1 /*
2 * Copyright (c) 2003, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "ci/ciUtilities.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/barrierSetNMethod.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "memory/universe.hpp"
34 #include "nativeInst_x86.hpp"
35 #include "oops/instanceOop.hpp"
36 #include "oops/method.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "prims/methodHandles.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/handles.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubCodeGenerator.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/thread.inline.hpp"
46 #ifdef COMPILER2
47 #include "opto/runtime.hpp"
48 #endif
49 #if INCLUDE_ZGC
50 #include "gc/z/zThreadLocalData.hpp"
51 #endif
52
53 // Declaration and definition of StubGenerator (no .hpp file).
54 // For a more detailed description of the stub routine structure
55 // see the comment in stubRoutines.hpp
56
57 #define __ _masm->
58 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
59 #define a__ ((Assembler*)_masm)->
60
61 #ifdef PRODUCT
62 #define BLOCK_COMMENT(str) /* nothing */
63 #else
64 #define BLOCK_COMMENT(str) __ block_comment(str)
65 #endif
66
67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
68 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
69
70 // Stub Code definitions
71
72 class StubGenerator: public StubCodeGenerator {
73 private:
74
75 #ifdef PRODUCT
76 #define inc_counter_np(counter) ((void)0)
77 #else
78 void inc_counter_np_(int& counter) {
79 // This can destroy rscratch1 if counter is far from the code cache
80 __ incrementl(ExternalAddress((address)&counter));
81 }
82 #define inc_counter_np(counter) \
83 BLOCK_COMMENT("inc_counter " #counter); \
84 inc_counter_np_(counter);
85 #endif
86
87 // Call stubs are used to call Java from C
88 //
89 // Linux Arguments:
90 // c_rarg0: call wrapper address address
91 // c_rarg1: result address
92 // c_rarg2: result type BasicType
93 // c_rarg3: method Method*
94 // c_rarg4: (interpreter) entry point address
95 // c_rarg5: parameters intptr_t*
96 // 16(rbp): parameter size (in words) int
97 // 24(rbp): thread Thread*
98 //
99 // [ return_from_Java ] <--- rsp
100 // [ argument word n ]
101 // ...
102 // -12 [ argument word 1 ]
103 // -11 [ saved r15 ] <--- rsp_after_call
104 // -10 [ saved r14 ]
105 // -9 [ saved r13 ]
106 // -8 [ saved r12 ]
107 // -7 [ saved rbx ]
108 // -6 [ call wrapper ]
109 // -5 [ result ]
110 // -4 [ result type ]
111 // -3 [ method ]
112 // -2 [ entry point ]
113 // -1 [ parameters ]
114 // 0 [ saved rbp ] <--- rbp
115 // 1 [ return address ]
116 // 2 [ parameter size ]
117 // 3 [ thread ]
118 //
119 // Windows Arguments:
120 // c_rarg0: call wrapper address address
121 // c_rarg1: result address
122 // c_rarg2: result type BasicType
123 // c_rarg3: method Method*
124 // 48(rbp): (interpreter) entry point address
125 // 56(rbp): parameters intptr_t*
126 // 64(rbp): parameter size (in words) int
127 // 72(rbp): thread Thread*
128 //
129 // [ return_from_Java ] <--- rsp
130 // [ argument word n ]
131 // ...
132 // -60 [ argument word 1 ]
133 // -59 [ saved xmm31 ] <--- rsp after_call
134 // [ saved xmm16-xmm30 ] (EVEX enabled, else the space is blank)
135 // -27 [ saved xmm15 ]
136 // [ saved xmm7-xmm14 ]
137 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
138 // -7 [ saved r15 ]
139 // -6 [ saved r14 ]
140 // -5 [ saved r13 ]
141 // -4 [ saved r12 ]
142 // -3 [ saved rdi ]
143 // -2 [ saved rsi ]
144 // -1 [ saved rbx ]
145 // 0 [ saved rbp ] <--- rbp
146 // 1 [ return address ]
147 // 2 [ call wrapper ]
148 // 3 [ result ]
149 // 4 [ result type ]
150 // 5 [ method ]
151 // 6 [ entry point ]
152 // 7 [ parameters ]
153 // 8 [ parameter size ]
154 // 9 [ thread ]
155 //
156 // Windows reserves the callers stack space for arguments 1-4.
157 // We spill c_rarg0-c_rarg3 to this space.
158
159 // Call stub stack layout word offsets from rbp
160 enum call_stub_layout {
161 #ifdef _WIN64
162 xmm_save_first = 6, // save from xmm6
163 xmm_save_last = 31, // to xmm31
164 xmm_save_base = -9,
165 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
166 r15_off = -7,
167 r14_off = -6,
168 r13_off = -5,
169 r12_off = -4,
170 rdi_off = -3,
171 rsi_off = -2,
172 rbx_off = -1,
173 rbp_off = 0,
174 retaddr_off = 1,
175 call_wrapper_off = 2,
176 result_off = 3,
177 result_type_off = 4,
178 method_off = 5,
179 entry_point_off = 6,
180 parameters_off = 7,
181 parameter_size_off = 8,
182 thread_off = 9
183 #else
184 rsp_after_call_off = -12,
185 mxcsr_off = rsp_after_call_off,
186 r15_off = -11,
187 r14_off = -10,
188 r13_off = -9,
189 r12_off = -8,
190 rbx_off = -7,
191 call_wrapper_off = -6,
192 result_off = -5,
193 result_type_off = -4,
194 method_off = -3,
195 entry_point_off = -2,
196 parameters_off = -1,
197 rbp_off = 0,
198 retaddr_off = 1,
199 parameter_size_off = 2,
200 thread_off = 3
201 #endif
202 };
203
204 #ifdef _WIN64
205 Address xmm_save(int reg) {
206 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
207 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
208 }
209 #endif
210
211 address generate_call_stub(address& return_address) {
212 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
213 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
214 "adjust this code");
215 StubCodeMark mark(this, "StubRoutines", "call_stub");
216 address start = __ pc();
217
218 // same as in generate_catch_exception()!
219 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
220
221 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
222 const Address result (rbp, result_off * wordSize);
223 const Address result_type (rbp, result_type_off * wordSize);
224 const Address method (rbp, method_off * wordSize);
225 const Address entry_point (rbp, entry_point_off * wordSize);
226 const Address parameters (rbp, parameters_off * wordSize);
227 const Address parameter_size(rbp, parameter_size_off * wordSize);
228
229 // same as in generate_catch_exception()!
230 const Address thread (rbp, thread_off * wordSize);
231
232 const Address r15_save(rbp, r15_off * wordSize);
233 const Address r14_save(rbp, r14_off * wordSize);
234 const Address r13_save(rbp, r13_off * wordSize);
235 const Address r12_save(rbp, r12_off * wordSize);
236 const Address rbx_save(rbp, rbx_off * wordSize);
237
238 // stub code
239 __ enter();
240 __ subptr(rsp, -rsp_after_call_off * wordSize);
241
242 // save register parameters
243 #ifndef _WIN64
244 __ movptr(parameters, c_rarg5); // parameters
245 __ movptr(entry_point, c_rarg4); // entry_point
246 #endif
247
248 __ movptr(method, c_rarg3); // method
249 __ movl(result_type, c_rarg2); // result type
250 __ movptr(result, c_rarg1); // result
251 __ movptr(call_wrapper, c_rarg0); // call wrapper
252
253 // save regs belonging to calling function
254 __ movptr(rbx_save, rbx);
255 __ movptr(r12_save, r12);
256 __ movptr(r13_save, r13);
257 __ movptr(r14_save, r14);
258 __ movptr(r15_save, r15);
259
260 #ifdef _WIN64
261 int last_reg = 15;
262 if (UseAVX > 2) {
263 last_reg = 31;
264 }
265 if (VM_Version::supports_evex()) {
266 for (int i = xmm_save_first; i <= last_reg; i++) {
267 __ vextractf32x4(xmm_save(i), as_XMMRegister(i), 0);
268 }
269 } else {
270 for (int i = xmm_save_first; i <= last_reg; i++) {
271 __ movdqu(xmm_save(i), as_XMMRegister(i));
272 }
273 }
274
275 const Address rdi_save(rbp, rdi_off * wordSize);
276 const Address rsi_save(rbp, rsi_off * wordSize);
277
278 __ movptr(rsi_save, rsi);
279 __ movptr(rdi_save, rdi);
280 #else
281 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
282 {
283 Label skip_ldmx;
284 __ stmxcsr(mxcsr_save);
285 __ movl(rax, mxcsr_save);
286 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
287 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
288 __ cmp32(rax, mxcsr_std);
289 __ jcc(Assembler::equal, skip_ldmx);
290 __ ldmxcsr(mxcsr_std);
291 __ bind(skip_ldmx);
292 }
293 #endif
294
295 // Load up thread register
296 __ movptr(r15_thread, thread);
297 __ reinit_heapbase();
298
299 #ifdef ASSERT
300 // make sure we have no pending exceptions
301 {
302 Label L;
303 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
304 __ jcc(Assembler::equal, L);
305 __ stop("StubRoutines::call_stub: entered with pending exception");
306 __ bind(L);
307 }
308 #endif
309
310 // pass parameters if any
311 BLOCK_COMMENT("pass parameters if any");
312 Label parameters_done;
313 __ movl(c_rarg3, parameter_size);
314 __ testl(c_rarg3, c_rarg3);
315 __ jcc(Assembler::zero, parameters_done);
316
317 Label loop;
318 __ movptr(c_rarg2, parameters); // parameter pointer
319 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
320 __ BIND(loop);
321 __ movptr(rax, Address(c_rarg2, 0));// get parameter
322 __ addptr(c_rarg2, wordSize); // advance to next parameter
323 __ decrementl(c_rarg1); // decrement counter
324 __ push(rax); // pass parameter
325 __ jcc(Assembler::notZero, loop);
326
327 // call Java function
328 __ BIND(parameters_done);
329 __ movptr(rbx, method); // get Method*
330 __ movptr(c_rarg1, entry_point); // get entry_point
331 __ mov(r13, rsp); // set sender sp
332 BLOCK_COMMENT("call Java function");
333 __ call(c_rarg1);
334
335 BLOCK_COMMENT("call_stub_return_address:");
336 return_address = __ pc();
337
338 // store result depending on type (everything that is not
339 // T_OBJECT, T_VALUETYPE, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
340 __ movptr(r13, result);
341 Label is_long, is_float, is_double, is_value, exit;
342 __ movl(rbx, result_type);
343 __ cmpl(rbx, T_OBJECT);
344 __ jcc(Assembler::equal, is_long);
345 __ cmpl(rbx, T_VALUETYPE);
346 __ jcc(Assembler::equal, is_value);
347 __ cmpl(rbx, T_LONG);
348 __ jcc(Assembler::equal, is_long);
349 __ cmpl(rbx, T_FLOAT);
350 __ jcc(Assembler::equal, is_float);
351 __ cmpl(rbx, T_DOUBLE);
352 __ jcc(Assembler::equal, is_double);
353
354 // handle T_INT case
355 __ movl(Address(r13, 0), rax);
356
357 __ BIND(exit);
358
359 // pop parameters
360 __ lea(rsp, rsp_after_call);
361
362 #ifdef ASSERT
363 // verify that threads correspond
364 {
365 Label L1, L2, L3;
366 __ cmpptr(r15_thread, thread);
367 __ jcc(Assembler::equal, L1);
368 __ stop("StubRoutines::call_stub: r15_thread is corrupted");
369 __ bind(L1);
370 __ get_thread(rbx);
371 __ cmpptr(r15_thread, thread);
372 __ jcc(Assembler::equal, L2);
373 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
374 __ bind(L2);
375 __ cmpptr(r15_thread, rbx);
376 __ jcc(Assembler::equal, L3);
377 __ stop("StubRoutines::call_stub: threads must correspond");
378 __ bind(L3);
379 }
380 #endif
381
382 // restore regs belonging to calling function
383 #ifdef _WIN64
384 // emit the restores for xmm regs
385 if (VM_Version::supports_evex()) {
386 for (int i = xmm_save_first; i <= last_reg; i++) {
387 __ vinsertf32x4(as_XMMRegister(i), as_XMMRegister(i), xmm_save(i), 0);
388 }
389 } else {
390 for (int i = xmm_save_first; i <= last_reg; i++) {
391 __ movdqu(as_XMMRegister(i), xmm_save(i));
392 }
393 }
394 #endif
395 __ movptr(r15, r15_save);
396 __ movptr(r14, r14_save);
397 __ movptr(r13, r13_save);
398 __ movptr(r12, r12_save);
399 __ movptr(rbx, rbx_save);
400
401 #ifdef _WIN64
402 __ movptr(rdi, rdi_save);
403 __ movptr(rsi, rsi_save);
404 #else
405 __ ldmxcsr(mxcsr_save);
406 #endif
407
408 // restore rsp
409 __ addptr(rsp, -rsp_after_call_off * wordSize);
410
411 // return
412 __ vzeroupper();
413 __ pop(rbp);
414 __ ret(0);
415
416 // handle return types different from T_INT
417 __ BIND(is_value);
418 if (ValueTypeReturnedAsFields) {
419 // Check for flattened return value
420 __ testptr(rax, 1);
421 __ jcc(Assembler::zero, is_long);
422 // Initialize pre-allocated buffer
423 __ mov(rbx, rax);
424 __ andptr(rbx, -2);
425 __ movptr(rbx, Address(rbx, InstanceKlass::adr_valueklass_fixed_block_offset()));
426 __ movptr(rbx, Address(rbx, ValueKlass::pack_handler_offset()));
427 __ movptr(rax, Address(r13, 0));
428 __ call(rbx);
429 __ jmp(exit);
430 }
431 __ BIND(is_long);
432 __ movq(Address(r13, 0), rax);
433 __ jmp(exit);
434
435 __ BIND(is_float);
436 __ movflt(Address(r13, 0), xmm0);
437 __ jmp(exit);
438
439 __ BIND(is_double);
440 __ movdbl(Address(r13, 0), xmm0);
441 __ jmp(exit);
442
443 return start;
444 }
445
446 // Return point for a Java call if there's an exception thrown in
447 // Java code. The exception is caught and transformed into a
448 // pending exception stored in JavaThread that can be tested from
449 // within the VM.
450 //
451 // Note: Usually the parameters are removed by the callee. In case
452 // of an exception crossing an activation frame boundary, that is
453 // not the case if the callee is compiled code => need to setup the
454 // rsp.
455 //
456 // rax: exception oop
457
458 address generate_catch_exception() {
459 StubCodeMark mark(this, "StubRoutines", "catch_exception");
460 address start = __ pc();
461
462 // same as in generate_call_stub():
463 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
464 const Address thread (rbp, thread_off * wordSize);
465
466 #ifdef ASSERT
467 // verify that threads correspond
468 {
469 Label L1, L2, L3;
470 __ cmpptr(r15_thread, thread);
471 __ jcc(Assembler::equal, L1);
472 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
473 __ bind(L1);
474 __ get_thread(rbx);
475 __ cmpptr(r15_thread, thread);
476 __ jcc(Assembler::equal, L2);
477 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
478 __ bind(L2);
479 __ cmpptr(r15_thread, rbx);
480 __ jcc(Assembler::equal, L3);
481 __ stop("StubRoutines::catch_exception: threads must correspond");
482 __ bind(L3);
483 }
484 #endif
485
486 // set pending exception
487 __ verify_oop(rax);
488
489 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
490 __ lea(rscratch1, ExternalAddress((address)__FILE__));
491 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
492 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
493
494 // complete return to VM
495 assert(StubRoutines::_call_stub_return_address != NULL,
496 "_call_stub_return_address must have been generated before");
497 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
498
499 return start;
500 }
501
502 // Continuation point for runtime calls returning with a pending
503 // exception. The pending exception check happened in the runtime
504 // or native call stub. The pending exception in Thread is
505 // converted into a Java-level exception.
506 //
507 // Contract with Java-level exception handlers:
508 // rax: exception
509 // rdx: throwing pc
510 //
511 // NOTE: At entry of this stub, exception-pc must be on stack !!
512
513 address generate_forward_exception() {
514 StubCodeMark mark(this, "StubRoutines", "forward exception");
515 address start = __ pc();
516
517 // Upon entry, the sp points to the return address returning into
518 // Java (interpreted or compiled) code; i.e., the return address
519 // becomes the throwing pc.
520 //
521 // Arguments pushed before the runtime call are still on the stack
522 // but the exception handler will reset the stack pointer ->
523 // ignore them. A potential result in registers can be ignored as
524 // well.
525
526 #ifdef ASSERT
527 // make sure this code is only executed if there is a pending exception
528 {
529 Label L;
530 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
531 __ jcc(Assembler::notEqual, L);
532 __ stop("StubRoutines::forward exception: no pending exception (1)");
533 __ bind(L);
534 }
535 #endif
536
537 // compute exception handler into rbx
538 __ movptr(c_rarg0, Address(rsp, 0));
539 BLOCK_COMMENT("call exception_handler_for_return_address");
540 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
541 SharedRuntime::exception_handler_for_return_address),
542 r15_thread, c_rarg0);
543 __ mov(rbx, rax);
544
545 // setup rax & rdx, remove return address & clear pending exception
546 __ pop(rdx);
547 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
548 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
549
550 #ifdef ASSERT
551 // make sure exception is set
552 {
553 Label L;
554 __ testptr(rax, rax);
555 __ jcc(Assembler::notEqual, L);
556 __ stop("StubRoutines::forward exception: no pending exception (2)");
557 __ bind(L);
558 }
559 #endif
560
561 // continue at exception handler (return address removed)
562 // rax: exception
563 // rbx: exception handler
564 // rdx: throwing pc
565 __ verify_oop(rax);
566 __ jmp(rbx);
567
568 return start;
569 }
570
571 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
572 //
573 // Arguments :
574 // c_rarg0: exchange_value
575 // c_rarg0: dest
576 //
577 // Result:
578 // *dest <- ex, return (orig *dest)
579 address generate_atomic_xchg() {
580 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
581 address start = __ pc();
582
583 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
584 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
585 __ ret(0);
586
587 return start;
588 }
589
590 // Support for intptr_t atomic::xchg_long(jlong exchange_value, volatile jlong* dest)
591 //
592 // Arguments :
593 // c_rarg0: exchange_value
594 // c_rarg1: dest
595 //
596 // Result:
597 // *dest <- ex, return (orig *dest)
598 address generate_atomic_xchg_long() {
599 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_long");
600 address start = __ pc();
601
602 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
603 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
604 __ ret(0);
605
606 return start;
607 }
608
609 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
610 // jint compare_value)
611 //
612 // Arguments :
613 // c_rarg0: exchange_value
614 // c_rarg1: dest
615 // c_rarg2: compare_value
616 //
617 // Result:
618 // if ( compare_value == *dest ) {
619 // *dest = exchange_value
620 // return compare_value;
621 // else
622 // return *dest;
623 address generate_atomic_cmpxchg() {
624 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
625 address start = __ pc();
626
627 __ movl(rax, c_rarg2);
628 __ lock();
629 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
630 __ ret(0);
631
632 return start;
633 }
634
635 // Support for int8_t atomic::atomic_cmpxchg(int8_t exchange_value, volatile int8_t* dest,
636 // int8_t compare_value)
637 //
638 // Arguments :
639 // c_rarg0: exchange_value
640 // c_rarg1: dest
641 // c_rarg2: compare_value
642 //
643 // Result:
644 // if ( compare_value == *dest ) {
645 // *dest = exchange_value
646 // return compare_value;
647 // else
648 // return *dest;
649 address generate_atomic_cmpxchg_byte() {
650 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_byte");
651 address start = __ pc();
652
653 __ movsbq(rax, c_rarg2);
654 __ lock();
655 __ cmpxchgb(c_rarg0, Address(c_rarg1, 0));
656 __ ret(0);
657
658 return start;
659 }
660
661 // Support for int64_t atomic::atomic_cmpxchg(int64_t exchange_value,
662 // volatile int64_t* dest,
663 // int64_t compare_value)
664 // Arguments :
665 // c_rarg0: exchange_value
666 // c_rarg1: dest
667 // c_rarg2: compare_value
668 //
669 // Result:
670 // if ( compare_value == *dest ) {
671 // *dest = exchange_value
672 // return compare_value;
673 // else
674 // return *dest;
675 address generate_atomic_cmpxchg_long() {
676 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
677 address start = __ pc();
678
679 __ movq(rax, c_rarg2);
680 __ lock();
681 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
682 __ ret(0);
683
684 return start;
685 }
686
687 // Support for jint atomic::add(jint add_value, volatile jint* dest)
688 //
689 // Arguments :
690 // c_rarg0: add_value
691 // c_rarg1: dest
692 //
693 // Result:
694 // *dest += add_value
695 // return *dest;
696 address generate_atomic_add() {
697 StubCodeMark mark(this, "StubRoutines", "atomic_add");
698 address start = __ pc();
699
700 __ movl(rax, c_rarg0);
701 __ lock();
702 __ xaddl(Address(c_rarg1, 0), c_rarg0);
703 __ addl(rax, c_rarg0);
704 __ ret(0);
705
706 return start;
707 }
708
709 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
710 //
711 // Arguments :
712 // c_rarg0: add_value
713 // c_rarg1: dest
714 //
715 // Result:
716 // *dest += add_value
717 // return *dest;
718 address generate_atomic_add_long() {
719 StubCodeMark mark(this, "StubRoutines", "atomic_add_long");
720 address start = __ pc();
721
722 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
723 __ lock();
724 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
725 __ addptr(rax, c_rarg0);
726 __ ret(0);
727
728 return start;
729 }
730
731 // Support for intptr_t OrderAccess::fence()
732 //
733 // Arguments :
734 //
735 // Result:
736 address generate_orderaccess_fence() {
737 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
738 address start = __ pc();
739 __ membar(Assembler::StoreLoad);
740 __ ret(0);
741
742 return start;
743 }
744
745 // Support for intptr_t get_previous_fp()
746 //
747 // This routine is used to find the previous frame pointer for the
748 // caller (current_frame_guess). This is used as part of debugging
749 // ps() is seemingly lost trying to find frames.
750 // This code assumes that caller current_frame_guess) has a frame.
751 address generate_get_previous_fp() {
752 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
753 const Address old_fp(rbp, 0);
754 const Address older_fp(rax, 0);
755 address start = __ pc();
756
757 __ enter();
758 __ movptr(rax, old_fp); // callers fp
759 __ movptr(rax, older_fp); // the frame for ps()
760 __ pop(rbp);
761 __ ret(0);
762
763 return start;
764 }
765
766 // Support for intptr_t get_previous_sp()
767 //
768 // This routine is used to find the previous stack pointer for the
769 // caller.
770 address generate_get_previous_sp() {
771 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
772 address start = __ pc();
773
774 __ movptr(rax, rsp);
775 __ addptr(rax, 8); // return address is at the top of the stack.
776 __ ret(0);
777
778 return start;
779 }
780
781 //----------------------------------------------------------------------------------------------------
782 // Support for void verify_mxcsr()
783 //
784 // This routine is used with -Xcheck:jni to verify that native
785 // JNI code does not return to Java code without restoring the
786 // MXCSR register to our expected state.
787
788 address generate_verify_mxcsr() {
789 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
790 address start = __ pc();
791
792 const Address mxcsr_save(rsp, 0);
793
794 if (CheckJNICalls) {
795 Label ok_ret;
796 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
797 __ push(rax);
798 __ subptr(rsp, wordSize); // allocate a temp location
799 __ stmxcsr(mxcsr_save);
800 __ movl(rax, mxcsr_save);
801 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
802 __ cmp32(rax, mxcsr_std);
803 __ jcc(Assembler::equal, ok_ret);
804
805 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
806
807 __ ldmxcsr(mxcsr_std);
808
809 __ bind(ok_ret);
810 __ addptr(rsp, wordSize);
811 __ pop(rax);
812 }
813
814 __ ret(0);
815
816 return start;
817 }
818
819 address generate_f2i_fixup() {
820 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
821 Address inout(rsp, 5 * wordSize); // return address + 4 saves
822
823 address start = __ pc();
824
825 Label L;
826
827 __ push(rax);
828 __ push(c_rarg3);
829 __ push(c_rarg2);
830 __ push(c_rarg1);
831
832 __ movl(rax, 0x7f800000);
833 __ xorl(c_rarg3, c_rarg3);
834 __ movl(c_rarg2, inout);
835 __ movl(c_rarg1, c_rarg2);
836 __ andl(c_rarg1, 0x7fffffff);
837 __ cmpl(rax, c_rarg1); // NaN? -> 0
838 __ jcc(Assembler::negative, L);
839 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
840 __ movl(c_rarg3, 0x80000000);
841 __ movl(rax, 0x7fffffff);
842 __ cmovl(Assembler::positive, c_rarg3, rax);
843
844 __ bind(L);
845 __ movptr(inout, c_rarg3);
846
847 __ pop(c_rarg1);
848 __ pop(c_rarg2);
849 __ pop(c_rarg3);
850 __ pop(rax);
851
852 __ ret(0);
853
854 return start;
855 }
856
857 address generate_f2l_fixup() {
858 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
859 Address inout(rsp, 5 * wordSize); // return address + 4 saves
860 address start = __ pc();
861
862 Label L;
863
864 __ push(rax);
865 __ push(c_rarg3);
866 __ push(c_rarg2);
867 __ push(c_rarg1);
868
869 __ movl(rax, 0x7f800000);
870 __ xorl(c_rarg3, c_rarg3);
871 __ movl(c_rarg2, inout);
872 __ movl(c_rarg1, c_rarg2);
873 __ andl(c_rarg1, 0x7fffffff);
874 __ cmpl(rax, c_rarg1); // NaN? -> 0
875 __ jcc(Assembler::negative, L);
876 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
877 __ mov64(c_rarg3, 0x8000000000000000);
878 __ mov64(rax, 0x7fffffffffffffff);
879 __ cmov(Assembler::positive, c_rarg3, rax);
880
881 __ bind(L);
882 __ movptr(inout, c_rarg3);
883
884 __ pop(c_rarg1);
885 __ pop(c_rarg2);
886 __ pop(c_rarg3);
887 __ pop(rax);
888
889 __ ret(0);
890
891 return start;
892 }
893
894 address generate_d2i_fixup() {
895 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
896 Address inout(rsp, 6 * wordSize); // return address + 5 saves
897
898 address start = __ pc();
899
900 Label L;
901
902 __ push(rax);
903 __ push(c_rarg3);
904 __ push(c_rarg2);
905 __ push(c_rarg1);
906 __ push(c_rarg0);
907
908 __ movl(rax, 0x7ff00000);
909 __ movq(c_rarg2, inout);
910 __ movl(c_rarg3, c_rarg2);
911 __ mov(c_rarg1, c_rarg2);
912 __ mov(c_rarg0, c_rarg2);
913 __ negl(c_rarg3);
914 __ shrptr(c_rarg1, 0x20);
915 __ orl(c_rarg3, c_rarg2);
916 __ andl(c_rarg1, 0x7fffffff);
917 __ xorl(c_rarg2, c_rarg2);
918 __ shrl(c_rarg3, 0x1f);
919 __ orl(c_rarg1, c_rarg3);
920 __ cmpl(rax, c_rarg1);
921 __ jcc(Assembler::negative, L); // NaN -> 0
922 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
923 __ movl(c_rarg2, 0x80000000);
924 __ movl(rax, 0x7fffffff);
925 __ cmov(Assembler::positive, c_rarg2, rax);
926
927 __ bind(L);
928 __ movptr(inout, c_rarg2);
929
930 __ pop(c_rarg0);
931 __ pop(c_rarg1);
932 __ pop(c_rarg2);
933 __ pop(c_rarg3);
934 __ pop(rax);
935
936 __ ret(0);
937
938 return start;
939 }
940
941 address generate_d2l_fixup() {
942 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
943 Address inout(rsp, 6 * wordSize); // return address + 5 saves
944
945 address start = __ pc();
946
947 Label L;
948
949 __ push(rax);
950 __ push(c_rarg3);
951 __ push(c_rarg2);
952 __ push(c_rarg1);
953 __ push(c_rarg0);
954
955 __ movl(rax, 0x7ff00000);
956 __ movq(c_rarg2, inout);
957 __ movl(c_rarg3, c_rarg2);
958 __ mov(c_rarg1, c_rarg2);
959 __ mov(c_rarg0, c_rarg2);
960 __ negl(c_rarg3);
961 __ shrptr(c_rarg1, 0x20);
962 __ orl(c_rarg3, c_rarg2);
963 __ andl(c_rarg1, 0x7fffffff);
964 __ xorl(c_rarg2, c_rarg2);
965 __ shrl(c_rarg3, 0x1f);
966 __ orl(c_rarg1, c_rarg3);
967 __ cmpl(rax, c_rarg1);
968 __ jcc(Assembler::negative, L); // NaN -> 0
969 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
970 __ mov64(c_rarg2, 0x8000000000000000);
971 __ mov64(rax, 0x7fffffffffffffff);
972 __ cmovq(Assembler::positive, c_rarg2, rax);
973
974 __ bind(L);
975 __ movq(inout, c_rarg2);
976
977 __ pop(c_rarg0);
978 __ pop(c_rarg1);
979 __ pop(c_rarg2);
980 __ pop(c_rarg3);
981 __ pop(rax);
982
983 __ ret(0);
984
985 return start;
986 }
987
988 address generate_fp_mask(const char *stub_name, int64_t mask) {
989 __ align(CodeEntryAlignment);
990 StubCodeMark mark(this, "StubRoutines", stub_name);
991 address start = __ pc();
992
993 __ emit_data64( mask, relocInfo::none );
994 __ emit_data64( mask, relocInfo::none );
995
996 return start;
997 }
998
999 address generate_vector_mask(const char *stub_name, int64_t mask) {
1000 __ align(CodeEntryAlignment);
1001 StubCodeMark mark(this, "StubRoutines", stub_name);
1002 address start = __ pc();
1003
1004 __ emit_data64(mask, relocInfo::none);
1005 __ emit_data64(mask, relocInfo::none);
1006 __ emit_data64(mask, relocInfo::none);
1007 __ emit_data64(mask, relocInfo::none);
1008 __ emit_data64(mask, relocInfo::none);
1009 __ emit_data64(mask, relocInfo::none);
1010 __ emit_data64(mask, relocInfo::none);
1011 __ emit_data64(mask, relocInfo::none);
1012
1013 return start;
1014 }
1015
1016 address generate_vector_byte_perm_mask(const char *stub_name) {
1017 __ align(CodeEntryAlignment);
1018 StubCodeMark mark(this, "StubRoutines", stub_name);
1019 address start = __ pc();
1020
1021 __ emit_data64(0x0000000000000001, relocInfo::none);
1022 __ emit_data64(0x0000000000000003, relocInfo::none);
1023 __ emit_data64(0x0000000000000005, relocInfo::none);
1024 __ emit_data64(0x0000000000000007, relocInfo::none);
1025 __ emit_data64(0x0000000000000000, relocInfo::none);
1026 __ emit_data64(0x0000000000000002, relocInfo::none);
1027 __ emit_data64(0x0000000000000004, relocInfo::none);
1028 __ emit_data64(0x0000000000000006, relocInfo::none);
1029
1030 return start;
1031 }
1032
1033 // Non-destructive plausibility checks for oops
1034 //
1035 // Arguments:
1036 // all args on stack!
