1 /*
2 * Copyright (c) 2003, 2025, 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 "asm/macroAssembler.hpp"
26 #include "classfile/javaClasses.hpp"
27 #include "classfile/vmIntrinsics.hpp"
28 #include "code/SCCache.hpp"
29 #include "compiler/oopMap.hpp"
30 #include "gc/shared/barrierSet.hpp"
31 #include "gc/shared/barrierSetAssembler.hpp"
32 #include "gc/shared/barrierSetNMethod.hpp"
33 #include "gc/shared/gc_globals.hpp"
34 #include "memory/universe.hpp"
35 #include "prims/jvmtiExport.hpp"
36 #include "prims/upcallLinker.hpp"
37 #include "runtime/arguments.hpp"
38 #include "runtime/continuationEntry.hpp"
39 #include "runtime/javaThread.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/stubRoutines.hpp"
42 #include "stubGenerator_x86_64.hpp"
43 #ifdef COMPILER2
44 #include "opto/runtime.hpp"
45 #include "opto/c2_globals.hpp"
46 #endif
47 #if INCLUDE_JVMCI
48 #include "jvmci/jvmci_globals.hpp"
49 #endif
50
51 // For a more detailed description of the stub routine structure
52 // see the comment in stubRoutines.hpp
53
54 #define __ _masm->
55 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
56
57 #ifdef PRODUCT
58 #define BLOCK_COMMENT(str) /* nothing */
59 #else
60 #define BLOCK_COMMENT(str) __ block_comment(str)
61 #endif // PRODUCT
62
63 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
64
65 //
66 // Linux Arguments:
67 // c_rarg0: call wrapper address address
68 // c_rarg1: result address
69 // c_rarg2: result type BasicType
70 // c_rarg3: method Method*
71 // c_rarg4: (interpreter) entry point address
72 // c_rarg5: parameters intptr_t*
73 // 16(rbp): parameter size (in words) int
74 // 24(rbp): thread Thread*
75 //
76 // [ return_from_Java ] <--- rsp
77 // [ argument word n ]
78 // ...
79 // -12 [ argument word 1 ]
80 // -11 [ saved r15 ] <--- rsp_after_call
81 // -10 [ saved r14 ]
82 // -9 [ saved r13 ]
83 // -8 [ saved r12 ]
84 // -7 [ saved rbx ]
85 // -6 [ call wrapper ]
86 // -5 [ result ]
87 // -4 [ result type ]
88 // -3 [ method ]
89 // -2 [ entry point ]
90 // -1 [ parameters ]
91 // 0 [ saved rbp ] <--- rbp
92 // 1 [ return address ]
93 // 2 [ parameter size ]
94 // 3 [ thread ]
95 //
96 // Windows Arguments:
97 // c_rarg0: call wrapper address address
98 // c_rarg1: result address
99 // c_rarg2: result type BasicType
100 // c_rarg3: method Method*
101 // 48(rbp): (interpreter) entry point address
102 // 56(rbp): parameters intptr_t*
103 // 64(rbp): parameter size (in words) int
104 // 72(rbp): thread Thread*
105 //
106 // [ return_from_Java ] <--- rsp
107 // [ argument word n ]
108 // ...
109 // -28 [ argument word 1 ]
110 // -27 [ saved xmm15 ] <--- rsp after_call
111 // [ saved xmm7-xmm14 ]
112 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
113 // -7 [ saved r15 ]
114 // -6 [ saved r14 ]
115 // -5 [ saved r13 ]
116 // -4 [ saved r12 ]
117 // -3 [ saved rdi ]
118 // -2 [ saved rsi ]
119 // -1 [ saved rbx ]
120 // 0 [ saved rbp ] <--- rbp
121 // 1 [ return address ]
122 // 2 [ call wrapper ]
123 // 3 [ result ]
124 // 4 [ result type ]
125 // 5 [ method ]
126 // 6 [ entry point ]
127 // 7 [ parameters ]
128 // 8 [ parameter size ]
129 // 9 [ thread ]
130 //
131 // Windows reserves the callers stack space for arguments 1-4.
132 // We spill c_rarg0-c_rarg3 to this space.
133
134 // Call stub stack layout word offsets from rbp
135 #ifdef _WIN64
136 enum call_stub_layout {
137 xmm_save_first = 6, // save from xmm6
138 xmm_save_last = 15, // to xmm15
139 xmm_save_base = -9,
140 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
141 r15_off = -7,
142 r14_off = -6,
143 r13_off = -5,
144 r12_off = -4,
145 rdi_off = -3,
146 rsi_off = -2,
147 rbx_off = -1,
148 rbp_off = 0,
149 retaddr_off = 1,
150 call_wrapper_off = 2,
151 result_off = 3,
152 result_type_off = 4,
153 method_off = 5,
154 entry_point_off = 6,
155 parameters_off = 7,
156 parameter_size_off = 8,
157 thread_off = 9
158 };
159
160 static Address xmm_save(int reg) {
161 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
162 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
163 }
164 #else // !_WIN64
165 enum call_stub_layout {
166 rsp_after_call_off = -12,
167 mxcsr_off = rsp_after_call_off,
168 r15_off = -11,
169 r14_off = -10,
170 r13_off = -9,
171 r12_off = -8,
172 rbx_off = -7,
173 call_wrapper_off = -6,
174 result_off = -5,
175 result_type_off = -4,
176 method_off = -3,
177 entry_point_off = -2,
178 parameters_off = -1,
179 rbp_off = 0,
180 retaddr_off = 1,
181 parameter_size_off = 2,
182 thread_off = 3
183 };
184 #endif // _WIN64
185
186 address StubGenerator::generate_call_stub(address& return_address) {
187
188 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
189 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
190 "adjust this code");
191 StubGenStubId stub_id = StubGenStubId::call_stub_id;
192 StubCodeMark mark(this, stub_id);
193 address start = __ pc();
194
195 // same as in generate_catch_exception()!
196 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
197
198 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
199 const Address result (rbp, result_off * wordSize);
200 const Address result_type (rbp, result_type_off * wordSize);
201 const Address method (rbp, method_off * wordSize);
202 const Address entry_point (rbp, entry_point_off * wordSize);
203 const Address parameters (rbp, parameters_off * wordSize);
204 const Address parameter_size(rbp, parameter_size_off * wordSize);
205
206 // same as in generate_catch_exception()!
207 const Address thread (rbp, thread_off * wordSize);
208
209 const Address r15_save(rbp, r15_off * wordSize);
210 const Address r14_save(rbp, r14_off * wordSize);
211 const Address r13_save(rbp, r13_off * wordSize);
212 const Address r12_save(rbp, r12_off * wordSize);
213 const Address rbx_save(rbp, rbx_off * wordSize);
214
215 // stub code
216 __ enter();
217 __ subptr(rsp, -rsp_after_call_off * wordSize);
218
219 // save register parameters
220 #ifndef _WIN64
221 __ movptr(parameters, c_rarg5); // parameters
222 __ movptr(entry_point, c_rarg4); // entry_point
223 #endif
224
225 __ movptr(method, c_rarg3); // method
226 __ movl(result_type, c_rarg2); // result type
227 __ movptr(result, c_rarg1); // result
228 __ movptr(call_wrapper, c_rarg0); // call wrapper
229
230 // save regs belonging to calling function
231 __ movptr(rbx_save, rbx);
232 __ movptr(r12_save, r12);
233 __ movptr(r13_save, r13);
234 __ movptr(r14_save, r14);
235 __ movptr(r15_save, r15);
236
237 #ifdef _WIN64
238 int last_reg = 15;
239 for (int i = xmm_save_first; i <= last_reg; i++) {
240 __ movdqu(xmm_save(i), as_XMMRegister(i));
241 }
242
243 const Address rdi_save(rbp, rdi_off * wordSize);
244 const Address rsi_save(rbp, rsi_off * wordSize);
245
246 __ movptr(rsi_save, rsi);
247 __ movptr(rdi_save, rdi);
248 #else
249 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
250 {
251 Label skip_ldmx;
252 __ cmp32_mxcsr_std(mxcsr_save, rax, rscratch1);
253 __ jcc(Assembler::equal, skip_ldmx);
254 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
255 __ ldmxcsr(mxcsr_std, rscratch1);
256 __ bind(skip_ldmx);
257 }
258 #endif
259
260 // Load up thread register
261 __ movptr(r15_thread, thread);
262 __ reinit_heapbase();
263
264 #ifdef ASSERT
265 // make sure we have no pending exceptions
266 {
267 Label L;
268 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
269 __ jcc(Assembler::equal, L);
270 __ stop("StubRoutines::call_stub: entered with pending exception");
271 __ bind(L);
272 }
273 #endif
274
275 // pass parameters if any
276 BLOCK_COMMENT("pass parameters if any");
277 Label parameters_done;
278 __ movl(c_rarg3, parameter_size);
279 __ testl(c_rarg3, c_rarg3);
280 __ jcc(Assembler::zero, parameters_done);
281
282 Label loop;
283 __ movptr(c_rarg2, parameters); // parameter pointer
284 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
285 __ BIND(loop);
286 __ movptr(rax, Address(c_rarg2, 0));// get parameter
287 __ addptr(c_rarg2, wordSize); // advance to next parameter
288 __ decrementl(c_rarg1); // decrement counter
289 __ push(rax); // pass parameter
290 __ jcc(Assembler::notZero, loop);
291
292 // call Java function
293 __ BIND(parameters_done);
294 __ movptr(rbx, method); // get Method*
295 __ movptr(c_rarg1, entry_point); // get entry_point
296 __ mov(r13, rsp); // set sender sp
297 BLOCK_COMMENT("call Java function");
298 __ call(c_rarg1);
299
300 BLOCK_COMMENT("call_stub_return_address:");
301 return_address = __ pc();
302
303 // store result depending on type (everything that is not
304 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
305 __ movptr(c_rarg0, result);
306 Label is_long, is_float, is_double, exit;
307 __ movl(c_rarg1, result_type);
308 __ cmpl(c_rarg1, T_OBJECT);
309 __ jcc(Assembler::equal, is_long);
310 __ cmpl(c_rarg1, T_LONG);
311 __ jcc(Assembler::equal, is_long);
312 __ cmpl(c_rarg1, T_FLOAT);
313 __ jcc(Assembler::equal, is_float);
314 __ cmpl(c_rarg1, T_DOUBLE);
315 __ jcc(Assembler::equal, is_double);
316 #ifdef ASSERT
317 // make sure the type is INT
318 {
319 Label L;
320 __ cmpl(c_rarg1, T_INT);
321 __ jcc(Assembler::equal, L);
322 __ stop("StubRoutines::call_stub: unexpected result type");
323 __ bind(L);
324 }
325 #endif
326
327 // handle T_INT case
328 __ movl(Address(c_rarg0, 0), rax);
329
330 __ BIND(exit);
331
332 // pop parameters
333 __ lea(rsp, rsp_after_call);
334
335 #ifdef ASSERT
336 // verify that threads correspond
337 {
338 Label L1, L2, L3;
339 __ cmpptr(r15_thread, thread);
340 __ jcc(Assembler::equal, L1);
341 __ stop("StubRoutines::call_stub: r15_thread is corrupted");
342 __ bind(L1);
343 __ get_thread(rbx);
344 __ cmpptr(r15_thread, thread);
345 __ jcc(Assembler::equal, L2);
346 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
347 __ bind(L2);
348 __ cmpptr(r15_thread, rbx);
349 __ jcc(Assembler::equal, L3);
350 __ stop("StubRoutines::call_stub: threads must correspond");
351 __ bind(L3);
352 }
353 #endif
354
355 __ pop_cont_fastpath();
356
357 // restore regs belonging to calling function
358 #ifdef _WIN64
359 // emit the restores for xmm regs
360 for (int i = xmm_save_first; i <= last_reg; i++) {
361 __ movdqu(as_XMMRegister(i), xmm_save(i));
362 }
363 #endif
364 __ movptr(r15, r15_save);
365 __ movptr(r14, r14_save);
366 __ movptr(r13, r13_save);
367 __ movptr(r12, r12_save);
368 __ movptr(rbx, rbx_save);
369
370 #ifdef _WIN64
371 __ movptr(rdi, rdi_save);
372 __ movptr(rsi, rsi_save);
373 #else
374 __ ldmxcsr(mxcsr_save);
375 #endif
376
377 // restore rsp
378 __ addptr(rsp, -rsp_after_call_off * wordSize);
379
380 // return
381 __ vzeroupper();
382 __ pop(rbp);
383 __ ret(0);
384
385 // handle return types different from T_INT
386 __ BIND(is_long);
387 __ movq(Address(c_rarg0, 0), rax);
388 __ jmp(exit);
389
390 __ BIND(is_float);
391 __ movflt(Address(c_rarg0, 0), xmm0);
392 __ jmp(exit);
393
394 __ BIND(is_double);
395 __ movdbl(Address(c_rarg0, 0), xmm0);
396 __ jmp(exit);
397
398 return start;
399 }
400
401 // Return point for a Java call if there's an exception thrown in
402 // Java code. The exception is caught and transformed into a
403 // pending exception stored in JavaThread that can be tested from
404 // within the VM.
405 //
406 // Note: Usually the parameters are removed by the callee. In case
407 // of an exception crossing an activation frame boundary, that is
408 // not the case if the callee is compiled code => need to setup the
409 // rsp.
410 //
411 // rax: exception oop
412
413 address StubGenerator::generate_catch_exception() {
414 StubGenStubId stub_id = StubGenStubId::catch_exception_id;
415 StubCodeMark mark(this, stub_id);
416 address start = __ pc();
417
418 // same as in generate_call_stub():
419 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
420 const Address thread (rbp, thread_off * wordSize);
421
422 #ifdef ASSERT
423 // verify that threads correspond
424 {
425 Label L1, L2, L3;
426 __ cmpptr(r15_thread, thread);
427 __ jcc(Assembler::equal, L1);
428 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
429 __ bind(L1);
430 __ get_thread(rbx);
431 __ cmpptr(r15_thread, thread);
432 __ jcc(Assembler::equal, L2);
433 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
434 __ bind(L2);
435 __ cmpptr(r15_thread, rbx);
436 __ jcc(Assembler::equal, L3);
437 __ stop("StubRoutines::catch_exception: threads must correspond");
438 __ bind(L3);
439 }
440 #endif
441
442 // set pending exception
443 __ verify_oop(rax);
444
445 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
446 __ lea(rscratch1, ExternalAddress((address)__FILE__));
447 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
448 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
449
450 // complete return to VM
451 assert(StubRoutines::_call_stub_return_address != nullptr,
452 "_call_stub_return_address must have been generated before");
453 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
454
455 return start;
456 }
457
458 // Continuation point for runtime calls returning with a pending
459 // exception. The pending exception check happened in the runtime
460 // or native call stub. The pending exception in Thread is
461 // converted into a Java-level exception.
462 //
463 // Contract with Java-level exception handlers:
464 // rax: exception
465 // rdx: throwing pc
466 //
467 // NOTE: At entry of this stub, exception-pc must be on stack !!
468
469 address StubGenerator::generate_forward_exception() {
470 StubGenStubId stub_id = StubGenStubId::forward_exception_id;
471 StubCodeMark mark(this, stub_id);
472 address start = __ pc();
473
474 // Upon entry, the sp points to the return address returning into
475 // Java (interpreted or compiled) code; i.e., the return address
476 // becomes the throwing pc.
477 //
478 // Arguments pushed before the runtime call are still on the stack
479 // but the exception handler will reset the stack pointer ->
480 // ignore them. A potential result in registers can be ignored as
481 // well.
482
483 #ifdef ASSERT
484 // make sure this code is only executed if there is a pending exception
485 {
486 Label L;
487 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
488 __ jcc(Assembler::notEqual, L);
489 __ stop("StubRoutines::forward exception: no pending exception (1)");
490 __ bind(L);
491 }
492 #endif
493
494 // compute exception handler into rbx
495 __ movptr(c_rarg0, Address(rsp, 0));
496 BLOCK_COMMENT("call exception_handler_for_return_address");
497 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
498 SharedRuntime::exception_handler_for_return_address),
499 r15_thread, c_rarg0);
500 __ mov(rbx, rax);
501
502 // setup rax & rdx, remove return address & clear pending exception
503 __ pop(rdx);
504 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
505 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
506
507 #ifdef ASSERT
508 // make sure exception is set
509 {
510 Label L;
511 __ testptr(rax, rax);
512 __ jcc(Assembler::notEqual, L);
513 __ stop("StubRoutines::forward exception: no pending exception (2)");
514 __ bind(L);
515 }
516 #endif
517
518 // continue at exception handler (return address removed)
519 // rax: exception
520 // rbx: exception handler
521 // rdx: throwing pc
522 __ verify_oop(rax);
523 __ jmp(rbx);
524
525 return start;
526 }
527
528 // Support for intptr_t OrderAccess::fence()
529 //
530 // Arguments :
531 //
532 // Result:
533 address StubGenerator::generate_orderaccess_fence() {
534 StubGenStubId stub_id = StubGenStubId::fence_id;
535 StubCodeMark mark(this, stub_id);
536 address start = __ pc();
537
538 __ membar(Assembler::StoreLoad);
539 __ ret(0);
540
541 return start;
542 }
543
544
545 // Support for intptr_t get_previous_sp()
546 //
547 // This routine is used to find the previous stack pointer for the
548 // caller.
549 address StubGenerator::generate_get_previous_sp() {
550 StubGenStubId stub_id = StubGenStubId::get_previous_sp_id;
551 StubCodeMark mark(this, stub_id);
552 address start = __ pc();
553
554 __ movptr(rax, rsp);
555 __ addptr(rax, 8); // return address is at the top of the stack.
556 __ ret(0);
557
558 return start;
559 }
560
561 //----------------------------------------------------------------------------------------------------
562 // Support for void verify_mxcsr()
563 //
564 // This routine is used with -Xcheck:jni to verify that native
565 // JNI code does not return to Java code without restoring the
566 // MXCSR register to our expected state.
