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