1037 //
1038 // Stack after saving c_rarg3:
1039 // [tos + 0]: saved c_rarg3
1040 // [tos + 1]: saved c_rarg2
1041 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1042 // [tos + 3]: saved flags
1043 // [tos + 4]: return address
1044 // * [tos + 5]: error message (char*)
1045 // * [tos + 6]: object to verify (oop)
1046 // * [tos + 7]: saved rax - saved by caller and bashed
1047 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1048 // * = popped on exit
1049 address generate_verify_oop() {
1050 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1051 address start = __ pc();
1052
1053 Label exit, error, in_Java_heap;
1054
1055 __ pushf();
1056 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
1057
1058 __ push(r12);
1059
1060 // save c_rarg2 and c_rarg3
1061 __ push(c_rarg2);
1062 __ push(c_rarg3);
1063
1064 enum {
1065 // After previous pushes.
1066 oop_to_verify = 6 * wordSize,
1067 saved_rax = 7 * wordSize,
1068 saved_r10 = 8 * wordSize,
1069
1070 // Before the call to MacroAssembler::debug(), see below.
1071 return_addr = 16 * wordSize,
1072 error_msg = 17 * wordSize
1073 };
1074
1075 // get object
1076 __ movptr(rax, Address(rsp, oop_to_verify));
1077
1078 // make sure object is 'reasonable'
1079 __ testptr(rax, rax);
1080 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1081
1082 #if INCLUDE_ZGC
1083 if (UseZGC) {
1084 // Check if metadata bits indicate a bad oop
1085 __ testptr(rax, Address(r15_thread, ZThreadLocalData::address_bad_mask_offset()));
1086 __ jcc(Assembler::notZero, error);
1087 }
1088 #endif
1089
1090 // Check if the oop is in the right area of memory
1091 __ movptr(c_rarg2, rax);
1092 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1093 __ andptr(c_rarg2, c_rarg3);
1094 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1095 __ cmpptr(c_rarg2, c_rarg3);
1096 __ jcc(Assembler::zero, in_Java_heap);
1097 // Not in Java heap, but could be valid if it's a bufferable value type
1098 __ load_klass(c_rarg2, rax);
1099 __ movbool(c_rarg2, Address(c_rarg2, InstanceKlass::extra_flags_offset()));
1100 __ andptr(c_rarg2, InstanceKlass::_extra_is_bufferable);
1101 __ testbool(c_rarg2);
1102 __ jcc(Assembler::zero, error);
1103 __ bind(in_Java_heap);
1104
1105 // set r12 to heapbase for load_klass()
1106 __ reinit_heapbase();
1107
1108 // make sure klass is 'reasonable', which is not zero.
1109 __ load_klass(rax, rax); // get klass
1110 __ testptr(rax, rax);
1111 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1112
1113 // return if everything seems ok
1114 __ bind(exit);
1115 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1116 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1117 __ pop(c_rarg3); // restore c_rarg3
1118 __ pop(c_rarg2); // restore c_rarg2
1119 __ pop(r12); // restore r12
1120 __ popf(); // restore flags
1121 __ ret(4 * wordSize); // pop caller saved stuff
1122
1123 // handle errors
1124 __ bind(error);
1125 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1126 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1127 __ pop(c_rarg3); // get saved c_rarg3 back
1128 __ pop(c_rarg2); // get saved c_rarg2 back
1129 __ pop(r12); // get saved r12 back
1130 __ popf(); // get saved flags off stack --
1131 // will be ignored
1132
1133 __ pusha(); // push registers
1134 // (rip is already
1135 // already pushed)
1136 // debug(char* msg, int64_t pc, int64_t regs[])
1137 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1138 // pushed all the registers, so now the stack looks like:
1139 // [tos + 0] 16 saved registers
1140 // [tos + 16] return address
1141 // * [tos + 17] error message (char*)
1142 // * [tos + 18] object to verify (oop)
1143 // * [tos + 19] saved rax - saved by caller and bashed
1144 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1145 // * = popped on exit
1146
1147 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1148 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1149 __ movq(c_rarg2, rsp); // pass address of regs on stack
1150 __ mov(r12, rsp); // remember rsp
1151 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1152 __ andptr(rsp, -16); // align stack as required by ABI
1153 BLOCK_COMMENT("call MacroAssembler::debug");
1154 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1155 __ mov(rsp, r12); // restore rsp
1156 __ popa(); // pop registers (includes r12)
1157 __ ret(4 * wordSize); // pop caller saved stuff
1158
1159 return start;
1160 }
1161
1162 //
1163 // Verify that a register contains clean 32-bits positive value
1164 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1165 //
1166 // Input:
1167 // Rint - 32-bits value
1168 // Rtmp - scratch
1169 //
1170 void assert_clean_int(Register Rint, Register Rtmp) {
1171 #ifdef ASSERT
1172 Label L;
1173 assert_different_registers(Rtmp, Rint);
1174 __ movslq(Rtmp, Rint);
1175 __ cmpq(Rtmp, Rint);
1176 __ jcc(Assembler::equal, L);
1177 __ stop("high 32-bits of int value are not 0");
1178 __ bind(L);
1179 #endif
1180 }
1181
1182 // Generate overlap test for array copy stubs
1183 //
1184 // Input:
1185 // c_rarg0 - from
1186 // c_rarg1 - to
1187 // c_rarg2 - element count
1188 //
1189 // Output:
1190 // rax - &from[element count - 1]
1191 //
1192 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1193 assert(no_overlap_target != NULL, "must be generated");
1194 array_overlap_test(no_overlap_target, NULL, sf);
1195 }
1196 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1197 array_overlap_test(NULL, &L_no_overlap, sf);
1198 }
1199 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1200 const Register from = c_rarg0;
1201 const Register to = c_rarg1;
1202 const Register count = c_rarg2;
1203 const Register end_from = rax;
1204
1205 __ cmpptr(to, from);
1206 __ lea(end_from, Address(from, count, sf, 0));
1207 if (NOLp == NULL) {
1208 ExternalAddress no_overlap(no_overlap_target);
1209 __ jump_cc(Assembler::belowEqual, no_overlap);
1210 __ cmpptr(to, end_from);
1211 __ jump_cc(Assembler::aboveEqual, no_overlap);
1212 } else {
1213 __ jcc(Assembler::belowEqual, (*NOLp));
1214 __ cmpptr(to, end_from);
1215 __ jcc(Assembler::aboveEqual, (*NOLp));
1216 }
1217 }
1218
1219 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1220 //
1221 // Outputs:
1222 // rdi - rcx
1223 // rsi - rdx
1224 // rdx - r8
1225 // rcx - r9
1226 //
1227 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1228 // are non-volatile. r9 and r10 should not be used by the caller.
1229 //
1230 DEBUG_ONLY(bool regs_in_thread;)
1231
1232 void setup_arg_regs(int nargs = 3) {
1233 const Register saved_rdi = r9;
1234 const Register saved_rsi = r10;
1235 assert(nargs == 3 || nargs == 4, "else fix");
1236 #ifdef _WIN64
1237 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1238 "unexpected argument registers");
1239 if (nargs >= 4)
1240 __ mov(rax, r9); // r9 is also saved_rdi
1241 __ movptr(saved_rdi, rdi);
1242 __ movptr(saved_rsi, rsi);
1243 __ mov(rdi, rcx); // c_rarg0
1244 __ mov(rsi, rdx); // c_rarg1
1245 __ mov(rdx, r8); // c_rarg2
1246 if (nargs >= 4)
1247 __ mov(rcx, rax); // c_rarg3 (via rax)
1248 #else
1249 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1250 "unexpected argument registers");
1251 #endif
1252 DEBUG_ONLY(regs_in_thread = false;)
1253 }
1254
1255 void restore_arg_regs() {
1256 assert(!regs_in_thread, "wrong call to restore_arg_regs");
1257 const Register saved_rdi = r9;
1258 const Register saved_rsi = r10;
1259 #ifdef _WIN64
1260 __ movptr(rdi, saved_rdi);
1261 __ movptr(rsi, saved_rsi);
1262 #endif
1263 }
1264
1265 // This is used in places where r10 is a scratch register, and can
1266 // be adapted if r9 is needed also.
1267 void setup_arg_regs_using_thread() {
1268 const Register saved_r15 = r9;
1269 #ifdef _WIN64
1270 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1271 __ get_thread(r15_thread);
1272 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1273 "unexpected argument registers");
1274 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1275 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1276
1277 __ mov(rdi, rcx); // c_rarg0
1278 __ mov(rsi, rdx); // c_rarg1
1279 __ mov(rdx, r8); // c_rarg2
1280 #else
1281 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1282 "unexpected argument registers");
1283 #endif
1284 DEBUG_ONLY(regs_in_thread = true;)
1285 }
1286
1287 void restore_arg_regs_using_thread() {
1288 assert(regs_in_thread, "wrong call to restore_arg_regs");
1289 const Register saved_r15 = r9;
1290 #ifdef _WIN64
1291 __ get_thread(r15_thread);
1292 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1293 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1294 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1295 #endif
1296 }
1297
1298 // Copy big chunks forward
1299 //
1300 // Inputs:
1301 // end_from - source arrays end address
1302 // end_to - destination array end address
1303 // qword_count - 64-bits element count, negative
1304 // to - scratch
1305 // L_copy_bytes - entry label
1306 // L_copy_8_bytes - exit label
1307 //
1308 void copy_bytes_forward(Register end_from, Register end_to,
1309 Register qword_count, Register to,
1310 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1311 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1312 Label L_loop;
1313 __ align(OptoLoopAlignment);
1314 if (UseUnalignedLoadStores) {
1315 Label L_end;
1316 // Copy 64-bytes per iteration
1317 __ BIND(L_loop);
1318 if (UseAVX > 2) {
1319 __ evmovdqul(xmm0, Address(end_from, qword_count, Address::times_8, -56), Assembler::AVX_512bit);
1320 __ evmovdqul(Address(end_to, qword_count, Address::times_8, -56), xmm0, Assembler::AVX_512bit);
1321 } else if (UseAVX == 2) {
1322 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1323 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1324 __ vmovdqu(xmm1, Address(end_from, qword_count, Address::times_8, -24));
1325 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm1);
1326 } else {
1327 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -56));
1328 __ movdqu(Address(end_to, qword_count, Address::times_8, -56), xmm0);
1329 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, -40));
1330 __ movdqu(Address(end_to, qword_count, Address::times_8, -40), xmm1);
1331 __ movdqu(xmm2, Address(end_from, qword_count, Address::times_8, -24));
1332 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm2);
1333 __ movdqu(xmm3, Address(end_from, qword_count, Address::times_8, - 8));
1334 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm3);
1335 }
1336 __ BIND(L_copy_bytes);
1337 __ addptr(qword_count, 8);
1338 __ jcc(Assembler::lessEqual, L_loop);
1339 __ subptr(qword_count, 4); // sub(8) and add(4)
1340 __ jccb(Assembler::greater, L_end);
1341 // Copy trailing 32 bytes
1342 if (UseAVX >= 2) {
1343 __ vmovdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1344 __ vmovdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1345 } else {
1346 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1347 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1348 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1349 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1350 }
1351 __ addptr(qword_count, 4);
1352 __ BIND(L_end);
1353 if (UseAVX >= 2) {
1354 // clean upper bits of YMM registers
1355 __ vpxor(xmm0, xmm0);
1356 __ vpxor(xmm1, xmm1);
1357 }
1358 } else {
1359 // Copy 32-bytes per iteration
1360 __ BIND(L_loop);
1361 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1362 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1363 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1364 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1365 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1366 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1367 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1368 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1369
1370 __ BIND(L_copy_bytes);
1371 __ addptr(qword_count, 4);
1372 __ jcc(Assembler::lessEqual, L_loop);
1373 }
1374 __ subptr(qword_count, 4);
1375 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1376 }
1377
1378 // Copy big chunks backward
1379 //
1380 // Inputs:
1381 // from - source arrays address
1382 // dest - destination array address
1383 // qword_count - 64-bits element count
1384 // to - scratch
1385 // L_copy_bytes - entry label
1386 // L_copy_8_bytes - exit label
1387 //
1388 void copy_bytes_backward(Register from, Register dest,
1389 Register qword_count, Register to,
1390 Label& L_copy_bytes, Label& L_copy_8_bytes) {
1391 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1392 Label L_loop;
1393 __ align(OptoLoopAlignment);
1394 if (UseUnalignedLoadStores) {
1395 Label L_end;
1396 // Copy 64-bytes per iteration
1397 __ BIND(L_loop);
1398 if (UseAVX > 2) {
1399 __ evmovdqul(xmm0, Address(from, qword_count, Address::times_8, 0), Assembler::AVX_512bit);
1400 __ evmovdqul(Address(dest, qword_count, Address::times_8, 0), xmm0, Assembler::AVX_512bit);
1401 } else if (UseAVX == 2) {
1402 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 32));
1403 __ vmovdqu(Address(dest, qword_count, Address::times_8, 32), xmm0);
1404 __ vmovdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1405 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1406 } else {
1407 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 48));
1408 __ movdqu(Address(dest, qword_count, Address::times_8, 48), xmm0);
1409 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 32));
1410 __ movdqu(Address(dest, qword_count, Address::times_8, 32), xmm1);
1411 __ movdqu(xmm2, Address(from, qword_count, Address::times_8, 16));
1412 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm2);
1413 __ movdqu(xmm3, Address(from, qword_count, Address::times_8, 0));
1414 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm3);
1415 }
1416 __ BIND(L_copy_bytes);
1417 __ subptr(qword_count, 8);
1418 __ jcc(Assembler::greaterEqual, L_loop);
1419
1420 __ addptr(qword_count, 4); // add(8) and sub(4)
1421 __ jccb(Assembler::less, L_end);
1422 // Copy trailing 32 bytes
1423 if (UseAVX >= 2) {
1424 __ vmovdqu(xmm0, Address(from, qword_count, Address::times_8, 0));
1425 __ vmovdqu(Address(dest, qword_count, Address::times_8, 0), xmm0);
1426 } else {
1427 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1428 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1429 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1430 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1431 }
1432 __ subptr(qword_count, 4);
1433 __ BIND(L_end);
1434 if (UseAVX >= 2) {
1435 // clean upper bits of YMM registers
1436 __ vpxor(xmm0, xmm0);
1437 __ vpxor(xmm1, xmm1);
1438 }
1439 } else {
1440 // Copy 32-bytes per iteration
1441 __ BIND(L_loop);
1442 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1443 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1444 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1445 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1446 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1447 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1448 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1449 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1450
1451 __ BIND(L_copy_bytes);
1452 __ subptr(qword_count, 4);
1453 __ jcc(Assembler::greaterEqual, L_loop);
1454 }
1455 __ addptr(qword_count, 4);
1456 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1457 }
1458
1459
1460 // Arguments:
1461 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1462 // ignored
1463 // name - stub name string
1464 //
1465 // Inputs:
1466 // c_rarg0 - source array address
1467 // c_rarg1 - destination array address
1468 // c_rarg2 - element count, treated as ssize_t, can be zero
1469 //
1470 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1471 // we let the hardware handle it. The one to eight bytes within words,
1472 // dwords or qwords that span cache line boundaries will still be loaded
1473 // and stored atomically.
1474 //
1475 // Side Effects:
1476 // disjoint_byte_copy_entry is set to the no-overlap entry point
1477 // used by generate_conjoint_byte_copy().
1478 //
1479 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1480 __ align(CodeEntryAlignment);
1481 StubCodeMark mark(this, "StubRoutines", name);
1482 address start = __ pc();
1483
1484 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1485 Label L_copy_byte, L_exit;
1486 const Register from = rdi; // source array address
1487 const Register to = rsi; // destination array address
1488 const Register count = rdx; // elements count
1489 const Register byte_count = rcx;
1490 const Register qword_count = count;
1491 const Register end_from = from; // source array end address
1492 const Register end_to = to; // destination array end address
1493 // End pointers are inclusive, and if count is not zero they point
1494 // to the last unit copied: end_to[0] := end_from[0]
1495
1496 __ enter(); // required for proper stackwalking of RuntimeStub frame
1497 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1498
1499 if (entry != NULL) {
1500 *entry = __ pc();
1501 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1502 BLOCK_COMMENT("Entry:");
1503 }
1504
1505 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1506 // r9 and r10 may be used to save non-volatile registers
1507
1508 // 'from', 'to' and 'count' are now valid
1509 __ movptr(byte_count, count);
1510 __ shrptr(count, 3); // count => qword_count
1511
1512 // Copy from low to high addresses. Use 'to' as scratch.
1513 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1514 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1515 __ negptr(qword_count); // make the count negative
1516 __ jmp(L_copy_bytes);
1517
1518 // Copy trailing qwords
1519 __ BIND(L_copy_8_bytes);
1520 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1521 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1522 __ increment(qword_count);
1523 __ jcc(Assembler::notZero, L_copy_8_bytes);
1524
1525 // Check for and copy trailing dword
1526 __ BIND(L_copy_4_bytes);
1527 __ testl(byte_count, 4);
1528 __ jccb(Assembler::zero, L_copy_2_bytes);
1529 __ movl(rax, Address(end_from, 8));
1530 __ movl(Address(end_to, 8), rax);
1531
1532 __ addptr(end_from, 4);
1533 __ addptr(end_to, 4);
1534
1535 // Check for and copy trailing word
1536 __ BIND(L_copy_2_bytes);
1537 __ testl(byte_count, 2);
1538 __ jccb(Assembler::zero, L_copy_byte);
1539 __ movw(rax, Address(end_from, 8));
1540 __ movw(Address(end_to, 8), rax);
1541
1542 __ addptr(end_from, 2);
1543 __ addptr(end_to, 2);
1544
1545 // Check for and copy trailing byte
1546 __ BIND(L_copy_byte);
1547 __ testl(byte_count, 1);
1548 __ jccb(Assembler::zero, L_exit);
1549 __ movb(rax, Address(end_from, 8));
1550 __ movb(Address(end_to, 8), rax);
1551
1552 __ BIND(L_exit);
1553 restore_arg_regs();
1554 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1555 __ xorptr(rax, rax); // return 0
1556 __ vzeroupper();
1557 __ leave(); // required for proper stackwalking of RuntimeStub frame
1558 __ ret(0);
1559
1560 // Copy in multi-bytes chunks
1561 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1562 __ jmp(L_copy_4_bytes);
1563
1564 return start;
1565 }
1566
1567 // Arguments:
1568 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1569 // ignored
1570 // name - stub name string
1571 //
1572 // Inputs:
1573 // c_rarg0 - source array address
1574 // c_rarg1 - destination array address
1575 // c_rarg2 - element count, treated as ssize_t, can be zero
1576 //
1577 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1578 // we let the hardware handle it. The one to eight bytes within words,
1579 // dwords or qwords that span cache line boundaries will still be loaded
1580 // and stored atomically.
1581 //
1582 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1583 address* entry, const char *name) {
1584 __ align(CodeEntryAlignment);
1585 StubCodeMark mark(this, "StubRoutines", name);
1586 address start = __ pc();
1587
1588 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1589 const Register from = rdi; // source array address
1590 const Register to = rsi; // destination array address
1591 const Register count = rdx; // elements count
1592 const Register byte_count = rcx;
1593 const Register qword_count = count;
1594
1595 __ enter(); // required for proper stackwalking of RuntimeStub frame
1596 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1597
1598 if (entry != NULL) {
1599 *entry = __ pc();
1600 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1601 BLOCK_COMMENT("Entry:");
1602 }
1603
1604 array_overlap_test(nooverlap_target, Address::times_1);
1605 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1606 // r9 and r10 may be used to save non-volatile registers
1607
1608 // 'from', 'to' and 'count' are now valid
1609 __ movptr(byte_count, count);
1610 __ shrptr(count, 3); // count => qword_count
1611
1612 // Copy from high to low addresses.
1613
1614 // Check for and copy trailing byte
1615 __ testl(byte_count, 1);
1616 __ jcc(Assembler::zero, L_copy_2_bytes);
1617 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1618 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1619 __ decrement(byte_count); // Adjust for possible trailing word
1620
1621 // Check for and copy trailing word
1622 __ BIND(L_copy_2_bytes);
1623 __ testl(byte_count, 2);
1624 __ jcc(Assembler::zero, L_copy_4_bytes);
1625 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1626 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1627
1628 // Check for and copy trailing dword
1629 __ BIND(L_copy_4_bytes);
1630 __ testl(byte_count, 4);
1631 __ jcc(Assembler::zero, L_copy_bytes);
1632 __ movl(rax, Address(from, qword_count, Address::times_8));
1633 __ movl(Address(to, qword_count, Address::times_8), rax);
1634 __ jmp(L_copy_bytes);
1635
1636 // Copy trailing qwords
1637 __ BIND(L_copy_8_bytes);
1638 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1639 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1640 __ decrement(qword_count);
1641 __ jcc(Assembler::notZero, L_copy_8_bytes);
1642
1643 restore_arg_regs();
1644 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1645 __ xorptr(rax, rax); // return 0
1646 __ vzeroupper();
1647 __ leave(); // required for proper stackwalking of RuntimeStub frame
1648 __ ret(0);
1649
1650 // Copy in multi-bytes chunks
1651 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1652
1653 restore_arg_regs();
1654 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1655 __ xorptr(rax, rax); // return 0
1656 __ vzeroupper();
1657 __ leave(); // required for proper stackwalking of RuntimeStub frame
1658 __ ret(0);
1659
1660 return start;
1661 }
1662
1663 // Arguments:
1664 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1665 // ignored
1666 // name - stub name string
1667 //
1668 // Inputs:
1669 // c_rarg0 - source array address
1670 // c_rarg1 - destination array address
1671 // c_rarg2 - element count, treated as ssize_t, can be zero
1672 //
1673 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1674 // let the hardware handle it. The two or four words within dwords
1675 // or qwords that span cache line boundaries will still be loaded
1676 // and stored atomically.
1677 //
1678 // Side Effects:
1679 // disjoint_short_copy_entry is set to the no-overlap entry point
1680 // used by generate_conjoint_short_copy().
1681 //
1682 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1683 __ align(CodeEntryAlignment);
1684 StubCodeMark mark(this, "StubRoutines", name);
1685 address start = __ pc();
1686
1687 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1688 const Register from = rdi; // source array address
1689 const Register to = rsi; // destination array address
1690 const Register count = rdx; // elements count
1691 const Register word_count = rcx;
1692 const Register qword_count = count;
1693 const Register end_from = from; // source array end address
1694 const Register end_to = to; // destination array end address
1695 // End pointers are inclusive, and if count is not zero they point
1696 // to the last unit copied: end_to[0] := end_from[0]
1697
1698 __ enter(); // required for proper stackwalking of RuntimeStub frame
1699 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1700
1701 if (entry != NULL) {
1702 *entry = __ pc();
1703 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1704 BLOCK_COMMENT("Entry:");
1705 }
1706
1707 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1708 // r9 and r10 may be used to save non-volatile registers
1709
1710 // 'from', 'to' and 'count' are now valid
1711 __ movptr(word_count, count);
1712 __ shrptr(count, 2); // count => qword_count
1713
1714 // Copy from low to high addresses. Use 'to' as scratch.
1715 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1716 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1717 __ negptr(qword_count);
1718 __ jmp(L_copy_bytes);
1719
1720 // Copy trailing qwords
1721 __ BIND(L_copy_8_bytes);
1722 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1723 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1724 __ increment(qword_count);
1725 __ jcc(Assembler::notZero, L_copy_8_bytes);
1726
1727 // Original 'dest' is trashed, so we can't use it as a
1728 // base register for a possible trailing word copy
1729
1730 // Check for and copy trailing dword
1731 __ BIND(L_copy_4_bytes);
1732 __ testl(word_count, 2);
1733 __ jccb(Assembler::zero, L_copy_2_bytes);
1734 __ movl(rax, Address(end_from, 8));
1735 __ movl(Address(end_to, 8), rax);
1736
1737 __ addptr(end_from, 4);
1738 __ addptr(end_to, 4);
1739
1740 // Check for and copy trailing word
1741 __ BIND(L_copy_2_bytes);
1742 __ testl(word_count, 1);
1743 __ jccb(Assembler::zero, L_exit);
1744 __ movw(rax, Address(end_from, 8));
1745 __ movw(Address(end_to, 8), rax);
1746
1747 __ BIND(L_exit);
1748 restore_arg_regs();
1749 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1750 __ xorptr(rax, rax); // return 0
1751 __ vzeroupper();
1752 __ leave(); // required for proper stackwalking of RuntimeStub frame
1753 __ ret(0);
1754
1755 // Copy in multi-bytes chunks
1756 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1757 __ jmp(L_copy_4_bytes);
1758
1759 return start;
1760 }
1761
1762 address generate_fill(BasicType t, bool aligned, const char *name) {
1763 __ align(CodeEntryAlignment);
1764 StubCodeMark mark(this, "StubRoutines", name);
1765 address start = __ pc();
1766
1767 BLOCK_COMMENT("Entry:");
1768
1769 const Register to = c_rarg0; // source array address
1770 const Register value = c_rarg1; // value
1771 const Register count = c_rarg2; // elements count
1772
1773 __ enter(); // required for proper stackwalking of RuntimeStub frame
1774
1775 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1776
1777 __ vzeroupper();
1778 __ leave(); // required for proper stackwalking of RuntimeStub frame
1779 __ ret(0);
1780 return start;
1781 }
1782
1783 // Arguments:
1784 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1785 // ignored
1786 // name - stub name string
1787 //
1788 // Inputs:
1789 // c_rarg0 - source array address
1790 // c_rarg1 - destination array address
1791 // c_rarg2 - element count, treated as ssize_t, can be zero
1792 //
1793 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1794 // let the hardware handle it. The two or four words within dwords
1795 // or qwords that span cache line boundaries will still be loaded
1796 // and stored atomically.