567
568 address StubGenerator::generate_verify_mxcsr() {
569 StubGenStubId stub_id = StubGenStubId::verify_mxcsr_id;
570 StubCodeMark mark(this, stub_id);
571 address start = __ pc();
572
573 const Address mxcsr_save(rsp, 0);
574
575 if (CheckJNICalls) {
576 Label ok_ret;
577 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
578 __ push(rax);
579 __ subptr(rsp, wordSize); // allocate a temp location
580 __ cmp32_mxcsr_std(mxcsr_save, rax, rscratch1);
581 __ jcc(Assembler::equal, ok_ret);
582
583 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
584
585 __ ldmxcsr(mxcsr_std, rscratch1);
586
587 __ bind(ok_ret);
588 __ addptr(rsp, wordSize);
589 __ pop(rax);
590 }
591
592 __ ret(0);
593
594 return start;
595 }
596
597 address StubGenerator::generate_f2i_fixup() {
598 StubGenStubId stub_id = StubGenStubId::f2i_fixup_id;
599 StubCodeMark mark(this, stub_id);
600 Address inout(rsp, 5 * wordSize); // return address + 4 saves
601
602 address start = __ pc();
603
604 Label L;
605
606 __ push(rax);
607 __ push(c_rarg3);
608 __ push(c_rarg2);
609 __ push(c_rarg1);
610
611 __ movl(rax, 0x7f800000);
612 __ xorl(c_rarg3, c_rarg3);
613 __ movl(c_rarg2, inout);
614 __ movl(c_rarg1, c_rarg2);
615 __ andl(c_rarg1, 0x7fffffff);
616 __ cmpl(rax, c_rarg1); // NaN? -> 0
617 __ jcc(Assembler::negative, L);
618 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
619 __ movl(c_rarg3, 0x80000000);
620 __ movl(rax, 0x7fffffff);
621 __ cmovl(Assembler::positive, c_rarg3, rax);
622
623 __ bind(L);
624 __ movptr(inout, c_rarg3);
625
626 __ pop(c_rarg1);
627 __ pop(c_rarg2);
628 __ pop(c_rarg3);
629 __ pop(rax);
630
631 __ ret(0);
632
633 return start;
634 }
635
636 address StubGenerator::generate_f2l_fixup() {
637 StubGenStubId stub_id = StubGenStubId::f2l_fixup_id;
638 StubCodeMark mark(this, stub_id);
639 Address inout(rsp, 5 * wordSize); // return address + 4 saves
640 address start = __ pc();
641
642 Label L;
643
644 __ push(rax);
645 __ push(c_rarg3);
646 __ push(c_rarg2);
647 __ push(c_rarg1);
648
649 __ movl(rax, 0x7f800000);
650 __ xorl(c_rarg3, c_rarg3);
651 __ movl(c_rarg2, inout);
652 __ movl(c_rarg1, c_rarg2);
653 __ andl(c_rarg1, 0x7fffffff);
654 __ cmpl(rax, c_rarg1); // NaN? -> 0
655 __ jcc(Assembler::negative, L);
656 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
657 __ mov64(c_rarg3, 0x8000000000000000);
658 __ mov64(rax, 0x7fffffffffffffff);
659 __ cmov(Assembler::positive, c_rarg3, rax);
660
661 __ bind(L);
662 __ movptr(inout, c_rarg3);
663
664 __ pop(c_rarg1);
665 __ pop(c_rarg2);
666 __ pop(c_rarg3);
667 __ pop(rax);
668
669 __ ret(0);
670
671 return start;
672 }
673
674 address StubGenerator::generate_d2i_fixup() {
675 StubGenStubId stub_id = StubGenStubId::d2i_fixup_id;
676 StubCodeMark mark(this, stub_id);
677 Address inout(rsp, 6 * wordSize); // return address + 5 saves
678
679 address start = __ pc();
680
681 Label L;
682
683 __ push(rax);
684 __ push(c_rarg3);
685 __ push(c_rarg2);
686 __ push(c_rarg1);
687 __ push(c_rarg0);
688
689 __ movl(rax, 0x7ff00000);
690 __ movq(c_rarg2, inout);
691 __ movl(c_rarg3, c_rarg2);
692 __ mov(c_rarg1, c_rarg2);
693 __ mov(c_rarg0, c_rarg2);
694 __ negl(c_rarg3);
695 __ shrptr(c_rarg1, 0x20);
696 __ orl(c_rarg3, c_rarg2);
697 __ andl(c_rarg1, 0x7fffffff);
698 __ xorl(c_rarg2, c_rarg2);
699 __ shrl(c_rarg3, 0x1f);
700 __ orl(c_rarg1, c_rarg3);
701 __ cmpl(rax, c_rarg1);
702 __ jcc(Assembler::negative, L); // NaN -> 0
703 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
704 __ movl(c_rarg2, 0x80000000);
705 __ movl(rax, 0x7fffffff);
706 __ cmov(Assembler::positive, c_rarg2, rax);
707
708 __ bind(L);
709 __ movptr(inout, c_rarg2);
710
711 __ pop(c_rarg0);
712 __ pop(c_rarg1);
713 __ pop(c_rarg2);
714 __ pop(c_rarg3);
715 __ pop(rax);
716
717 __ ret(0);
718
719 return start;
720 }
721
722 address StubGenerator::generate_d2l_fixup() {
723 StubGenStubId stub_id = StubGenStubId::d2l_fixup_id;
724 StubCodeMark mark(this, stub_id);
725 Address inout(rsp, 6 * wordSize); // return address + 5 saves
726
727 address start = __ pc();
728
729 Label L;
730
731 __ push(rax);
732 __ push(c_rarg3);
733 __ push(c_rarg2);
734 __ push(c_rarg1);
735 __ push(c_rarg0);
736
737 __ movl(rax, 0x7ff00000);
738 __ movq(c_rarg2, inout);
739 __ movl(c_rarg3, c_rarg2);
740 __ mov(c_rarg1, c_rarg2);
741 __ mov(c_rarg0, c_rarg2);
742 __ negl(c_rarg3);
743 __ shrptr(c_rarg1, 0x20);
744 __ orl(c_rarg3, c_rarg2);
745 __ andl(c_rarg1, 0x7fffffff);
746 __ xorl(c_rarg2, c_rarg2);
747 __ shrl(c_rarg3, 0x1f);
748 __ orl(c_rarg1, c_rarg3);
749 __ cmpl(rax, c_rarg1);
750 __ jcc(Assembler::negative, L); // NaN -> 0
751 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
752 __ mov64(c_rarg2, 0x8000000000000000);
753 __ mov64(rax, 0x7fffffffffffffff);
754 __ cmovq(Assembler::positive, c_rarg2, rax);
755
756 __ bind(L);
757 __ movq(inout, c_rarg2);
758
759 __ pop(c_rarg0);
760 __ pop(c_rarg1);
761 __ pop(c_rarg2);
762 __ pop(c_rarg3);
763 __ pop(rax);
764
765 __ ret(0);
766
767 return start;
768 }
769
770 address StubGenerator::generate_count_leading_zeros_lut() {
771 __ align64();
772 StubGenStubId stub_id = StubGenStubId::vector_count_leading_zeros_lut_id;
773 StubCodeMark mark(this, stub_id);
774 address start = __ pc();
775
776 __ emit_data64(0x0101010102020304, relocInfo::none);
777 __ emit_data64(0x0000000000000000, relocInfo::none);
778 __ emit_data64(0x0101010102020304, relocInfo::none);
779 __ emit_data64(0x0000000000000000, relocInfo::none);
780 __ emit_data64(0x0101010102020304, relocInfo::none);
781 __ emit_data64(0x0000000000000000, relocInfo::none);
782 __ emit_data64(0x0101010102020304, relocInfo::none);
783 __ emit_data64(0x0000000000000000, relocInfo::none);
784
785 return start;
786 }
787
788 address StubGenerator::generate_popcount_avx_lut() {
789 __ align64();
790 StubGenStubId stub_id = StubGenStubId::vector_popcount_lut_id;
791 StubCodeMark mark(this, stub_id);
792 address start = __ pc();
793
794 __ emit_data64(0x0302020102010100, relocInfo::none);
795 __ emit_data64(0x0403030203020201, relocInfo::none);
796 __ emit_data64(0x0302020102010100, relocInfo::none);
797 __ emit_data64(0x0403030203020201, relocInfo::none);
798 __ emit_data64(0x0302020102010100, relocInfo::none);
799 __ emit_data64(0x0403030203020201, relocInfo::none);
800 __ emit_data64(0x0302020102010100, relocInfo::none);
801 __ emit_data64(0x0403030203020201, relocInfo::none);
802
803 return start;
804 }
805
806 address StubGenerator::generate_iota_indices() {
807 __ align(CodeEntryAlignment);
808 StubGenStubId stub_id = StubGenStubId::vector_iota_indices_id;
809 StubCodeMark mark(this, stub_id);
810 address start = __ pc();
811 // B
812 __ emit_data64(0x0706050403020100, relocInfo::none);
813 __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
814 __ emit_data64(0x1716151413121110, relocInfo::none);
815 __ emit_data64(0x1F1E1D1C1B1A1918, relocInfo::none);
816 __ emit_data64(0x2726252423222120, relocInfo::none);
817 __ emit_data64(0x2F2E2D2C2B2A2928, relocInfo::none);
818 __ emit_data64(0x3736353433323130, relocInfo::none);
819 __ emit_data64(0x3F3E3D3C3B3A3938, relocInfo::none);
820 // W
821 __ emit_data64(0x0003000200010000, relocInfo::none);
822 __ emit_data64(0x0007000600050004, relocInfo::none);
823 __ emit_data64(0x000B000A00090008, relocInfo::none);
824 __ emit_data64(0x000F000E000D000C, relocInfo::none);
825 __ emit_data64(0x0013001200110010, relocInfo::none);
826 __ emit_data64(0x0017001600150014, relocInfo::none);
827 __ emit_data64(0x001B001A00190018, relocInfo::none);
828 __ emit_data64(0x001F001E001D001C, relocInfo::none);
829 // D
830 __ emit_data64(0x0000000100000000, relocInfo::none);
831 __ emit_data64(0x0000000300000002, relocInfo::none);
832 __ emit_data64(0x0000000500000004, relocInfo::none);
833 __ emit_data64(0x0000000700000006, relocInfo::none);
834 __ emit_data64(0x0000000900000008, relocInfo::none);
835 __ emit_data64(0x0000000B0000000A, relocInfo::none);
836 __ emit_data64(0x0000000D0000000C, relocInfo::none);
837 __ emit_data64(0x0000000F0000000E, relocInfo::none);
838 // Q
839 __ emit_data64(0x0000000000000000, relocInfo::none);
840 __ emit_data64(0x0000000000000001, relocInfo::none);
841 __ emit_data64(0x0000000000000002, relocInfo::none);
842 __ emit_data64(0x0000000000000003, relocInfo::none);
843 __ emit_data64(0x0000000000000004, relocInfo::none);
844 __ emit_data64(0x0000000000000005, relocInfo::none);
845 __ emit_data64(0x0000000000000006, relocInfo::none);
846 __ emit_data64(0x0000000000000007, relocInfo::none);
847 // D - FP
848 __ emit_data64(0x3F80000000000000, relocInfo::none); // 0.0f, 1.0f
849 __ emit_data64(0x4040000040000000, relocInfo::none); // 2.0f, 3.0f
850 __ emit_data64(0x40A0000040800000, relocInfo::none); // 4.0f, 5.0f
851 __ emit_data64(0x40E0000040C00000, relocInfo::none); // 6.0f, 7.0f
852 __ emit_data64(0x4110000041000000, relocInfo::none); // 8.0f, 9.0f
853 __ emit_data64(0x4130000041200000, relocInfo::none); // 10.0f, 11.0f
854 __ emit_data64(0x4150000041400000, relocInfo::none); // 12.0f, 13.0f
855 __ emit_data64(0x4170000041600000, relocInfo::none); // 14.0f, 15.0f
856 // Q - FP
857 __ emit_data64(0x0000000000000000, relocInfo::none); // 0.0d
858 __ emit_data64(0x3FF0000000000000, relocInfo::none); // 1.0d
859 __ emit_data64(0x4000000000000000, relocInfo::none); // 2.0d
860 __ emit_data64(0x4008000000000000, relocInfo::none); // 3.0d
861 __ emit_data64(0x4010000000000000, relocInfo::none); // 4.0d
862 __ emit_data64(0x4014000000000000, relocInfo::none); // 5.0d
863 __ emit_data64(0x4018000000000000, relocInfo::none); // 6.0d
864 __ emit_data64(0x401c000000000000, relocInfo::none); // 7.0d
865 return start;
866 }
867
868 address StubGenerator::generate_vector_reverse_bit_lut() {
869 __ align(CodeEntryAlignment);
870 StubGenStubId stub_id = StubGenStubId::vector_reverse_bit_lut_id;
871 StubCodeMark mark(this, stub_id);
872 address start = __ pc();
873
874 __ emit_data64(0x0E060A020C040800, relocInfo::none);
875 __ emit_data64(0x0F070B030D050901, relocInfo::none);
876 __ emit_data64(0x0E060A020C040800, relocInfo::none);
877 __ emit_data64(0x0F070B030D050901, relocInfo::none);
878 __ emit_data64(0x0E060A020C040800, relocInfo::none);
879 __ emit_data64(0x0F070B030D050901, relocInfo::none);
880 __ emit_data64(0x0E060A020C040800, relocInfo::none);
881 __ emit_data64(0x0F070B030D050901, relocInfo::none);
882
883 return start;
884 }
885
886 address StubGenerator::generate_vector_reverse_byte_perm_mask_long() {
887 __ align(CodeEntryAlignment);
888 StubGenStubId stub_id = StubGenStubId::vector_reverse_byte_perm_mask_long_id;
889 StubCodeMark mark(this, stub_id);
890 address start = __ pc();
891
892 __ emit_data64(0x0001020304050607, relocInfo::none);
893 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
894 __ emit_data64(0x0001020304050607, relocInfo::none);
895 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
896 __ emit_data64(0x0001020304050607, relocInfo::none);
897 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
898 __ emit_data64(0x0001020304050607, relocInfo::none);
899 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
900
901 return start;
902 }
903
904 address StubGenerator::generate_vector_reverse_byte_perm_mask_int() {
905 __ align(CodeEntryAlignment);
906 StubGenStubId stub_id = StubGenStubId::vector_reverse_byte_perm_mask_int_id;
907 StubCodeMark mark(this, stub_id);
908 address start = __ pc();
909
910 __ emit_data64(0x0405060700010203, relocInfo::none);
911 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
912 __ emit_data64(0x0405060700010203, relocInfo::none);
913 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
914 __ emit_data64(0x0405060700010203, relocInfo::none);
915 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
916 __ emit_data64(0x0405060700010203, relocInfo::none);
917 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
918
919 return start;
920 }
921
922 address StubGenerator::generate_vector_reverse_byte_perm_mask_short() {
923 __ align(CodeEntryAlignment);
924 StubGenStubId stub_id = StubGenStubId::vector_reverse_byte_perm_mask_short_id;
925 StubCodeMark mark(this, stub_id);
926 address start = __ pc();
927
928 __ emit_data64(0x0607040502030001, relocInfo::none);
929 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
930 __ emit_data64(0x0607040502030001, relocInfo::none);
931 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
932 __ emit_data64(0x0607040502030001, relocInfo::none);
933 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
934 __ emit_data64(0x0607040502030001, relocInfo::none);
935 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
936
937 return start;
938 }
939
940 address StubGenerator::generate_vector_byte_shuffle_mask() {
941 __ align(CodeEntryAlignment);
942 StubGenStubId stub_id = StubGenStubId::vector_byte_shuffle_mask_id;
943 StubCodeMark mark(this, stub_id);
944 address start = __ pc();
945
946 __ emit_data64(0x7070707070707070, relocInfo::none);
947 __ emit_data64(0x7070707070707070, relocInfo::none);
948 __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
949 __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
950
951 return start;
952 }
953
954 address StubGenerator::generate_fp_mask(StubGenStubId stub_id, int64_t mask) {
955 __ align(CodeEntryAlignment);
956 StubCodeMark mark(this, stub_id);
957 address start = __ pc();
958
959 __ emit_data64( mask, relocInfo::none );
960 __ emit_data64( mask, relocInfo::none );
961
962 return start;
963 }
964
965 address StubGenerator::generate_compress_perm_table(StubGenStubId stub_id) {
966 int esize;
967 switch (stub_id) {
968 case compress_perm_table32_id:
969 esize = 32;
970 break;
971 case compress_perm_table64_id:
972 esize = 64;
973 break;
974 default:
975 ShouldNotReachHere();
976 }
977 __ align(CodeEntryAlignment);
978 StubCodeMark mark(this, stub_id);
979 address start = __ pc();
980 if (esize == 32) {
981 // Loop to generate 256 x 8 int compression permute index table. A row is
982 // accessed using 8 bit index computed using vector mask. An entry in
983 // a row holds either a valid permute index corresponding to set bit position
984 // or a -1 (default) value.
985 for (int mask = 0; mask < 256; mask++) {
986 int ctr = 0;
987 for (int j = 0; j < 8; j++) {
988 if (mask & (1 << j)) {
989 __ emit_data(j, relocInfo::none);
990 ctr++;
991 }
992 }
993 for (; ctr < 8; ctr++) {
994 __ emit_data(-1, relocInfo::none);
995 }
996 }
997 } else {
998 assert(esize == 64, "");
999 // Loop to generate 16 x 4 long compression permute index table. A row is
1000 // accessed using 4 bit index computed using vector mask. An entry in
1001 // a row holds either a valid permute index pair for a quadword corresponding
1002 // to set bit position or a -1 (default) value.
1003 for (int mask = 0; mask < 16; mask++) {
1004 int ctr = 0;
1005 for (int j = 0; j < 4; j++) {
1006 if (mask & (1 << j)) {
1007 __ emit_data(2 * j, relocInfo::none);
1008 __ emit_data(2 * j + 1, relocInfo::none);
1009 ctr++;
1010 }
1011 }
1012 for (; ctr < 4; ctr++) {
1013 __ emit_data64(-1L, relocInfo::none);
1014 }
1015 }
1016 }
1017 return start;
1018 }
1019
1020 address StubGenerator::generate_expand_perm_table(StubGenStubId stub_id) {
1021 int esize;
1022 switch (stub_id) {
1023 case expand_perm_table32_id:
1024 esize = 32;
1025 break;
1026 case expand_perm_table64_id:
1027 esize = 64;
1028 break;
1029 default:
1030 ShouldNotReachHere();
1031 }
1032 __ align(CodeEntryAlignment);
1033 StubCodeMark mark(this, stub_id);
1034 address start = __ pc();
1035 if (esize == 32) {
1036 // Loop to generate 256 x 8 int expand permute index table. A row is accessed
1037 // using 8 bit index computed using vector mask. An entry in a row holds either
1038 // a valid permute index (starting from least significant lane) placed at poisition
1039 // corresponding to set bit position or a -1 (default) value.
1040 for (int mask = 0; mask < 256; mask++) {
1041 int ctr = 0;
1042 for (int j = 0; j < 8; j++) {
1043 if (mask & (1 << j)) {
1044 __ emit_data(ctr++, relocInfo::none);
1045 } else {
1046 __ emit_data(-1, relocInfo::none);
1047 }
1048 }
1049 }
1050 } else {
1051 assert(esize == 64, "");
1052 // Loop to generate 16 x 4 long expand permute index table. A row is accessed
1053 // using 4 bit index computed using vector mask. An entry in a row holds either
1054 // a valid doubleword permute index pair representing a quadword index (starting
1055 // from least significant lane) placed at poisition corresponding to set bit
1056 // position or a -1 (default) value.
1057 for (int mask = 0; mask < 16; mask++) {
1058 int ctr = 0;
1059 for (int j = 0; j < 4; j++) {
1060 if (mask & (1 << j)) {
1061 __ emit_data(2 * ctr, relocInfo::none);
1062 __ emit_data(2 * ctr + 1, relocInfo::none);
1063 ctr++;
1064 } else {
1065 __ emit_data64(-1L, relocInfo::none);
1066 }
1067 }
1068 }
1069 }
1070 return start;
1071 }
1072
1073 address StubGenerator::generate_vector_mask(StubGenStubId stub_id, int64_t mask) {
1074 __ align(CodeEntryAlignment);
1075 StubCodeMark mark(this, stub_id);
1076 address start = __ pc();
1077
1078 __ emit_data64(mask, relocInfo::none);
1079 __ emit_data64(mask, relocInfo::none);
1080 __ emit_data64(mask, relocInfo::none);
1081 __ emit_data64(mask, relocInfo::none);
1082 __ emit_data64(mask, relocInfo::none);
1083 __ emit_data64(mask, relocInfo::none);
1084 __ emit_data64(mask, relocInfo::none);
1085 __ emit_data64(mask, relocInfo::none);
1086
1087 return start;
1088 }
1089
1090 address StubGenerator::generate_vector_byte_perm_mask() {
1091 __ align(CodeEntryAlignment);
1092 StubGenStubId stub_id = StubGenStubId::vector_byte_perm_mask_id;
1093 StubCodeMark mark(this, stub_id);
1094 address start = __ pc();
1095
1096 __ emit_data64(0x0000000000000001, relocInfo::none);
1097 __ emit_data64(0x0000000000000003, relocInfo::none);
1098 __ emit_data64(0x0000000000000005, relocInfo::none);
1099 __ emit_data64(0x0000000000000007, relocInfo::none);
1100 __ emit_data64(0x0000000000000000, relocInfo::none);
1101 __ emit_data64(0x0000000000000002, relocInfo::none);
1102 __ emit_data64(0x0000000000000004, relocInfo::none);
1103 __ emit_data64(0x0000000000000006, relocInfo::none);
1104
1105 return start;
1106 }
1107
1108 address StubGenerator::generate_vector_fp_mask(StubGenStubId stub_id, int64_t mask) {
1109 __ align(CodeEntryAlignment);
1110 StubCodeMark mark(this, stub_id);
1111 address start = __ pc();
1112
1113 __ emit_data64(mask, relocInfo::none);
1114 __ emit_data64(mask, relocInfo::none);
1115 __ emit_data64(mask, relocInfo::none);
1116 __ emit_data64(mask, relocInfo::none);
1117 __ emit_data64(mask, relocInfo::none);
1118 __ emit_data64(mask, relocInfo::none);
1119 __ emit_data64(mask, relocInfo::none);
1120 __ emit_data64(mask, relocInfo::none);
1121
1122 return start;
1123 }
1124
1125 address StubGenerator::generate_vector_custom_i32(StubGenStubId stub_id, Assembler::AvxVectorLen len,
1126 int32_t val0, int32_t val1, int32_t val2, int32_t val3,
1127 int32_t val4, int32_t val5, int32_t val6, int32_t val7,
1128 int32_t val8, int32_t val9, int32_t val10, int32_t val11,
1129 int32_t val12, int32_t val13, int32_t val14, int32_t val15) {
1130 __ align(CodeEntryAlignment);
1131 StubCodeMark mark(this, stub_id);
1132 address start = __ pc();
1133
1134 assert(len != Assembler::AVX_NoVec, "vector len must be specified");
1135 __ emit_data(val0, relocInfo::none, 0);
1136 __ emit_data(val1, relocInfo::none, 0);
1137 __ emit_data(val2, relocInfo::none, 0);
1138 __ emit_data(val3, relocInfo::none, 0);
1139 if (len >= Assembler::AVX_256bit) {
1140 __ emit_data(val4, relocInfo::none, 0);
1141 __ emit_data(val5, relocInfo::none, 0);
1142 __ emit_data(val6, relocInfo::none, 0);
1143 __ emit_data(val7, relocInfo::none, 0);
1144 if (len >= Assembler::AVX_512bit) {
1145 __ emit_data(val8, relocInfo::none, 0);
1146 __ emit_data(val9, relocInfo::none, 0);
1147 __ emit_data(val10, relocInfo::none, 0);
1148 __ emit_data(val11, relocInfo::none, 0);
1149 __ emit_data(val12, relocInfo::none, 0);
1150 __ emit_data(val13, relocInfo::none, 0);
1151 __ emit_data(val14, relocInfo::none, 0);
1152 __ emit_data(val15, relocInfo::none, 0);
1153 }
1154 }
1155 return start;
1156 }
1157
1158 // Non-destructive plausibility checks for oops
1159 //
1160 // Arguments:
1161 // all args on stack!
