR(address,
6103 SharedRuntime::
6104 throw_StackOverflowError));
6105 StubRoutines::_throw_delayed_StackOverflowError_entry =
6106 generate_throw_exception("delayed StackOverflowError throw_exception",
6107 CAST_FROM_FN_PTR(address,
6108 SharedRuntime::
6109 throw_delayed_StackOverflowError));
6110 if (UseCRC32Intrinsics) {
6111 // set table address before stub generation which use it
6112 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
6113 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
6114 }
6115
6116 if (UseCRC32CIntrinsics) {
6117 bool supports_clmul = VM_Version::supports_clmul();
6118 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
6119 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
6120 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
6121 }
6122 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
6123 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
6124 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
6125 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6126 StubRoutines::x86::_ONEHALF_adr = (address)StubRoutines::x86::_ONEHALF;
6127 StubRoutines::x86::_P_2_adr = (address)StubRoutines::x86::_P_2;
6128 StubRoutines::x86::_SC_4_adr = (address)StubRoutines::x86::_SC_4;
6129 StubRoutines::x86::_Ctable_adr = (address)StubRoutines::x86::_Ctable;
6130 StubRoutines::x86::_SC_2_adr = (address)StubRoutines::x86::_SC_2;
6131 StubRoutines::x86::_SC_3_adr = (address)StubRoutines::x86::_SC_3;
6132 StubRoutines::x86::_SC_1_adr = (address)StubRoutines::x86::_SC_1;
6133 StubRoutines::x86::_PI_INV_TABLE_adr = (address)StubRoutines::x86::_PI_INV_TABLE;
6134 StubRoutines::x86::_PI_4_adr = (address)StubRoutines::x86::_PI_4;
6135 StubRoutines::x86::_PI32INV_adr = (address)StubRoutines::x86::_PI32INV;
6136 StubRoutines::x86::_SIGN_MASK_adr = (address)StubRoutines::x86::_SIGN_MASK;
6137 StubRoutines::x86::_P_1_adr = (address)StubRoutines::x86::_P_1;
6138 StubRoutines::x86::_P_3_adr = (address)StubRoutines::x86::_P_3;
6139 StubRoutines::x86::_NEG_ZERO_adr = (address)StubRoutines::x86::_NEG_ZERO;
6140 }
6141 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
6142 StubRoutines::_dexp = generate_libmExp();
6143 }
6144 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
6145 StubRoutines::_dlog = generate_libmLog();
6146 }
6147 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
6148 StubRoutines::_dlog10 = generate_libmLog10();
6149 }
6150 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
6151 StubRoutines::_dpow = generate_libmPow();
6152 }
6153 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
6154 StubRoutines::_dsin = generate_libmSin();
6155 }
6156 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
6157 StubRoutines::_dcos = generate_libmCos();
6158 }
6159 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6160 StubRoutines::_dtan = generate_libmTan();
6161 }
6162 }
6163 }
6164
6165 void generate_all() {
6166 // Generates all stubs and initializes the entry points
6167
6168 // These entry points require SharedInfo::stack0 to be set up in
6169 // non-core builds and need to be relocatable, so they each
6170 // fabricate a RuntimeStub internally.
6171 StubRoutines::_throw_AbstractMethodError_entry =
6172 generate_throw_exception("AbstractMethodError throw_exception",
6173 CAST_FROM_FN_PTR(address,
6174 SharedRuntime::
6175 throw_AbstractMethodError));
6176
6177 StubRoutines::_throw_IncompatibleClassChangeError_entry =
6178 generate_throw_exception("IncompatibleClassChangeError throw_exception",
6179 CAST_FROM_FN_PTR(address,
6180 SharedRuntime::
6181 throw_IncompatibleClassChangeError));
6182
6183 StubRoutines::_throw_NullPointerException_at_call_entry =
6184 generate_throw_exception("NullPointerException at call throw_exception",
6185 CAST_FROM_FN_PTR(address,
6186 SharedRuntime::
6187 throw_NullPointerException_at_call));
6188
6189 // entry points that are platform specific
6190 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
6191 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
6192 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
6193 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
6194
6195 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6196 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
6197 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6198 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
6199 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6200 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
6201 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6202 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
6203 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
6204 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
6205 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
6206
6207 // support for verify_oop (must happen after universe_init)
6208 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
6209
6210 // data cache line writeback
6211 StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
6212 StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
6213
6214 // arraycopy stubs used by compilers
6215 generate_arraycopy_stubs();
6216
6217 // don't bother generating these AES intrinsic stubs unless global flag is set