1797 //
1798 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1799 address *entry, const char *name) {
1800 __ align(CodeEntryAlignment);
1801 StubCodeMark mark(this, "StubRoutines", name);
1802 address start = __ pc();
1803
1804 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes;
1805 const Register from = rdi; // source array address
1806 const Register to = rsi; // destination array address
1807 const Register count = rdx; // elements count
1808 const Register word_count = rcx;
1809 const Register qword_count = count;
1810
1811 __ enter(); // required for proper stackwalking of RuntimeStub frame
1812 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1813
1814 if (entry != NULL) {
1815 *entry = __ pc();
1816 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1817 BLOCK_COMMENT("Entry:");
1818 }
1819
1820 array_overlap_test(nooverlap_target, Address::times_2);
1821 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1822 // r9 and r10 may be used to save non-volatile registers
1823
1824 // 'from', 'to' and 'count' are now valid
1825 __ movptr(word_count, count);
1826 __ shrptr(count, 2); // count => qword_count
1827
1828 // Copy from high to low addresses. Use 'to' as scratch.
1829
1830 // Check for and copy trailing word
1831 __ testl(word_count, 1);
1832 __ jccb(Assembler::zero, L_copy_4_bytes);
1833 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1834 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1835
1836 // Check for and copy trailing dword
1837 __ BIND(L_copy_4_bytes);
1838 __ testl(word_count, 2);
1839 __ jcc(Assembler::zero, L_copy_bytes);
1840 __ movl(rax, Address(from, qword_count, Address::times_8));
1841 __ movl(Address(to, qword_count, Address::times_8), rax);
1842 __ jmp(L_copy_bytes);
1843
1844 // Copy trailing qwords
1845 __ BIND(L_copy_8_bytes);
1846 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1847 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1848 __ decrement(qword_count);
1849 __ jcc(Assembler::notZero, L_copy_8_bytes);
1850
1851 restore_arg_regs();
1852 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1853 __ xorptr(rax, rax); // return 0
1854 __ vzeroupper();
1855 __ leave(); // required for proper stackwalking of RuntimeStub frame
1856 __ ret(0);
1857
1858 // Copy in multi-bytes chunks
1859 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1860
1861 restore_arg_regs();
1862 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1863 __ xorptr(rax, rax); // return 0
1864 __ vzeroupper();
1865 __ leave(); // required for proper stackwalking of RuntimeStub frame
1866 __ ret(0);
1867
1868 return start;
1869 }
1870
1871 // Arguments:
1872 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1873 // ignored
1874 // is_oop - true => oop array, so generate store check code
1875 // name - stub name string
1876 //
1877 // Inputs:
1878 // c_rarg0 - source array address
1879 // c_rarg1 - destination array address
1880 // c_rarg2 - element count, treated as ssize_t, can be zero
1881 //
1882 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1883 // the hardware handle it. The two dwords within qwords that span
1884 // cache line boundaries will still be loaded and stored atomicly.
1885 //
1886 // Side Effects:
1887 // disjoint_int_copy_entry is set to the no-overlap entry point
1888 // used by generate_conjoint_int_oop_copy().
1889 //
1890 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1891 const char *name, bool dest_uninitialized = false) {
1892 __ align(CodeEntryAlignment);
1893 StubCodeMark mark(this, "StubRoutines", name);
1894 address start = __ pc();
1895
1896 Label L_copy_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1897 const Register from = rdi; // source array address
1898 const Register to = rsi; // destination array address
1899 const Register count = rdx; // elements count
1900 const Register dword_count = rcx;
1901 const Register qword_count = count;
1902 const Register end_from = from; // source array end address
1903 const Register end_to = to; // destination array end address
1904 // End pointers are inclusive, and if count is not zero they point
1905 // to the last unit copied: end_to[0] := end_from[0]
1906
1907 __ enter(); // required for proper stackwalking of RuntimeStub frame
1908 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1909
1910 if (entry != NULL) {
1911 *entry = __ pc();
1912 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1913 BLOCK_COMMENT("Entry:");
1914 }
1915
1916 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1917 // r9 is used to save r15_thread
1918
1919 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
1920 if (dest_uninitialized) {
1921 decorators |= IS_DEST_UNINITIALIZED;
1922 }
1923 if (aligned) {
1924 decorators |= ARRAYCOPY_ALIGNED;
1925 }
1926
1927 BasicType type = is_oop ? T_OBJECT : T_INT;
1928 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1929 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1930
1931 // 'from', 'to' and 'count' are now valid
1932 __ movptr(dword_count, count);
1933 __ shrptr(count, 1); // count => qword_count
1934
1935 // Copy from low to high addresses. Use 'to' as scratch.
1936 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1937 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1938 __ negptr(qword_count);
1939 __ jmp(L_copy_bytes);
1940
1941 // Copy trailing qwords
1942 __ BIND(L_copy_8_bytes);
1943 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1944 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1945 __ increment(qword_count);
1946 __ jcc(Assembler::notZero, L_copy_8_bytes);
1947
1948 // Check for and copy trailing dword
1949 __ BIND(L_copy_4_bytes);
1950 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1951 __ jccb(Assembler::zero, L_exit);
1952 __ movl(rax, Address(end_from, 8));
1953 __ movl(Address(end_to, 8), rax);
1954
1955 __ BIND(L_exit);
1956 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
1957 restore_arg_regs_using_thread();
1958 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1959 __ vzeroupper();
1960 __ xorptr(rax, rax); // return 0
1961 __ leave(); // required for proper stackwalking of RuntimeStub frame
1962 __ ret(0);
1963
1964 // Copy in multi-bytes chunks
1965 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
1966 __ jmp(L_copy_4_bytes);
1967
1968 return start;
1969 }
1970
1971 // Arguments:
1972 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1973 // ignored
1974 // is_oop - true => oop array, so generate store check code
1975 // name - stub name string
1976 //
1977 // Inputs:
1978 // c_rarg0 - source array address
1979 // c_rarg1 - destination array address
1980 // c_rarg2 - element count, treated as ssize_t, can be zero
1981 //
1982 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1983 // the hardware handle it. The two dwords within qwords that span
1984 // cache line boundaries will still be loaded and stored atomicly.
1985 //
1986 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1987 address *entry, const char *name,
1988 bool dest_uninitialized = false) {
1989 __ align(CodeEntryAlignment);
1990 StubCodeMark mark(this, "StubRoutines", name);
1991 address start = __ pc();
1992
1993 Label L_copy_bytes, L_copy_8_bytes, L_exit;
1994 const Register from = rdi; // source array address
1995 const Register to = rsi; // destination array address
1996 const Register count = rdx; // elements count
1997 const Register dword_count = rcx;
1998 const Register qword_count = count;
1999
2000 __ enter(); // required for proper stackwalking of RuntimeStub frame
2001 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2002
2003 if (entry != NULL) {
2004 *entry = __ pc();
2005 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2006 BLOCK_COMMENT("Entry:");
2007 }
2008
2009 array_overlap_test(nooverlap_target, Address::times_4);
2010 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2011 // r9 is used to save r15_thread
2012
2013 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2014 if (dest_uninitialized) {
2015 decorators |= IS_DEST_UNINITIALIZED;
2016 }
2017 if (aligned) {
2018 decorators |= ARRAYCOPY_ALIGNED;
2019 }
2020
2021 BasicType type = is_oop ? T_OBJECT : T_INT;
2022 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2023 // no registers are destroyed by this call
2024 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2025
2026 assert_clean_int(count, rax); // Make sure 'count' is clean int.
2027 // 'from', 'to' and 'count' are now valid
2028 __ movptr(dword_count, count);
2029 __ shrptr(count, 1); // count => qword_count
2030
2031 // Copy from high to low addresses. Use 'to' as scratch.
2032
2033 // Check for and copy trailing dword
2034 __ testl(dword_count, 1);
2035 __ jcc(Assembler::zero, L_copy_bytes);
2036 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
2037 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
2038 __ jmp(L_copy_bytes);
2039
2040 // Copy trailing qwords
2041 __ BIND(L_copy_8_bytes);
2042 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2043 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2044 __ decrement(qword_count);
2045 __ jcc(Assembler::notZero, L_copy_8_bytes);
2046
2047 if (is_oop) {
2048 __ jmp(L_exit);
2049 }
2050 restore_arg_regs_using_thread();
2051 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2052 __ xorptr(rax, rax); // return 0
2053 __ vzeroupper();
2054 __ leave(); // required for proper stackwalking of RuntimeStub frame
2055 __ ret(0);
2056
2057 // Copy in multi-bytes chunks
2058 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2059
2060 __ BIND(L_exit);
2061 bs->arraycopy_epilogue(_masm, decorators, type, from, to, dword_count);
2062 restore_arg_regs_using_thread();
2063 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
2064 __ xorptr(rax, rax); // return 0
2065 __ vzeroupper();
2066 __ leave(); // required for proper stackwalking of RuntimeStub frame
2067 __ ret(0);
2068
2069 return start;
2070 }
2071
2072 // Arguments:
2073 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2074 // ignored
2075 // is_oop - true => oop array, so generate store check code
2076 // name - stub name string
2077 //
2078 // Inputs:
2079 // c_rarg0 - source array address
2080 // c_rarg1 - destination array address
2081 // c_rarg2 - element count, treated as ssize_t, can be zero
2082 //
2083 // Side Effects:
2084 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
2085 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2086 //
2087 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2088 const char *name, bool dest_uninitialized = false) {
2089 __ align(CodeEntryAlignment);
2090 StubCodeMark mark(this, "StubRoutines", name);
2091 address start = __ pc();
2092
2093 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2094 const Register from = rdi; // source array address
2095 const Register to = rsi; // destination array address
2096 const Register qword_count = rdx; // elements count
2097 const Register end_from = from; // source array end address
2098 const Register end_to = rcx; // destination array end address
2099 const Register saved_count = r11;
2100 // End pointers are inclusive, and if count is not zero they point
2101 // to the last unit copied: end_to[0] := end_from[0]
2102
2103 __ enter(); // required for proper stackwalking of RuntimeStub frame
2104 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2105 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2106
2107 if (entry != NULL) {
2108 *entry = __ pc();
2109 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2110 BLOCK_COMMENT("Entry:");
2111 }
2112
2113 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2114 // r9 is used to save r15_thread
2115 // 'from', 'to' and 'qword_count' are now valid
2116
2117 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
2118 if (dest_uninitialized) {
2119 decorators |= IS_DEST_UNINITIALIZED;
2120 }
2121 if (aligned) {
2122 decorators |= ARRAYCOPY_ALIGNED;
2123 }
2124
2125 BasicType type = is_oop ? T_OBJECT : T_LONG;
2126 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2127 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2128
2129 // Copy from low to high addresses. Use 'to' as scratch.
2130 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2131 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2132 __ negptr(qword_count);
2133 __ jmp(L_copy_bytes);
2134
2135 // Copy trailing qwords
2136 __ BIND(L_copy_8_bytes);
2137 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2138 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2139 __ increment(qword_count);
2140 __ jcc(Assembler::notZero, L_copy_8_bytes);
2141
2142 if (is_oop) {
2143 __ jmp(L_exit);
2144 } else {
2145 restore_arg_regs_using_thread();
2146 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2147 __ xorptr(rax, rax); // return 0
2148 __ vzeroupper();
2149 __ leave(); // required for proper stackwalking of RuntimeStub frame
2150 __ ret(0);
2151 }
2152
2153 // Copy in multi-bytes chunks
2154 copy_bytes_forward(end_from, end_to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2155
2156 __ BIND(L_exit);
2157 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2158 restore_arg_regs_using_thread();
2159 if (is_oop) {
2160 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2161 } else {
2162 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2163 }
2164 __ vzeroupper();
2165 __ xorptr(rax, rax); // return 0
2166 __ leave(); // required for proper stackwalking of RuntimeStub frame
2167 __ ret(0);
2168
2169 return start;
2170 }
2171
2172 // Arguments:
2173 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2174 // ignored
2175 // is_oop - true => oop array, so generate store check code
2176 // name - stub name string
2177 //
2178 // Inputs:
2179 // c_rarg0 - source array address
2180 // c_rarg1 - destination array address
2181 // c_rarg2 - element count, treated as ssize_t, can be zero
2182 //
2183 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2184 address nooverlap_target, address *entry,
2185 const char *name, bool dest_uninitialized = false) {
2186 __ align(CodeEntryAlignment);
2187 StubCodeMark mark(this, "StubRoutines", name);
2188 address start = __ pc();
2189
2190 Label L_copy_bytes, L_copy_8_bytes, L_exit;
2191 const Register from = rdi; // source array address
2192 const Register to = rsi; // destination array address
2193 const Register qword_count = rdx; // elements count
2194 const Register saved_count = rcx;
2195
2196 __ enter(); // required for proper stackwalking of RuntimeStub frame
2197 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2198
2199 if (entry != NULL) {
2200 *entry = __ pc();
2201 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2202 BLOCK_COMMENT("Entry:");
2203 }
2204
2205 array_overlap_test(nooverlap_target, Address::times_8);
2206 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
2207 // r9 is used to save r15_thread
2208 // 'from', 'to' and 'qword_count' are now valid
2209
2210 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
2211 if (dest_uninitialized) {
2212 decorators |= IS_DEST_UNINITIALIZED;
2213 }
2214 if (aligned) {
2215 decorators |= ARRAYCOPY_ALIGNED;
2216 }
2217
2218 BasicType type = is_oop ? T_OBJECT : T_LONG;
2219 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2220 bs->arraycopy_prologue(_masm, decorators, type, from, to, qword_count);
2221
2222 __ jmp(L_copy_bytes);
2223
2224 // Copy trailing qwords
2225 __ BIND(L_copy_8_bytes);
2226 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2227 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2228 __ decrement(qword_count);
2229 __ jcc(Assembler::notZero, L_copy_8_bytes);
2230
2231 if (is_oop) {
2232 __ jmp(L_exit);
2233 } else {
2234 restore_arg_regs_using_thread();
2235 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2236 __ xorptr(rax, rax); // return 0
2237 __ vzeroupper();
2238 __ leave(); // required for proper stackwalking of RuntimeStub frame
2239 __ ret(0);
2240 }
2241
2242 // Copy in multi-bytes chunks
2243 copy_bytes_backward(from, to, qword_count, rax, L_copy_bytes, L_copy_8_bytes);
2244
2245 __ BIND(L_exit);
2246 bs->arraycopy_epilogue(_masm, decorators, type, from, to, qword_count);
2247 restore_arg_regs_using_thread();
2248 if (is_oop) {
2249 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2250 } else {
2251 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2252 }
2253 __ vzeroupper();
2254 __ xorptr(rax, rax); // return 0
2255 __ leave(); // required for proper stackwalking of RuntimeStub frame
2256 __ ret(0);
2257
2258 return start;
2259 }
2260
2261
2262 // Helper for generating a dynamic type check.
2263 // Smashes no registers.
2264 void generate_type_check(Register sub_klass,
2265 Register super_check_offset,
2266 Register super_klass,
2267 Label& L_success) {
2268 assert_different_registers(sub_klass, super_check_offset, super_klass);
2269
2270 BLOCK_COMMENT("type_check:");
2271
2272 Label L_miss;
2273
2274 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2275 super_check_offset);
2276 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2277
2278 // Fall through on failure!
2279 __ BIND(L_miss);
2280 }
2281
2282 //
2283 // Generate checkcasting array copy stub
2284 //
2285 // Input:
2286 // c_rarg0 - source array address
2287 // c_rarg1 - destination array address
2288 // c_rarg2 - element count, treated as ssize_t, can be zero
2289 // c_rarg3 - size_t ckoff (super_check_offset)
2290 // not Win64
2291 // c_rarg4 - oop ckval (super_klass)
2292 // Win64
2293 // rsp+40 - oop ckval (super_klass)
2294 //
2295 // Output:
2296 // rax == 0 - success
2297 // rax == -1^K - failure, where K is partial transfer count
2298 //
2299 address generate_checkcast_copy(const char *name, address *entry,
2300 bool dest_uninitialized = false) {
2301
2302 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2303
2304 // Input registers (after setup_arg_regs)
2305 const Register from = rdi; // source array address
2306 const Register to = rsi; // destination array address
2307 const Register length = rdx; // elements count
2308 const Register ckoff = rcx; // super_check_offset
2309 const Register ckval = r8; // super_klass
2310
2311 // Registers used as temps (r13, r14 are save-on-entry)
2312 const Register end_from = from; // source array end address
2313 const Register end_to = r13; // destination array end address
2314 const Register count = rdx; // -(count_remaining)
2315 const Register r14_length = r14; // saved copy of length
2316 // End pointers are inclusive, and if length is not zero they point
2317 // to the last unit copied: end_to[0] := end_from[0]
2318
2319 const Register rax_oop = rax; // actual oop copied
2320 const Register r11_klass = r11; // oop._klass
2321
2322 //---------------------------------------------------------------
2323 // Assembler stub will be used for this call to arraycopy
2324 // if the two arrays are subtypes of Object[] but the
2325 // destination array type is not equal to or a supertype
2326 // of the source type. Each element must be separately
2327 // checked.
2328
2329 __ align(CodeEntryAlignment);
2330 StubCodeMark mark(this, "StubRoutines", name);
2331 address start = __ pc();
2332
2333 __ enter(); // required for proper stackwalking of RuntimeStub frame
2334
2335 #ifdef ASSERT
2336 // caller guarantees that the arrays really are different
2337 // otherwise, we would have to make conjoint checks
2338 { Label L;
2339 array_overlap_test(L, TIMES_OOP);
2340 __ stop("checkcast_copy within a single array");
2341 __ bind(L);
2342 }
2343 #endif //ASSERT
2344
2345 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2346 // ckoff => rcx, ckval => r8
2347 // r9 and r10 may be used to save non-volatile registers
2348 #ifdef _WIN64
2349 // last argument (#4) is on stack on Win64
2350 __ movptr(ckval, Address(rsp, 6 * wordSize));
2351 #endif
2352
2353 // Caller of this entry point must set up the argument registers.
2354 if (entry != NULL) {
2355 *entry = __ pc();
2356 BLOCK_COMMENT("Entry:");
2357 }
2358
2359 // allocate spill slots for r13, r14
2360 enum {
2361 saved_r13_offset,
2362 saved_r14_offset,
2363 saved_r10_offset,
2364 saved_rbp_offset
2365 };
2366 __ subptr(rsp, saved_rbp_offset * wordSize);
2367 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2368 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2369 __ movptr(Address(rsp, saved_r10_offset * wordSize), r10);
2370
2371 #ifdef ASSERT
2372 Label L2;
2373 __ get_thread(r14);
2374 __ cmpptr(r15_thread, r14);
2375 __ jcc(Assembler::equal, L2);
2376 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
2377 __ bind(L2);
2378 #endif // ASSERT
2379
2380 // check that int operands are properly extended to size_t
2381 assert_clean_int(length, rax);
2382 assert_clean_int(ckoff, rax);
2383
2384 #ifdef ASSERT
2385 BLOCK_COMMENT("assert consistent ckoff/ckval");
2386 // The ckoff and ckval must be mutually consistent,
2387 // even though caller generates both.
2388 { Label L;
2389 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2390 __ cmpl(ckoff, Address(ckval, sco_offset));
2391 __ jcc(Assembler::equal, L);
2392 __ stop("super_check_offset inconsistent");
2393 __ bind(L);
2394 }
2395 #endif //ASSERT
2396
2397 // Loop-invariant addresses. They are exclusive end pointers.
2398 Address end_from_addr(from, length, TIMES_OOP, 0);
2399 Address end_to_addr(to, length, TIMES_OOP, 0);
2400 // Loop-variant addresses. They assume post-incremented count < 0.
2401 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2402 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2403
2404 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST | ARRAYCOPY_DISJOINT;
2405 if (dest_uninitialized) {
2406 decorators |= IS_DEST_UNINITIALIZED;
2407 }
2408
2409 BasicType type = T_OBJECT;
2410 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
2411 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
2412
2413 // Copy from low to high addresses, indexed from the end of each array.
2414 __ lea(end_from, end_from_addr);
2415 __ lea(end_to, end_to_addr);
2416 __ movptr(r14_length, length); // save a copy of the length
2417 assert(length == count, ""); // else fix next line:
2418 __ negptr(count); // negate and test the length
2419 __ jcc(Assembler::notZero, L_load_element);
2420
2421 // Empty array: Nothing to do.
2422 __ xorptr(rax, rax); // return 0 on (trivial) success
2423 __ jmp(L_done);
2424
2425 // ======== begin loop ========
2426 // (Loop is rotated; its entry is L_load_element.)
2427 // Loop control:
2428 // for (count = -count; count != 0; count++)
2429 // Base pointers src, dst are biased by 8*(count-1),to last element.
2430 __ align(OptoLoopAlignment);
2431
2432 __ BIND(L_store_element);
2433 __ store_heap_oop(to_element_addr, rax_oop, noreg, noreg, noreg, AS_RAW); // store the oop
2434 __ increment(count); // increment the count toward zero
2435 __ jcc(Assembler::zero, L_do_card_marks);
2436
2437 // ======== loop entry is here ========
2438 __ BIND(L_load_element);
2439 __ load_heap_oop(rax_oop, from_element_addr, noreg, noreg, AS_RAW); // load the oop
2440 __ testptr(rax_oop, rax_oop);
2441 __ jcc(Assembler::zero, L_store_element);
2442
2443 __ load_klass(r11_klass, rax_oop);// query the object klass
2444 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2445 // ======== end loop ========
2446
2447 // It was a real error; we must depend on the caller to finish the job.
2448 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2449 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2450 // and report their number to the caller.
2451 assert_different_registers(rax, r14_length, count, to, end_to, rcx, rscratch1);
2452 Label L_post_barrier;
2453 __ addptr(r14_length, count); // K = (original - remaining) oops
2454 __ movptr(rax, r14_length); // save the value
2455 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
2456 __ jccb(Assembler::notZero, L_post_barrier);
2457 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
2458
2459 // Come here on success only.
2460 __ BIND(L_do_card_marks);
2461 __ xorptr(rax, rax); // return 0 on success
2462
2463 __ BIND(L_post_barrier);
2464 bs->arraycopy_epilogue(_masm, decorators, type, from, to, r14_length);
2465
2466 // Common exit point (success or failure).
2467 __ BIND(L_done);
2468 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2469 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2470 __ movptr(r10, Address(rsp, saved_r10_offset * wordSize));
2471 restore_arg_regs();
2472 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2473 __ leave(); // required for proper stackwalking of RuntimeStub frame
2474 __ ret(0);
2475
2476 return start;
2477 }
2478
2479 //
2480 // Generate 'unsafe' array copy stub
2481 // Though just as safe as the other stubs, it takes an unscaled
2482 // size_t argument instead of an element count.
2483 //
2484 // Input:
2485 // c_rarg0 - source array address
2486 // c_rarg1 - destination array address
2487 // c_rarg2 - byte count, treated as ssize_t, can be zero
2488 //
2489 // Examines the alignment of the operands and dispatches
2490 // to a long, int, short, or byte copy loop.
2491 //
2492 address generate_unsafe_copy(const char *name,
2493 address byte_copy_entry, address short_copy_entry,
2494 address int_copy_entry, address long_copy_entry) {
2495
2496 Label L_long_aligned, L_int_aligned, L_short_aligned;
2497
2498 // Input registers (before setup_arg_regs)
2499 const Register from = c_rarg0; // source array address
2500 const Register to = c_rarg1; // destination array address
2501 const Register size = c_rarg2; // byte count (size_t)
2502
2503 // Register used as a temp
2504 const Register bits = rax; // test copy of low bits
2505
2506 __ align(CodeEntryAlignment);
2507 StubCodeMark mark(this, "StubRoutines", name);
2508 address start = __ pc();
2509
2510 __ enter(); // required for proper stackwalking of RuntimeStub frame
2511
2512 // bump this on entry, not on exit:
2513 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2514
2515 __ mov(bits, from);
2516 __ orptr(bits, to);
2517 __ orptr(bits, size);
2518
2519 __ testb(bits, BytesPerLong-1);
2520 __ jccb(Assembler::zero, L_long_aligned);
2521
2522 __ testb(bits, BytesPerInt-1);
2523 __ jccb(Assembler::zero, L_int_aligned);
2524
2525 __ testb(bits, BytesPerShort-1);
2526 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2527
2528 __ BIND(L_short_aligned);
2529 __ shrptr(size, LogBytesPerShort); // size => short_count
2530 __ jump(RuntimeAddress(short_copy_entry));
2531
2532 __ BIND(L_int_aligned);
2533 __ shrptr(size, LogBytesPerInt); // size => int_count
2534 __ jump(RuntimeAddress(int_copy_entry));
2535
2536 __ BIND(L_long_aligned);
2537 __ shrptr(size, LogBytesPerLong); // size => qword_count
2538 __ jump(RuntimeAddress(long_copy_entry));
2539
2540 return start;
2541 }
2542
2543 // Perform range checks on the proposed arraycopy.
2544 // Kills temp, but nothing else.
2545 // Also, clean the sign bits of src_pos and dst_pos.
2546 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2547 Register src_pos, // source position (c_rarg1)
2548 Register dst, // destination array oo (c_rarg2)
2549 Register dst_pos, // destination position (c_rarg3)
2550 Register length,
2551 Register temp,
2552 Label& L_failed) {
2553 BLOCK_COMMENT("arraycopy_range_checks:");
2554
2555 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2556 __ movl(temp, length);
2557 __ addl(temp, src_pos); // src_pos + length
2558 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2559 __ jcc(Assembler::above, L_failed);
2560
2561 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2562 __ movl(temp, length);
2563 __ addl(temp, dst_pos); // dst_pos + length
2564 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2565 __ jcc(Assembler::above, L_failed);
2566
2567 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2568 // Move with sign extension can be used since they are positive.
2569 __ movslq(src_pos, src_pos);
2570 __ movslq(dst_pos, dst_pos);
2571
2572 BLOCK_COMMENT("arraycopy_range_checks done");
2573 }
2574
2575 //
2576 // Generate generic array copy stubs
2577 //
2578 // Input:
2579 // c_rarg0 - src oop
2580 // c_rarg1 - src_pos (32-bits)
2581 // c_rarg2 - dst oop
2582 // c_rarg3 - dst_pos (32-bits)
2583 // not Win64
2584 // c_rarg4 - element count (32-bits)
2585 // Win64
2586 // rsp+40 - element count (32-bits)
2587 //
2588 // Output:
2589 // rax == 0 - success
2590 // rax == -1^K - failure, where K is partial transfer count
2591 //
2592 address generate_generic_copy(const char *name,
2593 address byte_copy_entry, address short_copy_entry,
2594 address int_copy_entry, address oop_copy_entry,
2595 address long_copy_entry, address checkcast_copy_entry) {
2596
2597 Label L_failed, L_failed_0, L_objArray;
2598 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2599
2600 // Input registers
2601 const Register src = c_rarg0; // source array oop
2602 const Register src_pos = c_rarg1; // source position
2603 const Register dst = c_rarg2; // destination array oop
2604 const Register dst_pos = c_rarg3; // destination position
2605 #ifndef _WIN64
2606 const Register length = c_rarg4;
2607 #else
2608 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2609 #endif
2610
2611 { int modulus = CodeEntryAlignment;
2612 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2613 int advance = target - (__ offset() % modulus);
2614 if (advance < 0) advance += modulus;
2615 if (advance > 0) __ nop(advance);
2616 }
2617 StubCodeMark mark(this, "StubRoutines", name);
2618
2619 // Short-hop target to L_failed. Makes for denser prologue code.
2620 __ BIND(L_failed_0);
2621 __ jmp(L_failed);
2622 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2623
2624 __ align(CodeEntryAlignment);
2625 address start = __ pc();
2626
2627 __ enter(); // required for proper stackwalking of RuntimeStub frame
2628
2629 // bump this on entry, not on exit:
2630 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2631
2632 //-----------------------------------------------------------------------
2633 // Assembler stub will be used for this call to arraycopy
2634 // if the following conditions are met:
2635 //
2636 // (1) src and dst must not be null.