1162 //
1163 // Stack after saving c_rarg3:
1164 // [tos + 0]: saved c_rarg3
1165 // [tos + 1]: saved c_rarg2
1166 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1167 // [tos + 3]: saved flags
1168 // [tos + 4]: return address
1169 // * [tos + 5]: error message (char*)
1170 // * [tos + 6]: object to verify (oop)
1171 // * [tos + 7]: saved rax - saved by caller and bashed
1172 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1173 // * = popped on exit
1174 address StubGenerator::generate_verify_oop() {
1175 StubGenStubId stub_id = StubGenStubId::verify_oop_id;
1176 StubCodeMark mark(this, stub_id);
1177 address start = __ pc();
1178
1179 Label exit, error;
1180
1181 __ pushf();
1182 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()), rscratch1);
1183
1184 __ push(r12);
1185
1186 // save c_rarg2 and c_rarg3
1187 __ push(c_rarg2);
1188 __ push(c_rarg3);
1189
1190 enum {
1191 // After previous pushes.
1192 oop_to_verify = 6 * wordSize,
1193 saved_rax = 7 * wordSize,
1194 saved_r10 = 8 * wordSize,
1195
1196 // Before the call to MacroAssembler::debug(), see below.
1197 return_addr = 16 * wordSize,
1198 error_msg = 17 * wordSize
1199 };
1200
1201 // get object
1202 __ movptr(rax, Address(rsp, oop_to_verify));
1203
1204 // make sure object is 'reasonable'
1205 __ testptr(rax, rax);
1206 __ jcc(Assembler::zero, exit); // if obj is null it is OK
1207
1208 BarrierSetAssembler* bs_asm = BarrierSet::barrier_set()->barrier_set_assembler();
1209 bs_asm->check_oop(_masm, rax, c_rarg2, c_rarg3, error);
1210
1211 // return if everything seems ok
1212 __ bind(exit);
1213 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1214 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1215 __ pop(c_rarg3); // restore c_rarg3
1216 __ pop(c_rarg2); // restore c_rarg2
1217 __ pop(r12); // restore r12
1218 __ popf(); // restore flags
1219 __ ret(4 * wordSize); // pop caller saved stuff
1220
1221 // handle errors
1222 __ bind(error);
1223 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1224 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1225 __ pop(c_rarg3); // get saved c_rarg3 back
1226 __ pop(c_rarg2); // get saved c_rarg2 back
1227 __ pop(r12); // get saved r12 back
1228 __ popf(); // get saved flags off stack --
1229 // will be ignored
1230
1231 __ pusha(); // push registers
1232 // (rip is already
1233 // already pushed)
1234 // debug(char* msg, int64_t pc, int64_t regs[])
1235 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1236 // pushed all the registers, so now the stack looks like:
1237 // [tos + 0] 16 saved registers
1238 // [tos + 16] return address
1239 // * [tos + 17] error message (char*)
1240 // * [tos + 18] object to verify (oop)
1241 // * [tos + 19] saved rax - saved by caller and bashed
1242 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1243 // * = popped on exit
1244
1245 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1246 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1247 __ movq(c_rarg2, rsp); // pass address of regs on stack
1248 __ mov(r12, rsp); // remember rsp
1249 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1250 __ andptr(rsp, -16); // align stack as required by ABI
1251 BLOCK_COMMENT("call MacroAssembler::debug");
1252 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1253 __ hlt();
1254
1255 return start;
1256 }
1257
1258
1259 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1260 //
1261 // Outputs:
1262 // rdi - rcx
1263 // rsi - rdx
1264 // rdx - r8
1265 // rcx - r9
1266 //
1267 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1268 // are non-volatile. r9 and r10 should not be used by the caller.
1269 //
1270 void StubGenerator::setup_arg_regs(int nargs) {
1271 const Register saved_rdi = r9;
1272 const Register saved_rsi = r10;
1273 assert(nargs == 3 || nargs == 4, "else fix");
1274 #ifdef _WIN64
1275 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1276 "unexpected argument registers");
1277 if (nargs == 4) {
1278 __ mov(rax, r9); // r9 is also saved_rdi
1279 }
1280 __ movptr(saved_rdi, rdi);
1281 __ movptr(saved_rsi, rsi);
1282 __ mov(rdi, rcx); // c_rarg0
1283 __ mov(rsi, rdx); // c_rarg1
1284 __ mov(rdx, r8); // c_rarg2
1285 if (nargs == 4) {
1286 __ mov(rcx, rax); // c_rarg3 (via rax)
1287 }
1288 #else
1289 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1290 "unexpected argument registers");
1291 #endif
1292 DEBUG_ONLY(_regs_in_thread = false;)
1293 }
1294
1295
1296 void StubGenerator::restore_arg_regs() {
1297 assert(!_regs_in_thread, "wrong call to restore_arg_regs");
1298 const Register saved_rdi = r9;
1299 const Register saved_rsi = r10;
1300 #ifdef _WIN64
1301 __ movptr(rdi, saved_rdi);
1302 __ movptr(rsi, saved_rsi);
1303 #endif
1304 }
1305
1306
1307 // This is used in places where r10 is a scratch register, and can
1308 // be adapted if r9 is needed also.
1309 void StubGenerator::setup_arg_regs_using_thread(int nargs) {
1310 const Register saved_r15 = r9;
1311 assert(nargs == 3 || nargs == 4, "else fix");
1312 #ifdef _WIN64
1313 if (nargs == 4) {
1314 __ mov(rax, r9); // r9 is also saved_r15
1315 }
1316 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1317 __ get_thread(r15_thread);
1318 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1319 "unexpected argument registers");
1320 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1321 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1322
1323 __ mov(rdi, rcx); // c_rarg0
1324 __ mov(rsi, rdx); // c_rarg1
1325 __ mov(rdx, r8); // c_rarg2
1326 if (nargs == 4) {
1327 __ mov(rcx, rax); // c_rarg3 (via rax)
1328 }
1329 #else
1330 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1331 "unexpected argument registers");
1332 #endif
1333 DEBUG_ONLY(_regs_in_thread = true;)
1334 }
1335
1336
1337 void StubGenerator::restore_arg_regs_using_thread() {
1338 assert(_regs_in_thread, "wrong call to restore_arg_regs");
1339 const Register saved_r15 = r9;
1340 #ifdef _WIN64
1341 __ get_thread(r15_thread);
1342 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1343 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1344 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1345 #endif
1346 }
1347
1348
1349 void StubGenerator::setup_argument_regs(BasicType type) {
1350 if (type == T_BYTE || type == T_SHORT) {
1351 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1352 // r9 and r10 may be used to save non-volatile registers
1353 } else {
1354 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1355 // r9 is used to save r15_thread
1356 }
1357 }
1358
1359
1360 void StubGenerator::restore_argument_regs(BasicType type) {
1361 if (type == T_BYTE || type == T_SHORT) {
1362 restore_arg_regs();
1363 } else {
1364 restore_arg_regs_using_thread();
1365 }
1366 }
1367
1368 address StubGenerator::generate_data_cache_writeback() {
1369 const Register src = c_rarg0; // source address
1370
1371 __ align(CodeEntryAlignment);
1372
1373 StubGenStubId stub_id = StubGenStubId::data_cache_writeback_id;
1374 StubCodeMark mark(this, stub_id);
1375
1376 address start = __ pc();
1377
1378 __ enter();
1379 __ cache_wb(Address(src, 0));
1380 __ leave();
1381 __ ret(0);
1382
1383 return start;
1384 }
1385
1386 address StubGenerator::generate_data_cache_writeback_sync() {
1387 const Register is_pre = c_rarg0; // pre or post sync
1388
1389 __ align(CodeEntryAlignment);
1390
1391 StubGenStubId stub_id = StubGenStubId::data_cache_writeback_sync_id;
1392 StubCodeMark mark(this, stub_id);
1393
1394 // pre wbsync is a no-op
1395 // post wbsync translates to an sfence
1396
1397 Label skip;
1398 address start = __ pc();
1399
1400 __ enter();
1401 __ cmpl(is_pre, 0);
1402 __ jcc(Assembler::notEqual, skip);
1403 __ cache_wbsync(false);
1404 __ bind(skip);
1405 __ leave();
1406 __ ret(0);
1407
1408 return start;
1409 }
1410
1411 // ofs and limit are use for multi-block byte array.
1412 // int com.sun.security.provider.MD5.implCompress(byte[] b, int ofs)
1413 address StubGenerator::generate_md5_implCompress(StubGenStubId stub_id) {
1414 bool multi_block;
1415 switch (stub_id) {
1416 case md5_implCompress_id:
1417 multi_block = false;
1418 break;
1419 case md5_implCompressMB_id:
1420 multi_block = true;
1421 break;
1422 default:
1423 ShouldNotReachHere();
1424 }
1425 __ align(CodeEntryAlignment);
1426 StubCodeMark mark(this, stub_id);
1427 address start = __ pc();
1428
1429 const Register buf_param = r15;
1430 const Address state_param(rsp, 0 * wordSize);
1431 const Address ofs_param (rsp, 1 * wordSize );
1432 const Address limit_param(rsp, 1 * wordSize + 4);
1433
1434 __ enter();
1435 __ push(rbx);
1436 __ push(rdi);
1437 __ push(rsi);
1438 __ push(r15);
1439 __ subptr(rsp, 2 * wordSize);
1440
1441 __ movptr(buf_param, c_rarg0);
1442 __ movptr(state_param, c_rarg1);
1443 if (multi_block) {
1444 __ movl(ofs_param, c_rarg2);
1445 __ movl(limit_param, c_rarg3);
1446 }
1447 __ fast_md5(buf_param, state_param, ofs_param, limit_param, multi_block);
1448
1449 __ addptr(rsp, 2 * wordSize);
1450 __ pop(r15);
1451 __ pop(rsi);
1452 __ pop(rdi);
1453 __ pop(rbx);
1454 __ leave();
1455 __ ret(0);
1456
1457 return start;
1458 }
1459
1460 address StubGenerator::generate_upper_word_mask() {
1461 __ align64();
1462 StubGenStubId stub_id = StubGenStubId::upper_word_mask_id;
1463 StubCodeMark mark(this, stub_id);
1464 address start = __ pc();
1465
1466 __ emit_data64(0x0000000000000000, relocInfo::none);
1467 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
1468
1469 return start;
1470 }
1471
1472 address StubGenerator::generate_shuffle_byte_flip_mask() {
1473 __ align64();
1474 StubGenStubId stub_id = StubGenStubId::shuffle_byte_flip_mask_id;
1475 StubCodeMark mark(this, stub_id);
1476 address start = __ pc();
1477
1478 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1479 __ emit_data64(0x0001020304050607, relocInfo::none);
1480
1481 return start;
1482 }
1483
1484 // ofs and limit are use for multi-block byte array.
1485 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1486 address StubGenerator::generate_sha1_implCompress(StubGenStubId stub_id) {
1487 bool multi_block;
1488 switch (stub_id) {
1489 case sha1_implCompress_id:
1490 multi_block = false;
1491 break;
1492 case sha1_implCompressMB_id:
1493 multi_block = true;
1494 break;
1495 default:
1496 ShouldNotReachHere();
1497 }
1498 __ align(CodeEntryAlignment);
1499 StubCodeMark mark(this, stub_id);
1500 address start = __ pc();
1501
1502 Register buf = c_rarg0;
1503 Register state = c_rarg1;
1504 Register ofs = c_rarg2;
1505 Register limit = c_rarg3;
1506
1507 const XMMRegister abcd = xmm0;
1508 const XMMRegister e0 = xmm1;
1509 const XMMRegister e1 = xmm2;
1510 const XMMRegister msg0 = xmm3;
1511
1512 const XMMRegister msg1 = xmm4;
1513 const XMMRegister msg2 = xmm5;
1514 const XMMRegister msg3 = xmm6;
1515 const XMMRegister shuf_mask = xmm7;
1516
1517 __ enter();
1518
1519 __ subptr(rsp, 4 * wordSize);
1520
1521 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
1522 buf, state, ofs, limit, rsp, multi_block);
1523
1524 __ addptr(rsp, 4 * wordSize);
1525
1526 __ leave();
1527 __ ret(0);
1528
1529 return start;
1530 }
1531
1532 address StubGenerator::generate_pshuffle_byte_flip_mask() {
1533 __ align64();
1534 StubGenStubId stub_id = StubGenStubId::pshuffle_byte_flip_mask_id;
1535 StubCodeMark mark(this, stub_id);
1536 address start = __ pc();
1537
1538 __ emit_data64(0x0405060700010203, relocInfo::none);
1539 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1540
1541 if (VM_Version::supports_avx2()) {
1542 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
1543 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1544 // _SHUF_00BA
1545 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1546 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1547 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1548 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1549 // _SHUF_DC00
1550 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1551 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1552 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1553 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1554 }
1555
1556 return start;
1557 }
1558
1559 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
1560 address StubGenerator::generate_pshuffle_byte_flip_mask_sha512() {
1561 __ align32();
1562 StubGenStubId stub_id = StubGenStubId::pshuffle_byte_flip_mask_sha512_id;
1563 StubCodeMark mark(this, stub_id);
1564 address start = __ pc();
1565
1566 if (VM_Version::supports_avx2()) {
1567 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
1568 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1569 __ emit_data64(0x1011121314151617, relocInfo::none);
1570 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
1571 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
1572 __ emit_data64(0x0000000000000000, relocInfo::none);
1573 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1574 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1575 }
1576
1577 return start;
1578 }
1579
1580 // ofs and limit are use for multi-block byte array.
1581 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1582 address StubGenerator::generate_sha256_implCompress(StubGenStubId stub_id) {
1583 bool multi_block;
1584 switch (stub_id) {
1585 case sha256_implCompress_id:
1586 multi_block = false;
1587 break;
1588 case sha256_implCompressMB_id:
1589 multi_block = true;
1590 break;
1591 default:
1592 ShouldNotReachHere();
1593 }
1594 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
1595 __ align(CodeEntryAlignment);
1596 StubCodeMark mark(this, stub_id);
1597 address start = __ pc();
1598
1599 Register buf = c_rarg0;
1600 Register state = c_rarg1;
1601 Register ofs = c_rarg2;
1602 Register limit = c_rarg3;
1603
1604 const XMMRegister msg = xmm0;
1605 const XMMRegister state0 = xmm1;
1606 const XMMRegister state1 = xmm2;
1607 const XMMRegister msgtmp0 = xmm3;
1608
1609 const XMMRegister msgtmp1 = xmm4;
1610 const XMMRegister msgtmp2 = xmm5;
1611 const XMMRegister msgtmp3 = xmm6;
1612 const XMMRegister msgtmp4 = xmm7;
1613
1614 const XMMRegister shuf_mask = xmm8;
1615
1616 __ enter();
1617
1618 __ subptr(rsp, 4 * wordSize);
1619
1620 if (VM_Version::supports_sha()) {
1621 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1622 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1623 } else if (VM_Version::supports_avx2()) {
1624 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1625 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1626 }
1627 __ addptr(rsp, 4 * wordSize);
1628 __ vzeroupper();
1629 __ leave();
1630 __ ret(0);
1631
1632 return start;
1633 }
1634
1635 address StubGenerator::generate_sha512_implCompress(StubGenStubId stub_id) {
1636 bool multi_block;
1637 switch (stub_id) {
1638 case sha512_implCompress_id:
1639 multi_block = false;
1640 break;
1641 case sha512_implCompressMB_id:
1642 multi_block = true;
1643 break;
1644 default:
1645 ShouldNotReachHere();
1646 }
1647 assert(VM_Version::supports_avx2(), "");
1648 assert(VM_Version::supports_bmi2() || VM_Version::supports_sha512(), "");
1649 __ align(CodeEntryAlignment);
1650 StubCodeMark mark(this, stub_id);
1651 address start = __ pc();
1652
1653 Register buf = c_rarg0;
1654 Register state = c_rarg1;
1655 Register ofs = c_rarg2;
1656 Register limit = c_rarg3;
1657
1658 __ enter();
1659
1660 if (VM_Version::supports_sha512()) {
1661 __ sha512_update_ni_x1(state, buf, ofs, limit, multi_block);
1662 } else {
1663 const XMMRegister msg = xmm0;
1664 const XMMRegister state0 = xmm1;
1665 const XMMRegister state1 = xmm2;
1666 const XMMRegister msgtmp0 = xmm3;
1667 const XMMRegister msgtmp1 = xmm4;
1668 const XMMRegister msgtmp2 = xmm5;
1669 const XMMRegister msgtmp3 = xmm6;
1670 const XMMRegister msgtmp4 = xmm7;
1671
1672 const XMMRegister shuf_mask = xmm8;
1673 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1674 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1675 }
1676 __ vzeroupper();
1677 __ leave();
1678 __ ret(0);
1679
1680 return start;
1681 }
1682
1683 address StubGenerator::base64_shuffle_addr() {
1684 __ align64();
1685 StubGenStubId stub_id = StubGenStubId::shuffle_base64_id;
1686 StubCodeMark mark(this, stub_id);
1687 address start = __ pc();
1688
1689 assert(((unsigned long long)start & 0x3f) == 0,
1690 "Alignment problem (0x%08llx)", (unsigned long long)start);
1691 __ emit_data64(0x0405030401020001, relocInfo::none);
1692 __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1693 __ emit_data64(0x10110f100d0e0c0d, relocInfo::none);
1694 __ emit_data64(0x1617151613141213, relocInfo::none);
1695 __ emit_data64(0x1c1d1b1c191a1819, relocInfo::none);
1696 __ emit_data64(0x222321221f201e1f, relocInfo::none);
1697 __ emit_data64(0x2829272825262425, relocInfo::none);
1698 __ emit_data64(0x2e2f2d2e2b2c2a2b, relocInfo::none);
1699
1700 return start;
1701 }
1702
1703 address StubGenerator::base64_avx2_shuffle_addr() {
1704 __ align32();
1705 StubGenStubId stub_id = StubGenStubId::avx2_shuffle_base64_id;
1706 StubCodeMark mark(this, stub_id);
1707 address start = __ pc();
1708
1709 __ emit_data64(0x0809070805060405, relocInfo::none);
1710 __ emit_data64(0x0e0f0d0e0b0c0a0b, relocInfo::none);
1711 __ emit_data64(0x0405030401020001, relocInfo::none);
1712 __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1713
1714 return start;
1715 }
1716
1717 address StubGenerator::base64_avx2_input_mask_addr() {
1718 __ align32();
1719 StubGenStubId stub_id = StubGenStubId::avx2_input_mask_base64_id;
1720 StubCodeMark mark(this, stub_id);
1721 address start = __ pc();
1722
1723 __ emit_data64(0x8000000000000000, relocInfo::none);
1724 __ emit_data64(0x8000000080000000, relocInfo::none);
1725 __ emit_data64(0x8000000080000000, relocInfo::none);
1726 __ emit_data64(0x8000000080000000, relocInfo::none);
1727
1728 return start;
1729 }
1730
1731 address StubGenerator::base64_avx2_lut_addr() {
1732 __ align32();
1733 StubGenStubId stub_id = StubGenStubId::avx2_lut_base64_id;
1734 StubCodeMark mark(this, stub_id);
1735 address start = __ pc();
1736
1737 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1738 __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1739 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1740 __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1741
1742 // URL LUT
1743 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1744 __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1745 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1746 __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1747
1748 return start;
1749 }
1750
1751 address StubGenerator::base64_encoding_table_addr() {
1752 __ align64();
1753 StubGenStubId stub_id = StubGenStubId::encoding_table_base64_id;
1754 StubCodeMark mark(this, stub_id);
1755 address start = __ pc();
1756
1757 assert(((unsigned long long)start & 0x3f) == 0, "Alignment problem (0x%08llx)", (unsigned long long)start);
1758 __ emit_data64(0x4847464544434241, relocInfo::none);
1759 __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1760 __ emit_data64(0x5857565554535251, relocInfo::none);
1761 __ emit_data64(0x6665646362615a59, relocInfo::none);
1762 __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1763 __ emit_data64(0x767574737271706f, relocInfo::none);
1764 __ emit_data64(0x333231307a797877, relocInfo::none);
1765 __ emit_data64(0x2f2b393837363534, relocInfo::none);
1766
1767 // URL table
1768 __ emit_data64(0x4847464544434241, relocInfo::none);
1769 __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1770 __ emit_data64(0x5857565554535251, relocInfo::none);
1771 __ emit_data64(0x6665646362615a59, relocInfo::none);
1772 __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1773 __ emit_data64(0x767574737271706f, relocInfo::none);
1774 __ emit_data64(0x333231307a797877, relocInfo::none);
1775 __ emit_data64(0x5f2d393837363534, relocInfo::none);
1776
1777 return start;
1778 }
1779
1780 // Code for generating Base64 encoding.