6218 if (UseAESIntrinsics) {
6219 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
6220 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
6221 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
6222 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
6223 if (VM_Version::supports_vaes() && VM_Version::supports_avx512vl() && VM_Version::supports_avx512dq() ) {
6224 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptVectorAESCrypt();
6225 StubRoutines::_electronicCodeBook_encryptAESCrypt = generate_electronicCodeBook_encryptAESCrypt();
6226 StubRoutines::_electronicCodeBook_decryptAESCrypt = generate_electronicCodeBook_decryptAESCrypt();
6227 } else {
6228 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
6229 }
6230 }
6231 if (UseAESCTRIntrinsics) {
6232 if (VM_Version::supports_vaes() && VM_Version::supports_avx512bw() && VM_Version::supports_avx512vl()) {
6233 StubRoutines::x86::_counter_mask_addr = counter_mask_addr();
6234 StubRoutines::_counterMode_AESCrypt = generate_counterMode_VectorAESCrypt();
6235 } else {
6236 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
6237 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
6238 }
6239 }
6240
6241 if (UseSHA1Intrinsics) {
6242 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
6243 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
6244 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
6245 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
6246 }
6247 if (UseSHA256Intrinsics) {
6248 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
6249 char* dst = (char*)StubRoutines::x86::_k256_W;
6250 char* src = (char*)StubRoutines::x86::_k256;
6251 for (int ii = 0; ii < 16; ++ii) {
6252 memcpy(dst + 32 * ii, src + 16 * ii, 16);
6253 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
6254 }
6255 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
6256 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
6257 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
6258 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
6259 }
6260 if (UseSHA512Intrinsics) {
6261 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
6262 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
6263 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
6264 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
6265 }
6266
6267 // Generate GHASH intrinsics code
6268 if (UseGHASHIntrinsics) {
6269 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
6270 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
6271 if (VM_Version::supports_avx()) {
6272 StubRoutines::x86::_ghash_shuffmask_addr = ghash_shufflemask_addr();
6273 StubRoutines::x86::_ghash_poly_addr = ghash_polynomial_addr();
6274 StubRoutines::_ghash_processBlocks = generate_avx_ghash_processBlocks();
6275 } else {
6276 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
6277 }
6278 }
6279
6280 if (UseBASE64Intrinsics) {
6281 StubRoutines::x86::_and_mask = base64_and_mask_addr();
6282 StubRoutines::x86::_bswap_mask = base64_bswap_mask_addr();
6283 StubRoutines::x86::_base64_charset = base64_charset_addr();
6284 StubRoutines::x86::_url_charset = base64url_charset_addr();
6285 StubRoutines::x86::_gather_mask = base64_gather_mask_addr();
6286 StubRoutines::x86::_left_shift_mask = base64_left_shift_mask_addr();
6287 StubRoutines::x86::_right_shift_mask = base64_right_shift_mask_addr();
6288 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
6289 }
6290
6291 // Safefetch stubs.
6292 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
6293 &StubRoutines::_safefetch32_fault_pc,
6294 &StubRoutines::_safefetch32_continuation_pc);
6295 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
6296 &StubRoutines::_safefetchN_fault_pc,
6297 &StubRoutines::_safefetchN_continuation_pc);
6298
6299 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
6300 if (bs_nm != NULL) {
6301 StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
6302 }
6303 #ifdef COMPILER2
6304 if (UseMultiplyToLenIntrinsic) {
6305 StubRoutines::_multiplyToLen = generate_multiplyToLen();
6306 }
6307 if (UseSquareToLenIntrinsic) {
6308 StubRoutines::_squareToLen = generate_squareToLen();
6309 }
6310 if (UseMulAddIntrinsic) {
6311 StubRoutines::_mulAdd = generate_mulAdd();
6312 }
6313 #ifndef _WINDOWS
6314 if (UseMontgomeryMultiplyIntrinsic) {
6315 StubRoutines::_montgomeryMultiply
6316 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
6317 }
6318 if (UseMontgomerySquareIntrinsic) {
6319 StubRoutines::_montgomerySquare
6320 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
6321 }
6322 #endif // WINDOWS
6323 #endif // COMPILER2
6324
6325 if (UseVectorizedMismatchIntrinsic) {
6326 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
6327 }
6328 }
6329
6330 public:
6331 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
6332 if (all) {
6333 generate_all();
6334 } else {
6335 generate_initial();
6336 }
6337 }
6338 }; // end class declaration
6339
6340 #define UCM_TABLE_MAX_ENTRIES 16
6341 void StubGenerator_generate(CodeBuffer* code, bool all) {
6342 if (UnsafeCopyMemory::_table == NULL) {
6343 UnsafeCopyMemory::create_table(UCM_TABLE_MAX_ENTRIES);
6344 }
6345 StubGenerator g(code, all);
6346 }