2637 // (2) src_pos must not be negative.
2638 // (3) dst_pos must not be negative.
2639 // (4) length must not be negative.
2640 // (5) src klass and dst klass should be the same and not NULL.
2641 // (6) src and dst should be arrays.
2642 // (7) src_pos + length must not exceed length of src.
2643 // (8) dst_pos + length must not exceed length of dst.
2644 //
2645
2646 // if (src == NULL) return -1;
2647 __ testptr(src, src); // src oop
2648 size_t j1off = __ offset();
2649 __ jccb(Assembler::zero, L_failed_0);
2650
2651 // if (src_pos < 0) return -1;
2652 __ testl(src_pos, src_pos); // src_pos (32-bits)
2653 __ jccb(Assembler::negative, L_failed_0);
2654
2655 // if (dst == NULL) return -1;
2656 __ testptr(dst, dst); // dst oop
2657 __ jccb(Assembler::zero, L_failed_0);
2658
2659 // if (dst_pos < 0) return -1;
2660 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2661 size_t j4off = __ offset();
2662 __ jccb(Assembler::negative, L_failed_0);
2663
2664 // The first four tests are very dense code,
2665 // but not quite dense enough to put four
2666 // jumps in a 16-byte instruction fetch buffer.
2667 // That's good, because some branch predicters
2668 // do not like jumps so close together.
2669 // Make sure of this.
2670 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2671
2672 // registers used as temp
2673 const Register r11_length = r11; // elements count to copy
2674 const Register r10_src_klass = r10; // array klass
2675
2676 // if (length < 0) return -1;
2677 __ movl(r11_length, length); // length (elements count, 32-bits value)
2678 __ testl(r11_length, r11_length);
2679 __ jccb(Assembler::negative, L_failed_0);
2680
2681 __ load_klass(r10_src_klass, src);
2682 #ifdef ASSERT
2683 // assert(src->klass() != NULL);
2684 {
2685 BLOCK_COMMENT("assert klasses not null {");
2686 Label L1, L2;
2687 __ testptr(r10_src_klass, r10_src_klass);
2688 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2689 __ bind(L1);
2690 __ stop("broken null klass");
2691 __ bind(L2);
2692 __ load_klass(rax, dst);
2693 __ cmpq(rax, 0);
2694 __ jcc(Assembler::equal, L1); // this would be broken also
2695 BLOCK_COMMENT("} assert klasses not null done");
2696 }
2697 #endif
2698
2699 // Load layout helper (32-bits)
2700 //
2701 // |array_tag| | header_size | element_type | |log2_element_size|
2702 // 32 30 24 16 8 2 0
2703 //
2704 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2705 //
2706
2707 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2708
2709 // Handle objArrays completely differently...
2710 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2711 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2712 __ jcc(Assembler::equal, L_objArray);
2713
2714 // if (src->klass() != dst->klass()) return -1;
2715 __ load_klass(rax, dst);
2716 __ cmpq(r10_src_klass, rax);
2717 __ jcc(Assembler::notEqual, L_failed);
2718
2719 const Register rax_lh = rax; // layout helper
2720 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2721
2722 // if (!src->is_Array()) return -1;
2723 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2724 __ jcc(Assembler::greaterEqual, L_failed);
2725
2726 // At this point, it is known to be a typeArray (array_tag 0x3).
2727 #ifdef ASSERT
2728 {
2729 BLOCK_COMMENT("assert primitive array {");
2730 Label L;
2731 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2732 __ jcc(Assembler::greaterEqual, L);
2733 __ stop("must be a primitive array");
2734 __ bind(L);
2735 BLOCK_COMMENT("} assert primitive array done");
2736 }
2737 #endif
2738
2739 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2740 r10, L_failed);
2741
2742 // TypeArrayKlass
2743 //
2744 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2745 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2746 //
2747
2748 const Register r10_offset = r10; // array offset
2749 const Register rax_elsize = rax_lh; // element size
2750
2751 __ movl(r10_offset, rax_lh);
2752 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2753 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2754 __ addptr(src, r10_offset); // src array offset
2755 __ addptr(dst, r10_offset); // dst array offset
2756 BLOCK_COMMENT("choose copy loop based on element size");
2757 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2758
2759 // next registers should be set before the jump to corresponding stub
2760 const Register from = c_rarg0; // source array address
2761 const Register to = c_rarg1; // destination array address
2762 const Register count = c_rarg2; // elements count
2763
2764 // 'from', 'to', 'count' registers should be set in such order
2765 // since they are the same as 'src', 'src_pos', 'dst'.
2766
2767 __ BIND(L_copy_bytes);
2768 __ cmpl(rax_elsize, 0);
2769 __ jccb(Assembler::notEqual, L_copy_shorts);
2770 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2771 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2772 __ movl2ptr(count, r11_length); // length
2773 __ jump(RuntimeAddress(byte_copy_entry));
2774
2775 __ BIND(L_copy_shorts);
2776 __ cmpl(rax_elsize, LogBytesPerShort);
2777 __ jccb(Assembler::notEqual, L_copy_ints);
2778 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2779 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2780 __ movl2ptr(count, r11_length); // length
2781 __ jump(RuntimeAddress(short_copy_entry));
2782
2783 __ BIND(L_copy_ints);
2784 __ cmpl(rax_elsize, LogBytesPerInt);
2785 __ jccb(Assembler::notEqual, L_copy_longs);
2786 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2787 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2788 __ movl2ptr(count, r11_length); // length
2789 __ jump(RuntimeAddress(int_copy_entry));
2790
2791 __ BIND(L_copy_longs);
2792 #ifdef ASSERT
2793 {
2794 BLOCK_COMMENT("assert long copy {");
2795 Label L;
2796 __ cmpl(rax_elsize, LogBytesPerLong);
2797 __ jcc(Assembler::equal, L);
2798 __ stop("must be long copy, but elsize is wrong");
2799 __ bind(L);
2800 BLOCK_COMMENT("} assert long copy done");
2801 }
2802 #endif
2803 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2804 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2805 __ movl2ptr(count, r11_length); // length
2806 __ jump(RuntimeAddress(long_copy_entry));
2807
2808 // ObjArrayKlass
2809 __ BIND(L_objArray);
2810 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2811
2812 Label L_plain_copy, L_checkcast_copy;
2813 // test array classes for subtyping
2814 __ load_klass(rax, dst);
2815 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2816 __ jcc(Assembler::notEqual, L_checkcast_copy);
2817
2818 // Identically typed arrays can be copied without element-wise checks.
2819 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2820 r10, L_failed);
2821
2822 __ lea(from, Address(src, src_pos, TIMES_OOP,
2823 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2824 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2825 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2826 __ movl2ptr(count, r11_length); // length
2827 __ BIND(L_plain_copy);
2828 __ jump(RuntimeAddress(oop_copy_entry));
2829
2830 __ BIND(L_checkcast_copy);
2831 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2832 {
2833 // Before looking at dst.length, make sure dst is also an objArray.
2834 __ cmpl(Address(rax, lh_offset), objArray_lh);
2835 __ jcc(Assembler::notEqual, L_failed);
2836
2837 // It is safe to examine both src.length and dst.length.
2838 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2839 rax, L_failed);
2840
2841 const Register r11_dst_klass = r11;
2842 __ load_klass(r11_dst_klass, dst); // reload
2843
2844 // Marshal the base address arguments now, freeing registers.
2845 __ lea(from, Address(src, src_pos, TIMES_OOP,
2846 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2847 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2848 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2849 __ movl(count, length); // length (reloaded)
2850 Register sco_temp = c_rarg3; // this register is free now
2851 assert_different_registers(from, to, count, sco_temp,
2852 r11_dst_klass, r10_src_klass);
2853 assert_clean_int(count, sco_temp);
2854
2855 // Generate the type check.
2856 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2857 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2858 assert_clean_int(sco_temp, rax);
2859 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2860
2861 // Fetch destination element klass from the ObjArrayKlass header.
2862 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
2863 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2864 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2865 assert_clean_int(sco_temp, rax);
2866
2867 // the checkcast_copy loop needs two extra arguments:
2868 assert(c_rarg3 == sco_temp, "#3 already in place");
2869 // Set up arguments for checkcast_copy_entry.
2870 setup_arg_regs(4);
2871 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2872 __ jump(RuntimeAddress(checkcast_copy_entry));
2873 }
2874
2875 __ BIND(L_failed);
2876 __ xorptr(rax, rax);
2877 __ notptr(rax); // return -1
2878 __ leave(); // required for proper stackwalking of RuntimeStub frame
2879 __ ret(0);
2880
2881 return start;
2882 }
2883
2884 void generate_arraycopy_stubs() {
2885 address entry;
2886 address entry_jbyte_arraycopy;
2887 address entry_jshort_arraycopy;
2888 address entry_jint_arraycopy;
2889 address entry_oop_arraycopy;
2890 address entry_jlong_arraycopy;
2891 address entry_checkcast_arraycopy;
2892
2893 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2894 "jbyte_disjoint_arraycopy");
2895 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2896 "jbyte_arraycopy");
2897
2898 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2899 "jshort_disjoint_arraycopy");
2900 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2901 "jshort_arraycopy");
2902
2903 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2904 "jint_disjoint_arraycopy");
2905 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2906 &entry_jint_arraycopy, "jint_arraycopy");
2907
2908 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2909 "jlong_disjoint_arraycopy");
2910 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2911 &entry_jlong_arraycopy, "jlong_arraycopy");
2912
2913
2914 if (UseCompressedOops) {
2915 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2916 "oop_disjoint_arraycopy");
2917 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2918 &entry_oop_arraycopy, "oop_arraycopy");
2919 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2920 "oop_disjoint_arraycopy_uninit",
2921 /*dest_uninitialized*/true);
2922 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2923 NULL, "oop_arraycopy_uninit",
2924 /*dest_uninitialized*/true);
2925 } else {
2926 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2927 "oop_disjoint_arraycopy");
2928 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2929 &entry_oop_arraycopy, "oop_arraycopy");
2930 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2931 "oop_disjoint_arraycopy_uninit",
2932 /*dest_uninitialized*/true);
2933 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2934 NULL, "oop_arraycopy_uninit",
2935 /*dest_uninitialized*/true);
2936 }
2937
2938 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2939 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2940 /*dest_uninitialized*/true);
2941
2942 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2943 entry_jbyte_arraycopy,
2944 entry_jshort_arraycopy,
2945 entry_jint_arraycopy,
2946 entry_jlong_arraycopy);
2947 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2948 entry_jbyte_arraycopy,
2949 entry_jshort_arraycopy,
2950 entry_jint_arraycopy,
2951 entry_oop_arraycopy,
2952 entry_jlong_arraycopy,
2953 entry_checkcast_arraycopy);
2954
2955 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2956 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2957 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2958 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2959 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2960 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2961
2962 // We don't generate specialized code for HeapWord-aligned source
2963 // arrays, so just use the code we've already generated
2964 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2965 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2966
2967 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2968 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2969
2970 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2971 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2972
2973 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2974 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2975
2976 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2977 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2978
2979 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2980 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2981 }
2982
2983 // AES intrinsic stubs
2984 enum {AESBlockSize = 16};
2985
2986 address generate_key_shuffle_mask() {
2987 __ align(16);
2988 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2989 address start = __ pc();
2990 __ emit_data64( 0x0405060700010203, relocInfo::none );
2991 __ emit_data64( 0x0c0d0e0f08090a0b, relocInfo::none );
2992 return start;
2993 }
2994
2995 address generate_counter_shuffle_mask() {
2996 __ align(16);
2997 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2998 address start = __ pc();
2999 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3000 __ emit_data64(0x0001020304050607, relocInfo::none);
3001 return start;
3002 }
3003
3004 // Utility routine for loading a 128-bit key word in little endian format
3005 // can optionally specify that the shuffle mask is already in an xmmregister
3006 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
3007 __ movdqu(xmmdst, Address(key, offset));
3008 if (xmm_shuf_mask != NULL) {
3009 __ pshufb(xmmdst, xmm_shuf_mask);
3010 } else {
3011 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3012 }
3013 }
3014
3015 // Utility routine for increase 128bit counter (iv in CTR mode)
3016 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
3017 __ pextrq(reg, xmmdst, 0x0);
3018 __ addq(reg, inc_delta);
3019 __ pinsrq(xmmdst, reg, 0x0);
3020 __ jcc(Assembler::carryClear, next_block); // jump if no carry
3021 __ pextrq(reg, xmmdst, 0x01); // Carry
3022 __ addq(reg, 0x01);
3023 __ pinsrq(xmmdst, reg, 0x01); //Carry end
3024 __ BIND(next_block); // next instruction
3025 }
3026
3027 // Arguments:
3028 //
3029 // Inputs:
3030 // c_rarg0 - source byte array address
3031 // c_rarg1 - destination byte array address
3032 // c_rarg2 - K (key) in little endian int array
3033 //
3034 address generate_aescrypt_encryptBlock() {
3035 assert(UseAES, "need AES instructions and misaligned SSE support");
3036 __ align(CodeEntryAlignment);
3037 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
3038 Label L_doLast;
3039 address start = __ pc();
3040
3041 const Register from = c_rarg0; // source array address
3042 const Register to = c_rarg1; // destination array address
3043 const Register key = c_rarg2; // key array address
3044 const Register keylen = rax;
3045
3046 const XMMRegister xmm_result = xmm0;
3047 const XMMRegister xmm_key_shuf_mask = xmm1;
3048 // On win64 xmm6-xmm15 must be preserved so don't use them.
3049 const XMMRegister xmm_temp1 = xmm2;
3050 const XMMRegister xmm_temp2 = xmm3;
3051 const XMMRegister xmm_temp3 = xmm4;
3052 const XMMRegister xmm_temp4 = xmm5;
3053
3054 __ enter(); // required for proper stackwalking of RuntimeStub frame
3055
3056 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3057 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3058
3059 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3060 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
3061
3062 // For encryption, the java expanded key ordering is just what we need
3063 // we don't know if the key is aligned, hence not using load-execute form
3064
3065 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
3066 __ pxor(xmm_result, xmm_temp1);
3067
3068 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3069 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3070 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3071 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3072
3073 __ aesenc(xmm_result, xmm_temp1);
3074 __ aesenc(xmm_result, xmm_temp2);
3075 __ aesenc(xmm_result, xmm_temp3);
3076 __ aesenc(xmm_result, xmm_temp4);
3077
3078 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3079 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3080 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3081 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3082
3083 __ aesenc(xmm_result, xmm_temp1);
3084 __ aesenc(xmm_result, xmm_temp2);
3085 __ aesenc(xmm_result, xmm_temp3);
3086 __ aesenc(xmm_result, xmm_temp4);
3087
3088 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3089 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3090
3091 __ cmpl(keylen, 44);
3092 __ jccb(Assembler::equal, L_doLast);
3093
3094 __ aesenc(xmm_result, xmm_temp1);
3095 __ aesenc(xmm_result, xmm_temp2);
3096
3097 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3098 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3099
3100 __ cmpl(keylen, 52);
3101 __ jccb(Assembler::equal, L_doLast);
3102
3103 __ aesenc(xmm_result, xmm_temp1);
3104 __ aesenc(xmm_result, xmm_temp2);
3105
3106 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3107 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3108
3109 __ BIND(L_doLast);
3110 __ aesenc(xmm_result, xmm_temp1);
3111 __ aesenclast(xmm_result, xmm_temp2);
3112 __ movdqu(Address(to, 0), xmm_result); // store the result
3113 __ xorptr(rax, rax); // return 0
3114 __ leave(); // required for proper stackwalking of RuntimeStub frame
3115 __ ret(0);
3116
3117 return start;
3118 }
3119
3120
3121 // Arguments:
3122 //
3123 // Inputs:
3124 // c_rarg0 - source byte array address
3125 // c_rarg1 - destination byte array address
3126 // c_rarg2 - K (key) in little endian int array
3127 //
3128 address generate_aescrypt_decryptBlock() {
3129 assert(UseAES, "need AES instructions and misaligned SSE support");
3130 __ align(CodeEntryAlignment);
3131 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
3132 Label L_doLast;
3133 address start = __ pc();
3134
3135 const Register from = c_rarg0; // source array address
3136 const Register to = c_rarg1; // destination array address
3137 const Register key = c_rarg2; // key array address
3138 const Register keylen = rax;
3139
3140 const XMMRegister xmm_result = xmm0;
3141 const XMMRegister xmm_key_shuf_mask = xmm1;
3142 // On win64 xmm6-xmm15 must be preserved so don't use them.
3143 const XMMRegister xmm_temp1 = xmm2;
3144 const XMMRegister xmm_temp2 = xmm3;
3145 const XMMRegister xmm_temp3 = xmm4;
3146 const XMMRegister xmm_temp4 = xmm5;
3147
3148 __ enter(); // required for proper stackwalking of RuntimeStub frame
3149
3150 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
3151 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3152
3153 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3154 __ movdqu(xmm_result, Address(from, 0));
3155
3156 // for decryption java expanded key ordering is rotated one position from what we want
3157 // so we start from 0x10 here and hit 0x00 last
3158 // we don't know if the key is aligned, hence not using load-execute form
3159 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
3160 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
3161 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
3162 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
3163
3164 __ pxor (xmm_result, xmm_temp1);
3165 __ aesdec(xmm_result, xmm_temp2);
3166 __ aesdec(xmm_result, xmm_temp3);
3167 __ aesdec(xmm_result, xmm_temp4);
3168
3169 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
3170 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
3171 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
3172 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
3173
3174 __ aesdec(xmm_result, xmm_temp1);
3175 __ aesdec(xmm_result, xmm_temp2);
3176 __ aesdec(xmm_result, xmm_temp3);
3177 __ aesdec(xmm_result, xmm_temp4);
3178
3179 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
3180 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
3181 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
3182
3183 __ cmpl(keylen, 44);
3184 __ jccb(Assembler::equal, L_doLast);
3185
3186 __ aesdec(xmm_result, xmm_temp1);
3187 __ aesdec(xmm_result, xmm_temp2);
3188
3189 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
3190 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
3191
3192 __ cmpl(keylen, 52);
3193 __ jccb(Assembler::equal, L_doLast);
3194
3195 __ aesdec(xmm_result, xmm_temp1);
3196 __ aesdec(xmm_result, xmm_temp2);
3197
3198 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
3199 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
3200
3201 __ BIND(L_doLast);
3202 __ aesdec(xmm_result, xmm_temp1);
3203 __ aesdec(xmm_result, xmm_temp2);
3204
3205 // for decryption the aesdeclast operation is always on key+0x00
3206 __ aesdeclast(xmm_result, xmm_temp3);
3207 __ movdqu(Address(to, 0), xmm_result); // store the result
3208 __ xorptr(rax, rax); // return 0
3209 __ leave(); // required for proper stackwalking of RuntimeStub frame
3210 __ ret(0);
3211
3212 return start;
3213 }
3214
3215
3216 // Arguments:
3217 //
3218 // Inputs:
3219 // c_rarg0 - source byte array address
3220 // c_rarg1 - destination byte array address
3221 // c_rarg2 - K (key) in little endian int array
3222 // c_rarg3 - r vector byte array address
3223 // c_rarg4 - input length
3224 //
3225 // Output:
3226 // rax - input length
3227 //
3228 address generate_cipherBlockChaining_encryptAESCrypt() {
3229 assert(UseAES, "need AES instructions and misaligned SSE support");
3230 __ align(CodeEntryAlignment);
3231 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
3232 address start = __ pc();
3233
3234 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
3235 const Register from = c_rarg0; // source array address
3236 const Register to = c_rarg1; // destination array address
3237 const Register key = c_rarg2; // key array address
3238 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3239 // and left with the results of the last encryption block
3240 #ifndef _WIN64
3241 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3242 #else
3243 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3244 const Register len_reg = r11; // pick the volatile windows register
3245 #endif
3246 const Register pos = rax;
3247
3248 // xmm register assignments for the loops below
3249 const XMMRegister xmm_result = xmm0;
3250 const XMMRegister xmm_temp = xmm1;
3251 // keys 0-10 preloaded into xmm2-xmm12
3252 const int XMM_REG_NUM_KEY_FIRST = 2;
3253 const int XMM_REG_NUM_KEY_LAST = 15;
3254 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3255 const XMMRegister xmm_key10 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+10);
3256 const XMMRegister xmm_key11 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+11);
3257 const XMMRegister xmm_key12 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+12);
3258 const XMMRegister xmm_key13 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST+13);
3259
3260 __ enter(); // required for proper stackwalking of RuntimeStub frame
3261
3262 #ifdef _WIN64
3263 // on win64, fill len_reg from stack position
3264 __ movl(len_reg, len_mem);
3265 #else
3266 __ push(len_reg); // Save
3267 #endif
3268
3269 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
3270 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3271 // load up xmm regs xmm2 thru xmm12 with key 0x00 - 0xa0
3272 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_FIRST+10; rnum++) {
3273 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3274 offset += 0x10;
3275 }
3276 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
3277
3278 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3279 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3280 __ cmpl(rax, 44);
3281 __ jcc(Assembler::notEqual, L_key_192_256);
3282
3283 // 128 bit code follows here
3284 __ movptr(pos, 0);
3285 __ align(OptoLoopAlignment);
3286
3287 __ BIND(L_loopTop_128);
3288 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3289 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3290 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3291 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 9; rnum++) {
3292 __ aesenc(xmm_result, as_XMMRegister(rnum));
3293 }
3294 __ aesenclast(xmm_result, xmm_key10);
3295 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3296 // no need to store r to memory until we exit
3297 __ addptr(pos, AESBlockSize);
3298 __ subptr(len_reg, AESBlockSize);
3299 __ jcc(Assembler::notEqual, L_loopTop_128);
3300
3301 __ BIND(L_exit);
3302 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
3303
3304 #ifdef _WIN64
3305 __ movl(rax, len_mem);
3306 #else
3307 __ pop(rax); // return length
3308 #endif
3309 __ leave(); // required for proper stackwalking of RuntimeStub frame
3310 __ ret(0);
3311
3312 __ BIND(L_key_192_256);
3313 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
3314 load_key(xmm_key11, key, 0xb0, xmm_key_shuf_mask);
3315 load_key(xmm_key12, key, 0xc0, xmm_key_shuf_mask);
3316 __ cmpl(rax, 52);
3317 __ jcc(Assembler::notEqual, L_key_256);
3318
3319 // 192-bit code follows here (could be changed to use more xmm registers)
3320 __ movptr(pos, 0);
3321 __ align(OptoLoopAlignment);
3322
3323 __ BIND(L_loopTop_192);
3324 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3325 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3326 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3327 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 11; rnum++) {
3328 __ aesenc(xmm_result, as_XMMRegister(rnum));
3329 }
3330 __ aesenclast(xmm_result, xmm_key12);
3331 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3332 // no need to store r to memory until we exit
3333 __ addptr(pos, AESBlockSize);
3334 __ subptr(len_reg, AESBlockSize);
3335 __ jcc(Assembler::notEqual, L_loopTop_192);
3336 __ jmp(L_exit);
3337
3338 __ BIND(L_key_256);
3339 // 256-bit code follows here (could be changed to use more xmm registers)
3340 load_key(xmm_key13, key, 0xd0, xmm_key_shuf_mask);
3341 __ movptr(pos, 0);
3342 __ align(OptoLoopAlignment);
3343
3344 __ BIND(L_loopTop_256);
3345 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
3346 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
3347 __ pxor (xmm_result, xmm_key0); // do the aes rounds
3348 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_FIRST + 13; rnum++) {
3349 __ aesenc(xmm_result, as_XMMRegister(rnum));
3350 }
3351 load_key(xmm_temp, key, 0xe0);
3352 __ aesenclast(xmm_result, xmm_temp);
3353 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3354 // no need to store r to memory until we exit
3355 __ addptr(pos, AESBlockSize);
3356 __ subptr(len_reg, AESBlockSize);
3357 __ jcc(Assembler::notEqual, L_loopTop_256);
3358 __ jmp(L_exit);
3359
3360 return start;
3361 }
3362
3363 // Safefetch stubs.
3364 void generate_safefetch(const char* name, int size, address* entry,
3365 address* fault_pc, address* continuation_pc) {
3366 // safefetch signatures:
3367 // int SafeFetch32(int* adr, int errValue);
3368 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3369 //
3370 // arguments:
3371 // c_rarg0 = adr
3372 // c_rarg1 = errValue
3373 //
3374 // result:
3375 // PPC_RET = *adr or errValue
3376
3377 StubCodeMark mark(this, "StubRoutines", name);
3378
3379 // Entry point, pc or function descriptor.
3380 *entry = __ pc();
3381
3382 // Load *adr into c_rarg1, may fault.
3383 *fault_pc = __ pc();
3384 switch (size) {
3385 case 4:
3386 // int32_t
3387 __ movl(c_rarg1, Address(c_rarg0, 0));
3388 break;
3389 case 8:
3390 // int64_t
3391 __ movq(c_rarg1, Address(c_rarg0, 0));
3392 break;
3393 default:
3394 ShouldNotReachHere();
3395 }
3396
3397 // return errValue or *adr
3398 *continuation_pc = __ pc();
3399 __ movq(rax, c_rarg1);
3400 __ ret(0);
3401 }
3402
3403 // This is a version of CBC/AES Decrypt which does 4 blocks in a loop at a time
3404 // to hide instruction latency
3405 //
3406 // Arguments:
3407 //
3408 // Inputs:
3409 // c_rarg0 - source byte array address
3410 // c_rarg1 - destination byte array address
3411 // c_rarg2 - K (key) in little endian int array
3412 // c_rarg3 - r vector byte array address
3413 // c_rarg4 - input length
3414 //
3415 // Output:
3416 // rax - input length
3417 //
3418 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
3419 assert(UseAES, "need AES instructions and misaligned SSE support");
3420 __ align(CodeEntryAlignment);
3421 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
3422 address start = __ pc();
3423
3424 const Register from = c_rarg0; // source array address
3425 const Register to = c_rarg1; // destination array address
3426 const Register key = c_rarg2; // key array address
3427 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
3428 // and left with the results of the last encryption block
3429 #ifndef _WIN64
3430 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
3431 #else
3432 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3433 const Register len_reg = r11; // pick the volatile windows register
3434 #endif
3435 const Register pos = rax;
3436
3437 const int PARALLEL_FACTOR = 4;
3438 const int ROUNDS[3] = { 10, 12, 14 }; // aes rounds for key128, key192, key256
3439
3440 Label L_exit;
3441 Label L_singleBlock_loopTopHead[3]; // 128, 192, 256
3442 Label L_singleBlock_loopTopHead2[3]; // 128, 192, 256
3443 Label L_singleBlock_loopTop[3]; // 128, 192, 256
3444 Label L_multiBlock_loopTopHead[3]; // 128, 192, 256
3445 Label L_multiBlock_loopTop[3]; // 128, 192, 256
3446
3447 // keys 0-10 preloaded into xmm5-xmm15
3448 const int XMM_REG_NUM_KEY_FIRST = 5;
3449 const int XMM_REG_NUM_KEY_LAST = 15;
3450 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
3451 const XMMRegister xmm_key_last = as_XMMRegister(XMM_REG_NUM_KEY_LAST);
3452
3453 __ enter(); // required for proper stackwalking of RuntimeStub frame
3454
3455 #ifdef _WIN64
3456 // on win64, fill len_reg from stack position
3457 __ movl(len_reg, len_mem);
3458 #else
3459 __ push(len_reg); // Save
3460 #endif
3461 __ push(rbx);
3462 // the java expanded key ordering is rotated one position from what we want
3463 // so we start from 0x10 here and hit 0x00 last
3464 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3465 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
3466 // load up xmm regs 5 thru 15 with key 0x10 - 0xa0 - 0x00
3467 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum < XMM_REG_NUM_KEY_LAST; rnum++) {
3468 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
3469 offset += 0x10;
3470 }
3471 load_key(xmm_key_last, key, 0x00, xmm_key_shuf_mask);
3472
3473 const XMMRegister xmm_prev_block_cipher = xmm1; // holds cipher of previous block
3474
3475 // registers holding the four results in the parallelized loop
3476 const XMMRegister xmm_result0 = xmm0;
3477 const XMMRegister xmm_result1 = xmm2;
3478 const XMMRegister xmm_result2 = xmm3;
3479 const XMMRegister xmm_result3 = xmm4;
3480
3481 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
3482
3483 __ xorptr(pos, pos);
3484
3485 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
3486 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3487 __ cmpl(rbx, 52);
3488 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[1]);
3489 __ cmpl(rbx, 60);
3490 __ jcc(Assembler::equal, L_multiBlock_loopTopHead[2]);
3491
3492 #define DoFour(opc, src_reg) \
3493 __ opc(xmm_result0, src_reg); \
3494 __ opc(xmm_result1, src_reg); \
3495 __ opc(xmm_result2, src_reg); \
3496 __ opc(xmm_result3, src_reg); \
3497
3498 for (int k = 0; k < 3; ++k) {
3499 __ BIND(L_multiBlock_loopTopHead[k]);
3500 if (k != 0) {
3501 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3502 __ jcc(Assembler::less, L_singleBlock_loopTopHead2[k]);
3503 }
3504 if (k == 1) {
3505 __ subptr(rsp, 6 * wordSize);
3506 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3507 load_key(xmm15, key, 0xb0); // 0xb0; 192-bit key goes up to 0xc0
3508 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3509 load_key(xmm1, key, 0xc0); // 0xc0;
3510 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3511 } else if (k == 2) {
3512 __ subptr(rsp, 10 * wordSize);
3513 __ movdqu(Address(rsp, 0), xmm15); //save last_key from xmm15
3514 load_key(xmm15, key, 0xd0); // 0xd0; 256-bit key goes upto 0xe0
3515 __ movdqu(Address(rsp, 6 * wordSize), xmm15);
3516 load_key(xmm1, key, 0xe0); // 0xe0;
3517 __ movdqu(Address(rsp, 8 * wordSize), xmm1);
3518 load_key(xmm15, key, 0xb0); // 0xb0;
3519 __ movdqu(Address(rsp, 2 * wordSize), xmm15);
3520 load_key(xmm1, key, 0xc0); // 0xc0;
3521 __ movdqu(Address(rsp, 4 * wordSize), xmm1);
3522 }
3523 __ align(OptoLoopAlignment);
3524 __ BIND(L_multiBlock_loopTop[k]);
3525 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
3526 __ jcc(Assembler::less, L_singleBlock_loopTopHead[k]);
3527
3528 if (k != 0) {
3529 __ movdqu(xmm15, Address(rsp, 2 * wordSize));
3530 __ movdqu(xmm1, Address(rsp, 4 * wordSize));
3531 }
3532
3533 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
3534 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3535 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3536 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
3537
3538 DoFour(pxor, xmm_key_first);
3539 if (k == 0) {
3540 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
3541 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3542 }
3543 DoFour(aesdeclast, xmm_key_last);
3544 } else if (k == 1) {
3545 for (int rnum = 1; rnum <= ROUNDS[k]-2; rnum++) {
3546 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3547 }
3548 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3549 DoFour(aesdec, xmm1); // key : 0xc0
3550 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3551 DoFour(aesdeclast, xmm_key_last);
3552 } else if (k == 2) {
3553 for (int rnum = 1; rnum <= ROUNDS[k] - 4; rnum++) {
3554 DoFour(aesdec, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3555 }
3556 DoFour(aesdec, xmm1); // key : 0xc0
3557 __ movdqu(xmm15, Address(rsp, 6 * wordSize));
3558 __ movdqu(xmm1, Address(rsp, 8 * wordSize));
3559 DoFour(aesdec, xmm15); // key : 0xd0
3560 __ movdqu(xmm_key_last, Address(rsp, 0)); // xmm15 needs to be loaded again.