1781 // Intrinsic function prototype in Base64.java:
1782 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp,
1783 // boolean isURL) {
1784 address StubGenerator::generate_base64_encodeBlock()
1785 {
1786 __ align(CodeEntryAlignment);
1787 StubGenStubId stub_id = StubGenStubId::base64_encodeBlock_id;
1788 StubCodeMark mark(this, stub_id);
1789 address start = __ pc();
1790
1791 __ enter();
1792
1793 // Save callee-saved registers before using them
1794 __ push(r12);
1795 __ push(r13);
1796 __ push(r14);
1797 __ push(r15);
1798
1799 // arguments
1800 const Register source = c_rarg0; // Source Array
1801 const Register start_offset = c_rarg1; // start offset
1802 const Register end_offset = c_rarg2; // end offset
1803 const Register dest = c_rarg3; // destination array
1804
1805 #ifndef _WIN64
1806 const Register dp = c_rarg4; // Position for writing to dest array
1807 const Register isURL = c_rarg5; // Base64 or URL character set
1808 #else
1809 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
1810 const Address isURL_mem(rbp, 7 * wordSize);
1811 const Register isURL = r10; // pick the volatile windows register
1812 const Register dp = r12;
1813 __ movl(dp, dp_mem);
1814 __ movl(isURL, isURL_mem);
1815 #endif
1816
1817 const Register length = r14;
1818 const Register encode_table = r13;
1819 Label L_process3, L_exit, L_processdata, L_vbmiLoop, L_not512, L_32byteLoop;
1820
1821 // calculate length from offsets
1822 __ movl(length, end_offset);
1823 __ subl(length, start_offset);
1824 __ jcc(Assembler::lessEqual, L_exit);
1825
1826 // Code for 512-bit VBMI encoding. Encodes 48 input bytes into 64
1827 // output bytes. We read 64 input bytes and ignore the last 16, so be
1828 // sure not to read past the end of the input buffer.
1829 if (VM_Version::supports_avx512_vbmi()) {
1830 __ cmpl(length, 64); // Do not overrun input buffer.
1831 __ jcc(Assembler::below, L_not512);
1832
1833 __ shll(isURL, 6); // index into decode table based on isURL
1834 __ lea(encode_table, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
1835 __ addptr(encode_table, isURL);
1836 __ shrl(isURL, 6); // restore isURL
1837
1838 __ mov64(rax, 0x3036242a1016040aull); // Shifts
1839 __ evmovdquq(xmm3, ExternalAddress(StubRoutines::x86::base64_shuffle_addr()), Assembler::AVX_512bit, r15);
1840 __ evmovdquq(xmm2, Address(encode_table, 0), Assembler::AVX_512bit);
1841 __ evpbroadcastq(xmm1, rax, Assembler::AVX_512bit);
1842
1843 __ align32();
1844 __ BIND(L_vbmiLoop);
1845
1846 __ vpermb(xmm0, xmm3, Address(source, start_offset), Assembler::AVX_512bit);
1847 __ subl(length, 48);
1848
1849 // Put the input bytes into the proper lanes for writing, then
1850 // encode them.
1851 __ evpmultishiftqb(xmm0, xmm1, xmm0, Assembler::AVX_512bit);
1852 __ vpermb(xmm0, xmm0, xmm2, Assembler::AVX_512bit);
1853
1854 // Write to destination
1855 __ evmovdquq(Address(dest, dp), xmm0, Assembler::AVX_512bit);
1856
1857 __ addptr(dest, 64);
1858 __ addptr(source, 48);
1859 __ cmpl(length, 64);
1860 __ jcc(Assembler::aboveEqual, L_vbmiLoop);
1861
1862 __ vzeroupper();
1863 }
1864
1865 __ BIND(L_not512);
1866 if (VM_Version::supports_avx2()) {
1867 /*
1868 ** This AVX2 encoder is based off the paper at:
1869 ** https://dl.acm.org/doi/10.1145/3132709
1870 **
1871 ** We use AVX2 SIMD instructions to encode 24 bytes into 32
1872 ** output bytes.
1873 **
1874 */
1875 // Lengths under 32 bytes are done with scalar routine
1876 __ cmpl(length, 31);
1877 __ jcc(Assembler::belowEqual, L_process3);
1878
1879 // Set up supporting constant table data
1880 __ vmovdqu(xmm9, ExternalAddress(StubRoutines::x86::base64_avx2_shuffle_addr()), rax);
1881 // 6-bit mask for 2nd and 4th (and multiples) 6-bit values
1882 __ movl(rax, 0x0fc0fc00);
1883 __ movdl(xmm8, rax);
1884 __ vmovdqu(xmm1, ExternalAddress(StubRoutines::x86::base64_avx2_input_mask_addr()), rax);
1885 __ vpbroadcastd(xmm8, xmm8, Assembler::AVX_256bit);
1886
1887 // Multiplication constant for "shifting" right by 6 and 10
1888 // bits
1889 __ movl(rax, 0x04000040);
1890
1891 __ subl(length, 24);
1892 __ movdl(xmm7, rax);
1893 __ vpbroadcastd(xmm7, xmm7, Assembler::AVX_256bit);
1894
1895 // For the first load, we mask off reading of the first 4
1896 // bytes into the register. This is so we can get 4 3-byte
1897 // chunks into each lane of the register, avoiding having to
1898 // handle end conditions. We then shuffle these bytes into a
1899 // specific order so that manipulation is easier.
1900 //
1901 // The initial read loads the XMM register like this:
1902 //
1903 // Lower 128-bit lane:
1904 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1905 // | XX | XX | XX | XX | A0 | A1 | A2 | B0 | B1 | B2 | C0 | C1
1906 // | C2 | D0 | D1 | D2 |
1907 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1908 //
1909 // Upper 128-bit lane:
1910 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1911 // | E0 | E1 | E2 | F0 | F1 | F2 | G0 | G1 | G2 | H0 | H1 | H2
1912 // | XX | XX | XX | XX |
1913 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1914 //
1915 // Where A0 is the first input byte, B0 is the fourth, etc.
1916 // The alphabetical significance denotes the 3 bytes to be
1917 // consumed and encoded into 4 bytes.
1918 //
1919 // We then shuffle the register so each 32-bit word contains
1920 // the sequence:
1921 // A1 A0 A2 A1, B1, B0, B2, B1, etc.
1922 // Each of these byte sequences are then manipulated into 4
1923 // 6-bit values ready for encoding.
1924 //
1925 // If we focus on one set of 3-byte chunks, changing the
1926 // nomenclature such that A0 => a, A1 => b, and A2 => c, we
1927 // shuffle such that each 24-bit chunk contains:
1928 //
1929 // b7 b6 b5 b4 b3 b2 b1 b0 | a7 a6 a5 a4 a3 a2 a1 a0 | c7 c6
1930 // c5 c4 c3 c2 c1 c0 | b7 b6 b5 b4 b3 b2 b1 b0
1931 // Explain this step.
1932 // b3 b2 b1 b0 c5 c4 c3 c2 | c1 c0 d5 d4 d3 d2 d1 d0 | a5 a4
1933 // a3 a2 a1 a0 b5 b4 | b3 b2 b1 b0 c5 c4 c3 c2
1934 //
1935 // W first and off all but bits 4-9 and 16-21 (c5..c0 and
1936 // a5..a0) and shift them using a vector multiplication
1937 // operation (vpmulhuw) which effectively shifts c right by 6
1938 // bits and a right by 10 bits. We similarly mask bits 10-15
1939 // (d5..d0) and 22-27 (b5..b0) and shift them left by 8 and 4
1940 // bits respectively. This is done using vpmullw. We end up
1941 // with 4 6-bit values, thus splitting the 3 input bytes,
1942 // ready for encoding:
1943 // 0 0 d5..d0 0 0 c5..c0 0 0 b5..b0 0 0 a5..a0
1944 //
1945 // For translation, we recognize that there are 5 distinct
1946 // ranges of legal Base64 characters as below:
1947 //
1948 // +-------------+-------------+------------+
1949 // | 6-bit value | ASCII range | offset |
1950 // +-------------+-------------+------------+
1951 // | 0..25 | A..Z | 65 |
1952 // | 26..51 | a..z | 71 |
1953 // | 52..61 | 0..9 | -4 |
1954 // | 62 | + or - | -19 or -17 |
1955 // | 63 | / or _ | -16 or 32 |
1956 // +-------------+-------------+------------+
1957 //
1958 // We note that vpshufb does a parallel lookup in a
1959 // destination register using the lower 4 bits of bytes from a
1960 // source register. If we use a saturated subtraction and
1961 // subtract 51 from each 6-bit value, bytes from [0,51]
1962 // saturate to 0, and [52,63] map to a range of [1,12]. We
1963 // distinguish the [0,25] and [26,51] ranges by assigning a
1964 // value of 13 for all 6-bit values less than 26. We end up
1965 // with:
1966 //
1967 // +-------------+-------------+------------+
1968 // | 6-bit value | Reduced | offset |
1969 // +-------------+-------------+------------+
1970 // | 0..25 | 13 | 65 |
1971 // | 26..51 | 0 | 71 |
1972 // | 52..61 | 0..9 | -4 |
1973 // | 62 | 11 | -19 or -17 |
1974 // | 63 | 12 | -16 or 32 |
1975 // +-------------+-------------+------------+
1976 //
1977 // We then use a final vpshufb to add the appropriate offset,
1978 // translating the bytes.
1979 //
1980 // Load input bytes - only 28 bytes. Mask the first load to
1981 // not load into the full register.
1982 __ vpmaskmovd(xmm1, xmm1, Address(source, start_offset, Address::times_1, -4), Assembler::AVX_256bit);
1983
1984 // Move 3-byte chunks of input (12 bytes) into 16 bytes,
1985 // ordering by:
1986 // 1, 0, 2, 1; 4, 3, 5, 4; etc. This groups 6-bit chunks
1987 // for easy masking
1988 __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1989
1990 __ addl(start_offset, 24);
1991
1992 // Load masking register for first and third (and multiples)
1993 // 6-bit values.
1994 __ movl(rax, 0x003f03f0);
1995 __ movdl(xmm6, rax);
1996 __ vpbroadcastd(xmm6, xmm6, Assembler::AVX_256bit);
1997 // Multiplication constant for "shifting" left by 4 and 8 bits
1998 __ movl(rax, 0x01000010);
1999 __ movdl(xmm5, rax);
2000 __ vpbroadcastd(xmm5, xmm5, Assembler::AVX_256bit);
2001
2002 // Isolate 6-bit chunks of interest
2003 __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
2004
2005 // Load constants for encoding
2006 __ movl(rax, 0x19191919);
2007 __ movdl(xmm3, rax);
2008 __ vpbroadcastd(xmm3, xmm3, Assembler::AVX_256bit);
2009 __ movl(rax, 0x33333333);
2010 __ movdl(xmm4, rax);
2011 __ vpbroadcastd(xmm4, xmm4, Assembler::AVX_256bit);
2012
2013 // Shift output bytes 0 and 2 into proper lanes
2014 __ vpmulhuw(xmm2, xmm0, xmm7, Assembler::AVX_256bit);
2015
2016 // Mask and shift output bytes 1 and 3 into proper lanes and
2017 // combine
2018 __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
2019 __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
2020 __ vpor(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
2021
2022 // Find out which are 0..25. This indicates which input
2023 // values fall in the range of 'A'-'Z', which require an
2024 // additional offset (see comments above)
2025 __ vpcmpgtb(xmm2, xmm0, xmm3, Assembler::AVX_256bit);
2026 __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
2027 __ vpsubb(xmm1, xmm1, xmm2, Assembler::AVX_256bit);
2028
2029 // Load the proper lookup table
2030 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_avx2_lut_addr()));
2031 __ movl(r15, isURL);
2032 __ shll(r15, 5);
2033 __ vmovdqu(xmm2, Address(r11, r15));
2034
2035 // Shuffle the offsets based on the range calculation done
2036 // above. This allows us to add the correct offset to the
2037 // 6-bit value corresponding to the range documented above.
2038 __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
2039 __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
2040
2041 // Store the encoded bytes
2042 __ vmovdqu(Address(dest, dp), xmm0);
2043 __ addl(dp, 32);
2044
2045 __ cmpl(length, 31);
2046 __ jcc(Assembler::belowEqual, L_process3);
2047
2048 __ align32();
2049 __ BIND(L_32byteLoop);
2050
2051 // Get next 32 bytes
2052 __ vmovdqu(xmm1, Address(source, start_offset, Address::times_1, -4));
2053
2054 __ subl(length, 24);
2055 __ addl(start_offset, 24);
2056
2057 // This logic is identical to the above, with only constant
2058 // register loads removed. Shuffle the input, mask off 6-bit
2059 // chunks, shift them into place, then add the offset to
2060 // encode.
2061 __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
2062
2063 __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
2064 __ vpmulhuw(xmm10, xmm0, xmm7, Assembler::AVX_256bit);
2065 __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
2066 __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
2067 __ vpor(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
2068 __ vpcmpgtb(xmm10, xmm0, xmm3, Assembler::AVX_256bit);
2069 __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
2070 __ vpsubb(xmm1, xmm1, xmm10, Assembler::AVX_256bit);
2071 __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
2072 __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
2073
2074 // Store the encoded bytes
2075 __ vmovdqu(Address(dest, dp), xmm0);
2076 __ addl(dp, 32);
2077
2078 __ cmpl(length, 31);
2079 __ jcc(Assembler::above, L_32byteLoop);
2080
2081 __ BIND(L_process3);
2082 __ vzeroupper();
2083 } else {
2084 __ BIND(L_process3);
2085 }
2086
2087 __ cmpl(length, 3);
2088 __ jcc(Assembler::below, L_exit);
2089
2090 // Load the encoding table based on isURL
2091 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
2092 __ movl(r15, isURL);
2093 __ shll(r15, 6);
2094 __ addptr(r11, r15);
2095
2096 __ BIND(L_processdata);
2097
2098 // Load 3 bytes
2099 __ load_unsigned_byte(r15, Address(source, start_offset));
2100 __ load_unsigned_byte(r10, Address(source, start_offset, Address::times_1, 1));
2101 __ load_unsigned_byte(r13, Address(source, start_offset, Address::times_1, 2));
2102
2103 // Build a 32-bit word with bytes 1, 2, 0, 1
2104 __ movl(rax, r10);
2105 __ shll(r10, 24);
2106 __ orl(rax, r10);
2107
2108 __ subl(length, 3);
2109
2110 __ shll(r15, 8);
2111 __ shll(r13, 16);
2112 __ orl(rax, r15);
2113
2114 __ addl(start_offset, 3);
2115
2116 __ orl(rax, r13);
2117 // At this point, rax contains | byte1 | byte2 | byte0 | byte1
2118 // r13 has byte2 << 16 - need low-order 6 bits to translate.
2119 // This translated byte is the fourth output byte.
2120 __ shrl(r13, 16);
2121 __ andl(r13, 0x3f);
2122
2123 // The high-order 6 bits of r15 (byte0) is translated.
2124 // The translated byte is the first output byte.
2125 __ shrl(r15, 10);
2126
2127 __ load_unsigned_byte(r13, Address(r11, r13));
2128 __ load_unsigned_byte(r15, Address(r11, r15));
2129
2130 __ movb(Address(dest, dp, Address::times_1, 3), r13);
2131
2132 // Extract high-order 4 bits of byte1 and low-order 2 bits of byte0.
2133 // This translated byte is the second output byte.
2134 __ shrl(rax, 4);
2135 __ movl(r10, rax);
2136 __ andl(rax, 0x3f);
2137
2138 __ movb(Address(dest, dp, Address::times_1, 0), r15);
2139
2140 __ load_unsigned_byte(rax, Address(r11, rax));
2141
2142 // Extract low-order 2 bits of byte1 and high-order 4 bits of byte2.
2143 // This translated byte is the third output byte.