3561 DoFour(aesdec, xmm1); // key : 0xe0
3562 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // xmm1 needs to be loaded again
3563 DoFour(aesdeclast, xmm_key_last);
3564 }
3565
3566 // for each result, xor with the r vector of previous cipher block
3567 __ pxor(xmm_result0, xmm_prev_block_cipher);
3568 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
3569 __ pxor(xmm_result1, xmm_prev_block_cipher);
3570 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
3571 __ pxor(xmm_result2, xmm_prev_block_cipher);
3572 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
3573 __ pxor(xmm_result3, xmm_prev_block_cipher);
3574 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
3575 if (k != 0) {
3576 __ movdqu(Address(rvec, 0x00), xmm_prev_block_cipher);
3577 }
3578
3579 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0); // store 4 results into the next 64 bytes of output
3580 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
3581 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
3582 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
3583
3584 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize);
3585 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize);
3586 __ jmp(L_multiBlock_loopTop[k]);
3587
3588 // registers used in the non-parallelized loops
3589 // xmm register assignments for the loops below
3590 const XMMRegister xmm_result = xmm0;
3591 const XMMRegister xmm_prev_block_cipher_save = xmm2;
3592 const XMMRegister xmm_key11 = xmm3;
3593 const XMMRegister xmm_key12 = xmm4;
3594 const XMMRegister key_tmp = xmm4;
3595
3596 __ BIND(L_singleBlock_loopTopHead[k]);
3597 if (k == 1) {
3598 __ addptr(rsp, 6 * wordSize);
3599 } else if (k == 2) {
3600 __ addptr(rsp, 10 * wordSize);
3601 }
3602 __ cmpptr(len_reg, 0); // any blocks left??
3603 __ jcc(Assembler::equal, L_exit);
3604 __ BIND(L_singleBlock_loopTopHead2[k]);
3605 if (k == 1) {
3606 load_key(xmm_key11, key, 0xb0); // 0xb0; 192-bit key goes upto 0xc0
3607 load_key(xmm_key12, key, 0xc0); // 0xc0; 192-bit key goes upto 0xc0
3608 }
3609 if (k == 2) {
3610 load_key(xmm_key11, key, 0xb0); // 0xb0; 256-bit key goes upto 0xe0
3611 }
3612 __ align(OptoLoopAlignment);
3613 __ BIND(L_singleBlock_loopTop[k]);
3614 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
3615 __ movdqa(xmm_prev_block_cipher_save, xmm_result); // save for next r vector
3616 __ pxor(xmm_result, xmm_key_first); // do the aes dec rounds
3617 for (int rnum = 1; rnum <= 9 ; rnum++) {
3618 __ aesdec(xmm_result, as_XMMRegister(rnum + XMM_REG_NUM_KEY_FIRST));
3619 }
3620 if (k == 1) {
3621 __ aesdec(xmm_result, xmm_key11);
3622 __ aesdec(xmm_result, xmm_key12);
3623 }
3624 if (k == 2) {
3625 __ aesdec(xmm_result, xmm_key11);
3626 load_key(key_tmp, key, 0xc0);
3627 __ aesdec(xmm_result, key_tmp);
3628 load_key(key_tmp, key, 0xd0);
3629 __ aesdec(xmm_result, key_tmp);
3630 load_key(key_tmp, key, 0xe0);
3631 __ aesdec(xmm_result, key_tmp);
3632 }
3633
3634 __ aesdeclast(xmm_result, xmm_key_last); // xmm15 always came from key+0
3635 __ pxor(xmm_result, xmm_prev_block_cipher); // xor with the current r vector
3636 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
3637 // no need to store r to memory until we exit
3638 __ movdqa(xmm_prev_block_cipher, xmm_prev_block_cipher_save); // set up next r vector with cipher input from this block
3639 __ addptr(pos, AESBlockSize);
3640 __ subptr(len_reg, AESBlockSize);
3641 __ jcc(Assembler::notEqual, L_singleBlock_loopTop[k]);
3642 if (k != 2) {
3643 __ jmp(L_exit);
3644 }
3645 } //for 128/192/256
3646
3647 __ BIND(L_exit);
3648 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
3649 __ pop(rbx);
3650 #ifdef _WIN64
3651 __ movl(rax, len_mem);
3652 #else
3653 __ pop(rax); // return length
3654 #endif
3655 __ leave(); // required for proper stackwalking of RuntimeStub frame
3656 __ ret(0);
3657 return start;
3658 }
3659
3660 address generate_upper_word_mask() {
3661 __ align(64);
3662 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3663 address start = __ pc();
3664 __ emit_data64(0x0000000000000000, relocInfo::none);
3665 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
3666 return start;
3667 }
3668
3669 address generate_shuffle_byte_flip_mask() {
3670 __ align(64);
3671 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3672 address start = __ pc();
3673 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3674 __ emit_data64(0x0001020304050607, relocInfo::none);
3675 return start;
3676 }
3677
3678 // ofs and limit are use for multi-block byte array.
3679 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3680 address generate_sha1_implCompress(bool multi_block, const char *name) {
3681 __ align(CodeEntryAlignment);
3682 StubCodeMark mark(this, "StubRoutines", name);
3683 address start = __ pc();
3684
3685 Register buf = c_rarg0;
3686 Register state = c_rarg1;
3687 Register ofs = c_rarg2;
3688 Register limit = c_rarg3;
3689
3690 const XMMRegister abcd = xmm0;
3691 const XMMRegister e0 = xmm1;
3692 const XMMRegister e1 = xmm2;
3693 const XMMRegister msg0 = xmm3;
3694
3695 const XMMRegister msg1 = xmm4;
3696 const XMMRegister msg2 = xmm5;
3697 const XMMRegister msg3 = xmm6;
3698 const XMMRegister shuf_mask = xmm7;
3699
3700 __ enter();
3701
3702 __ subptr(rsp, 4 * wordSize);
3703
3704 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3705 buf, state, ofs, limit, rsp, multi_block);
3706
3707 __ addptr(rsp, 4 * wordSize);
3708
3709 __ leave();
3710 __ ret(0);
3711 return start;
3712 }
3713
3714 address generate_pshuffle_byte_flip_mask() {
3715 __ align(64);
3716 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3717 address start = __ pc();
3718 __ emit_data64(0x0405060700010203, relocInfo::none);
3719 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3720
3721 if (VM_Version::supports_avx2()) {
3722 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
3723 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
3724 // _SHUF_00BA
3725 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3726 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3727 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3728 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3729 // _SHUF_DC00
3730 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3731 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3732 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3733 __ emit_data64(0x0b0a090803020100, relocInfo::none);
3734 }
3735
3736 return start;
3737 }
3738
3739 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
3740 address generate_pshuffle_byte_flip_mask_sha512() {
3741 __ align(32);
3742 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
3743 address start = __ pc();
3744 if (VM_Version::supports_avx2()) {
3745 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
3746 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
3747 __ emit_data64(0x1011121314151617, relocInfo::none);
3748 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
3749 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
3750 __ emit_data64(0x0000000000000000, relocInfo::none);
3751 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3752 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
3753 }
3754
3755 return start;
3756 }
3757
3758 // ofs and limit are use for multi-block byte array.
3759 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3760 address generate_sha256_implCompress(bool multi_block, const char *name) {
3761 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
3762 __ align(CodeEntryAlignment);
3763 StubCodeMark mark(this, "StubRoutines", name);
3764 address start = __ pc();
3765
3766 Register buf = c_rarg0;
3767 Register state = c_rarg1;
3768 Register ofs = c_rarg2;
3769 Register limit = c_rarg3;
3770
3771 const XMMRegister msg = xmm0;
3772 const XMMRegister state0 = xmm1;
3773 const XMMRegister state1 = xmm2;
3774 const XMMRegister msgtmp0 = xmm3;
3775
3776 const XMMRegister msgtmp1 = xmm4;
3777 const XMMRegister msgtmp2 = xmm5;
3778 const XMMRegister msgtmp3 = xmm6;
3779 const XMMRegister msgtmp4 = xmm7;
3780
3781 const XMMRegister shuf_mask = xmm8;
3782
3783 __ enter();
3784
3785 __ subptr(rsp, 4 * wordSize);
3786
3787 if (VM_Version::supports_sha()) {
3788 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3789 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3790 } else if (VM_Version::supports_avx2()) {
3791 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3792 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3793 }
3794 __ addptr(rsp, 4 * wordSize);
3795 __ vzeroupper();
3796 __ leave();
3797 __ ret(0);
3798 return start;
3799 }
3800
3801 address generate_sha512_implCompress(bool multi_block, const char *name) {
3802 assert(VM_Version::supports_avx2(), "");
3803 assert(VM_Version::supports_bmi2(), "");
3804 __ align(CodeEntryAlignment);
3805 StubCodeMark mark(this, "StubRoutines", name);
3806 address start = __ pc();
3807
3808 Register buf = c_rarg0;
3809 Register state = c_rarg1;
3810 Register ofs = c_rarg2;
3811 Register limit = c_rarg3;
3812
3813 const XMMRegister msg = xmm0;
3814 const XMMRegister state0 = xmm1;
3815 const XMMRegister state1 = xmm2;
3816 const XMMRegister msgtmp0 = xmm3;
3817 const XMMRegister msgtmp1 = xmm4;
3818 const XMMRegister msgtmp2 = xmm5;
3819 const XMMRegister msgtmp3 = xmm6;
3820 const XMMRegister msgtmp4 = xmm7;
3821
3822 const XMMRegister shuf_mask = xmm8;
3823
3824 __ enter();
3825
3826 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3827 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
3828
3829 __ vzeroupper();
3830 __ leave();
3831 __ ret(0);
3832 return start;
3833 }
3834
3835 // This is a version of CTR/AES crypt which does 6 blocks in a loop at a time
3836 // to hide instruction latency
3837 //
3838 // Arguments:
3839 //
3840 // Inputs:
3841 // c_rarg0 - source byte array address
3842 // c_rarg1 - destination byte array address
3843 // c_rarg2 - K (key) in little endian int array
3844 // c_rarg3 - counter vector byte array address
3845 // Linux
3846 // c_rarg4 - input length
3847 // c_rarg5 - saved encryptedCounter start
3848 // rbp + 6 * wordSize - saved used length
3849 // Windows
3850 // rbp + 6 * wordSize - input length
3851 // rbp + 7 * wordSize - saved encryptedCounter start
3852 // rbp + 8 * wordSize - saved used length
3853 //
3854 // Output:
3855 // rax - input length
3856 //
3857 address generate_counterMode_AESCrypt_Parallel() {
3858 assert(UseAES, "need AES instructions and misaligned SSE support");
3859 __ align(CodeEntryAlignment);
3860 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
3861 address start = __ pc();
3862 const Register from = c_rarg0; // source array address
3863 const Register to = c_rarg1; // destination array address
3864 const Register key = c_rarg2; // key array address
3865 const Register counter = c_rarg3; // counter byte array initialized from counter array address
3866 // and updated with the incremented counter in the end
3867 #ifndef _WIN64
3868 const Register len_reg = c_rarg4;
3869 const Register saved_encCounter_start = c_rarg5;
3870 const Register used_addr = r10;
3871 const Address used_mem(rbp, 2 * wordSize);
3872 const Register used = r11;
3873 #else
3874 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
3875 const Address saved_encCounter_mem(rbp, 7 * wordSize); // length is on stack on Win64
3876 const Address used_mem(rbp, 8 * wordSize); // length is on stack on Win64
3877 const Register len_reg = r10; // pick the first volatile windows register
3878 const Register saved_encCounter_start = r11;
3879 const Register used_addr = r13;
3880 const Register used = r14;
3881 #endif
3882 const Register pos = rax;
3883
3884 const int PARALLEL_FACTOR = 6;
3885 const XMMRegister xmm_counter_shuf_mask = xmm0;
3886 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
3887 const XMMRegister xmm_curr_counter = xmm2;
3888
3889 const XMMRegister xmm_key_tmp0 = xmm3;
3890 const XMMRegister xmm_key_tmp1 = xmm4;
3891
3892 // registers holding the four results in the parallelized loop
3893 const XMMRegister xmm_result0 = xmm5;
3894 const XMMRegister xmm_result1 = xmm6;
3895 const XMMRegister xmm_result2 = xmm7;
3896 const XMMRegister xmm_result3 = xmm8;
3897 const XMMRegister xmm_result4 = xmm9;
3898 const XMMRegister xmm_result5 = xmm10;
3899
3900 const XMMRegister xmm_from0 = xmm11;
3901 const XMMRegister xmm_from1 = xmm12;
3902 const XMMRegister xmm_from2 = xmm13;
3903 const XMMRegister xmm_from3 = xmm14; //the last one is xmm14. we have to preserve it on WIN64.
3904 const XMMRegister xmm_from4 = xmm3; //reuse xmm3~4. Because xmm_key_tmp0~1 are useless when loading input text
3905 const XMMRegister xmm_from5 = xmm4;
3906
3907 //for key_128, key_192, key_256
3908 const int rounds[3] = {10, 12, 14};
3909 Label L_exit_preLoop, L_preLoop_start;
3910 Label L_multiBlock_loopTop[3];
3911 Label L_singleBlockLoopTop[3];
3912 Label L__incCounter[3][6]; //for 6 blocks
3913 Label L__incCounter_single[3]; //for single block, key128, key192, key256
3914 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];
3915 Label L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
3916
3917 Label L_exit;
3918
3919 __ enter(); // required for proper stackwalking of RuntimeStub frame
3920
3921 #ifdef _WIN64
3922 // allocate spill slots for r13, r14
3923 enum {
3924 saved_r13_offset,
3925 saved_r14_offset
3926 };
3927 __ subptr(rsp, 2 * wordSize);
3928 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
3929 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
3930
3931 // on win64, fill len_reg from stack position
3932 __ movl(len_reg, len_mem);
3933 __ movptr(saved_encCounter_start, saved_encCounter_mem);
3934 __ movptr(used_addr, used_mem);
3935 __ movl(used, Address(used_addr, 0));
3936 #else
3937 __ push(len_reg); // Save
3938 __ movptr(used_addr, used_mem);
3939 __ movl(used, Address(used_addr, 0));
3940 #endif
3941
3942 __ push(rbx); // Save RBX
3943 __ movdqu(xmm_curr_counter, Address(counter, 0x00)); // initialize counter with initial counter
3944 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()), pos); // pos as scratch
3945 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled
3946 __ movptr(pos, 0);
3947
3948 // Use the partially used encrpyted counter from last invocation
3949 __ BIND(L_preLoop_start);
3950 __ cmpptr(used, 16);
3951 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
3952 __ cmpptr(len_reg, 0);
3953 __ jcc(Assembler::lessEqual, L_exit_preLoop);
3954 __ movb(rbx, Address(saved_encCounter_start, used));
3955 __ xorb(rbx, Address(from, pos));
3956 __ movb(Address(to, pos), rbx);
3957 __ addptr(pos, 1);
3958 __ addptr(used, 1);
3959 __ subptr(len_reg, 1);
3960
3961 __ jmp(L_preLoop_start);
3962
3963 __ BIND(L_exit_preLoop);
3964 __ movl(Address(used_addr, 0), used);
3965
3966 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
3967 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()), rbx); // rbx as scratch
3968 __ movl(rbx, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
3969 __ cmpl(rbx, 52);
3970 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
3971 __ cmpl(rbx, 60);
3972 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
3973
3974 #define CTR_DoSix(opc, src_reg) \
3975 __ opc(xmm_result0, src_reg); \
3976 __ opc(xmm_result1, src_reg); \
3977 __ opc(xmm_result2, src_reg); \
3978 __ opc(xmm_result3, src_reg); \
3979 __ opc(xmm_result4, src_reg); \
3980 __ opc(xmm_result5, src_reg);
3981
3982 // k == 0 : generate code for key_128
3983 // k == 1 : generate code for key_192
3984 // k == 2 : generate code for key_256
3985 for (int k = 0; k < 3; ++k) {
3986 //multi blocks starts here
3987 __ align(OptoLoopAlignment);
3988 __ BIND(L_multiBlock_loopTop[k]);
3989 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
3990 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
3991 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
3992
3993 //load, then increase counters
3994 CTR_DoSix(movdqa, xmm_curr_counter);
3995 inc_counter(rbx, xmm_result1, 0x01, L__incCounter[k][0]);
3996 inc_counter(rbx, xmm_result2, 0x02, L__incCounter[k][1]);
3997 inc_counter(rbx, xmm_result3, 0x03, L__incCounter[k][2]);
3998 inc_counter(rbx, xmm_result4, 0x04, L__incCounter[k][3]);
3999 inc_counter(rbx, xmm_result5, 0x05, L__incCounter[k][4]);
4000 inc_counter(rbx, xmm_curr_counter, 0x06, L__incCounter[k][5]);
4001 CTR_DoSix(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
4002 CTR_DoSix(pxor, xmm_key_tmp0); //PXOR with Round 0 key
4003
4004 //load two ROUND_KEYs at a time
4005 for (int i = 1; i < rounds[k]; ) {
4006 load_key(xmm_key_tmp1, key, (0x10 * i), xmm_key_shuf_mask);
4007 load_key(xmm_key_tmp0, key, (0x10 * (i+1)), xmm_key_shuf_mask);
4008 CTR_DoSix(aesenc, xmm_key_tmp1);
4009 i++;
4010 if (i != rounds[k]) {
4011 CTR_DoSix(aesenc, xmm_key_tmp0);
4012 } else {
4013 CTR_DoSix(aesenclast, xmm_key_tmp0);
4014 }
4015 i++;
4016 }
4017
4018 // get next PARALLEL_FACTOR blocks into xmm_result registers
4019 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4020 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
4021 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
4022 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
4023 __ movdqu(xmm_from4, Address(from, pos, Address::times_1, 4 * AESBlockSize));
4024 __ movdqu(xmm_from5, Address(from, pos, Address::times_1, 5 * AESBlockSize));
4025
4026 __ pxor(xmm_result0, xmm_from0);
4027 __ pxor(xmm_result1, xmm_from1);
4028 __ pxor(xmm_result2, xmm_from2);
4029 __ pxor(xmm_result3, xmm_from3);
4030 __ pxor(xmm_result4, xmm_from4);
4031 __ pxor(xmm_result5, xmm_from5);
4032
4033 // store 6 results into the next 64 bytes of output
4034 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4035 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
4036 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
4037 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
4038 __ movdqu(Address(to, pos, Address::times_1, 4 * AESBlockSize), xmm_result4);
4039 __ movdqu(Address(to, pos, Address::times_1, 5 * AESBlockSize), xmm_result5);
4040
4041 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
4042 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
4043 __ jmp(L_multiBlock_loopTop[k]);
4044
4045 // singleBlock starts here
4046 __ align(OptoLoopAlignment);
4047 __ BIND(L_singleBlockLoopTop[k]);
4048 __ cmpptr(len_reg, 0);
4049 __ jcc(Assembler::lessEqual, L_exit);
4050 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
4051 __ movdqa(xmm_result0, xmm_curr_counter);
4052 inc_counter(rbx, xmm_curr_counter, 0x01, L__incCounter_single[k]);
4053 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
4054 __ pxor(xmm_result0, xmm_key_tmp0);
4055 for (int i = 1; i < rounds[k]; i++) {
4056 load_key(xmm_key_tmp0, key, (0x10 * i), xmm_key_shuf_mask);
4057 __ aesenc(xmm_result0, xmm_key_tmp0);
4058 }
4059 load_key(xmm_key_tmp0, key, (rounds[k] * 0x10), xmm_key_shuf_mask);
4060 __ aesenclast(xmm_result0, xmm_key_tmp0);
4061 __ cmpptr(len_reg, AESBlockSize);
4062 __ jcc(Assembler::less, L_processTail_insr[k]);
4063 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
4064 __ pxor(xmm_result0, xmm_from0);
4065 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
4066 __ addptr(pos, AESBlockSize);
4067 __ subptr(len_reg, AESBlockSize);
4068 __ jmp(L_singleBlockLoopTop[k]);
4069 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
4070 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
4071 __ testptr(len_reg, 8);
4072 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
4073 __ subptr(pos,8);
4074 __ pinsrq(xmm_from0, Address(from, pos), 0);
4075 __ BIND(L_processTail_4_insr[k]);
4076 __ testptr(len_reg, 4);
4077 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
4078 __ subptr(pos,4);
4079 __ pslldq(xmm_from0, 4);
4080 __ pinsrd(xmm_from0, Address(from, pos), 0);
4081 __ BIND(L_processTail_2_insr[k]);
4082 __ testptr(len_reg, 2);
4083 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
4084 __ subptr(pos, 2);
4085 __ pslldq(xmm_from0, 2);
4086 __ pinsrw(xmm_from0, Address(from, pos), 0);
4087 __ BIND(L_processTail_1_insr[k]);
4088 __ testptr(len_reg, 1);
4089 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
4090 __ subptr(pos, 1);
4091 __ pslldq(xmm_from0, 1);
4092 __ pinsrb(xmm_from0, Address(from, pos), 0);
4093 __ BIND(L_processTail_exit_insr[k]);
4094
4095 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
4096 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
4097
4098 __ testptr(len_reg, 8);
4099 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
4100 __ pextrq(Address(to, pos), xmm_result0, 0);
4101 __ psrldq(xmm_result0, 8);
4102 __ addptr(pos, 8);
4103 __ BIND(L_processTail_4_extr[k]);
4104 __ testptr(len_reg, 4);
4105 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
4106 __ pextrd(Address(to, pos), xmm_result0, 0);
4107 __ psrldq(xmm_result0, 4);
4108 __ addptr(pos, 4);
4109 __ BIND(L_processTail_2_extr[k]);
4110 __ testptr(len_reg, 2);
4111 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
4112 __ pextrw(Address(to, pos), xmm_result0, 0);
4113 __ psrldq(xmm_result0, 2);
4114 __ addptr(pos, 2);
4115 __ BIND(L_processTail_1_extr[k]);
4116 __ testptr(len_reg, 1);
4117 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
4118 __ pextrb(Address(to, pos), xmm_result0, 0);
4119
4120 __ BIND(L_processTail_exit_extr[k]);
4121 __ movl(Address(used_addr, 0), len_reg);
4122 __ jmp(L_exit);
4123
4124 }
4125
4126 __ BIND(L_exit);
4127 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
4128 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
4129 __ pop(rbx); // pop the saved RBX.
4130 #ifdef _WIN64
4131 __ movl(rax, len_mem);
4132 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
4133 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
4134 __ addptr(rsp, 2 * wordSize);
4135 #else
4136 __ pop(rax); // return 'len'
4137 #endif
4138 __ leave(); // required for proper stackwalking of RuntimeStub frame
4139 __ ret(0);
4140 return start;
4141 }
4142
4143 void roundDec(XMMRegister xmm_reg) {
4144 __ vaesdec(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4145 __ vaesdec(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4146 __ vaesdec(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4147 __ vaesdec(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4148 __ vaesdec(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4149 __ vaesdec(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4150 __ vaesdec(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4151 __ vaesdec(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4152 }
4153
4154 void roundDeclast(XMMRegister xmm_reg) {
4155 __ vaesdeclast(xmm1, xmm1, xmm_reg, Assembler::AVX_512bit);
4156 __ vaesdeclast(xmm2, xmm2, xmm_reg, Assembler::AVX_512bit);
4157 __ vaesdeclast(xmm3, xmm3, xmm_reg, Assembler::AVX_512bit);
4158 __ vaesdeclast(xmm4, xmm4, xmm_reg, Assembler::AVX_512bit);
4159 __ vaesdeclast(xmm5, xmm5, xmm_reg, Assembler::AVX_512bit);
4160 __ vaesdeclast(xmm6, xmm6, xmm_reg, Assembler::AVX_512bit);
4161 __ vaesdeclast(xmm7, xmm7, xmm_reg, Assembler::AVX_512bit);
4162 __ vaesdeclast(xmm8, xmm8, xmm_reg, Assembler::AVX_512bit);
4163 }
4164
4165 void ev_load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask = NULL) {
4166 __ movdqu(xmmdst, Address(key, offset));
4167 if (xmm_shuf_mask != NULL) {
4168 __ pshufb(xmmdst, xmm_shuf_mask);
4169 } else {
4170 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4171 }
4172 __ evshufi64x2(xmmdst, xmmdst, xmmdst, 0x0, Assembler::AVX_512bit);
4173
4174 }
4175
4176 address generate_cipherBlockChaining_decryptVectorAESCrypt() {
4177 assert(VM_Version::supports_vaes(), "need AES instructions and misaligned SSE support");
4178 __ align(CodeEntryAlignment);
4179 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
4180 address start = __ pc();
4181
4182 const Register from = c_rarg0; // source array address
4183 const Register to = c_rarg1; // destination array address
4184 const Register key = c_rarg2; // key array address
4185 const Register rvec = c_rarg3; // r byte array initialized from initvector array address
4186 // and left with the results of the last encryption block
4187 #ifndef _WIN64
4188 const Register len_reg = c_rarg4; // src len (must be multiple of blocksize 16)
4189 #else
4190 const Address len_mem(rbp, 6 * wordSize); // length is on stack on Win64
4191 const Register len_reg = r11; // pick the volatile windows register
4192 #endif
4193
4194 Label Loop, Loop1, L_128, L_256, L_192, KEY_192, KEY_256, Loop2, Lcbc_dec_rem_loop,
4195 Lcbc_dec_rem_last, Lcbc_dec_ret, Lcbc_dec_rem, Lcbc_exit;
4196
4197 __ enter();
4198
4199 #ifdef _WIN64
4200 // on win64, fill len_reg from stack position
4201 __ movl(len_reg, len_mem);
4202 #else
4203 __ push(len_reg); // Save
4204 #endif
4205 __ push(rbx);
4206 __ vzeroupper();
4207
4208 // Temporary variable declaration for swapping key bytes
4209 const XMMRegister xmm_key_shuf_mask = xmm1;
4210 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
4211
4212 // Calculate number of rounds from key size: 44 for 10-rounds, 52 for 12-rounds, 60 for 14-rounds
4213 const Register rounds = rbx;
4214 __ movl(rounds, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
4215
4216 const XMMRegister IV = xmm0;
4217 // Load IV and broadcast value to 512-bits
4218 __ evbroadcasti64x2(IV, Address(rvec, 0), Assembler::AVX_512bit);
4219
4220 // Temporary variables for storing round keys
4221 const XMMRegister RK0 = xmm30;
4222 const XMMRegister RK1 = xmm9;
4223 const XMMRegister RK2 = xmm18;
4224 const XMMRegister RK3 = xmm19;
4225 const XMMRegister RK4 = xmm20;
4226 const XMMRegister RK5 = xmm21;
4227 const XMMRegister RK6 = xmm22;
4228 const XMMRegister RK7 = xmm23;
4229 const XMMRegister RK8 = xmm24;
4230 const XMMRegister RK9 = xmm25;
4231 const XMMRegister RK10 = xmm26;
4232
4233 // Load and shuffle key
4234 // the java expanded key ordering is rotated one position from what we want
4235 // so we start from 1*16 here and hit 0*16 last
4236 ev_load_key(RK1, key, 1 * 16, xmm_key_shuf_mask);
4237 ev_load_key(RK2, key, 2 * 16, xmm_key_shuf_mask);
4238 ev_load_key(RK3, key, 3 * 16, xmm_key_shuf_mask);
4239 ev_load_key(RK4, key, 4 * 16, xmm_key_shuf_mask);
4240 ev_load_key(RK5, key, 5 * 16, xmm_key_shuf_mask);
4241 ev_load_key(RK6, key, 6 * 16, xmm_key_shuf_mask);
4242 ev_load_key(RK7, key, 7 * 16, xmm_key_shuf_mask);
4243 ev_load_key(RK8, key, 8 * 16, xmm_key_shuf_mask);
4244 ev_load_key(RK9, key, 9 * 16, xmm_key_shuf_mask);
4245 ev_load_key(RK10, key, 10 * 16, xmm_key_shuf_mask);
4246 ev_load_key(RK0, key, 0*16, xmm_key_shuf_mask);
4247
4248 // Variables for storing source cipher text
4249 const XMMRegister S0 = xmm10;
4250 const XMMRegister S1 = xmm11;
4251 const XMMRegister S2 = xmm12;
4252 const XMMRegister S3 = xmm13;
4253 const XMMRegister S4 = xmm14;
4254 const XMMRegister S5 = xmm15;
4255 const XMMRegister S6 = xmm16;
4256 const XMMRegister S7 = xmm17;
4257
4258 // Variables for storing decrypted text
4259 const XMMRegister B0 = xmm1;
4260 const XMMRegister B1 = xmm2;
4261 const XMMRegister B2 = xmm3;
4262 const XMMRegister B3 = xmm4;
4263 const XMMRegister B4 = xmm5;
4264 const XMMRegister B5 = xmm6;
4265 const XMMRegister B6 = xmm7;
4266 const XMMRegister B7 = xmm8;
4267
4268 __ cmpl(rounds, 44);
4269 __ jcc(Assembler::greater, KEY_192);
4270 __ jmp(Loop);
4271
4272 __ BIND(KEY_192);
4273 const XMMRegister RK11 = xmm27;
4274 const XMMRegister RK12 = xmm28;
4275 ev_load_key(RK11, key, 11*16, xmm_key_shuf_mask);
4276 ev_load_key(RK12, key, 12*16, xmm_key_shuf_mask);
4277
4278 __ cmpl(rounds, 52);
4279 __ jcc(Assembler::greater, KEY_256);
4280 __ jmp(Loop);
4281
4282 __ BIND(KEY_256);
4283 const XMMRegister RK13 = xmm29;
4284 const XMMRegister RK14 = xmm31;
4285 ev_load_key(RK13, key, 13*16, xmm_key_shuf_mask);
4286 ev_load_key(RK14, key, 14*16, xmm_key_shuf_mask);
4287
4288 __ BIND(Loop);
4289 __ cmpl(len_reg, 512);
4290 __ jcc(Assembler::below, Lcbc_dec_rem);
4291 __ BIND(Loop1);
4292 __ subl(len_reg, 512);
4293 __ evmovdquq(S0, Address(from, 0 * 64), Assembler::AVX_512bit);
4294 __ evmovdquq(S1, Address(from, 1 * 64), Assembler::AVX_512bit);
4295 __ evmovdquq(S2, Address(from, 2 * 64), Assembler::AVX_512bit);
4296 __ evmovdquq(S3, Address(from, 3 * 64), Assembler::AVX_512bit);
4297 __ evmovdquq(S4, Address(from, 4 * 64), Assembler::AVX_512bit);
4298 __ evmovdquq(S5, Address(from, 5 * 64), Assembler::AVX_512bit);
4299 __ evmovdquq(S6, Address(from, 6 * 64), Assembler::AVX_512bit);
4300 __ evmovdquq(S7, Address(from, 7 * 64), Assembler::AVX_512bit);
4301 __ leaq(from, Address(from, 8 * 64));
4302
4303 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4304 __ evpxorq(B1, S1, RK1, Assembler::AVX_512bit);
4305 __ evpxorq(B2, S2, RK1, Assembler::AVX_512bit);
4306 __ evpxorq(B3, S3, RK1, Assembler::AVX_512bit);
4307 __ evpxorq(B4, S4, RK1, Assembler::AVX_512bit);
4308 __ evpxorq(B5, S5, RK1, Assembler::AVX_512bit);
4309 __ evpxorq(B6, S6, RK1, Assembler::AVX_512bit);
4310 __ evpxorq(B7, S7, RK1, Assembler::AVX_512bit);
4311
4312 __ evalignq(IV, S0, IV, 0x06);
4313 __ evalignq(S0, S1, S0, 0x06);
4314 __ evalignq(S1, S2, S1, 0x06);
4315 __ evalignq(S2, S3, S2, 0x06);
4316 __ evalignq(S3, S4, S3, 0x06);
4317 __ evalignq(S4, S5, S4, 0x06);
4318 __ evalignq(S5, S6, S5, 0x06);
4319 __ evalignq(S6, S7, S6, 0x06);
4320
4321 roundDec(RK2);
4322 roundDec(RK3);
4323 roundDec(RK4);
4324 roundDec(RK5);
4325 roundDec(RK6);
4326 roundDec(RK7);
4327 roundDec(RK8);
4328 roundDec(RK9);
4329 roundDec(RK10);
4330
4331 __ cmpl(rounds, 44);
4332 __ jcc(Assembler::belowEqual, L_128);
4333 roundDec(RK11);
4334 roundDec(RK12);
4335
4336 __ cmpl(rounds, 52);
4337 __ jcc(Assembler::belowEqual, L_192);
4338 roundDec(RK13);
4339 roundDec(RK14);
4340
4341 __ BIND(L_256);
4342 roundDeclast(RK0);
4343 __ jmp(Loop2);
4344
4345 __ BIND(L_128);
4346 roundDeclast(RK0);
4347 __ jmp(Loop2);
4348
4349 __ BIND(L_192);
4350 roundDeclast(RK0);
4351
4352 __ BIND(Loop2);
4353 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4354 __ evpxorq(B1, B1, S0, Assembler::AVX_512bit);
4355 __ evpxorq(B2, B2, S1, Assembler::AVX_512bit);
4356 __ evpxorq(B3, B3, S2, Assembler::AVX_512bit);
4357 __ evpxorq(B4, B4, S3, Assembler::AVX_512bit);
4358 __ evpxorq(B5, B5, S4, Assembler::AVX_512bit);
4359 __ evpxorq(B6, B6, S5, Assembler::AVX_512bit);
4360 __ evpxorq(B7, B7, S6, Assembler::AVX_512bit);
4361 __ evmovdquq(IV, S7, Assembler::AVX_512bit);
4362
4363 __ evmovdquq(Address(to, 0 * 64), B0, Assembler::AVX_512bit);
4364 __ evmovdquq(Address(to, 1 * 64), B1, Assembler::AVX_512bit);
4365 __ evmovdquq(Address(to, 2 * 64), B2, Assembler::AVX_512bit);
4366 __ evmovdquq(Address(to, 3 * 64), B3, Assembler::AVX_512bit);
4367 __ evmovdquq(Address(to, 4 * 64), B4, Assembler::AVX_512bit);
4368 __ evmovdquq(Address(to, 5 * 64), B5, Assembler::AVX_512bit);
4369 __ evmovdquq(Address(to, 6 * 64), B6, Assembler::AVX_512bit);
4370 __ evmovdquq(Address(to, 7 * 64), B7, Assembler::AVX_512bit);
4371 __ leaq(to, Address(to, 8 * 64));
4372 __ jmp(Loop);
4373
4374 __ BIND(Lcbc_dec_rem);
4375 __ evshufi64x2(IV, IV, IV, 0x03, Assembler::AVX_512bit);
4376
4377 __ BIND(Lcbc_dec_rem_loop);
4378 __ subl(len_reg, 16);
4379 __ jcc(Assembler::carrySet, Lcbc_dec_ret);
4380
4381 __ movdqu(S0, Address(from, 0));
4382 __ evpxorq(B0, S0, RK1, Assembler::AVX_512bit);
4383 __ vaesdec(B0, B0, RK2, Assembler::AVX_512bit);
4384 __ vaesdec(B0, B0, RK3, Assembler::AVX_512bit);
4385 __ vaesdec(B0, B0, RK4, Assembler::AVX_512bit);
4386 __ vaesdec(B0, B0, RK5, Assembler::AVX_512bit);
4387 __ vaesdec(B0, B0, RK6, Assembler::AVX_512bit);
4388 __ vaesdec(B0, B0, RK7, Assembler::AVX_512bit);
4389 __ vaesdec(B0, B0, RK8, Assembler::AVX_512bit);
4390 __ vaesdec(B0, B0, RK9, Assembler::AVX_512bit);
4391 __ vaesdec(B0, B0, RK10, Assembler::AVX_512bit);
4392 __ cmpl(rounds, 44);
4393 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4394
4395 __ vaesdec(B0, B0, RK11, Assembler::AVX_512bit);
4396 __ vaesdec(B0, B0, RK12, Assembler::AVX_512bit);
4397 __ cmpl(rounds, 52);
4398 __ jcc(Assembler::belowEqual, Lcbc_dec_rem_last);
4399
4400 __ vaesdec(B0, B0, RK13, Assembler::AVX_512bit);
4401 __ vaesdec(B0, B0, RK14, Assembler::AVX_512bit);
4402
4403 __ BIND(Lcbc_dec_rem_last);
4404 __ vaesdeclast(B0, B0, RK0, Assembler::AVX_512bit);
4405
4406 __ evpxorq(B0, B0, IV, Assembler::AVX_512bit);
4407 __ evmovdquq(IV, S0, Assembler::AVX_512bit);
4408 __ movdqu(Address(to, 0), B0);
4409 __ leaq(from, Address(from, 16));
4410 __ leaq(to, Address(to, 16));
4411 __ jmp(Lcbc_dec_rem_loop);
4412
4413 __ BIND(Lcbc_dec_ret);
4414 __ movdqu(Address(rvec, 0), IV);
4415
4416 // Zero out the round keys
4417 __ evpxorq(RK0, RK0, RK0, Assembler::AVX_512bit);
4418 __ evpxorq(RK1, RK1, RK1, Assembler::AVX_512bit);
4419 __ evpxorq(RK2, RK2, RK2, Assembler::AVX_512bit);
4420 __ evpxorq(RK3, RK3, RK3, Assembler::AVX_512bit);
4421 __ evpxorq(RK4, RK4, RK4, Assembler::AVX_512bit);
4422 __ evpxorq(RK5, RK5, RK5, Assembler::AVX_512bit);
4423 __ evpxorq(RK6, RK6, RK6, Assembler::AVX_512bit);
4424 __ evpxorq(RK7, RK7, RK7, Assembler::AVX_512bit);
4425 __ evpxorq(RK8, RK8, RK8, Assembler::AVX_512bit);
4426 __ evpxorq(RK9, RK9, RK9, Assembler::AVX_512bit);
4427 __ evpxorq(RK10, RK10, RK10, Assembler::AVX_512bit);
4428 __ cmpl(rounds, 44);
4429 __ jcc(Assembler::belowEqual, Lcbc_exit);
4430 __ evpxorq(RK11, RK11, RK11, Assembler::AVX_512bit);
4431 __ evpxorq(RK12, RK12, RK12, Assembler::AVX_512bit);
4432 __ cmpl(rounds, 52);
4433 __ jcc(Assembler::belowEqual, Lcbc_exit);
4434 __ evpxorq(RK13, RK13, RK13, Assembler::AVX_512bit);
4435 __ evpxorq(RK14, RK14, RK14, Assembler::AVX_512bit);
4436
4437 __ BIND(Lcbc_exit);
4438 __ pop(rbx);
4439 #ifdef _WIN64
4440 __ movl(rax, len_mem);
4441 #else
4442 __ pop(rax); // return length
4443 #endif
4444 __ leave(); // required for proper stackwalking of RuntimeStub frame
4445 __ ret(0);
4446 return start;
4447 }
4448
4449 // Polynomial x^128+x^127+x^126+x^121+1
4450 address ghash_polynomial_addr() {
4451 __ align(CodeEntryAlignment);
4452 StubCodeMark mark(this, "StubRoutines", "_ghash_poly_addr");
4453 address start = __ pc();
4454 __ emit_data64(0x0000000000000001, relocInfo::none);
4455 __ emit_data64(0xc200000000000000, relocInfo::none);
4456 return start;
4457 }
4458
4459 address ghash_shufflemask_addr() {
4460 __ align(CodeEntryAlignment);
4461 StubCodeMark mark(this, "StubRoutines", "_ghash_shuffmask_addr");
4462 address start = __ pc();
4463 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4464 __ emit_data64(0x0f0f0f0f0f0f0f0f, relocInfo::none);
4465 return start;
4466 }
4467
4468 // Ghash single and multi block operations using AVX instructions
4469 address generate_avx_ghash_processBlocks() {
4470 __ align(CodeEntryAlignment);
4471
4472 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4473 address start = __ pc();
4474
4475 // arguments
4476 const Register state = c_rarg0;
4477 const Register htbl = c_rarg1;
4478 const Register data = c_rarg2;
4479 const Register blocks = c_rarg3;
4480 __ enter();
4481 // Save state before entering routine
4482 __ avx_ghash(state, htbl, data, blocks);
4483 __ leave(); // required for proper stackwalking of RuntimeStub frame
4484 __ ret(0);
4485 return start;
4486 }
4487
4488 // byte swap x86 long
4489 address generate_ghash_long_swap_mask() {
4490 __ align(CodeEntryAlignment);
4491 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
4492 address start = __ pc();
4493 __ emit_data64(0x0f0e0d0c0b0a0908, relocInfo::none );
4494 __ emit_data64(0x0706050403020100, relocInfo::none );
4495 return start;
4496 }
4497
4498 // byte swap x86 byte array
4499 address generate_ghash_byte_swap_mask() {
4500 __ align(CodeEntryAlignment);
4501 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
4502 address start = __ pc();
4503 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none );
4504 __ emit_data64(0x0001020304050607, relocInfo::none );
4505 return start;
4506 }
4507
4508 /* Single and multi-block ghash operations */
4509 address generate_ghash_processBlocks() {
4510 __ align(CodeEntryAlignment);
4511 Label L_ghash_loop, L_exit;
4512 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
4513 address start = __ pc();
4514
4515 const Register state = c_rarg0;
4516 const Register subkeyH = c_rarg1;
4517 const Register data = c_rarg2;
4518 const Register blocks = c_rarg3;
4519
4520 const XMMRegister xmm_temp0 = xmm0;
4521 const XMMRegister xmm_temp1 = xmm1;
4522 const XMMRegister xmm_temp2 = xmm2;
4523 const XMMRegister xmm_temp3 = xmm3;
4524 const XMMRegister xmm_temp4 = xmm4;
4525 const XMMRegister xmm_temp5 = xmm5;
4526 const XMMRegister xmm_temp6 = xmm6;
4527 const XMMRegister xmm_temp7 = xmm7;
4528 const XMMRegister xmm_temp8 = xmm8;
4529 const XMMRegister xmm_temp9 = xmm9;
4530 const XMMRegister xmm_temp10 = xmm10;
4531
4532 __ enter();
4533
4534 __ movdqu(xmm_temp10, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
4535
4536 __ movdqu(xmm_temp0, Address(state, 0));
4537 __ pshufb(xmm_temp0, xmm_temp10);
4538
4539
4540 __ BIND(L_ghash_loop);
4541 __ movdqu(xmm_temp2, Address(data, 0));
4542 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
4543
4544 __ movdqu(xmm_temp1, Address(subkeyH, 0));
4545 __ pshufb(xmm_temp1, xmm_temp10);
4546
4547 __ pxor(xmm_temp0, xmm_temp2);
4548
4549 //
4550 // Multiply with the hash key
4551 //
4552 __ movdqu(xmm_temp3, xmm_temp0);
4553 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
4554 __ movdqu(xmm_temp4, xmm_temp0);
4555 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
4556
4557 __ movdqu(xmm_temp5, xmm_temp0);
4558 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
4559 __ movdqu(xmm_temp6, xmm_temp0);
4560 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
4561
4562 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
4563
4564 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
4565 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
4566 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
4567 __ pxor(xmm_temp3, xmm_temp5);
4568 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
4569 // of the carry-less multiplication of
4570 // xmm0 by xmm1.
4571
4572 // We shift the result of the multiplication by one bit position
4573 // to the left to cope for the fact that the bits are reversed.
4574 __ movdqu(xmm_temp7, xmm_temp3);
4575 __ movdqu(xmm_temp8, xmm_temp6);
4576 __ pslld(xmm_temp3, 1);
4577 __ pslld(xmm_temp6, 1);
4578 __ psrld(xmm_temp7, 31);
4579 __ psrld(xmm_temp8, 31);
4580 __ movdqu(xmm_temp9, xmm_temp7);
4581 __ pslldq(xmm_temp8, 4);
4582 __ pslldq(xmm_temp7, 4);
4583 __ psrldq(xmm_temp9, 12);
4584 __ por(xmm_temp3, xmm_temp7);
4585 __ por(xmm_temp6, xmm_temp8);
4586 __ por(xmm_temp6, xmm_temp9);
4587
4588 //
4589 // First phase of the reduction
4590 //
4591 // Move xmm3 into xmm7, xmm8, xmm9 in order to perform the shifts
4592 // independently.
4593 __ movdqu(xmm_temp7, xmm_temp3);
4594 __ movdqu(xmm_temp8, xmm_temp3);
4595 __ movdqu(xmm_temp9, xmm_temp3);
4596 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
4597 __ pslld(xmm_temp8, 30); // packed right shift shifting << 30
4598 __ pslld(xmm_temp9, 25); // packed right shift shifting << 25
4599 __ pxor(xmm_temp7, xmm_temp8); // xor the shifted versions
4600 __ pxor(xmm_temp7, xmm_temp9);
4601 __ movdqu(xmm_temp8, xmm_temp7);
4602 __ pslldq(xmm_temp7, 12);
4603 __ psrldq(xmm_temp8, 4);
4604 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
4605
4606 //
4607 // Second phase of the reduction
4608 //
4609 // Make 3 copies of xmm3 in xmm2, xmm4, xmm5 for doing these
4610 // shift operations.
4611 __ movdqu(xmm_temp2, xmm_temp3);
4612 __ movdqu(xmm_temp4, xmm_temp3);
4613 __ movdqu(xmm_temp5, xmm_temp3);
4614 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
4615 __ psrld(xmm_temp4, 2); // packed left shifting >> 2
4616 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
4617 __ pxor(xmm_temp2, xmm_temp4); // xor the shifted versions
4618 __ pxor(xmm_temp2, xmm_temp5);
4619 __ pxor(xmm_temp2, xmm_temp8);
4620 __ pxor(xmm_temp3, xmm_temp2);
4621 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
4622
4623 __ decrement(blocks);
4624 __ jcc(Assembler::zero, L_exit);
4625 __ movdqu(xmm_temp0, xmm_temp6);
4626 __ addptr(data, 16);
4627 __ jmp(L_ghash_loop);
4628
4629 __ BIND(L_exit);
4630 __ pshufb(xmm_temp6, xmm_temp10); // Byte swap 16-byte result
4631 __ movdqu(Address(state, 0), xmm_temp6); // store the result
4632 __ leave();
4633 __ ret(0);
4634 return start;
4635 }
4636
4637 //base64 character set
4638 address base64_charset_addr() {
4639 __ align(CodeEntryAlignment);
4640 StubCodeMark mark(this, "StubRoutines", "base64_charset");
4641 address start = __ pc();
4642 __ emit_data64(0x0000004200000041, relocInfo::none);
4643 __ emit_data64(0x0000004400000043, relocInfo::none);
4644 __ emit_data64(0x0000004600000045, relocInfo::none);
4645 __ emit_data64(0x0000004800000047, relocInfo::none);
4646 __ emit_data64(0x0000004a00000049, relocInfo::none);
4647 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4648 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4649 __ emit_data64(0x000000500000004f, relocInfo::none);
4650 __ emit_data64(0x0000005200000051, relocInfo::none);
4651 __ emit_data64(0x0000005400000053, relocInfo::none);
4652 __ emit_data64(0x0000005600000055, relocInfo::none);
4653 __ emit_data64(0x0000005800000057, relocInfo::none);
4654 __ emit_data64(0x0000005a00000059, relocInfo::none);
4655 __ emit_data64(0x0000006200000061, relocInfo::none);
4656 __ emit_data64(0x0000006400000063, relocInfo::none);
4657 __ emit_data64(0x0000006600000065, relocInfo::none);
4658 __ emit_data64(0x0000006800000067, relocInfo::none);
4659 __ emit_data64(0x0000006a00000069, relocInfo::none);
4660 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4661 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4662 __ emit_data64(0x000000700000006f, relocInfo::none);
4663 __ emit_data64(0x0000007200000071, relocInfo::none);
4664 __ emit_data64(0x0000007400000073, relocInfo::none);
4665 __ emit_data64(0x0000007600000075, relocInfo::none);
4666 __ emit_data64(0x0000007800000077, relocInfo::none);
4667 __ emit_data64(0x0000007a00000079, relocInfo::none);
4668 __ emit_data64(0x0000003100000030, relocInfo::none);
4669 __ emit_data64(0x0000003300000032, relocInfo::none);
4670 __ emit_data64(0x0000003500000034, relocInfo::none);
4671 __ emit_data64(0x0000003700000036, relocInfo::none);
4672 __ emit_data64(0x0000003900000038, relocInfo::none);
4673 __ emit_data64(0x0000002f0000002b, relocInfo::none);
4674 return start;
4675 }
4676
4677 //base64 url character set
4678 address base64url_charset_addr() {
4679 __ align(CodeEntryAlignment);
4680 StubCodeMark mark(this, "StubRoutines", "base64url_charset");
4681 address start = __ pc();
4682 __ emit_data64(0x0000004200000041, relocInfo::none);
4683 __ emit_data64(0x0000004400000043, relocInfo::none);
4684 __ emit_data64(0x0000004600000045, relocInfo::none);
4685 __ emit_data64(0x0000004800000047, relocInfo::none);
4686 __ emit_data64(0x0000004a00000049, relocInfo::none);
4687 __ emit_data64(0x0000004c0000004b, relocInfo::none);
4688 __ emit_data64(0x0000004e0000004d, relocInfo::none);
4689 __ emit_data64(0x000000500000004f, relocInfo::none);
4690 __ emit_data64(0x0000005200000051, relocInfo::none);
4691 __ emit_data64(0x0000005400000053, relocInfo::none);
4692 __ emit_data64(0x0000005600000055, relocInfo::none);
4693 __ emit_data64(0x0000005800000057, relocInfo::none);
4694 __ emit_data64(0x0000005a00000059, relocInfo::none);
4695 __ emit_data64(0x0000006200000061, relocInfo::none);
4696 __ emit_data64(0x0000006400000063, relocInfo::none);
4697 __ emit_data64(0x0000006600000065, relocInfo::none);
4698 __ emit_data64(0x0000006800000067, relocInfo::none);
4699 __ emit_data64(0x0000006a00000069, relocInfo::none);
4700 __ emit_data64(0x0000006c0000006b, relocInfo::none);
4701 __ emit_data64(0x0000006e0000006d, relocInfo::none);
4702 __ emit_data64(0x000000700000006f, relocInfo::none);
4703 __ emit_data64(0x0000007200000071, relocInfo::none);
4704 __ emit_data64(0x0000007400000073, relocInfo::none);
4705 __ emit_data64(0x0000007600000075, relocInfo::none);
4706 __ emit_data64(0x0000007800000077, relocInfo::none);
4707 __ emit_data64(0x0000007a00000079, relocInfo::none);
4708 __ emit_data64(0x0000003100000030, relocInfo::none);
4709 __ emit_data64(0x0000003300000032, relocInfo::none);
4710 __ emit_data64(0x0000003500000034, relocInfo::none);
4711 __ emit_data64(0x0000003700000036, relocInfo::none);
4712 __ emit_data64(0x0000003900000038, relocInfo::none);
4713 __ emit_data64(0x0000005f0000002d, relocInfo::none);
4714
4715 return start;
4716 }
4717
4718 address base64_bswap_mask_addr() {
4719 __ align(CodeEntryAlignment);
4720 StubCodeMark mark(this, "StubRoutines", "bswap_mask_base64");
4721 address start = __ pc();
4722 __ emit_data64(0x0504038002010080, relocInfo::none);
4723 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4724 __ emit_data64(0x0908078006050480, relocInfo::none);
4725 __ emit_data64(0x0f0e0d800c0b0a80, relocInfo::none);
4726 __ emit_data64(0x0605048003020180, relocInfo::none);
4727 __ emit_data64(0x0c0b0a8009080780, relocInfo::none);
4728 __ emit_data64(0x0504038002010080, relocInfo::none);
4729 __ emit_data64(0x0b0a098008070680, relocInfo::none);
4730
4731 return start;
4732 }
4733
4734 address base64_right_shift_mask_addr() {
4735 __ align(CodeEntryAlignment);
4736 StubCodeMark mark(this, "StubRoutines", "right_shift_mask");
4737 address start = __ pc();
4738 __ emit_data64(0x0006000400020000, relocInfo::none);
4739 __ emit_data64(0x0006000400020000, relocInfo::none);
4740 __ emit_data64(0x0006000400020000, relocInfo::none);
4741 __ emit_data64(0x0006000400020000, relocInfo::none);
4742 __ emit_data64(0x0006000400020000, relocInfo::none);
4743 __ emit_data64(0x0006000400020000, relocInfo::none);
4744 __ emit_data64(0x0006000400020000, relocInfo::none);
4745 __ emit_data64(0x0006000400020000, relocInfo::none);
4746
4747 return start;
4748 }
4749
4750 address base64_left_shift_mask_addr() {
4751 __ align(CodeEntryAlignment);
4752 StubCodeMark mark(this, "StubRoutines", "left_shift_mask");
4753 address start = __ pc();
4754 __ emit_data64(0x0000000200040000, relocInfo::none);
4755 __ emit_data64(0x0000000200040000, relocInfo::none);
4756 __ emit_data64(0x0000000200040000, relocInfo::none);
4757 __ emit_data64(0x0000000200040000, relocInfo::none);
4758 __ emit_data64(0x0000000200040000, relocInfo::none);
4759 __ emit_data64(0x0000000200040000, relocInfo::none);
4760 __ emit_data64(0x0000000200040000, relocInfo::none);
4761 __ emit_data64(0x0000000200040000, relocInfo::none);
4762
4763 return start;
4764 }
4765
4766 address base64_and_mask_addr() {
4767 __ align(CodeEntryAlignment);
4768 StubCodeMark mark(this, "StubRoutines", "and_mask");
4769 address start = __ pc();
4770 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4771 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4772 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4773 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4774 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4775 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4776 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4777 __ emit_data64(0x3f003f003f000000, relocInfo::none);
4778 return start;
4779 }
4780
4781 address base64_gather_mask_addr() {
4782 __ align(CodeEntryAlignment);
4783 StubCodeMark mark(this, "StubRoutines", "gather_mask");
4784 address start = __ pc();
4785 __ emit_data64(0xffffffffffffffff, relocInfo::none);
4786 return start;
4787 }
4788
4789 // Code for generating Base64 encoding.