2144 __ shrl(r10, 18);
2145 __ andl(r10, 0x3f);
2146
2147 __ load_unsigned_byte(r10, Address(r11, r10));
2148
2149 __ movb(Address(dest, dp, Address::times_1, 1), rax);
2150 __ movb(Address(dest, dp, Address::times_1, 2), r10);
2151
2152 __ addl(dp, 4);
2153 __ cmpl(length, 3);
2154 __ jcc(Assembler::aboveEqual, L_processdata);
2155
2156 __ BIND(L_exit);
2157 __ pop(r15);
2158 __ pop(r14);
2159 __ pop(r13);
2160 __ pop(r12);
2161 __ leave();
2162 __ ret(0);
2163
2164 return start;
2165 }
2166
2167 // base64 AVX512vbmi tables
2168 address StubGenerator::base64_vbmi_lookup_lo_addr() {
2169 __ align64();
2170 StubGenStubId stub_id = StubGenStubId::lookup_lo_base64_id;
2171 StubCodeMark mark(this, stub_id);
2172 address start = __ pc();
2173
2174 assert(((unsigned long long)start & 0x3f) == 0,
2175 "Alignment problem (0x%08llx)", (unsigned long long)start);
2176 __ emit_data64(0x8080808080808080, relocInfo::none);
2177 __ emit_data64(0x8080808080808080, relocInfo::none);
2178 __ emit_data64(0x8080808080808080, relocInfo::none);
2179 __ emit_data64(0x8080808080808080, relocInfo::none);
2180 __ emit_data64(0x8080808080808080, relocInfo::none);
2181 __ emit_data64(0x3f8080803e808080, relocInfo::none);
2182 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2183 __ emit_data64(0x8080808080803d3c, relocInfo::none);
2184
2185 return start;
2186 }
2187
2188 address StubGenerator::base64_vbmi_lookup_hi_addr() {
2189 __ align64();
2190 StubGenStubId stub_id = StubGenStubId::lookup_hi_base64_id;
2191 StubCodeMark mark(this, stub_id);
2192 address start = __ pc();
2193
2194 assert(((unsigned long long)start & 0x3f) == 0,
2195 "Alignment problem (0x%08llx)", (unsigned long long)start);
2196 __ emit_data64(0x0605040302010080, relocInfo::none);
2197 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2198 __ emit_data64(0x161514131211100f, relocInfo::none);
2199 __ emit_data64(0x8080808080191817, relocInfo::none);
2200 __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2201 __ emit_data64(0x2827262524232221, relocInfo::none);
2202 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2203 __ emit_data64(0x8080808080333231, relocInfo::none);
2204
2205 return start;
2206 }
2207 address StubGenerator::base64_vbmi_lookup_lo_url_addr() {
2208 __ align64();
2209 StubGenStubId stub_id = StubGenStubId::lookup_lo_base64url_id;
2210 StubCodeMark mark(this, stub_id);
2211 address start = __ pc();
2212
2213 assert(((unsigned long long)start & 0x3f) == 0,
2214 "Alignment problem (0x%08llx)", (unsigned long long)start);
2215 __ emit_data64(0x8080808080808080, relocInfo::none);
2216 __ emit_data64(0x8080808080808080, relocInfo::none);
2217 __ emit_data64(0x8080808080808080, relocInfo::none);
2218 __ emit_data64(0x8080808080808080, relocInfo::none);
2219 __ emit_data64(0x8080808080808080, relocInfo::none);
2220 __ emit_data64(0x80803e8080808080, relocInfo::none);
2221 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2222 __ emit_data64(0x8080808080803d3c, relocInfo::none);
2223
2224 return start;
2225 }
2226
2227 address StubGenerator::base64_vbmi_lookup_hi_url_addr() {
2228 __ align64();
2229 StubGenStubId stub_id = StubGenStubId::lookup_hi_base64url_id;
2230 StubCodeMark mark(this, stub_id);
2231 address start = __ pc();
2232
2233 assert(((unsigned long long)start & 0x3f) == 0,
2234 "Alignment problem (0x%08llx)", (unsigned long long)start);
2235 __ emit_data64(0x0605040302010080, relocInfo::none);
2236 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2237 __ emit_data64(0x161514131211100f, relocInfo::none);
2238 __ emit_data64(0x3f80808080191817, relocInfo::none);
2239 __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2240 __ emit_data64(0x2827262524232221, relocInfo::none);
2241 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2242 __ emit_data64(0x8080808080333231, relocInfo::none);
2243
2244 return start;
2245 }
2246
2247 address StubGenerator::base64_vbmi_pack_vec_addr() {
2248 __ align64();
2249 StubGenStubId stub_id = StubGenStubId::pack_vec_base64_id;
2250 StubCodeMark mark(this, stub_id);
2251 address start = __ pc();
2252
2253 assert(((unsigned long long)start & 0x3f) == 0,
2254 "Alignment problem (0x%08llx)", (unsigned long long)start);
2255 __ emit_data64(0x090a040506000102, relocInfo::none);
2256 __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2257 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2258 __ emit_data64(0x292a242526202122, relocInfo::none);
2259 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2260 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2261 __ emit_data64(0x0000000000000000, relocInfo::none);
2262 __ emit_data64(0x0000000000000000, relocInfo::none);
2263
2264 return start;
2265 }
2266
2267 address StubGenerator::base64_vbmi_join_0_1_addr() {
2268 __ align64();
2269 StubGenStubId stub_id = StubGenStubId::join_0_1_base64_id;
2270 StubCodeMark mark(this, stub_id);
2271 address start = __ pc();
2272
2273 assert(((unsigned long long)start & 0x3f) == 0,
2274 "Alignment problem (0x%08llx)", (unsigned long long)start);
2275 __ emit_data64(0x090a040506000102, relocInfo::none);
2276 __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2277 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2278 __ emit_data64(0x292a242526202122, relocInfo::none);
2279 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2280 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2281 __ emit_data64(0x494a444546404142, relocInfo::none);
2282 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2283
2284 return start;
2285 }
2286
2287 address StubGenerator::base64_vbmi_join_1_2_addr() {
2288 __ align64();
2289 StubGenStubId stub_id = StubGenStubId::join_1_2_base64_id;
2290 StubCodeMark mark(this, stub_id);
2291 address start = __ pc();
2292
2293 assert(((unsigned long long)start & 0x3f) == 0,
2294 "Alignment problem (0x%08llx)", (unsigned long long)start);
2295 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2296 __ emit_data64(0x292a242526202122, relocInfo::none);
2297 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2298 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2299 __ emit_data64(0x494a444546404142, relocInfo::none);
2300 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2301 __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2302 __ emit_data64(0x696a646566606162, relocInfo::none);
2303
2304 return start;
2305 }
2306
2307 address StubGenerator::base64_vbmi_join_2_3_addr() {
2308 __ align64();
2309 StubGenStubId stub_id = StubGenStubId::join_2_3_base64_id;
2310 StubCodeMark mark(this, stub_id);
2311 address start = __ pc();
2312
2313 assert(((unsigned long long)start & 0x3f) == 0,
2314 "Alignment problem (0x%08llx)", (unsigned long long)start);
2315 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2316 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2317 __ emit_data64(0x494a444546404142, relocInfo::none);
2318 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2319 __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2320 __ emit_data64(0x696a646566606162, relocInfo::none);
2321 __ emit_data64(0x767071726c6d6e68, relocInfo::none);
2322 __ emit_data64(0x7c7d7e78797a7475, relocInfo::none);
2323
2324 return start;
2325 }
2326
2327 address StubGenerator::base64_AVX2_decode_tables_addr() {
2328 __ align64();
2329 StubGenStubId stub_id = StubGenStubId::avx2_decode_tables_base64_id;
2330 StubCodeMark mark(this, stub_id);
2331 address start = __ pc();
2332
2333 assert(((unsigned long long)start & 0x3f) == 0,
2334 "Alignment problem (0x%08llx)", (unsigned long long)start);
2335 __ emit_data(0x2f2f2f2f, relocInfo::none, 0);
2336 __ emit_data(0x5f5f5f5f, relocInfo::none, 0); // for URL
2337
2338 __ emit_data(0xffffffff, relocInfo::none, 0);
2339 __ emit_data(0xfcfcfcfc, relocInfo::none, 0); // for URL
2340
2341 // Permute table
2342 __ emit_data64(0x0000000100000000, relocInfo::none);
2343 __ emit_data64(0x0000000400000002, relocInfo::none);
2344 __ emit_data64(0x0000000600000005, relocInfo::none);
2345 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2346
2347 // Shuffle table
2348 __ emit_data64(0x090a040506000102, relocInfo::none);
2349 __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2350 __ emit_data64(0x090a040506000102, relocInfo::none);
2351 __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2352
2353 // merge table
2354 __ emit_data(0x01400140, relocInfo::none, 0);
2355
2356 // merge multiplier
2357 __ emit_data(0x00011000, relocInfo::none, 0);
2358
2359 return start;
2360 }
2361
2362 address StubGenerator::base64_AVX2_decode_LUT_tables_addr() {
2363 __ align64();
2364 StubGenStubId stub_id = StubGenStubId::avx2_decode_lut_tables_base64_id;
2365 StubCodeMark mark(this, stub_id);
2366 address start = __ pc();
2367
2368 assert(((unsigned long long)start & 0x3f) == 0,
2369 "Alignment problem (0x%08llx)", (unsigned long long)start);
2370 // lut_lo
2371 __ emit_data64(0x1111111111111115, relocInfo::none);
2372 __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2373 __ emit_data64(0x1111111111111115, relocInfo::none);
2374 __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2375
2376 // lut_roll
2377 __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2378 __ emit_data64(0x0000000000000000, relocInfo::none);
2379 __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2380 __ emit_data64(0x0000000000000000, relocInfo::none);
2381
2382 // lut_lo URL
2383 __ emit_data64(0x1111111111111115, relocInfo::none);
2384 __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2385 __ emit_data64(0x1111111111111115, relocInfo::none);
2386 __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2387
2388 // lut_roll URL
2389 __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2390 __ emit_data64(0x0000000000000000, relocInfo::none);
2391 __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2392 __ emit_data64(0x0000000000000000, relocInfo::none);
2393
2394 // lut_hi
2395 __ emit_data64(0x0804080402011010, relocInfo::none);
2396 __ emit_data64(0x1010101010101010, relocInfo::none);
2397 __ emit_data64(0x0804080402011010, relocInfo::none);
2398 __ emit_data64(0x1010101010101010, relocInfo::none);
2399
2400 return start;
2401 }
2402
2403 address StubGenerator::base64_decoding_table_addr() {
2404 StubGenStubId stub_id = StubGenStubId::decoding_table_base64_id;
2405 StubCodeMark mark(this, stub_id);
2406 address start = __ pc();
2407
2408 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2409 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2410 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2411 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2412 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2413 __ emit_data64(0x3fffffff3effffff, relocInfo::none);
2414 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2415 __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2416 __ emit_data64(0x06050403020100ff, relocInfo::none);
2417 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2418 __ emit_data64(0x161514131211100f, relocInfo::none);
2419 __ emit_data64(0xffffffffff191817, relocInfo::none);
2420 __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2421 __ emit_data64(0x2827262524232221, relocInfo::none);
2422 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2423 __ emit_data64(0xffffffffff333231, relocInfo::none);
2424 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2425 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2426 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2427 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2428 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2429 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2430 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2431 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2432 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2433 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2434 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2435 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2436 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2437 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2438 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2439 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2440
2441 // URL table
2442 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2443 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2444 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2445 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2446 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2447 __ emit_data64(0xffff3effffffffff, relocInfo::none);
2448 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2449 __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2450 __ emit_data64(0x06050403020100ff, relocInfo::none);
2451 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2452 __ emit_data64(0x161514131211100f, relocInfo::none);
2453 __ emit_data64(0x3fffffffff191817, relocInfo::none);
2454 __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2455 __ emit_data64(0x2827262524232221, relocInfo::none);
2456 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2457 __ emit_data64(0xffffffffff333231, relocInfo::none);
2458 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2459 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2460 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2461 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2462 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2463 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2464 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2465 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2466 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2467 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2468 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2469 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2470 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2471 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2472 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2473 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2474
2475 return start;
2476 }
2477
2478
2479 // Code for generating Base64 decoding.
2480 //
2481 // Based on the article (and associated code) from https://arxiv.org/abs/1910.05109.
2482 //
2483 // Intrinsic function prototype in Base64.java:
2484 // private void decodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL, isMIME) {
2485 address StubGenerator::generate_base64_decodeBlock() {
2486 __ align(CodeEntryAlignment);
2487 StubGenStubId stub_id = StubGenStubId::base64_decodeBlock_id;
2488 StubCodeMark mark(this, stub_id);
2489 address start = __ pc();
2490
2491 __ enter();
2492
2493 // Save callee-saved registers before using them
2494 __ push(r12);
2495 __ push(r13);
2496 __ push(r14);
2497 __ push(r15);
2498 __ push(rbx);
2499
2500 // arguments
2501 const Register source = c_rarg0; // Source Array
2502 const Register start_offset = c_rarg1; // start offset
2503 const Register end_offset = c_rarg2; // end offset
2504 const Register dest = c_rarg3; // destination array
2505 const Register isMIME = rbx;
2506
2507 #ifndef _WIN64
2508 const Register dp = c_rarg4; // Position for writing to dest array
2509 const Register isURL = c_rarg5;// Base64 or URL character set
2510 __ movl(isMIME, Address(rbp, 2 * wordSize));
2511 #else
2512 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
2513 const Address isURL_mem(rbp, 7 * wordSize);
2514 const Register isURL = r10; // pick the volatile windows register
2515 const Register dp = r12;
2516 __ movl(dp, dp_mem);
2517 __ movl(isURL, isURL_mem);
2518 __ movl(isMIME, Address(rbp, 8 * wordSize));
2519 #endif
2520
2521 const XMMRegister lookup_lo = xmm5;
2522 const XMMRegister lookup_hi = xmm6;
2523 const XMMRegister errorvec = xmm7;
2524 const XMMRegister pack16_op = xmm9;
2525 const XMMRegister pack32_op = xmm8;
2526 const XMMRegister input0 = xmm3;
2527 const XMMRegister input1 = xmm20;
2528 const XMMRegister input2 = xmm21;
2529 const XMMRegister input3 = xmm19;
2530 const XMMRegister join01 = xmm12;
2531 const XMMRegister join12 = xmm11;
2532 const XMMRegister join23 = xmm10;
2533 const XMMRegister translated0 = xmm2;
2534 const XMMRegister translated1 = xmm1;
2535 const XMMRegister translated2 = xmm0;
2536 const XMMRegister translated3 = xmm4;
2537
2538 const XMMRegister merged0 = xmm2;
2539 const XMMRegister merged1 = xmm1;
2540 const XMMRegister merged2 = xmm0;
2541 const XMMRegister merged3 = xmm4;
2542 const XMMRegister merge_ab_bc0 = xmm2;
2543 const XMMRegister merge_ab_bc1 = xmm1;
2544 const XMMRegister merge_ab_bc2 = xmm0;
2545 const XMMRegister merge_ab_bc3 = xmm4;
2546
2547 const XMMRegister pack24bits = xmm4;
2548
2549 const Register length = r14;
2550 const Register output_size = r13;
2551 const Register output_mask = r15;
2552 const KRegister input_mask = k1;
2553
2554 const XMMRegister input_initial_valid_b64 = xmm0;
2555 const XMMRegister tmp = xmm10;
2556 const XMMRegister mask = xmm0;
2557 const XMMRegister invalid_b64 = xmm1;
2558
2559 Label L_process256, L_process64, L_process64Loop, L_exit, L_processdata, L_loadURL;
2560 Label L_continue, L_finalBit, L_padding, L_donePadding, L_bruteForce;
2561 Label L_forceLoop, L_bottomLoop, L_checkMIME, L_exit_no_vzero, L_lastChunk;
2562
2563 // calculate length from offsets
2564 __ movl(length, end_offset);
2565 __ subl(length, start_offset);
2566 __ push(dest); // Save for return value calc
2567
2568 // If AVX512 VBMI not supported, just compile non-AVX code
2569 if(VM_Version::supports_avx512_vbmi() &&
2570 VM_Version::supports_avx512bw()) {
2571 __ cmpl(length, 31); // 32-bytes is break-even for AVX-512
2572 __ jcc(Assembler::lessEqual, L_lastChunk);
2573
2574 __ cmpl(isMIME, 0);
2575 __ jcc(Assembler::notEqual, L_lastChunk);
2576
2577 // Load lookup tables based on isURL
2578 __ cmpl(isURL, 0);
2579 __ jcc(Assembler::notZero, L_loadURL);
2580
2581 __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_addr()), Assembler::AVX_512bit, r13);
2582 __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_addr()), Assembler::AVX_512bit, r13);
2583
2584 __ BIND(L_continue);
2585
2586 __ movl(r15, 0x01400140);
2587 __ evpbroadcastd(pack16_op, r15, Assembler::AVX_512bit);
2588
2589 __ movl(r15, 0x00011000);
2590 __ evpbroadcastd(pack32_op, r15, Assembler::AVX_512bit);
2591
2592 __ cmpl(length, 0xff);
2593 __ jcc(Assembler::lessEqual, L_process64);
2594
2595 // load masks required for decoding data
2596 __ BIND(L_processdata);
2597 __ evmovdquq(join01, ExternalAddress(StubRoutines::x86::base64_vbmi_join_0_1_addr()), Assembler::AVX_512bit,r13);
2598 __ evmovdquq(join12, ExternalAddress(StubRoutines::x86::base64_vbmi_join_1_2_addr()), Assembler::AVX_512bit, r13);
2599 __ evmovdquq(join23, ExternalAddress(StubRoutines::x86::base64_vbmi_join_2_3_addr()), Assembler::AVX_512bit, r13);
2600
2601 __ align32();
2602 __ BIND(L_process256);
2603 // Grab input data
2604 __ evmovdquq(input0, Address(source, start_offset, Address::times_1, 0x00), Assembler::AVX_512bit);
2605 __ evmovdquq(input1, Address(source, start_offset, Address::times_1, 0x40), Assembler::AVX_512bit);
2606 __ evmovdquq(input2, Address(source, start_offset, Address::times_1, 0x80), Assembler::AVX_512bit);
2607 __ evmovdquq(input3, Address(source, start_offset, Address::times_1, 0xc0), Assembler::AVX_512bit);
2608
2609 // Copy the low part of the lookup table into the destination of the permutation
2610 __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2611 __ evmovdquq(translated1, lookup_lo, Assembler::AVX_512bit);
2612 __ evmovdquq(translated2, lookup_lo, Assembler::AVX_512bit);
2613 __ evmovdquq(translated3, lookup_lo, Assembler::AVX_512bit);
2614
2615 // Translate the base64 input into "decoded" bytes
2616 __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2617 __ evpermt2b(translated1, input1, lookup_hi, Assembler::AVX_512bit);
2618 __ evpermt2b(translated2, input2, lookup_hi, Assembler::AVX_512bit);
2619 __ evpermt2b(translated3, input3, lookup_hi, Assembler::AVX_512bit);
2620
2621 // OR all of the translations together to check for errors (high-order bit of byte set)
2622 __ vpternlogd(input0, 0xfe, input1, input2, Assembler::AVX_512bit);
2623
2624 __ vpternlogd(input3, 0xfe, translated0, translated1, Assembler::AVX_512bit);
2625 __ vpternlogd(input0, 0xfe, translated2, translated3, Assembler::AVX_512bit);
2626 __ vpor(errorvec, input3, input0, Assembler::AVX_512bit);
2627
2628 // Check if there was an error - if so, try 64-byte chunks
2629 __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2630 __ kortestql(k3, k3);
2631 __ jcc(Assembler::notZero, L_process64);
2632
2633 // The merging and shuffling happens here
2634 // We multiply each byte pair [00dddddd | 00cccccc | 00bbbbbb | 00aaaaaa]
2635 // Multiply [00cccccc] by 2^6 added to [00dddddd] to get [0000cccc | ccdddddd]
2636 // The pack16_op is a vector of 0x01400140, so multiply D by 1 and C by 0x40
2637 __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2638 __ vpmaddubsw(merge_ab_bc1, translated1, pack16_op, Assembler::AVX_512bit);
2639 __ vpmaddubsw(merge_ab_bc2, translated2, pack16_op, Assembler::AVX_512bit);
2640 __ vpmaddubsw(merge_ab_bc3, translated3, pack16_op, Assembler::AVX_512bit);
2641
2642 // Now do the same with packed 16-bit values.
2643 // We start with [0000cccc | ccdddddd | 0000aaaa | aabbbbbb]
2644 // pack32_op is 0x00011000 (2^12, 1), so this multiplies [0000aaaa | aabbbbbb] by 2^12
2645 // and adds [0000cccc | ccdddddd] to yield [00000000 | aaaaaabb | bbbbcccc | ccdddddd]
2646 __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2647 __ vpmaddwd(merged1, merge_ab_bc1, pack32_op, Assembler::AVX_512bit);
2648 __ vpmaddwd(merged2, merge_ab_bc2, pack32_op, Assembler::AVX_512bit);
2649 __ vpmaddwd(merged3, merge_ab_bc3, pack32_op, Assembler::AVX_512bit);
2650
2651 // The join vectors specify which byte from which vector goes into the outputs
2652 // One of every 4 bytes in the extended vector is zero, so we pack them into their
2653 // final positions in the register for storing (256 bytes in, 192 bytes out)
2654 __ evpermt2b(merged0, join01, merged1, Assembler::AVX_512bit);
2655 __ evpermt2b(merged1, join12, merged2, Assembler::AVX_512bit);
2656 __ evpermt2b(merged2, join23, merged3, Assembler::AVX_512bit);
2657
2658 // Store result
2659 __ evmovdquq(Address(dest, dp, Address::times_1, 0x00), merged0, Assembler::AVX_512bit);
2660 __ evmovdquq(Address(dest, dp, Address::times_1, 0x40), merged1, Assembler::AVX_512bit);
2661 __ evmovdquq(Address(dest, dp, Address::times_1, 0x80), merged2, Assembler::AVX_512bit);
2662
2663 __ addptr(source, 0x100);
2664 __ addptr(dest, 0xc0);
2665 __ subl(length, 0x100);
2666 __ cmpl(length, 64 * 4);
2667 __ jcc(Assembler::greaterEqual, L_process256);
2668
2669 // At this point, we've decoded 64 * 4 * n bytes.
2670 // The remaining length will be <= 64 * 4 - 1.
2671 // UNLESS there was an error decoding the first 256-byte chunk. In this
2672 // case, the length will be arbitrarily long.
2673 //
2674 // Note that this will be the path for MIME-encoded strings.