4790 // Intrinsic function prototype in Base64.java:
4791 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL) {
4792 address generate_base64_encodeBlock() {
4793 __ align(CodeEntryAlignment);
4794 StubCodeMark mark(this, "StubRoutines", "implEncode");
4795 address start = __ pc();
4796 __ enter();
4797
4798 // Save callee-saved registers before using them
4799 __ push(r12);
4800 __ push(r13);
4801 __ push(r14);
4802 __ push(r15);
4803
4804 // arguments
4805 const Register source = c_rarg0; // Source Array
4806 const Register start_offset = c_rarg1; // start offset
4807 const Register end_offset = c_rarg2; // end offset
4808 const Register dest = c_rarg3; // destination array
4809
4810 #ifndef _WIN64
4811 const Register dp = c_rarg4; // Position for writing to dest array
4812 const Register isURL = c_rarg5;// Base64 or URL character set
4813 #else
4814 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
4815 const Address isURL_mem(rbp, 7 * wordSize);
4816 const Register isURL = r10; // pick the volatile windows register
4817 const Register dp = r12;
4818 __ movl(dp, dp_mem);
4819 __ movl(isURL, isURL_mem);
4820 #endif
4821
4822 const Register length = r14;
4823 Label L_process80, L_process32, L_process3, L_exit, L_processdata;
4824
4825 // calculate length from offsets
4826 __ movl(length, end_offset);
4827 __ subl(length, start_offset);
4828 __ cmpl(length, 0);
4829 __ jcc(Assembler::lessEqual, L_exit);
4830
4831 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_charset_addr()));
4832 // check if base64 charset(isURL=0) or base64 url charset(isURL=1) needs to be loaded
4833 __ cmpl(isURL, 0);
4834 __ jcc(Assembler::equal, L_processdata);
4835 __ lea(r11, ExternalAddress(StubRoutines::x86::base64url_charset_addr()));
4836
4837 // load masks required for encoding data
4838 __ BIND(L_processdata);
4839 __ movdqu(xmm16, ExternalAddress(StubRoutines::x86::base64_gather_mask_addr()));
4840 // Set 64 bits of K register.
4841 __ evpcmpeqb(k3, xmm16, xmm16, Assembler::AVX_512bit);
4842 __ evmovdquq(xmm12, ExternalAddress(StubRoutines::x86::base64_bswap_mask_addr()), Assembler::AVX_256bit, r13);
4843 __ evmovdquq(xmm13, ExternalAddress(StubRoutines::x86::base64_right_shift_mask_addr()), Assembler::AVX_512bit, r13);
4844 __ evmovdquq(xmm14, ExternalAddress(StubRoutines::x86::base64_left_shift_mask_addr()), Assembler::AVX_512bit, r13);
4845 __ evmovdquq(xmm15, ExternalAddress(StubRoutines::x86::base64_and_mask_addr()), Assembler::AVX_512bit, r13);
4846
4847 // Vector Base64 implementation, producing 96 bytes of encoded data
4848 __ BIND(L_process80);
4849 __ cmpl(length, 80);
4850 __ jcc(Assembler::below, L_process32);
4851 __ evmovdquq(xmm0, Address(source, start_offset, Address::times_1, 0), Assembler::AVX_256bit);
4852 __ evmovdquq(xmm1, Address(source, start_offset, Address::times_1, 24), Assembler::AVX_256bit);
4853 __ evmovdquq(xmm2, Address(source, start_offset, Address::times_1, 48), Assembler::AVX_256bit);
4854
4855 //permute the input data in such a manner that we have continuity of the source
4856 __ vpermq(xmm3, xmm0, 148, Assembler::AVX_256bit);
4857 __ vpermq(xmm4, xmm1, 148, Assembler::AVX_256bit);
4858 __ vpermq(xmm5, xmm2, 148, Assembler::AVX_256bit);
4859
4860 //shuffle input and group 3 bytes of data and to it add 0 as the 4th byte.
4861 //we can deal with 12 bytes at a time in a 128 bit register
4862 __ vpshufb(xmm3, xmm3, xmm12, Assembler::AVX_256bit);
4863 __ vpshufb(xmm4, xmm4, xmm12, Assembler::AVX_256bit);
4864 __ vpshufb(xmm5, xmm5, xmm12, Assembler::AVX_256bit);
4865
4866 //convert byte to word. Each 128 bit register will have 6 bytes for processing
4867 __ vpmovzxbw(xmm3, xmm3, Assembler::AVX_512bit);
4868 __ vpmovzxbw(xmm4, xmm4, Assembler::AVX_512bit);
4869 __ vpmovzxbw(xmm5, xmm5, Assembler::AVX_512bit);
4870
4871 // Extract bits in the following pattern 6, 4+2, 2+4, 6 to convert 3, 8 bit numbers to 4, 6 bit numbers
4872 __ evpsrlvw(xmm0, xmm3, xmm13, Assembler::AVX_512bit);
4873 __ evpsrlvw(xmm1, xmm4, xmm13, Assembler::AVX_512bit);
4874 __ evpsrlvw(xmm2, xmm5, xmm13, Assembler::AVX_512bit);
4875
4876 __ evpsllvw(xmm3, xmm3, xmm14, Assembler::AVX_512bit);
4877 __ evpsllvw(xmm4, xmm4, xmm14, Assembler::AVX_512bit);
4878 __ evpsllvw(xmm5, xmm5, xmm14, Assembler::AVX_512bit);
4879
4880 __ vpsrlq(xmm0, xmm0, 8, Assembler::AVX_512bit);
4881 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
4882 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
4883
4884 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4885 __ vpsllq(xmm4, xmm4, 8, Assembler::AVX_512bit);
4886 __ vpsllq(xmm5, xmm5, 8, Assembler::AVX_512bit);
4887
4888 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
4889 __ vpandq(xmm4, xmm4, xmm15, Assembler::AVX_512bit);
4890 __ vpandq(xmm5, xmm5, xmm15, Assembler::AVX_512bit);
4891
4892 // Get the final 4*6 bits base64 encoding
4893 __ vporq(xmm3, xmm3, xmm0, Assembler::AVX_512bit);
4894 __ vporq(xmm4, xmm4, xmm1, Assembler::AVX_512bit);
4895 __ vporq(xmm5, xmm5, xmm2, Assembler::AVX_512bit);
4896
4897 // Shift
4898 __ vpsrlq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4899 __ vpsrlq(xmm4, xmm4, 8, Assembler::AVX_512bit);
4900 __ vpsrlq(xmm5, xmm5, 8, Assembler::AVX_512bit);
4901
4902 // look up 6 bits in the base64 character set to fetch the encoding
4903 // we are converting word to dword as gather instructions need dword indices for looking up encoding
4904 __ vextracti64x4(xmm6, xmm3, 0);
4905 __ vpmovzxwd(xmm0, xmm6, Assembler::AVX_512bit);
4906 __ vextracti64x4(xmm6, xmm3, 1);
4907 __ vpmovzxwd(xmm1, xmm6, Assembler::AVX_512bit);
4908
4909 __ vextracti64x4(xmm6, xmm4, 0);
4910 __ vpmovzxwd(xmm2, xmm6, Assembler::AVX_512bit);
4911 __ vextracti64x4(xmm6, xmm4, 1);
4912 __ vpmovzxwd(xmm3, xmm6, Assembler::AVX_512bit);
4913
4914 __ vextracti64x4(xmm4, xmm5, 0);
4915 __ vpmovzxwd(xmm6, xmm4, Assembler::AVX_512bit);
4916
4917 __ vextracti64x4(xmm4, xmm5, 1);
4918 __ vpmovzxwd(xmm7, xmm4, Assembler::AVX_512bit);
4919
4920 __ kmovql(k2, k3);
4921 __ evpgatherdd(xmm4, k2, Address(r11, xmm0, Address::times_4, 0), Assembler::AVX_512bit);
4922 __ kmovql(k2, k3);
4923 __ evpgatherdd(xmm5, k2, Address(r11, xmm1, Address::times_4, 0), Assembler::AVX_512bit);
4924 __ kmovql(k2, k3);
4925 __ evpgatherdd(xmm8, k2, Address(r11, xmm2, Address::times_4, 0), Assembler::AVX_512bit);
4926 __ kmovql(k2, k3);
4927 __ evpgatherdd(xmm9, k2, Address(r11, xmm3, Address::times_4, 0), Assembler::AVX_512bit);
4928 __ kmovql(k2, k3);
4929 __ evpgatherdd(xmm10, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
4930 __ kmovql(k2, k3);
4931 __ evpgatherdd(xmm11, k2, Address(r11, xmm7, Address::times_4, 0), Assembler::AVX_512bit);
4932
4933 //Down convert dword to byte. Final output is 16*6 = 96 bytes long
4934 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm4, Assembler::AVX_512bit);
4935 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm5, Assembler::AVX_512bit);
4936 __ evpmovdb(Address(dest, dp, Address::times_1, 32), xmm8, Assembler::AVX_512bit);
4937 __ evpmovdb(Address(dest, dp, Address::times_1, 48), xmm9, Assembler::AVX_512bit);
4938 __ evpmovdb(Address(dest, dp, Address::times_1, 64), xmm10, Assembler::AVX_512bit);
4939 __ evpmovdb(Address(dest, dp, Address::times_1, 80), xmm11, Assembler::AVX_512bit);
4940
4941 __ addq(dest, 96);
4942 __ addq(source, 72);
4943 __ subq(length, 72);
4944 __ jmp(L_process80);
4945
4946 // Vector Base64 implementation generating 32 bytes of encoded data
4947 __ BIND(L_process32);
4948 __ cmpl(length, 32);
4949 __ jcc(Assembler::below, L_process3);
4950 __ evmovdquq(xmm0, Address(source, start_offset), Assembler::AVX_256bit);
4951 __ vpermq(xmm0, xmm0, 148, Assembler::AVX_256bit);
4952 __ vpshufb(xmm6, xmm0, xmm12, Assembler::AVX_256bit);
4953 __ vpmovzxbw(xmm6, xmm6, Assembler::AVX_512bit);
4954 __ evpsrlvw(xmm2, xmm6, xmm13, Assembler::AVX_512bit);
4955 __ evpsllvw(xmm3, xmm6, xmm14, Assembler::AVX_512bit);
4956
4957 __ vpsrlq(xmm2, xmm2, 8, Assembler::AVX_512bit);
4958 __ vpsllq(xmm3, xmm3, 8, Assembler::AVX_512bit);
4959 __ vpandq(xmm3, xmm3, xmm15, Assembler::AVX_512bit);
4960 __ vporq(xmm1, xmm2, xmm3, Assembler::AVX_512bit);
4961 __ vpsrlq(xmm1, xmm1, 8, Assembler::AVX_512bit);
4962 __ vextracti64x4(xmm9, xmm1, 0);
4963 __ vpmovzxwd(xmm6, xmm9, Assembler::AVX_512bit);
4964 __ vextracti64x4(xmm9, xmm1, 1);
4965 __ vpmovzxwd(xmm5, xmm9, Assembler::AVX_512bit);
4966 __ kmovql(k2, k3);
4967 __ evpgatherdd(xmm8, k2, Address(r11, xmm6, Address::times_4, 0), Assembler::AVX_512bit);
4968 __ kmovql(k2, k3);
4969 __ evpgatherdd(xmm10, k2, Address(r11, xmm5, Address::times_4, 0), Assembler::AVX_512bit);
4970 __ evpmovdb(Address(dest, dp, Address::times_1, 0), xmm8, Assembler::AVX_512bit);
4971 __ evpmovdb(Address(dest, dp, Address::times_1, 16), xmm10, Assembler::AVX_512bit);
4972 __ subq(length, 24);
4973 __ addq(dest, 32);
4974 __ addq(source, 24);
4975 __ jmp(L_process32);
4976
4977 // Scalar data processing takes 3 bytes at a time and produces 4 bytes of encoded data
4978 /* This code corresponds to the scalar version of the following snippet in Base64.java
4979 ** int bits = (src[sp0++] & 0xff) << 16 |(src[sp0++] & 0xff) << 8 |(src[sp0++] & 0xff);
4980 ** dst[dp0++] = (byte)base64[(bits >> > 18) & 0x3f];
4981 ** dst[dp0++] = (byte)base64[(bits >> > 12) & 0x3f];
4982 ** dst[dp0++] = (byte)base64[(bits >> > 6) & 0x3f];
4983 ** dst[dp0++] = (byte)base64[bits & 0x3f];*/
4984 __ BIND(L_process3);
4985 __ cmpl(length, 3);
4986 __ jcc(Assembler::below, L_exit);
4987 // Read 1 byte at a time
4988 __ movzbl(rax, Address(source, start_offset));
4989 __ shll(rax, 0x10);
4990 __ movl(r15, rax);
4991 __ movzbl(rax, Address(source, start_offset, Address::times_1, 1));
4992 __ shll(rax, 0x8);
4993 __ movzwl(rax, rax);
4994 __ orl(r15, rax);
4995 __ movzbl(rax, Address(source, start_offset, Address::times_1, 2));
4996 __ orl(rax, r15);
4997 // Save 3 bytes read in r15
4998 __ movl(r15, rax);
4999 __ shrl(rax, 0x12);
5000 __ andl(rax, 0x3f);
5001 // rax contains the index, r11 contains base64 lookup table
5002 __ movb(rax, Address(r11, rax, Address::times_4));
5003 // Write the encoded byte to destination
5004 __ movb(Address(dest, dp, Address::times_1, 0), rax);
5005 __ movl(rax, r15);
5006 __ shrl(rax, 0xc);
5007 __ andl(rax, 0x3f);
5008 __ movb(rax, Address(r11, rax, Address::times_4));
5009 __ movb(Address(dest, dp, Address::times_1, 1), rax);
5010 __ movl(rax, r15);
5011 __ shrl(rax, 0x6);
5012 __ andl(rax, 0x3f);
5013 __ movb(rax, Address(r11, rax, Address::times_4));
5014 __ movb(Address(dest, dp, Address::times_1, 2), rax);
5015 __ movl(rax, r15);
5016 __ andl(rax, 0x3f);
5017 __ movb(rax, Address(r11, rax, Address::times_4));
5018 __ movb(Address(dest, dp, Address::times_1, 3), rax);
5019 __ subl(length, 3);
5020 __ addq(dest, 4);
5021 __ addq(source, 3);
5022 __ jmp(L_process3);
5023 __ BIND(L_exit);
5024 __ pop(r15);
5025 __ pop(r14);
5026 __ pop(r13);
5027 __ pop(r12);
5028 __ leave();
5029 __ ret(0);
5030 return start;
5031 }
5032
5033 /**
5034 * Arguments:
5035 *
5036 * Inputs:
5037 * c_rarg0 - int crc
5038 * c_rarg1 - byte* buf
5039 * c_rarg2 - int length
5040 *
5041 * Ouput:
5042 * rax - int crc result
5043 */
5044 address generate_updateBytesCRC32() {
5045 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
5046
5047 __ align(CodeEntryAlignment);
5048 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
5049
5050 address start = __ pc();
5051 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5052 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5053 // rscratch1: r10
5054 const Register crc = c_rarg0; // crc
5055 const Register buf = c_rarg1; // source java byte array address
5056 const Register len = c_rarg2; // length
5057 const Register table = c_rarg3; // crc_table address (reuse register)
5058 const Register tmp = r11;
5059 assert_different_registers(crc, buf, len, table, tmp, rax);
5060
5061 BLOCK_COMMENT("Entry:");
5062 __ enter(); // required for proper stackwalking of RuntimeStub frame
5063
5064 __ kernel_crc32(crc, buf, len, table, tmp);
5065
5066 __ movl(rax, crc);
5067 __ vzeroupper();
5068 __ leave(); // required for proper stackwalking of RuntimeStub frame
5069 __ ret(0);
5070
5071 return start;
5072 }
5073
5074 /**
5075 * Arguments:
5076 *
5077 * Inputs:
5078 * c_rarg0 - int crc
5079 * c_rarg1 - byte* buf
5080 * c_rarg2 - long length
5081 * c_rarg3 - table_start - optional (present only when doing a library_call,
5082 * not used by x86 algorithm)
5083 *
5084 * Ouput:
5085 * rax - int crc result
5086 */
5087 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
5088 assert(UseCRC32CIntrinsics, "need SSE4_2");
5089 __ align(CodeEntryAlignment);
5090 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
5091 address start = __ pc();
5092 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
5093 //Windows RCX RDX R8 R9 none none XMM0..XMM3
5094 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
5095 const Register crc = c_rarg0; // crc
5096 const Register buf = c_rarg1; // source java byte array address
5097 const Register len = c_rarg2; // length
5098 const Register a = rax;
5099 const Register j = r9;
5100 const Register k = r10;
5101 const Register l = r11;
5102 #ifdef _WIN64
5103 const Register y = rdi;
5104 const Register z = rsi;
5105 #else
5106 const Register y = rcx;
5107 const Register z = r8;
5108 #endif
5109 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
5110
5111 BLOCK_COMMENT("Entry:");
5112 __ enter(); // required for proper stackwalking of RuntimeStub frame
5113 #ifdef _WIN64
5114 __ push(y);
5115 __ push(z);
5116 #endif
5117 __ crc32c_ipl_alg2_alt2(crc, buf, len,
5118 a, j, k,
5119 l, y, z,
5120 c_farg0, c_farg1, c_farg2,
5121 is_pclmulqdq_supported);
5122 __ movl(rax, crc);
5123 #ifdef _WIN64
5124 __ pop(z);
5125 __ pop(y);
5126 #endif
5127 __ vzeroupper();
5128 __ leave(); // required for proper stackwalking of RuntimeStub frame
5129 __ ret(0);
5130
5131 return start;
5132 }
5133
5134 /**
5135 * Arguments:
5136 *
5137 * Input:
5138 * c_rarg0 - x address
5139 * c_rarg1 - x length
5140 * c_rarg2 - y address
5141 * c_rarg3 - y length
5142 * not Win64
5143 * c_rarg4 - z address
5144 * c_rarg5 - z length
5145 * Win64
5146 * rsp+40 - z address
5147 * rsp+48 - z length
5148 */
5149 address generate_multiplyToLen() {
5150 __ align(CodeEntryAlignment);
5151 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
5152
5153 address start = __ pc();
5154 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5155 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5156 const Register x = rdi;
5157 const Register xlen = rax;
5158 const Register y = rsi;
5159 const Register ylen = rcx;
5160 const Register z = r8;
5161 const Register zlen = r11;
5162
5163 // Next registers will be saved on stack in multiply_to_len().
5164 const Register tmp1 = r12;
5165 const Register tmp2 = r13;
5166 const Register tmp3 = r14;
5167 const Register tmp4 = r15;
5168 const Register tmp5 = rbx;
5169
5170 BLOCK_COMMENT("Entry:");
5171 __ enter(); // required for proper stackwalking of RuntimeStub frame
5172
5173 #ifndef _WIN64
5174 __ movptr(zlen, r9); // Save r9 in r11 - zlen
5175 #endif
5176 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
5177 // ylen => rcx, z => r8, zlen => r11
5178 // r9 and r10 may be used to save non-volatile registers
5179 #ifdef _WIN64
5180 // last 2 arguments (#4, #5) are on stack on Win64
5181 __ movptr(z, Address(rsp, 6 * wordSize));
5182 __ movptr(zlen, Address(rsp, 7 * wordSize));
5183 #endif
5184
5185 __ movptr(xlen, rsi);
5186 __ movptr(y, rdx);
5187 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
5188
5189 restore_arg_regs();
5190
5191 __ leave(); // required for proper stackwalking of RuntimeStub frame
5192 __ ret(0);
5193
5194 return start;
5195 }
5196
5197 /**
5198 * Arguments:
5199 *
5200 * Input:
5201 * c_rarg0 - obja address
5202 * c_rarg1 - objb address
5203 * c_rarg3 - length length
5204 * c_rarg4 - scale log2_array_indxscale
5205 *
5206 * Output:
5207 * rax - int >= mismatched index, < 0 bitwise complement of tail
5208 */
5209 address generate_vectorizedMismatch() {
5210 __ align(CodeEntryAlignment);
5211 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
5212 address start = __ pc();
5213
5214 BLOCK_COMMENT("Entry:");
5215 __ enter();
5216
5217 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5218 const Register scale = c_rarg0; //rcx, will exchange with r9
5219 const Register objb = c_rarg1; //rdx
5220 const Register length = c_rarg2; //r8
5221 const Register obja = c_rarg3; //r9
5222 __ xchgq(obja, scale); //now obja and scale contains the correct contents
5223
5224 const Register tmp1 = r10;
5225 const Register tmp2 = r11;
5226 #endif
5227 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5228 const Register obja = c_rarg0; //U:rdi
5229 const Register objb = c_rarg1; //U:rsi
5230 const Register length = c_rarg2; //U:rdx
5231 const Register scale = c_rarg3; //U:rcx
5232 const Register tmp1 = r8;
5233 const Register tmp2 = r9;
5234 #endif
5235 const Register result = rax; //return value
5236 const XMMRegister vec0 = xmm0;
5237 const XMMRegister vec1 = xmm1;
5238 const XMMRegister vec2 = xmm2;
5239
5240 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
5241
5242 __ vzeroupper();
5243 __ leave();
5244 __ ret(0);
5245
5246 return start;
5247 }
5248
5249 /**
5250 * Arguments:
5251 *
5252 // Input:
5253 // c_rarg0 - x address
5254 // c_rarg1 - x length
5255 // c_rarg2 - z address
5256 // c_rarg3 - z lenth
5257 *
5258 */
5259 address generate_squareToLen() {
5260
5261 __ align(CodeEntryAlignment);
5262 StubCodeMark mark(this, "StubRoutines", "squareToLen");
5263
5264 address start = __ pc();
5265 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5266 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
5267 const Register x = rdi;
5268 const Register len = rsi;
5269 const Register z = r8;
5270 const Register zlen = rcx;
5271
5272 const Register tmp1 = r12;
5273 const Register tmp2 = r13;
5274 const Register tmp3 = r14;
5275 const Register tmp4 = r15;
5276 const Register tmp5 = rbx;
5277
5278 BLOCK_COMMENT("Entry:");
5279 __ enter(); // required for proper stackwalking of RuntimeStub frame
5280
5281 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
5282 // zlen => rcx
5283 // r9 and r10 may be used to save non-volatile registers
5284 __ movptr(r8, rdx);
5285 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5286
5287 restore_arg_regs();
5288
5289 __ leave(); // required for proper stackwalking of RuntimeStub frame
5290 __ ret(0);
5291
5292 return start;
5293 }
5294
5295 address generate_method_entry_barrier() {
5296 __ align(CodeEntryAlignment);
5297 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
5298
5299 Label deoptimize_label;
5300
5301 address start = __ pc();
5302
5303 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
5304
5305 BLOCK_COMMENT("Entry:");
5306 __ enter(); // save rbp
5307
5308 // save c_rarg0, because we want to use that value.
5309 // We could do without it but then we depend on the number of slots used by pusha
5310 __ push(c_rarg0);
5311
5312 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
5313
5314 __ pusha();
5315
5316 // The method may have floats as arguments, and we must spill them before calling
5317 // the VM runtime.
5318 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
5319 const int xmm_size = wordSize * 2;
5320 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
5321 __ subptr(rsp, xmm_spill_size);
5322 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
5323 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
5324 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
5325 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
5326 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
5327 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
5328 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
5329 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
5330
5331 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
5332
5333 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
5334 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
5335 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
5336 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
5337 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
5338 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
5339 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
5340 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
5341 __ addptr(rsp, xmm_spill_size);
5342
5343 __ cmpl(rax, 1); // 1 means deoptimize
5344 __ jcc(Assembler::equal, deoptimize_label);
5345
5346 __ popa();
5347 __ pop(c_rarg0);
5348
5349 __ leave();
5350
5351 __ addptr(rsp, 1 * wordSize); // cookie
5352 __ ret(0);
5353
5354
5355 __ BIND(deoptimize_label);
5356
5357 __ popa();
5358 __ pop(c_rarg0);
5359
5360 __ leave();
5361
5362 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
5363 // here while still having a correct stack is valuable
5364 __ testptr(rsp, Address(rsp, 0));
5365
5366 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
5367 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
5368
5369 return start;
5370 }
5371
5372 /**
5373 * Arguments:
5374 *
5375 * Input:
5376 * c_rarg0 - out address
5377 * c_rarg1 - in address
5378 * c_rarg2 - offset
5379 * c_rarg3 - len
5380 * not Win64
5381 * c_rarg4 - k
5382 * Win64
5383 * rsp+40 - k
5384 */
5385 address generate_mulAdd() {
5386 __ align(CodeEntryAlignment);
5387 StubCodeMark mark(this, "StubRoutines", "mulAdd");
5388
5389 address start = __ pc();
5390 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
5391 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
5392 const Register out = rdi;
5393 const Register in = rsi;
5394 const Register offset = r11;
5395 const Register len = rcx;
5396 const Register k = r8;
5397
5398 // Next registers will be saved on stack in mul_add().