2675
2676 __ BIND(L_process64);
2677
2678 __ evmovdquq(pack24bits, ExternalAddress(StubRoutines::x86::base64_vbmi_pack_vec_addr()), Assembler::AVX_512bit, r13);
2679
2680 __ cmpl(length, 63);
2681 __ jcc(Assembler::lessEqual, L_finalBit);
2682
2683 __ mov64(rax, 0x0000ffffffffffff);
2684 __ kmovql(k2, rax);
2685
2686 __ align32();
2687 __ BIND(L_process64Loop);
2688
2689 // Handle first 64-byte block
2690
2691 __ evmovdquq(input0, Address(source, start_offset), Assembler::AVX_512bit);
2692 __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2693 __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2694
2695 __ vpor(errorvec, translated0, input0, Assembler::AVX_512bit);
2696
2697 // Check for error and bomb out before updating dest
2698 __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2699 __ kortestql(k3, k3);
2700 __ jcc(Assembler::notZero, L_exit);
2701
2702 // Pack output register, selecting correct byte ordering
2703 __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2704 __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2705 __ vpermb(merged0, pack24bits, merged0, Assembler::AVX_512bit);
2706
2707 __ evmovdqub(Address(dest, dp), k2, merged0, true, Assembler::AVX_512bit);
2708
2709 __ subl(length, 64);
2710 __ addptr(source, 64);
2711 __ addptr(dest, 48);
2712
2713 __ cmpl(length, 64);
2714 __ jcc(Assembler::greaterEqual, L_process64Loop);
2715
2716 __ cmpl(length, 0);
2717 __ jcc(Assembler::lessEqual, L_exit);
2718
2719 __ BIND(L_finalBit);
2720 // Now have 1 to 63 bytes left to decode
2721
2722 // I was going to let Java take care of the final fragment
2723 // however it will repeatedly call this routine for every 4 bytes
2724 // of input data, so handle the rest here.
2725 __ movq(rax, -1);
2726 __ bzhiq(rax, rax, length); // Input mask in rax
2727
2728 __ movl(output_size, length);
2729 __ shrl(output_size, 2); // Find (len / 4) * 3 (output length)
2730 __ lea(output_size, Address(output_size, output_size, Address::times_2, 0));
2731 // output_size in r13
2732
2733 // Strip pad characters, if any, and adjust length and mask
2734 __ addq(length, start_offset);
2735 __ cmpb(Address(source, length, Address::times_1, -1), '=');
2736 __ jcc(Assembler::equal, L_padding);
2737
2738 __ BIND(L_donePadding);
2739 __ subq(length, start_offset);
2740
2741 // Output size is (64 - output_size), output mask is (all 1s >> output_size).
2742 __ kmovql(input_mask, rax);
2743 __ movq(output_mask, -1);
2744 __ bzhiq(output_mask, output_mask, output_size);
2745
2746 // Load initial input with all valid base64 characters. Will be used
2747 // in merging source bytes to avoid masking when determining if an error occurred.
2748 __ movl(rax, 0x61616161);
2749 __ evpbroadcastd(input_initial_valid_b64, rax, Assembler::AVX_512bit);
2750
2751 // A register containing all invalid base64 decoded values
2752 __ movl(rax, 0x80808080);
2753 __ evpbroadcastd(invalid_b64, rax, Assembler::AVX_512bit);
2754
2755 // input_mask is in k1
2756 // output_size is in r13
2757 // output_mask is in r15
2758 // zmm0 - free
2759 // zmm1 - 0x00011000
2760 // zmm2 - 0x01400140
2761 // zmm3 - errorvec
2762 // zmm4 - pack vector
2763 // zmm5 - lookup_lo
2764 // zmm6 - lookup_hi
2765 // zmm7 - errorvec
2766 // zmm8 - 0x61616161
2767 // zmm9 - 0x80808080
2768
2769 // Load only the bytes from source, merging into our "fully-valid" register
2770 __ evmovdqub(input_initial_valid_b64, input_mask, Address(source, start_offset, Address::times_1, 0x0), true, Assembler::AVX_512bit);
2771
2772 // Decode all bytes within our merged input
2773 __ evmovdquq(tmp, lookup_lo, Assembler::AVX_512bit);
2774 __ evpermt2b(tmp, input_initial_valid_b64, lookup_hi, Assembler::AVX_512bit);
2775 __ evporq(mask, tmp, input_initial_valid_b64, Assembler::AVX_512bit);
2776
2777 // Check for error. Compare (decoded | initial) to all invalid.
2778 // If any bytes have their high-order bit set, then we have an error.
2779 __ evptestmb(k2, mask, invalid_b64, Assembler::AVX_512bit);
2780 __ kortestql(k2, k2);
2781
2782 // If we have an error, use the brute force loop to decode what we can (4-byte chunks).
2783 __ jcc(Assembler::notZero, L_bruteForce);
2784
2785 // Shuffle output bytes
2786 __ vpmaddubsw(tmp, tmp, pack16_op, Assembler::AVX_512bit);
2787 __ vpmaddwd(tmp, tmp, pack32_op, Assembler::AVX_512bit);
2788
2789 __ vpermb(tmp, pack24bits, tmp, Assembler::AVX_512bit);
2790 __ kmovql(k1, output_mask);
2791 __ evmovdqub(Address(dest, dp), k1, tmp, true, Assembler::AVX_512bit);
2792
2793 __ addptr(dest, output_size);
2794
2795 __ BIND(L_exit);
2796 __ vzeroupper();
2797 __ pop(rax); // Get original dest value
2798 __ subptr(dest, rax); // Number of bytes converted
2799 __ movptr(rax, dest);
2800 __ pop(rbx);
2801 __ pop(r15);
2802 __ pop(r14);
2803 __ pop(r13);
2804 __ pop(r12);
2805 __ leave();
2806 __ ret(0);
2807
2808 __ BIND(L_loadURL);
2809 __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_url_addr()), Assembler::AVX_512bit, r13);
2810 __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_url_addr()), Assembler::AVX_512bit, r13);
2811 __ jmp(L_continue);
2812
2813 __ BIND(L_padding);
2814 __ decrementq(output_size, 1);
2815 __ shrq(rax, 1);
2816
2817 __ cmpb(Address(source, length, Address::times_1, -2), '=');
2818 __ jcc(Assembler::notEqual, L_donePadding);
2819
2820 __ decrementq(output_size, 1);
2821 __ shrq(rax, 1);
2822 __ jmp(L_donePadding);
2823
2824 __ align32();
2825 __ BIND(L_bruteForce);
2826 } // End of if(avx512_vbmi)
2827
2828 if (VM_Version::supports_avx2()) {
2829 Label L_tailProc, L_topLoop, L_enterLoop;
2830
2831 __ cmpl(isMIME, 0);
2832 __ jcc(Assembler::notEqual, L_lastChunk);
2833
2834 // Check for buffer too small (for algorithm)
2835 __ subl(length, 0x2c);
2836 __ jcc(Assembler::less, L_tailProc);
2837
2838 __ shll(isURL, 2);
2839
2840 // Algorithm adapted from https://arxiv.org/abs/1704.00605, "Faster Base64
2841 // Encoding and Decoding using AVX2 Instructions". URL modifications added.
2842
2843 // Set up constants
2844 __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_tables_addr()));
2845 __ vpbroadcastd(xmm4, Address(r13, isURL, Address::times_1), Assembler::AVX_256bit); // 2F or 5F
2846 __ vpbroadcastd(xmm10, Address(r13, isURL, Address::times_1, 0x08), Assembler::AVX_256bit); // -1 or -4
2847 __ vmovdqu(xmm12, Address(r13, 0x10)); // permute
2848 __ vmovdqu(xmm13, Address(r13, 0x30)); // shuffle
2849 __ vpbroadcastd(xmm7, Address(r13, 0x50), Assembler::AVX_256bit); // merge
2850 __ vpbroadcastd(xmm6, Address(r13, 0x54), Assembler::AVX_256bit); // merge mult
2851
2852 __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_LUT_tables_addr()));
2853 __ shll(isURL, 4);
2854 __ vmovdqu(xmm11, Address(r13, isURL, Address::times_1, 0x00)); // lut_lo
2855 __ vmovdqu(xmm8, Address(r13, isURL, Address::times_1, 0x20)); // lut_roll
2856 __ shrl(isURL, 6); // restore isURL
2857 __ vmovdqu(xmm9, Address(r13, 0x80)); // lut_hi
2858 __ jmp(L_enterLoop);
2859
2860 __ align32();
2861 __ bind(L_topLoop);
2862 // Add in the offset value (roll) to get 6-bit out values
2863 __ vpaddb(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
2864 // Merge and permute the output bits into appropriate output byte lanes
2865 __ vpmaddubsw(xmm0, xmm0, xmm7, Assembler::AVX_256bit);
2866 __ vpmaddwd(xmm0, xmm0, xmm6, Assembler::AVX_256bit);
2867 __ vpshufb(xmm0, xmm0, xmm13, Assembler::AVX_256bit);
2868 __ vpermd(xmm0, xmm12, xmm0, Assembler::AVX_256bit);
2869 // Store the output bytes
2870 __ vmovdqu(Address(dest, dp, Address::times_1, 0), xmm0);
2871 __ addptr(source, 0x20);
2872 __ addptr(dest, 0x18);
2873 __ subl(length, 0x20);
2874 __ jcc(Assembler::less, L_tailProc);
2875
2876 __ bind(L_enterLoop);
2877
2878 // Load in encoded string (32 bytes)
2879 __ vmovdqu(xmm2, Address(source, start_offset, Address::times_1, 0x0));
2880 // Extract the high nibble for indexing into the lut tables. High 4 bits are don't care.
2881 __ vpsrld(xmm1, xmm2, 0x4, Assembler::AVX_256bit);
2882 __ vpand(xmm1, xmm4, xmm1, Assembler::AVX_256bit);
2883 // Extract the low nibble. 5F/2F will isolate the low-order 4 bits. High 4 bits are don't care.
2884 __ vpand(xmm3, xmm2, xmm4, Assembler::AVX_256bit);
2885 // Check for special-case (0x2F or 0x5F (URL))
2886 __ vpcmpeqb(xmm0, xmm4, xmm2, Assembler::AVX_256bit);
2887 // Get the bitset based on the low nibble. vpshufb uses low-order 4 bits only.
2888 __ vpshufb(xmm3, xmm11, xmm3, Assembler::AVX_256bit);
2889 // Get the bit value of the high nibble
2890 __ vpshufb(xmm5, xmm9, xmm1, Assembler::AVX_256bit);
2891 // Make sure 2F / 5F shows as valid
2892 __ vpandn(xmm3, xmm0, xmm3, Assembler::AVX_256bit);
2893 // Make adjustment for roll index. For non-URL, this is a no-op,
2894 // for URL, this adjusts by -4. This is to properly index the
2895 // roll value for 2F / 5F.
2896 __ vpand(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
2897 // If the and of the two is non-zero, we have an invalid input character
2898 __ vptest(xmm3, xmm5);
2899 // Extract the "roll" value - value to add to the input to get 6-bit out value
2900 __ vpaddb(xmm0, xmm0, xmm1, Assembler::AVX_256bit); // Handle 2F / 5F
2901 __ vpshufb(xmm0, xmm8, xmm0, Assembler::AVX_256bit);
2902 __ jcc(Assembler::equal, L_topLoop); // Fall through on error
2903
2904 __ bind(L_tailProc);
2905
2906 __ addl(length, 0x2c);
2907
2908 __ vzeroupper();
2909 }
2910
2911 // Use non-AVX code to decode 4-byte chunks into 3 bytes of output
2912
2913 // Register state (Linux):
2914 // r12-15 - saved on stack
2915 // rdi - src
2916 // rsi - sp
2917 // rdx - sl
2918 // rcx - dst
2919 // r8 - dp
2920 // r9 - isURL
2921
2922 // Register state (Windows):
2923 // r12-15 - saved on stack
2924 // rcx - src
2925 // rdx - sp
2926 // r8 - sl
2927 // r9 - dst
2928 // r12 - dp
2929 // r10 - isURL
2930
2931 // Registers (common):
2932 // length (r14) - bytes in src
2933
2934 const Register decode_table = r11;
2935 const Register out_byte_count = rbx;
2936 const Register byte1 = r13;
2937 const Register byte2 = r15;
2938 const Register byte3 = WIN64_ONLY(r8) NOT_WIN64(rdx);
2939 const Register byte4 = WIN64_ONLY(r10) NOT_WIN64(r9);
2940
2941 __ bind(L_lastChunk);
2942
2943 __ shrl(length, 2); // Multiple of 4 bytes only - length is # 4-byte chunks
2944 __ cmpl(length, 0);
2945 __ jcc(Assembler::lessEqual, L_exit_no_vzero);
2946
2947 __ shll(isURL, 8); // index into decode table based on isURL
2948 __ lea(decode_table, ExternalAddress(StubRoutines::x86::base64_decoding_table_addr()));
2949 __ addptr(decode_table, isURL);
2950
2951 __ jmp(L_bottomLoop);
2952
2953 __ align32();
2954 __ BIND(L_forceLoop);
2955 __ shll(byte1, 18);
2956 __ shll(byte2, 12);
2957 __ shll(byte3, 6);
2958 __ orl(byte1, byte2);
2959 __ orl(byte1, byte3);
2960 __ orl(byte1, byte4);
2961
2962 __ addptr(source, 4);
2963
2964 __ movb(Address(dest, dp, Address::times_1, 2), byte1);
2965 __ shrl(byte1, 8);
2966 __ movb(Address(dest, dp, Address::times_1, 1), byte1);
2967 __ shrl(byte1, 8);
2968 __ movb(Address(dest, dp, Address::times_1, 0), byte1);
2969
2970 __ addptr(dest, 3);
2971 __ decrementl(length, 1);
2972 __ jcc(Assembler::zero, L_exit_no_vzero);
2973
2974 __ BIND(L_bottomLoop);
2975 __ load_unsigned_byte(byte1, Address(source, start_offset, Address::times_1, 0x00));
2976 __ load_unsigned_byte(byte2, Address(source, start_offset, Address::times_1, 0x01));
2977 __ load_signed_byte(byte1, Address(decode_table, byte1));
2978 __ load_signed_byte(byte2, Address(decode_table, byte2));
2979 __ load_unsigned_byte(byte3, Address(source, start_offset, Address::times_1, 0x02));
2980 __ load_unsigned_byte(byte4, Address(source, start_offset, Address::times_1, 0x03));
2981 __ load_signed_byte(byte3, Address(decode_table, byte3));
2982 __ load_signed_byte(byte4, Address(decode_table, byte4));
2983
2984 __ mov(rax, byte1);
2985 __ orl(rax, byte2);
2986 __ orl(rax, byte3);
2987 __ orl(rax, byte4);
2988 __ jcc(Assembler::positive, L_forceLoop);
2989
2990 __ BIND(L_exit_no_vzero);
2991 __ pop(rax); // Get original dest value
2992 __ subptr(dest, rax); // Number of bytes converted
2993 __ movptr(rax, dest);
2994 __ pop(rbx);
2995 __ pop(r15);
2996 __ pop(r14);
2997 __ pop(r13);
2998 __ pop(r12);
2999 __ leave();
3000 __ ret(0);
3001
3002 return start;
3003 }
3004
3005
3006 /**
3007 * Arguments:
3008 *
3009 * Inputs:
3010 * c_rarg0 - int crc
3011 * c_rarg1 - byte* buf
3012 * c_rarg2 - int length
3013 *
3014 * Output:
3015 * rax - int crc result
3016 */
3017 address StubGenerator::generate_updateBytesCRC32() {
3018 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3019
3020 __ align(CodeEntryAlignment);
3021 StubGenStubId stub_id = StubGenStubId::updateBytesCRC32_id;
3022 StubCodeMark mark(this, stub_id);
3023
3024 address start = __ pc();
3025
3026 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3027 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3028 // rscratch1: r10
3029 const Register crc = c_rarg0; // crc
3030 const Register buf = c_rarg1; // source java byte array address
3031 const Register len = c_rarg2; // length
3032 const Register table = c_rarg3; // crc_table address (reuse register)
3033 const Register tmp1 = r11;
3034 const Register tmp2 = r10;
3035 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax);
3036
3037 BLOCK_COMMENT("Entry:");
3038 __ enter(); // required for proper stackwalking of RuntimeStub frame
3039
3040 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
3041 VM_Version::supports_avx512bw() &&
3042 VM_Version::supports_avx512vl()) {
3043 // The constants used in the CRC32 algorithm requires the 1's compliment of the initial crc value.
3044 // However, the constant table for CRC32-C assumes the original crc value. Account for this
3045 // difference before calling and after returning.
3046 __ lea(table, ExternalAddress(StubRoutines::x86::crc_table_avx512_addr()));
3047 __ notl(crc);
3048 __ kernel_crc32_avx512(crc, buf, len, table, tmp1, tmp2);
3049 __ notl(crc);
3050 } else {
3051 __ kernel_crc32(crc, buf, len, table, tmp1);
3052 }
3053
3054 __ movl(rax, crc);
3055 __ vzeroupper();
3056 __ leave(); // required for proper stackwalking of RuntimeStub frame
3057 __ ret(0);
3058
3059 return start;
3060 }
3061
3062 /**
3063 * Arguments:
3064 *
3065 * Inputs:
3066 * c_rarg0 - int crc
3067 * c_rarg1 - byte* buf
3068 * c_rarg2 - long length
3069 * c_rarg3 - table_start - optional (present only when doing a library_call,
3070 * not used by x86 algorithm)
3071 *
3072 * Output:
3073 * rax - int crc result
3074 */
3075 address StubGenerator::generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
3076 assert(UseCRC32CIntrinsics, "need SSE4_2");
3077 __ align(CodeEntryAlignment);
3078 StubGenStubId stub_id = StubGenStubId::updateBytesCRC32C_id;
3079 StubCodeMark mark(this, stub_id);
3080 address start = __ pc();
3081
3082 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
3083 //Windows RCX RDX R8 R9 none none XMM0..XMM3
3084 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
3085 const Register crc = c_rarg0; // crc
3086 const Register buf = c_rarg1; // source java byte array address
3087 const Register len = c_rarg2; // length
3088 const Register a = rax;
3089 const Register j = r9;
3090 const Register k = r10;
3091 const Register l = r11;
3092 #ifdef _WIN64
3093 const Register y = rdi;
3094 const Register z = rsi;
3095 #else
3096 const Register y = rcx;
3097 const Register z = r8;
3098 #endif
3099 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
3100
3101 BLOCK_COMMENT("Entry:");
3102 __ enter(); // required for proper stackwalking of RuntimeStub frame
3103 Label L_continue;
3104
3105 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
3106 VM_Version::supports_avx512bw() &&
3107 VM_Version::supports_avx512vl()) {
3108 Label L_doSmall;
3109
3110 __ cmpl(len, 384);
3111 __ jcc(Assembler::lessEqual, L_doSmall);
3112
3113 __ lea(j, ExternalAddress(StubRoutines::x86::crc32c_table_avx512_addr()));
3114 __ kernel_crc32_avx512(crc, buf, len, j, l, k);
3115
3116 __ jmp(L_continue);
3117
3118 __ bind(L_doSmall);
3119 }
3120 #ifdef _WIN64
3121 __ push(y);
3122 __ push(z);
3123 #endif
3124 __ crc32c_ipl_alg2_alt2(crc, buf, len,
3125 a, j, k,
3126 l, y, z,
3127 c_farg0, c_farg1, c_farg2,
3128 is_pclmulqdq_supported);
3129 #ifdef _WIN64
3130 __ pop(z);
3131 __ pop(y);
3132 #endif
3133
3134 __ bind(L_continue);
3135 __ movl(rax, crc);
3136 __ vzeroupper();
3137 __ leave(); // required for proper stackwalking of RuntimeStub frame
3138 __ ret(0);
3139
3140 return start;
3141 }
3142
3143
3144 /**
3145 * Arguments:
3146 *
3147 * Input:
3148 * c_rarg0 - x address
3149 * c_rarg1 - x length
3150 * c_rarg2 - y address
3151 * c_rarg3 - y length
3152 * not Win64
3153 * c_rarg4 - z address
3154 * Win64
3155 * rsp+40 - z address
3156 */
3157 address StubGenerator::generate_multiplyToLen() {
3158 __ align(CodeEntryAlignment);
3159 StubGenStubId stub_id = StubGenStubId::multiplyToLen_id;
3160 StubCodeMark mark(this, stub_id);
3161 address start = __ pc();
3162
3163 if (SCCache::load_stub(this, vmIntrinsics::_multiplyToLen, "multiplyToLen", start)) {
3164 return start;
3165 }
3166
3167 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3168 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3169 const Register x = rdi;
3170 const Register xlen = rax;
3171 const Register y = rsi;
3172 const Register ylen = rcx;
3173 const Register z = r8;
3174
3175 // Next registers will be saved on stack in multiply_to_len().
3176 const Register tmp0 = r11;
3177 const Register tmp1 = r12;
3178 const Register tmp2 = r13;
3179 const Register tmp3 = r14;
3180 const Register tmp4 = r15;
3181 const Register tmp5 = rbx;
3182
3183 BLOCK_COMMENT("Entry:");
3184 __ enter(); // required for proper stackwalking of RuntimeStub frame
3185
3186 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
3187 // ylen => rcx, z => r8
3188 // r9 and r10 may be used to save non-volatile registers
3189 #ifdef _WIN64
3190 // last argument (#4) is on stack on Win64
3191 __ movptr(z, Address(rsp, 6 * wordSize));
3192 #endif
3193
3194 __ movptr(xlen, rsi);
3195 __ movptr(y, rdx);
3196 __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
3197
3198 restore_arg_regs();
3199
3200 __ leave(); // required for proper stackwalking of RuntimeStub frame
3201 __ ret(0);
3202
3203 SCCache::store_stub(this, vmIntrinsics::_multiplyToLen, "multiplyToLen", start);
3204 return start;
3205 }
3206
3207 /**
3208 * Arguments:
3209 *
3210 * Input:
3211 * c_rarg0 - obja address
3212 * c_rarg1 - objb address
3213 * c_rarg3 - length length
3214 * c_rarg4 - scale log2_array_indxscale
3215 *
3216 * Output:
3217 * rax - int >= mismatched index, < 0 bitwise complement of tail
3218 */
3219 address StubGenerator::generate_vectorizedMismatch() {
3220 __ align(CodeEntryAlignment);
3221 StubGenStubId stub_id = StubGenStubId::vectorizedMismatch_id;
3222 StubCodeMark mark(this, stub_id);
3223 address start = __ pc();
3224
3225 BLOCK_COMMENT("Entry:");
3226 __ enter();
3227
3228 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3229 const Register scale = c_rarg0; //rcx, will exchange with r9
3230 const Register objb = c_rarg1; //rdx
3231 const Register length = c_rarg2; //r8
3232 const Register obja = c_rarg3; //r9
3233 __ xchgq(obja, scale); //now obja and scale contains the correct contents
3234
3235 const Register tmp1 = r10;
3236 const Register tmp2 = r11;
3237 #endif
3238 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3239 const Register obja = c_rarg0; //U:rdi
3240 const Register objb = c_rarg1; //U:rsi
3241 const Register length = c_rarg2; //U:rdx
3242 const Register scale = c_rarg3; //U:rcx
3243 const Register tmp1 = r8;
3244 const Register tmp2 = r9;
3245 #endif
3246 const Register result = rax; //return value
3247 const XMMRegister vec0 = xmm0;
3248 const XMMRegister vec1 = xmm1;
3249 const XMMRegister vec2 = xmm2;
3250
3251 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
3252
3253 __ vzeroupper();
3254 __ leave();
3255 __ ret(0);
3256
3257 return start;
3258 }
3259
3260 /**
3261 * Arguments:
3262 *
3263 // Input:
3264 // c_rarg0 - x address
3265 // c_rarg1 - x length
3266 // c_rarg2 - z address
3267 // c_rarg3 - z length
3268 *
3269 */
3270 address StubGenerator::generate_squareToLen() {
3271
3272 __ align(CodeEntryAlignment);
3273 StubGenStubId stub_id = StubGenStubId::squareToLen_id;
3274 StubCodeMark mark(this, stub_id);
3275 address start = __ pc();
3276
3277 if (SCCache::load_stub(this, vmIntrinsics::_squareToLen, "squareToLen", start)) {
3278 return start;
3279 }
3280
3281 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3282 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
3283 const Register x = rdi;
3284 const Register len = rsi;
3285 const Register z = r8;
3286 const Register zlen = rcx;
3287
3288 const Register tmp1 = r12;
3289 const Register tmp2 = r13;
3290 const Register tmp3 = r14;
3291 const Register tmp4 = r15;
3292 const Register tmp5 = rbx;
3293
3294 BLOCK_COMMENT("Entry:");
3295 __ enter(); // required for proper stackwalking of RuntimeStub frame
3296
3297 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
3298 // zlen => rcx
3299 // r9 and r10 may be used to save non-volatile registers
3300 __ movptr(r8, rdx);
3301 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3302
3303 restore_arg_regs();
3304
3305 __ leave(); // required for proper stackwalking of RuntimeStub frame
3306 __ ret(0);
3307
3308 SCCache::store_stub(this, vmIntrinsics::_squareToLen, "squareToLen", start);
3309 return start;
3310 }
3311
3312 address StubGenerator::generate_method_entry_barrier() {
3313 __ align(CodeEntryAlignment);
3314 StubGenStubId stub_id = StubGenStubId::method_entry_barrier_id;
3315 StubCodeMark mark(this, stub_id);
3316 address start = __ pc();
3317
3318 Label deoptimize_label;
3319
3320 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
3321
3322 BLOCK_COMMENT("Entry:");
3323 __ enter(); // save rbp
3324
3325 // save c_rarg0, because we want to use that value.
3326 // We could do without it but then we depend on the number of slots used by pusha
3327 __ push(c_rarg0);
3328
3329 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
3330
3331 __ pusha();
3332
3333 // The method may have floats as arguments, and we must spill them before calling
3334 // the VM runtime.
3335 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
3336 const int xmm_size = wordSize * 2;
3337 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
3338 __ subptr(rsp, xmm_spill_size);
3339 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
3340 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
3341 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
3342 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
3343 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
3344 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
3345 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
3346 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
3347
3348 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
3349
3350 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
3351 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
3352 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
3353 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
3354 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
3355 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
3356 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
3357 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
3358 __ addptr(rsp, xmm_spill_size);
3359
3360 __ cmpl(rax, 1); // 1 means deoptimize
3361 __ jcc(Assembler::equal, deoptimize_label);
3362
3363 __ popa();
3364 __ pop(c_rarg0);
3365
3366 __ leave();
3367
3368 __ addptr(rsp, 1 * wordSize); // cookie
3369 __ ret(0);
3370
3371
3372 __ BIND(deoptimize_label);
3373
3374 __ popa();
3375 __ pop(c_rarg0);
3376
3377 __ leave();
3378
3379 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
3380 // here while still having a correct stack is valuable
3381 __ testptr(rsp, Address(rsp, 0));
3382
3383 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
3384 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
3385
3386 return start;
3387 }
3388
3389 /**
3390 * Arguments:
3391 *
3392 * Input:
3393 * c_rarg0 - out address
3394 * c_rarg1 - in address
3395 * c_rarg2 - offset
3396 * c_rarg3 - len
3397 * not Win64
3398 * c_rarg4 - k
3399 * Win64
3400 * rsp+40 - k
3401 */
3402 address StubGenerator::generate_mulAdd() {
3403 __ align(CodeEntryAlignment);
3404 StubGenStubId stub_id = StubGenStubId::mulAdd_id;
3405 StubCodeMark mark(this, stub_id);
3406 address start = __ pc();
3407
3408 if (SCCache::load_stub(this, vmIntrinsics::_mulAdd, "mulAdd", start)) {
3409 return start;
3410 }
3411
3412 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3413 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3414 const Register out = rdi;
3415 const Register in = rsi;
3416 const Register offset = r11;
3417 const Register len = rcx;
3418 const Register k = r8;
3419
3420 // Next registers will be saved on stack in mul_add().
3421 const Register tmp1 = r12;
3422 const Register tmp2 = r13;
3423 const Register tmp3 = r14;
3424 const Register tmp4 = r15;
3425 const Register tmp5 = rbx;
3426
3427 BLOCK_COMMENT("Entry:");
3428 __ enter(); // required for proper stackwalking of RuntimeStub frame
3429
3430 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
3431 // len => rcx, k => r8
3432 // r9 and r10 may be used to save non-volatile registers
3433 #ifdef _WIN64
3434 // last argument is on stack on Win64
3435 __ movl(k, Address(rsp, 6 * wordSize));
3436 #endif
3437 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
3438 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3439
3440 restore_arg_regs();
3441
3442 __ leave(); // required for proper stackwalking of RuntimeStub frame
3443 __ ret(0);
3444
3445 SCCache::store_stub(this, vmIntrinsics::_mulAdd, "mulAdd", start);
3446 return start;
3447 }
3448
3449 address StubGenerator::generate_bigIntegerRightShift() {
3450 __ align(CodeEntryAlignment);
3451 StubGenStubId stub_id = StubGenStubId::bigIntegerRightShiftWorker_id;
3452 StubCodeMark mark(this, stub_id);
3453 address start = __ pc();
3454
3455 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3456 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3457 const Register newArr = rdi;
3458 const Register oldArr = rsi;
3459 const Register newIdx = rdx;
3460 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3461 const Register totalNumIter = r8;
3462
3463 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3464 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3465 const Register tmp1 = r11; // Caller save.
3466 const Register tmp2 = rax; // Caller save.
3467 const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9); // Windows: Callee save. Linux: Caller save.
3468 const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10); // Windows: Callee save. Linux: Caller save.
3469 const Register tmp5 = r14; // Callee save.
3470 const Register tmp6 = r15;
3471
3472 const XMMRegister x0 = xmm0;
3473 const XMMRegister x1 = xmm1;
3474 const XMMRegister x2 = xmm2;
3475
3476 BLOCK_COMMENT("Entry:");
3477 __ enter(); // required for proper stackwalking of RuntimeStub frame
3478
3479 #ifdef _WIN64
3480 setup_arg_regs(4);
3481 // For windows, since last argument is on stack, we need to move it to the appropriate register.
3482 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3483 // Save callee save registers.
3484 __ push(tmp3);
3485 __ push(tmp4);
3486 #endif
3487 __ push(tmp5);
3488
3489 // Rename temps used throughout the code.
3490 const Register idx = tmp1;
3491 const Register nIdx = tmp2;
3492
3493 __ xorl(idx, idx);
3494
3495 // Start right shift from end of the array.
3496 // For example, if #iteration = 4 and newIdx = 1
3497 // then dest[4] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
3498 // if #iteration = 4 and newIdx = 0
3499 // then dest[3] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
3500 __ movl(idx, totalNumIter);
3501 __ movl(nIdx, idx);
3502 __ addl(nIdx, newIdx);
3503
3504 // If vectorization is enabled, check if the number of iterations is at least 64
3505 // If not, then go to ShifTwo processing 2 iterations
3506 if (VM_Version::supports_avx512_vbmi2()) {
3507 __ cmpptr(totalNumIter, (AVX3Threshold/64));
3508 __ jcc(Assembler::less, ShiftTwo);
3509
3510 if (AVX3Threshold < 16 * 64) {
3511 __ cmpl(totalNumIter, 16);
3512 __ jcc(Assembler::less, ShiftTwo);
3513 }
3514 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3515 __ subl(idx, 16);
3516 __ subl(nIdx, 16);
3517 __ BIND(Shift512Loop);
3518 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 4), Assembler::AVX_512bit);
3519 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3520 __ vpshrdvd(x2, x1, x0, Assembler::AVX_512bit);
3521 __ evmovdqul(Address(newArr, nIdx, Address::times_4), x2, Assembler::AVX_512bit);
3522 __ subl(nIdx, 16);
3523 __ subl(idx, 16);
3524 __ jcc(Assembler::greaterEqual, Shift512Loop);
3525 __ addl(idx, 16);
3526 __ addl(nIdx, 16);
3527 }
3528 __ BIND(ShiftTwo);
3529 __ cmpl(idx, 2);
3530 __ jcc(Assembler::less, ShiftOne);
3531 __ subl(idx, 2);
3532 __ subl(nIdx, 2);
3533 __ BIND(ShiftTwoLoop);
3534 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 8));
3535 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3536 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3537 __ shrdl(tmp5, tmp4);
3538 __ shrdl(tmp4, tmp3);
3539 __ movl(Address(newArr, nIdx, Address::times_4, 4), tmp5);
3540 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3541 __ subl(nIdx, 2);
3542 __ subl(idx, 2);
3543 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3544 __ addl(idx, 2);
3545 __ addl(nIdx, 2);
3546
3547 // Do the last iteration
3548 __ BIND(ShiftOne);
3549 __ cmpl(idx, 1);
3550 __ jcc(Assembler::less, Exit);
3551 __ subl(idx, 1);
3552 __ subl(nIdx, 1);
3553 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3554 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3555 __ shrdl(tmp4, tmp3);
3556 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3557 __ BIND(Exit);
3558 __ vzeroupper();
3559 // Restore callee save registers.
3560 __ pop(tmp5);
3561 #ifdef _WIN64
3562 __ pop(tmp4);
3563 __ pop(tmp3);
3564 restore_arg_regs();
3565 #endif
3566 __ leave(); // required for proper stackwalking of RuntimeStub frame
3567 __ ret(0);
3568
3569 return start;
3570 }
3571
3572 /**
3573 * Arguments:
3574 *
3575 * Input:
3576 * c_rarg0 - newArr address
3577 * c_rarg1 - oldArr address
3578 * c_rarg2 - newIdx
3579 * c_rarg3 - shiftCount
3580 * not Win64
3581 * c_rarg4 - numIter
3582 * Win64
3583 * rsp40 - numIter
3584 */
3585 address StubGenerator::generate_bigIntegerLeftShift() {
3586 __ align(CodeEntryAlignment);
3587 StubGenStubId stub_id = StubGenStubId::bigIntegerLeftShiftWorker_id;
3588 StubCodeMark mark(this, stub_id);
3589 address start = __ pc();
3590
3591 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3592 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3593 const Register newArr = rdi;
3594 const Register oldArr = rsi;
3595 const Register newIdx = rdx;
3596 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3597 const Register totalNumIter = r8;
3598 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3599 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3600 const Register tmp1 = r11; // Caller save.
3601 const Register tmp2 = rax; // Caller save.
3602 const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9); // Windows: Callee save. Linux: Caller save.
3603 const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10); // Windows: Callee save. Linux: Caller save.
3604 const Register tmp5 = r14; // Callee save.
3605
3606 const XMMRegister x0 = xmm0;
3607 const XMMRegister x1 = xmm1;
3608 const XMMRegister x2 = xmm2;
3609 BLOCK_COMMENT("Entry:");
3610 __ enter(); // required for proper stackwalking of RuntimeStub frame
3611
3612 #ifdef _WIN64
3613 setup_arg_regs(4);
3614 // For windows, since last argument is on stack, we need to move it to the appropriate register.
3615 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3616 // Save callee save registers.
3617 __ push(tmp3);
3618 __ push(tmp4);
3619 #endif
3620 __ push(tmp5);
3621
3622 // Rename temps used throughout the code
3623 const Register idx = tmp1;
3624 const Register numIterTmp = tmp2;
3625
3626 // Start idx from zero.
3627 __ xorl(idx, idx);
3628 // Compute interior pointer for new array. We do this so that we can use same index for both old and new arrays.
3629 __ lea(newArr, Address(newArr, newIdx, Address::times_4));
3630 __ movl(numIterTmp, totalNumIter);
3631
3632 // If vectorization is enabled, check if the number of iterations is at least 64
3633 // If not, then go to ShiftTwo shifting two numbers at a time
3634 if (VM_Version::supports_avx512_vbmi2()) {
3635 __ cmpl(totalNumIter, (AVX3Threshold/64));
3636 __ jcc(Assembler::less, ShiftTwo);
3637
3638 if (AVX3Threshold < 16 * 64) {
3639 __ cmpl(totalNumIter, 16);
3640 __ jcc(Assembler::less, ShiftTwo);
3641 }
3642 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3643 __ subl(numIterTmp, 16);
3644 __ BIND(Shift512Loop);
3645 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3646 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 0x4), Assembler::AVX_512bit);
3647 __ vpshldvd(x1, x2, x0, Assembler::AVX_512bit);
3648 __ evmovdqul(Address(newArr, idx, Address::times_4), x1, Assembler::AVX_512bit);
3649 __ addl(idx, 16);
3650 __ subl(numIterTmp, 16);
3651 __ jcc(Assembler::greaterEqual, Shift512Loop);
3652 __ addl(numIterTmp, 16);
3653 }
3654 __ BIND(ShiftTwo);
3655 __ cmpl(totalNumIter, 1);
3656 __ jcc(Assembler::less, Exit);
3657 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3658 __ subl(numIterTmp, 2);
3659 __ jcc(Assembler::less, ShiftOne);
3660
3661 __ BIND(ShiftTwoLoop);
3662 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3663 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 0x8));
3664 __ shldl(tmp3, tmp4);
3665 __ shldl(tmp4, tmp5);
3666 __ movl(Address(newArr, idx, Address::times_4), tmp3);
3667 __ movl(Address(newArr, idx, Address::times_4, 0x4), tmp4);
3668 __ movl(tmp3, tmp5);
3669 __ addl(idx, 2);
3670 __ subl(numIterTmp, 2);
3671 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3672
3673 // Do the last iteration
3674 __ BIND(ShiftOne);
3675 __ addl(numIterTmp, 2);
3676 __ cmpl(numIterTmp, 1);
3677 __ jcc(Assembler::less, Exit);
3678 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3679 __ shldl(tmp3, tmp4);
3680 __ movl(Address(newArr, idx, Address::times_4), tmp3);
3681
3682 __ BIND(Exit);
3683 __ vzeroupper();
3684 // Restore callee save registers.
3685 __ pop(tmp5);
3686 #ifdef _WIN64
3687 __ pop(tmp4);
3688 __ pop(tmp3);
3689 restore_arg_regs();
3690 #endif
3691 __ leave(); // required for proper stackwalking of RuntimeStub frame
3692 __ ret(0);
3693
3694 return start;
3695 }
3696
3697 void StubGenerator::generate_libm_stubs() {
3698 if (UseLibmIntrinsic && InlineIntrinsics) {
3699 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3700 StubRoutines::_dsin = generate_libmSin(); // from stubGenerator_x86_64_sin.cpp
3701 }
3702 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3703 StubRoutines::_dcos = generate_libmCos(); // from stubGenerator_x86_64_cos.cpp
3704 }
3705 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3706 StubRoutines::_dtan = generate_libmTan(); // from stubGenerator_x86_64_tan.cpp
3707 }
3708 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtanh)) {
3709 StubRoutines::_dtanh = generate_libmTanh(); // from stubGenerator_x86_64_tanh.cpp
3710 }
3711 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3712 StubRoutines::_dexp = generate_libmExp(); // from stubGenerator_x86_64_exp.cpp
3713 }
3714 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3715 StubRoutines::_dpow = generate_libmPow(); // from stubGenerator_x86_64_pow.cpp
3716 }
3717 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3718 StubRoutines::_dlog = generate_libmLog(); // from stubGenerator_x86_64_log.cpp
3719 }
3720 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3721 StubRoutines::_dlog10 = generate_libmLog10(); // from stubGenerator_x86_64_log.cpp
3722 }
3723 }
3724 }
3725
3726 /**
3727 * Arguments:
3728 *
3729 * Input:
3730 * c_rarg0 - float16 jshort
3731 *
3732 * Output:
3733 * xmm0 - float
3734 */
3735 address StubGenerator::generate_float16ToFloat() {
3736 StubGenStubId stub_id = StubGenStubId::hf2f_id;
3737 StubCodeMark mark(this, stub_id);
3738
3739 address start = __ pc();
3740
3741 BLOCK_COMMENT("Entry:");
3742 // No need for RuntimeStub frame since it is called only during JIT compilation
3743
3744 // Load value into xmm0 and convert
3745 __ flt16_to_flt(xmm0, c_rarg0);
3746
3747 __ ret(0);
3748
3749 return start;
3750 }
3751
3752 /**
3753 * Arguments:
3754 *
3755 * Input:
3756 * xmm0 - float
3757 *
3758 * Output:
3759 * rax - float16 jshort
3760 */
3761 address StubGenerator::generate_floatToFloat16() {
3762 StubGenStubId stub_id = StubGenStubId::f2hf_id;
3763 StubCodeMark mark(this, stub_id);
3764
3765 address start = __ pc();
3766
3767 BLOCK_COMMENT("Entry:");
3768 // No need for RuntimeStub frame since it is called only during JIT compilation
3769
3770 // Convert and put result into rax
3771 __ flt_to_flt16(rax, xmm0, xmm1);
3772
3773 __ ret(0);
3774
3775 return start;
3776 }
3777
3778 address StubGenerator::generate_cont_thaw(StubGenStubId stub_id) {
3779 if (!Continuations::enabled()) return nullptr;
3780
3781 bool return_barrier;
3782 bool return_barrier_exception;
3783 Continuation::thaw_kind kind;
3784
3785 switch (stub_id) {
3786 case cont_thaw_id:
3787 return_barrier = false;
3788 return_barrier_exception = false;
3789 kind = Continuation::thaw_top;
3790 break;
3791 case cont_returnBarrier_id:
3792 return_barrier = true;
3793 return_barrier_exception = false;
3794 kind = Continuation::thaw_return_barrier;
3795 break;
3796 case cont_returnBarrierExc_id:
3797 return_barrier = true;
3798 return_barrier_exception = true;
3799 kind = Continuation::thaw_return_barrier_exception;
3800 break;
3801 default:
3802 ShouldNotReachHere();
3803 }
3804 StubCodeMark mark(this, stub_id);
3805 address start = __ pc();
3806
3807 // TODO: Handle Valhalla return types. May require generating different return barriers.