5399 const Register tmp1 = r12;
5400 const Register tmp2 = r13;
5401 const Register tmp3 = r14;
5402 const Register tmp4 = r15;
5403 const Register tmp5 = rbx;
5404
5405 BLOCK_COMMENT("Entry:");
5406 __ enter(); // required for proper stackwalking of RuntimeStub frame
5407
5408 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
5409 // len => rcx, k => r8
5410 // r9 and r10 may be used to save non-volatile registers
5411 #ifdef _WIN64
5412 // last argument is on stack on Win64
5413 __ movl(k, Address(rsp, 6 * wordSize));
5414 #endif
5415 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
5416 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
5417
5418 restore_arg_regs();
5419
5420 __ leave(); // required for proper stackwalking of RuntimeStub frame
5421 __ ret(0);
5422
5423 return start;
5424 }
5425
5426 address generate_libmExp() {
5427 StubCodeMark mark(this, "StubRoutines", "libmExp");
5428
5429 address start = __ pc();
5430
5431 const XMMRegister x0 = xmm0;
5432 const XMMRegister x1 = xmm1;
5433 const XMMRegister x2 = xmm2;
5434 const XMMRegister x3 = xmm3;
5435
5436 const XMMRegister x4 = xmm4;
5437 const XMMRegister x5 = xmm5;
5438 const XMMRegister x6 = xmm6;
5439 const XMMRegister x7 = xmm7;
5440
5441 const Register tmp = r11;
5442
5443 BLOCK_COMMENT("Entry:");
5444 __ enter(); // required for proper stackwalking of RuntimeStub frame
5445
5446 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5447
5448 __ leave(); // required for proper stackwalking of RuntimeStub frame
5449 __ ret(0);
5450
5451 return start;
5452
5453 }
5454
5455 address generate_libmLog() {
5456 StubCodeMark mark(this, "StubRoutines", "libmLog");
5457
5458 address start = __ pc();
5459
5460 const XMMRegister x0 = xmm0;
5461 const XMMRegister x1 = xmm1;
5462 const XMMRegister x2 = xmm2;
5463 const XMMRegister x3 = xmm3;
5464
5465 const XMMRegister x4 = xmm4;
5466 const XMMRegister x5 = xmm5;
5467 const XMMRegister x6 = xmm6;
5468 const XMMRegister x7 = xmm7;
5469
5470 const Register tmp1 = r11;
5471 const Register tmp2 = r8;
5472
5473 BLOCK_COMMENT("Entry:");
5474 __ enter(); // required for proper stackwalking of RuntimeStub frame
5475
5476 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2);
5477
5478 __ leave(); // required for proper stackwalking of RuntimeStub frame
5479 __ ret(0);
5480
5481 return start;
5482
5483 }
5484
5485 address generate_libmLog10() {
5486 StubCodeMark mark(this, "StubRoutines", "libmLog10");
5487
5488 address start = __ pc();
5489
5490 const XMMRegister x0 = xmm0;
5491 const XMMRegister x1 = xmm1;
5492 const XMMRegister x2 = xmm2;
5493 const XMMRegister x3 = xmm3;
5494
5495 const XMMRegister x4 = xmm4;
5496 const XMMRegister x5 = xmm5;
5497 const XMMRegister x6 = xmm6;
5498 const XMMRegister x7 = xmm7;
5499
5500 const Register tmp = r11;
5501
5502 BLOCK_COMMENT("Entry:");
5503 __ enter(); // required for proper stackwalking of RuntimeStub frame
5504
5505 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
5506
5507 __ leave(); // required for proper stackwalking of RuntimeStub frame
5508 __ ret(0);
5509
5510 return start;
5511
5512 }
5513
5514 address generate_libmPow() {
5515 StubCodeMark mark(this, "StubRoutines", "libmPow");
5516
5517 address start = __ pc();
5518
5519 const XMMRegister x0 = xmm0;
5520 const XMMRegister x1 = xmm1;
5521 const XMMRegister x2 = xmm2;
5522 const XMMRegister x3 = xmm3;
5523
5524 const XMMRegister x4 = xmm4;
5525 const XMMRegister x5 = xmm5;
5526 const XMMRegister x6 = xmm6;
5527 const XMMRegister x7 = xmm7;
5528
5529 const Register tmp1 = r8;
5530 const Register tmp2 = r9;
5531 const Register tmp3 = r10;
5532 const Register tmp4 = r11;
5533
5534 BLOCK_COMMENT("Entry:");
5535 __ enter(); // required for proper stackwalking of RuntimeStub frame
5536
5537 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5538
5539 __ leave(); // required for proper stackwalking of RuntimeStub frame
5540 __ ret(0);
5541
5542 return start;
5543
5544 }
5545
5546 address generate_libmSin() {
5547 StubCodeMark mark(this, "StubRoutines", "libmSin");
5548
5549 address start = __ pc();
5550
5551 const XMMRegister x0 = xmm0;
5552 const XMMRegister x1 = xmm1;
5553 const XMMRegister x2 = xmm2;
5554 const XMMRegister x3 = xmm3;
5555
5556 const XMMRegister x4 = xmm4;
5557 const XMMRegister x5 = xmm5;
5558 const XMMRegister x6 = xmm6;
5559 const XMMRegister x7 = xmm7;
5560
5561 const Register tmp1 = r8;
5562 const Register tmp2 = r9;
5563 const Register tmp3 = r10;
5564 const Register tmp4 = r11;
5565
5566 BLOCK_COMMENT("Entry:");
5567 __ enter(); // required for proper stackwalking of RuntimeStub frame
5568
5569 #ifdef _WIN64
5570 __ push(rsi);
5571 __ push(rdi);
5572 #endif
5573 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5574
5575 #ifdef _WIN64
5576 __ pop(rdi);
5577 __ pop(rsi);
5578 #endif
5579
5580 __ leave(); // required for proper stackwalking of RuntimeStub frame
5581 __ ret(0);
5582
5583 return start;
5584
5585 }
5586
5587 address generate_libmCos() {
5588 StubCodeMark mark(this, "StubRoutines", "libmCos");
5589
5590 address start = __ pc();
5591
5592 const XMMRegister x0 = xmm0;
5593 const XMMRegister x1 = xmm1;
5594 const XMMRegister x2 = xmm2;
5595 const XMMRegister x3 = xmm3;
5596
5597 const XMMRegister x4 = xmm4;
5598 const XMMRegister x5 = xmm5;
5599 const XMMRegister x6 = xmm6;
5600 const XMMRegister x7 = xmm7;
5601
5602 const Register tmp1 = r8;
5603 const Register tmp2 = r9;
5604 const Register tmp3 = r10;
5605 const Register tmp4 = r11;
5606
5607 BLOCK_COMMENT("Entry:");
5608 __ enter(); // required for proper stackwalking of RuntimeStub frame
5609
5610 #ifdef _WIN64
5611 __ push(rsi);
5612 __ push(rdi);
5613 #endif
5614 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5615
5616 #ifdef _WIN64
5617 __ pop(rdi);
5618 __ pop(rsi);
5619 #endif
5620
5621 __ leave(); // required for proper stackwalking of RuntimeStub frame
5622 __ ret(0);
5623
5624 return start;
5625
5626 }
5627
5628 address generate_libmTan() {
5629 StubCodeMark mark(this, "StubRoutines", "libmTan");
5630
5631 address start = __ pc();
5632
5633 const XMMRegister x0 = xmm0;
5634 const XMMRegister x1 = xmm1;
5635 const XMMRegister x2 = xmm2;
5636 const XMMRegister x3 = xmm3;
5637
5638 const XMMRegister x4 = xmm4;
5639 const XMMRegister x5 = xmm5;
5640 const XMMRegister x6 = xmm6;
5641 const XMMRegister x7 = xmm7;
5642
5643 const Register tmp1 = r8;
5644 const Register tmp2 = r9;
5645 const Register tmp3 = r10;
5646 const Register tmp4 = r11;
5647
5648 BLOCK_COMMENT("Entry:");
5649 __ enter(); // required for proper stackwalking of RuntimeStub frame
5650
5651 #ifdef _WIN64
5652 __ push(rsi);
5653 __ push(rdi);
5654 #endif
5655 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp1, tmp2, tmp3, tmp4);
5656
5657 #ifdef _WIN64
5658 __ pop(rdi);
5659 __ pop(rsi);
5660 #endif
5661
5662 __ leave(); // required for proper stackwalking of RuntimeStub frame
5663 __ ret(0);
5664
5665 return start;
5666
5667 }
5668
5669 #undef __
5670 #define __ masm->
5671
5672 // Continuation point for throwing of implicit exceptions that are
5673 // not handled in the current activation. Fabricates an exception
5674 // oop and initiates normal exception dispatching in this
5675 // frame. Since we need to preserve callee-saved values (currently
5676 // only for C2, but done for C1 as well) we need a callee-saved oop
5677 // map and therefore have to make these stubs into RuntimeStubs
5678 // rather than BufferBlobs. If the compiler needs all registers to
5679 // be preserved between the fault point and the exception handler
5680 // then it must assume responsibility for that in
5681 // AbstractCompiler::continuation_for_implicit_null_exception or
5682 // continuation_for_implicit_division_by_zero_exception. All other
5683 // implicit exceptions (e.g., NullPointerException or
5684 // AbstractMethodError on entry) are either at call sites or
5685 // otherwise assume that stack unwinding will be initiated, so
5686 // caller saved registers were assumed volatile in the compiler.
5687 address generate_throw_exception(const char* name,
5688 address runtime_entry,
5689 Register arg1 = noreg,
5690 Register arg2 = noreg) {
5691 // Information about frame layout at time of blocking runtime call.
5692 // Note that we only have to preserve callee-saved registers since
5693 // the compilers are responsible for supplying a continuation point
5694 // if they expect all registers to be preserved.
5695 enum layout {
5696 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
5697 rbp_off2,
5698 return_off,
5699 return_off2,
5700 framesize // inclusive of return address
5701 };
5702
5703 int insts_size = 512;
5704 int locs_size = 64;
5705
5706 CodeBuffer code(name, insts_size, locs_size);
5707 OopMapSet* oop_maps = new OopMapSet();
5708 MacroAssembler* masm = new MacroAssembler(&code);
5709
5710 address start = __ pc();
5711
5712 // This is an inlined and slightly modified version of call_VM
5713 // which has the ability to fetch the return PC out of
5714 // thread-local storage and also sets up last_Java_sp slightly
5715 // differently than the real call_VM
5716
5717 __ enter(); // required for proper stackwalking of RuntimeStub frame
5718
5719 assert(is_even(framesize/2), "sp not 16-byte aligned");
5720
5721 // return address and rbp are already in place
5722 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
5723
5724 int frame_complete = __ pc() - start;
5725
5726 // Set up last_Java_sp and last_Java_fp
5727 address the_pc = __ pc();
5728 __ set_last_Java_frame(rsp, rbp, the_pc);
5729 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
5730
5731 // Call runtime
5732 if (arg1 != noreg) {
5733 assert(arg2 != c_rarg1, "clobbered");
5734 __ movptr(c_rarg1, arg1);
5735 }
5736 if (arg2 != noreg) {
5737 __ movptr(c_rarg2, arg2);
5738 }
5739 __ movptr(c_rarg0, r15_thread);
5740 BLOCK_COMMENT("call runtime_entry");
5741 __ call(RuntimeAddress(runtime_entry));
5742
5743 // Generate oop map
5744 OopMap* map = new OopMap(framesize, 0);
5745
5746 oop_maps->add_gc_map(the_pc - start, map);
5747
5748 __ reset_last_Java_frame(true);
5749
5750 __ leave(); // required for proper stackwalking of RuntimeStub frame
5751
5752 // check for pending exceptions
5753 #ifdef ASSERT
5754 Label L;
5755 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
5756 (int32_t) NULL_WORD);
5757 __ jcc(Assembler::notEqual, L);
5758 __ should_not_reach_here();
5759 __ bind(L);
5760 #endif // ASSERT
5761 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5762
5763
5764 // codeBlob framesize is in words (not VMRegImpl::slot_size)
5765 RuntimeStub* stub =
5766 RuntimeStub::new_runtime_stub(name,
5767 &code,
5768 frame_complete,
5769 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
5770 oop_maps, false);
5771 return stub->entry_point();
5772 }
5773
5774 void create_control_words() {
5775 // Round to nearest, 53-bit mode, exceptions masked
5776 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
5777 // Round to zero, 53-bit mode, exception mased
5778 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
5779 // Round to nearest, 24-bit mode, exceptions masked
5780 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
5781 // Round to nearest, 64-bit mode, exceptions masked
5782 StubRoutines::_mxcsr_std = 0x1F80;
5783 // Note: the following two constants are 80-bit values
5784 // layout is critical for correct loading by FPU.
5785 // Bias for strict fp multiply/divide
5786 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
5787 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
5788 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
5789 // Un-Bias for strict fp multiply/divide
5790 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
5791 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
5792 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
5793 }
5794
5795 // Call here from the interpreter or compiled code to either load
5796 // multiple returned values from the value type instance being
5797 // returned to registers or to store returned values to a newly
5798 // allocated value type instance.
5799 address generate_return_value_stub(address destination, const char* name, bool has_res) {
5800 // We need to save all registers the calling convention may use so
5801 // the runtime calls read or update those registers. This needs to
5802 // be in sync with SharedRuntime::java_return_convention().
5803 enum layout {
5804 pad_off = frame::arg_reg_save_area_bytes/BytesPerInt, pad_off_2,
5805 rax_off, rax_off_2,
5806 j_rarg5_off, j_rarg5_2,
5807 j_rarg4_off, j_rarg4_2,
5808 j_rarg3_off, j_rarg3_2,
5809 j_rarg2_off, j_rarg2_2,
5810 j_rarg1_off, j_rarg1_2,
5811 j_rarg0_off, j_rarg0_2,
5812 j_farg0_off, j_farg0_2,
5813 j_farg1_off, j_farg1_2,
5814 j_farg2_off, j_farg2_2,
5815 j_farg3_off, j_farg3_2,
5816 j_farg4_off, j_farg4_2,
5817 j_farg5_off, j_farg5_2,
5818 j_farg6_off, j_farg6_2,
5819 j_farg7_off, j_farg7_2,
5820 rbp_off, rbp_off_2,
5821 return_off, return_off_2,
5822
5823 framesize
5824 };
5825
5826 CodeBuffer buffer(name, 1000, 512);
5827 MacroAssembler* masm = new MacroAssembler(&buffer);
5828
5829 int frame_size_in_bytes = align_up(framesize*BytesPerInt, 16);
5830 assert(frame_size_in_bytes == framesize*BytesPerInt, "misaligned");
5831 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
5832 int frame_size_in_words = frame_size_in_bytes / wordSize;
5833
5834 OopMapSet *oop_maps = new OopMapSet();
5835 OopMap* map = new OopMap(frame_size_in_slots, 0);
5836
5837 map->set_callee_saved(VMRegImpl::stack2reg(rax_off), rax->as_VMReg());
5838 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg5_off), j_rarg5->as_VMReg());
5839 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg4_off), j_rarg4->as_VMReg());
5840 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg3_off), j_rarg3->as_VMReg());
5841 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg2_off), j_rarg2->as_VMReg());
5842 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg1_off), j_rarg1->as_VMReg());
5843 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg0_off), j_rarg0->as_VMReg());
5844 map->set_callee_saved(VMRegImpl::stack2reg(j_farg0_off), j_farg0->as_VMReg());
5845 map->set_callee_saved(VMRegImpl::stack2reg(j_farg1_off), j_farg1->as_VMReg());
5846 map->set_callee_saved(VMRegImpl::stack2reg(j_farg2_off), j_farg2->as_VMReg());
5847 map->set_callee_saved(VMRegImpl::stack2reg(j_farg3_off), j_farg3->as_VMReg());
5848 map->set_callee_saved(VMRegImpl::stack2reg(j_farg4_off), j_farg4->as_VMReg());
5849 map->set_callee_saved(VMRegImpl::stack2reg(j_farg5_off), j_farg5->as_VMReg());
5850 map->set_callee_saved(VMRegImpl::stack2reg(j_farg6_off), j_farg6->as_VMReg());
5851 map->set_callee_saved(VMRegImpl::stack2reg(j_farg7_off), j_farg7->as_VMReg());
5852
5853 int start = __ offset();
5854
5855 __ subptr(rsp, frame_size_in_bytes - 8 /* return address*/);
5856
5857 __ movptr(Address(rsp, rbp_off * BytesPerInt), rbp);
5858 __ movdbl(Address(rsp, j_farg7_off * BytesPerInt), j_farg7);
5859 __ movdbl(Address(rsp, j_farg6_off * BytesPerInt), j_farg6);
5860 __ movdbl(Address(rsp, j_farg5_off * BytesPerInt), j_farg5);
5861 __ movdbl(Address(rsp, j_farg4_off * BytesPerInt), j_farg4);
5862 __ movdbl(Address(rsp, j_farg3_off * BytesPerInt), j_farg3);
5863 __ movdbl(Address(rsp, j_farg2_off * BytesPerInt), j_farg2);
5864 __ movdbl(Address(rsp, j_farg1_off * BytesPerInt), j_farg1);
5865 __ movdbl(Address(rsp, j_farg0_off * BytesPerInt), j_farg0);
5866
5867 __ movptr(Address(rsp, j_rarg0_off * BytesPerInt), j_rarg0);
5868 __ movptr(Address(rsp, j_rarg1_off * BytesPerInt), j_rarg1);
5869 __ movptr(Address(rsp, j_rarg2_off * BytesPerInt), j_rarg2);
5870 __ movptr(Address(rsp, j_rarg3_off * BytesPerInt), j_rarg3);
5871 __ movptr(Address(rsp, j_rarg4_off * BytesPerInt), j_rarg4);
5872 __ movptr(Address(rsp, j_rarg5_off * BytesPerInt), j_rarg5);
5873 __ movptr(Address(rsp, rax_off * BytesPerInt), rax);
5874
5875 int frame_complete = __ offset();
5876
5877 __ set_last_Java_frame(noreg, noreg, NULL);
5878
5879 __ mov(c_rarg0, r15_thread);
5880 __ mov(c_rarg1, rax);
5881
5882 __ call(RuntimeAddress(destination));
5883
5884 // Set an oopmap for the call site.
5885
5886 oop_maps->add_gc_map( __ offset() - start, map);
5887
5888 // clear last_Java_sp
5889 __ reset_last_Java_frame(false);
5890
5891 __ movptr(rbp, Address(rsp, rbp_off * BytesPerInt));
5892 __ movdbl(j_farg7, Address(rsp, j_farg7_off * BytesPerInt));
5893 __ movdbl(j_farg6, Address(rsp, j_farg6_off * BytesPerInt));
5894 __ movdbl(j_farg5, Address(rsp, j_farg5_off * BytesPerInt));
5895 __ movdbl(j_farg4, Address(rsp, j_farg4_off * BytesPerInt));
5896 __ movdbl(j_farg3, Address(rsp, j_farg3_off * BytesPerInt));
5897 __ movdbl(j_farg2, Address(rsp, j_farg2_off * BytesPerInt));
5898 __ movdbl(j_farg1, Address(rsp, j_farg1_off * BytesPerInt));
5899 __ movdbl(j_farg0, Address(rsp, j_farg0_off * BytesPerInt));
5900
5901 __ movptr(j_rarg0, Address(rsp, j_rarg0_off * BytesPerInt));
5902 __ movptr(j_rarg1, Address(rsp, j_rarg1_off * BytesPerInt));
5903 __ movptr(j_rarg2, Address(rsp, j_rarg2_off * BytesPerInt));
5904 __ movptr(j_rarg3, Address(rsp, j_rarg3_off * BytesPerInt));
5905 __ movptr(j_rarg4, Address(rsp, j_rarg4_off * BytesPerInt));
5906 __ movptr(j_rarg5, Address(rsp, j_rarg5_off * BytesPerInt));
5907 __ movptr(rax, Address(rsp, rax_off * BytesPerInt));
5908
5909 __ addptr(rsp, frame_size_in_bytes-8);
5910
5911 // check for pending exceptions
5912 Label pending;
5913 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
5914 __ jcc(Assembler::notEqual, pending);
5915
5916 if (has_res) {
5917 __ get_vm_result(rax, r15_thread);
5918 }
5919
5920 __ ret(0);
5921
5922 __ bind(pending);
5923
5924 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
5925 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
5926
5927 // -------------
5928 // make sure all code is generated
5929 masm->flush();
5930
5931 // The caller may not know the register mapping of the fields of the returned value
5932 // object, so it won't generate a valid oopmap for the call site. Hence, we can't
5933 // do InterfaceSupport::verify_stack().
5934 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, false,
5935 /*can_verify_stack =*/false);
5936 return stub->entry_point();
5937 }
5938
5939 // Initialization
5940 void generate_initial() {
5941 // Generates all stubs and initializes the entry points
5942
5943 // This platform-specific settings are needed by generate_call_stub()
5944 create_control_words();
5945
5946 // entry points that exist in all platforms Note: This is code
5947 // that could be shared among different platforms - however the
5948 // benefit seems to be smaller than the disadvantage of having a
5949 // much more complicated generator structure. See also comment in
5950 // stubRoutines.hpp.
5951
5952 StubRoutines::_forward_exception_entry = generate_forward_exception();
5953
5954 // Generate these first because they are called from other stubs
5955 StubRoutines::_load_value_type_fields_in_regs = generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::load_value_type_fields_in_regs), "load_value_type_fields_in_regs", false);
5956 StubRoutines::_store_value_type_fields_to_buf = generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::store_value_type_fields_to_buf), "store_value_type_fields_to_buf", true);
5957
5958 StubRoutines::_call_stub_entry = generate_call_stub(StubRoutines::_call_stub_return_address);
5959
5960 // is referenced by megamorphic call
5961 StubRoutines::_catch_exception_entry = generate_catch_exception();
5962
5963 // atomic calls
5964 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
5965 StubRoutines::_atomic_xchg_long_entry = generate_atomic_xchg_long();
5966 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
5967 StubRoutines::_atomic_cmpxchg_byte_entry = generate_atomic_cmpxchg_byte();
5968 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
5969 StubRoutines::_atomic_add_entry = generate_atomic_add();
5970 StubRoutines::_atomic_add_long_entry = generate_atomic_add_long();
5971 StubRoutines::_fence_entry = generate_orderaccess_fence();
5972
5973 // platform dependent
5974 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
5975 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
5976
5977 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
5978
5979 // Build this early so it's available for the interpreter.
5980 StubRoutines::_throw_StackOverflowError_entry =
5981 generate_throw_exception("StackOverflowError throw_exception",
5982 CAST_FROM_FN_PTR(address,
5983 SharedRuntime::
5984 throw_StackOverflowError));
5985 StubRoutines::_throw_delayed_StackOverflowError_entry =
5986 generate_throw_exception("delayed StackOverflowError throw_exception",
5987 CAST_FROM_FN_PTR(address,
5988 SharedRuntime::
5989 throw_delayed_StackOverflowError));
5990 if (UseCRC32Intrinsics) {
5991 // set table address before stub generation which use it
5992 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
5993 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
5994 }
5995
5996 if (UseCRC32CIntrinsics) {
5997 bool supports_clmul = VM_Version::supports_clmul();
5998 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
5999 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
6000 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
6001 }
6002 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
6003 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
6004 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
6005 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6006 StubRoutines::x86::_ONEHALF_adr = (address)StubRoutines::x86::_ONEHALF;
6007 StubRoutines::x86::_P_2_adr = (address)StubRoutines::x86::_P_2;
6008 StubRoutines::x86::_SC_4_adr = (address)StubRoutines::x86::_SC_4;
6009 StubRoutines::x86::_Ctable_adr = (address)StubRoutines::x86::_Ctable;
6010 StubRoutines::x86::_SC_2_adr = (address)StubRoutines::x86::_SC_2;
6011 StubRoutines::x86::_SC_3_adr = (address)StubRoutines::x86::_SC_3;
6012 StubRoutines::x86::_SC_1_adr = (address)StubRoutines::x86::_SC_1;
6013 StubRoutines::x86::_PI_INV_TABLE_adr = (address)StubRoutines::x86::_PI_INV_TABLE;
6014 StubRoutines::x86::_PI_4_adr = (address)StubRoutines::x86::_PI_4;
6015 StubRoutines::x86::_PI32INV_adr = (address)StubRoutines::x86::_PI32INV;
6016 StubRoutines::x86::_SIGN_MASK_adr = (address)StubRoutines::x86::_SIGN_MASK;
6017 StubRoutines::x86::_P_1_adr = (address)StubRoutines::x86::_P_1;
6018 StubRoutines::x86::_P_3_adr = (address)StubRoutines::x86::_P_3;
6019 StubRoutines::x86::_NEG_ZERO_adr = (address)StubRoutines::x86::_NEG_ZERO;
6020 }
6021 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
6022 StubRoutines::_dexp = generate_libmExp();
6023 }
6024 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
6025 StubRoutines::_dlog = generate_libmLog();
6026 }
6027 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
6028 StubRoutines::_dlog10 = generate_libmLog10();
6029 }
6030 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
6031 StubRoutines::_dpow = generate_libmPow();
6032 }
6033 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
6034 StubRoutines::_dsin = generate_libmSin();
6035 }
6036 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
6037 StubRoutines::_dcos = generate_libmCos();
6038 }
6039 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
6040 StubRoutines::_dtan = generate_libmTan();
6041 }
6042 }
6043 }
6044
6045 void generate_all() {
6046 // Generates all stubs and initializes the entry points
6047
6048 // These entry points require SharedInfo::stack0 to be set up in
6049 // non-core builds and need to be relocatable, so they each
6050 // fabricate a RuntimeStub internally.
6051 StubRoutines::_throw_AbstractMethodError_entry =
6052 generate_throw_exception("AbstractMethodError throw_exception",
6053 CAST_FROM_FN_PTR(address,
6054 SharedRuntime::
6055 throw_AbstractMethodError));
6056
6057 StubRoutines::_throw_IncompatibleClassChangeError_entry =
6058 generate_throw_exception("IncompatibleClassChangeError throw_exception",
6059 CAST_FROM_FN_PTR(address,
6060 SharedRuntime::
6061 throw_IncompatibleClassChangeError));
6062
6063 StubRoutines::_throw_NullPointerException_at_call_entry =
6064 generate_throw_exception("NullPointerException at call throw_exception",
6065 CAST_FROM_FN_PTR(address,
6066 SharedRuntime::
6067 throw_NullPointerException_at_call));
6068
6069 // entry points that are platform specific
6070 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
6071 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
6072 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
6073 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
6074
6075 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6076 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
6077 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6078 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
6079 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
6080 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
6081 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
6082 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
6083 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
6084 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
6085 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
6086
6087 // support for verify_oop (must happen after universe_init)
6088 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
6089
6090 // arraycopy stubs used by compilers
6091 generate_arraycopy_stubs();
6092
6093 // don't bother generating these AES intrinsic stubs unless global flag is set
6094 if (UseAESIntrinsics) {
6095 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // needed by the others
6096 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
6097 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
6098 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
6099 if (VM_Version::supports_vaes() && VM_Version::supports_avx512vl() && VM_Version::supports_avx512dq() ) {
6100 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptVectorAESCrypt();
6101 } else {
6102 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
6103 }
6104 }
6105 if (UseAESCTRIntrinsics){
6106 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
6107 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
6108 }
6109
6110 if (UseSHA1Intrinsics) {
6111 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
6112 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
6113 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
6114 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
6115 }
6116 if (UseSHA256Intrinsics) {
6117 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
6118 char* dst = (char*)StubRoutines::x86::_k256_W;
6119 char* src = (char*)StubRoutines::x86::_k256;
6120 for (int ii = 0; ii < 16; ++ii) {
6121 memcpy(dst + 32 * ii, src + 16 * ii, 16);
6122 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
6123 }
6124 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
6125 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
6126 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
6127 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
6128 }
6129 if (UseSHA512Intrinsics) {
6130 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
6131 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
6132 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
6133 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
6134 }
6135
6136 // Generate GHASH intrinsics code
6137 if (UseGHASHIntrinsics) {
6138 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
6139 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
6140 if (VM_Version::supports_avx()) {
6141 StubRoutines::x86::_ghash_shuffmask_addr = ghash_shufflemask_addr();
6142 StubRoutines::x86::_ghash_poly_addr = ghash_polynomial_addr();
6143 StubRoutines::_ghash_processBlocks = generate_avx_ghash_processBlocks();
6144 } else {
6145 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
6146 }
6147 }
6148
6149 if (UseBASE64Intrinsics) {
6150 StubRoutines::x86::_and_mask = base64_and_mask_addr();
6151 StubRoutines::x86::_bswap_mask = base64_bswap_mask_addr();
6152 StubRoutines::x86::_base64_charset = base64_charset_addr();
6153 StubRoutines::x86::_url_charset = base64url_charset_addr();
6154 StubRoutines::x86::_gather_mask = base64_gather_mask_addr();
6155 StubRoutines::x86::_left_shift_mask = base64_left_shift_mask_addr();
6156 StubRoutines::x86::_right_shift_mask = base64_right_shift_mask_addr();
6157 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
6158 }
6159
6160 // Safefetch stubs.
6161 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
6162 &StubRoutines::_safefetch32_fault_pc,
6163 &StubRoutines::_safefetch32_continuation_pc);
6164 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
6165 &StubRoutines::_safefetchN_fault_pc,
6166 &StubRoutines::_safefetchN_continuation_pc);
6167
6168 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
6169 if (bs_nm != NULL) {
6170 StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
6171 }
6172 #ifdef COMPILER2
6173 if (UseMultiplyToLenIntrinsic) {
6174 StubRoutines::_multiplyToLen = generate_multiplyToLen();
6175 }
6176 if (UseSquareToLenIntrinsic) {
6177 StubRoutines::_squareToLen = generate_squareToLen();
6178 }
6179 if (UseMulAddIntrinsic) {
6180 StubRoutines::_mulAdd = generate_mulAdd();
6181 }
6182 #ifndef _WINDOWS
6183 if (UseMontgomeryMultiplyIntrinsic) {
6184 StubRoutines::_montgomeryMultiply
6185 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
6186 }
6187 if (UseMontgomerySquareIntrinsic) {
6188 StubRoutines::_montgomerySquare
6189 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
6190 }
6191 #endif // WINDOWS
6192 #endif // COMPILER2
6193
6194 if (UseVectorizedMismatchIntrinsic) {
6195 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
6196 }
6197 }
6198
6199 public:
6200 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
6201 if (all) {
6202 generate_all();
6203 } else {
6204 generate_initial();
6205 }
6206 }
6207 }; // end class declaration
6208
6209 void StubGenerator_generate(CodeBuffer* code, bool all) {
6210 StubGenerator g(code, all);
6211 }