3808
3809 if (!return_barrier) {
3810 // Pop return address. If we don't do this, we get a drift,
3811 // where the bottom-most frozen frame continuously grows.
3812 __ pop(c_rarg3);
3813 } else {
3814 __ movptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3815 }
3816
3817 #ifdef ASSERT
3818 {
3819 Label L_good_sp;
3820 __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3821 __ jcc(Assembler::equal, L_good_sp);
3822 __ stop("Incorrect rsp at thaw entry");
3823 __ BIND(L_good_sp);
3824 }
3825 #endif // ASSERT
3826
3827 if (return_barrier) {
3828 // Preserve possible return value from a method returning to the return barrier.
3829 __ push(rax);
3830 __ push_d(xmm0);
3831 }
3832
3833 __ movptr(c_rarg0, r15_thread);
3834 __ movptr(c_rarg1, (return_barrier ? 1 : 0));
3835 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Continuation::prepare_thaw), 2);
3836 __ movptr(rbx, rax);
3837
3838 if (return_barrier) {
3839 // Restore return value from a method returning to the return barrier.
3840 // No safepoint in the call to thaw, so even an oop return value should be OK.
3841 __ pop_d(xmm0);
3842 __ pop(rax);
3843 }
3844
3845 #ifdef ASSERT
3846 {
3847 Label L_good_sp;
3848 __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3849 __ jcc(Assembler::equal, L_good_sp);
3850 __ stop("Incorrect rsp after prepare thaw");
3851 __ BIND(L_good_sp);
3852 }
3853 #endif // ASSERT
3854
3855 // rbx contains the size of the frames to thaw, 0 if overflow or no more frames
3856 Label L_thaw_success;
3857 __ testptr(rbx, rbx);
3858 __ jccb(Assembler::notZero, L_thaw_success);
3859 __ jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
3860 __ bind(L_thaw_success);
3861
3862 // Make room for the thawed frames and align the stack.
3863 __ subptr(rsp, rbx);
3864 __ andptr(rsp, -StackAlignmentInBytes);
3865
3866 if (return_barrier) {
3867 // Preserve possible return value from a method returning to the return barrier. (Again.)
3868 __ push(rax);
3869 __ push_d(xmm0);
3870 }
3871
3872 // If we want, we can templatize thaw by kind, and have three different entries.
3873 __ movptr(c_rarg0, r15_thread);
3874 __ movptr(c_rarg1, kind);
3875 __ call_VM_leaf(Continuation::thaw_entry(), 2);
3876 __ movptr(rbx, rax);
3877
3878 if (return_barrier) {
3879 // Restore return value from a method returning to the return barrier. (Again.)
3880 // No safepoint in the call to thaw, so even an oop return value should be OK.
3881 __ pop_d(xmm0);
3882 __ pop(rax);
3883 } else {
3884 // Return 0 (success) from doYield.
3885 __ xorptr(rax, rax);
3886 }
3887
3888 // After thawing, rbx is the SP of the yielding frame.
3889 // Move there, and then to saved RBP slot.
3890 __ movptr(rsp, rbx);
3891 __ subptr(rsp, 2*wordSize);
3892
3893 if (return_barrier_exception) {
3894 __ movptr(c_rarg0, r15_thread);
3895 __ movptr(c_rarg1, Address(rsp, wordSize)); // return address
3896
3897 // rax still holds the original exception oop, save it before the call
3898 __ push(rax);
3899
3900 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 2);
3901 __ movptr(rbx, rax);
3902
3903 // Continue at exception handler:
3904 // rax: exception oop
3905 // rbx: exception handler
3906 // rdx: exception pc
3907 __ pop(rax);
3908 __ verify_oop(rax);
3909 __ pop(rbp); // pop out RBP here too
3910 __ pop(rdx);
3911 __ jmp(rbx);
3912 } else {
3913 // We are "returning" into the topmost thawed frame; see Thaw::push_return_frame
3914 __ pop(rbp);
3915 __ ret(0);
3916 }
3917
3918 return start;
3919 }
3920
3921 address StubGenerator::generate_cont_thaw() {
3922 return generate_cont_thaw(StubGenStubId::cont_thaw_id);
3923 }
3924
3925 // TODO: will probably need multiple return barriers depending on return type
3926
3927 address StubGenerator::generate_cont_returnBarrier() {
3928 return generate_cont_thaw(StubGenStubId::cont_returnBarrier_id);
3929 }
3930
3931 address StubGenerator::generate_cont_returnBarrier_exception() {
3932 return generate_cont_thaw(StubGenStubId::cont_returnBarrierExc_id);
3933 }
3934
3935 address StubGenerator::generate_cont_preempt_stub() {
3936 if (!Continuations::enabled()) return nullptr;
3937 StubGenStubId stub_id = StubGenStubId::cont_preempt_id;
3938 StubCodeMark mark(this, stub_id);
3939 address start = __ pc();
3940
3941 __ reset_last_Java_frame(true);
3942
3943 // Set rsp to enterSpecial frame, i.e. remove all frames copied into the heap.
3944 __ movptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3945
3946 Label preemption_cancelled;
3947 __ movbool(rscratch1, Address(r15_thread, JavaThread::preemption_cancelled_offset()));
3948 __ testbool(rscratch1);
3949 __ jcc(Assembler::notZero, preemption_cancelled);
3950
3951 // Remove enterSpecial frame from the stack and return to Continuation.run() to unmount.
3952 SharedRuntime::continuation_enter_cleanup(_masm);
3953 __ pop(rbp);
3954 __ ret(0);
3955
3956 // We acquired the monitor after freezing the frames so call thaw to continue execution.
3957 __ bind(preemption_cancelled);
3958 __ movbool(Address(r15_thread, JavaThread::preemption_cancelled_offset()), false);
3959 __ lea(rbp, Address(rsp, checked_cast<int32_t>(ContinuationEntry::size())));
3960 __ movptr(rscratch1, ExternalAddress(ContinuationEntry::thaw_call_pc_address()));
3961 __ jmp(rscratch1);
3962
3963 return start;
3964 }
3965
3966 // exception handler for upcall stubs
3967 address StubGenerator::generate_upcall_stub_exception_handler() {
3968 StubGenStubId stub_id = StubGenStubId::upcall_stub_exception_handler_id;
3969 StubCodeMark mark(this, stub_id);
3970 address start = __ pc();
3971
3972 // native caller has no idea how to handle exceptions
3973 // we just crash here. Up to callee to catch exceptions.
3974 __ verify_oop(rax);
3975 __ vzeroupper();
3976 __ mov(c_rarg0, rax);
3977 __ andptr(rsp, -StackAlignmentInBytes); // align stack as required by ABI
3978 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
3979 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, UpcallLinker::handle_uncaught_exception)));
3980 __ should_not_reach_here();
3981
3982 return start;
3983 }
3984
3985 // load Method* target of MethodHandle
3986 // j_rarg0 = jobject receiver
3987 // rbx = result
3988 address StubGenerator::generate_upcall_stub_load_target() {
3989 StubGenStubId stub_id = StubGenStubId::upcall_stub_load_target_id;
3990 StubCodeMark mark(this, stub_id);
3991 address start = __ pc();
3992
3993 __ resolve_global_jobject(j_rarg0, r15_thread, rscratch1);
3994 // Load target method from receiver
3995 __ load_heap_oop(rbx, Address(j_rarg0, java_lang_invoke_MethodHandle::form_offset()), rscratch1);
3996 __ load_heap_oop(rbx, Address(rbx, java_lang_invoke_LambdaForm::vmentry_offset()), rscratch1);
3997 __ load_heap_oop(rbx, Address(rbx, java_lang_invoke_MemberName::method_offset()), rscratch1);
3998 __ access_load_at(T_ADDRESS, IN_HEAP, rbx,
3999 Address(rbx, java_lang_invoke_ResolvedMethodName::vmtarget_offset()),
4000 noreg, noreg);
4001 __ movptr(Address(r15_thread, JavaThread::callee_target_offset()), rbx); // just in case callee is deoptimized
4002
4003 __ ret(0);
4004
4005 return start;
4006 }
4007
4008 void StubGenerator::generate_lookup_secondary_supers_table_stub() {
4009 StubGenStubId stub_id = StubGenStubId::lookup_secondary_supers_table_id;
4010 StubCodeMark mark(this, stub_id);
4011
4012 const Register
4013 r_super_klass = rax,
4014 r_sub_klass = rsi,
4015 result = rdi;
4016
4017 for (int slot = 0; slot < Klass::SECONDARY_SUPERS_TABLE_SIZE; slot++) {
4018 StubRoutines::_lookup_secondary_supers_table_stubs[slot] = __ pc();
4019 __ lookup_secondary_supers_table_const(r_sub_klass, r_super_klass,
4020 rdx, rcx, rbx, r11, // temps
4021 result,
4022 slot);
4023 __ ret(0);
4024 }
4025 }
4026
4027 // Slow path implementation for UseSecondarySupersTable.
4028 address StubGenerator::generate_lookup_secondary_supers_table_slow_path_stub() {
4029 StubGenStubId stub_id = StubGenStubId::lookup_secondary_supers_table_slow_path_id;
4030 StubCodeMark mark(this, stub_id);
4031
4032 address start = __ pc();
4033
4034 const Register
4035 r_super_klass = rax,
4036 r_array_base = rbx,
4037 r_array_index = rdx,
4038 r_sub_klass = rsi,
4039 r_bitmap = r11,
4040 result = rdi;
4041
4042 Label L_success;
4043 __ lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index, r_bitmap,
4044 rcx, rdi, // temps
4045 &L_success);
4046 // bind(L_failure);
4047 __ movl(result, 1);
4048 __ ret(0);
4049
4050 __ bind(L_success);
4051 __ movl(result, 0);
4052 __ ret(0);
4053
4054 return start;
4055 }
4056
4057 void StubGenerator::create_control_words() {
4058 // Round to nearest, 64-bit mode, exceptions masked, flags specialized
4059 StubRoutines::x86::_mxcsr_std = EnableX86ECoreOpts ? 0x1FBF : 0x1F80;
4060 // Round to zero, 64-bit mode, exceptions masked, flags specialized
4061 StubRoutines::x86::_mxcsr_rz = EnableX86ECoreOpts ? 0x7FBF : 0x7F80;
4062 }
4063
4064 // Initialization
4065 void StubGenerator::generate_initial_stubs() {
4066 // Generates all stubs and initializes the entry points
4067
4068 // This platform-specific settings are needed by generate_call_stub()
4069 create_control_words();
4070
4071 // Initialize table for unsafe copy memeory check.
4072 if (UnsafeMemoryAccess::_table == nullptr) {
4073 UnsafeMemoryAccess::create_table(16 + 4); // 16 for copyMemory; 4 for setMemory
4074 }
4075
4076 // entry points that exist in all platforms Note: This is code
4077 // that could be shared among different platforms - however the
4078 // benefit seems to be smaller than the disadvantage of having a
4079 // much more complicated generator structure. See also comment in
4080 // stubRoutines.hpp.
4081
4082 StubRoutines::_forward_exception_entry = generate_forward_exception();
4083
4084 StubRoutines::_call_stub_entry =
4085 generate_call_stub(StubRoutines::_call_stub_return_address);
4086
4087 // is referenced by megamorphic call
4088 StubRoutines::_catch_exception_entry = generate_catch_exception();
4089
4090 // atomic calls
4091 StubRoutines::_fence_entry = generate_orderaccess_fence();
4092
4093 // platform dependent
4094 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
4095
4096 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
4097
4098 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
4099 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
4100 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
4101 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
4102
4103 StubRoutines::x86::_float_sign_mask = generate_fp_mask(StubGenStubId::float_sign_mask_id, 0x7FFFFFFF7FFFFFFF);
4104 StubRoutines::x86::_float_sign_flip = generate_fp_mask(StubGenStubId::float_sign_flip_id, 0x8000000080000000);
4105 StubRoutines::x86::_double_sign_mask = generate_fp_mask(StubGenStubId::double_sign_mask_id, 0x7FFFFFFFFFFFFFFF);
4106 StubRoutines::x86::_double_sign_flip = generate_fp_mask(StubGenStubId::double_sign_flip_id, 0x8000000000000000);
4107
4108 if (UseCRC32Intrinsics) {
4109 // set table address before stub generation which use it
4110 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
4111 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
4112 }
4113
4114 if (UseCRC32CIntrinsics) {
4115 bool supports_clmul = VM_Version::supports_clmul();
4116 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
4117 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
4118 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
4119 }
4120
4121 if (VM_Version::supports_float16()) {
4122 // For results consistency both intrinsics should be enabled.
4123 // vmIntrinsics checks InlineIntrinsics flag, no need to check it here.
4124 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_float16ToFloat) &&
4125 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_floatToFloat16)) {
4126 StubRoutines::_hf2f = generate_float16ToFloat();
4127 StubRoutines::_f2hf = generate_floatToFloat16();
4128 }
4129 }
4130
4131 generate_libm_stubs();
4132
4133 StubRoutines::_fmod = generate_libmFmod(); // from stubGenerator_x86_64_fmod.cpp
4134 }
4135
4136 void StubGenerator::generate_continuation_stubs() {
4137 // Continuation stubs:
4138 StubRoutines::_cont_thaw = generate_cont_thaw();
4139 StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
4140 StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
4141 StubRoutines::_cont_preempt_stub = generate_cont_preempt_stub();
4142 }
4143
4144 void StubGenerator::generate_final_stubs() {
4145 // Generates the rest of stubs and initializes the entry points
4146
4147 // support for verify_oop (must happen after universe_init)
4148 if (VerifyOops) {
4149 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
4150 }
4151
4152 // arraycopy stubs used by compilers
4153 generate_arraycopy_stubs();
4154
4155 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
4156 if (bs_nm != nullptr) {
4157 StubRoutines::_method_entry_barrier = generate_method_entry_barrier();
4158 }
4159
4160 #ifdef COMPILER2
4161 if (UseSecondarySupersTable) {
4162 StubRoutines::_lookup_secondary_supers_table_slow_path_stub = generate_lookup_secondary_supers_table_slow_path_stub();
4163 if (! InlineSecondarySupersTest) {
4164 generate_lookup_secondary_supers_table_stub();
4165 }
4166 }
4167 #endif // COMPILER2
4168
4169 if (UseVectorizedMismatchIntrinsic) {
4170 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
4171 }
4172
4173 StubRoutines::_upcall_stub_exception_handler = generate_upcall_stub_exception_handler();
4174 StubRoutines::_upcall_stub_load_target = generate_upcall_stub_load_target();
4175 }
4176
4177 void StubGenerator::generate_compiler_stubs() {
4178 #if COMPILER2_OR_JVMCI
4179
4180 // Entry points that are C2 compiler specific.
4181
4182 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask(StubGenStubId::vector_float_sign_mask_id, 0x7FFFFFFF7FFFFFFF);
4183 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask(StubGenStubId::vector_float_sign_flip_id, 0x8000000080000000);
4184 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask(StubGenStubId::vector_double_sign_mask_id, 0x7FFFFFFFFFFFFFFF);
4185 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask(StubGenStubId::vector_double_sign_flip_id, 0x8000000000000000);
4186 StubRoutines::x86::_vector_all_bits_set = generate_vector_mask(StubGenStubId::vector_all_bits_set_id, 0xFFFFFFFFFFFFFFFF);
4187 StubRoutines::x86::_vector_int_mask_cmp_bits = generate_vector_mask(StubGenStubId::vector_int_mask_cmp_bits_id, 0x0000000100000001);
4188 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask(StubGenStubId::vector_short_to_byte_mask_id, 0x00ff00ff00ff00ff);
4189 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask();
4190 StubRoutines::x86::_vector_int_to_byte_mask = generate_vector_mask(StubGenStubId::vector_int_to_byte_mask_id, 0x000000ff000000ff);
4191 StubRoutines::x86::_vector_int_to_short_mask = generate_vector_mask(StubGenStubId::vector_int_to_short_mask_id, 0x0000ffff0000ffff);
4192 StubRoutines::x86::_vector_32_bit_mask = generate_vector_custom_i32(StubGenStubId::vector_32_bit_mask_id, Assembler::AVX_512bit,
4193 0xFFFFFFFF, 0, 0, 0);
4194 StubRoutines::x86::_vector_64_bit_mask = generate_vector_custom_i32(StubGenStubId::vector_64_bit_mask_id, Assembler::AVX_512bit,
4195 0xFFFFFFFF, 0xFFFFFFFF, 0, 0);
4196 StubRoutines::x86::_vector_int_shuffle_mask = generate_vector_mask(StubGenStubId::vector_int_shuffle_mask_id, 0x0302010003020100);
4197 StubRoutines::x86::_vector_byte_shuffle_mask = generate_vector_byte_shuffle_mask();
4198 StubRoutines::x86::_vector_short_shuffle_mask = generate_vector_mask(StubGenStubId::vector_short_shuffle_mask_id, 0x0100010001000100);
4199 StubRoutines::x86::_vector_long_shuffle_mask = generate_vector_mask(StubGenStubId::vector_long_shuffle_mask_id, 0x0000000100000000);
4200 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask(StubGenStubId::vector_long_sign_mask_id, 0x8000000000000000);
4201 StubRoutines::x86::_vector_iota_indices = generate_iota_indices();
4202 StubRoutines::x86::_vector_count_leading_zeros_lut = generate_count_leading_zeros_lut();
4203 StubRoutines::x86::_vector_reverse_bit_lut = generate_vector_reverse_bit_lut();
4204 StubRoutines::x86::_vector_reverse_byte_perm_mask_long = generate_vector_reverse_byte_perm_mask_long();
4205 StubRoutines::x86::_vector_reverse_byte_perm_mask_int = generate_vector_reverse_byte_perm_mask_int();
4206 StubRoutines::x86::_vector_reverse_byte_perm_mask_short = generate_vector_reverse_byte_perm_mask_short();
4207
4208 if (VM_Version::supports_avx2() && !VM_Version::supports_avx512vl()) {
4209 StubRoutines::x86::_compress_perm_table32 = generate_compress_perm_table(StubGenStubId::compress_perm_table32_id);
4210 StubRoutines::x86::_compress_perm_table64 = generate_compress_perm_table(StubGenStubId::compress_perm_table64_id);
4211 StubRoutines::x86::_expand_perm_table32 = generate_expand_perm_table(StubGenStubId::expand_perm_table32_id);
4212 StubRoutines::x86::_expand_perm_table64 = generate_expand_perm_table(StubGenStubId::expand_perm_table64_id);
4213 }
4214
4215 if (VM_Version::supports_avx2() && !VM_Version::supports_avx512_vpopcntdq()) {
4216 // lut implementation influenced by counting 1s algorithm from section 5-1 of Hackers' Delight.
4217 StubRoutines::x86::_vector_popcount_lut = generate_popcount_avx_lut();
4218 }
4219
4220 generate_aes_stubs();
4221
4222 generate_ghash_stubs();
4223
4224 generate_chacha_stubs();
4225
4226 generate_sha3_stubs();
4227
4228 // data cache line writeback
4229 StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
4230 StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
4231
4232 #ifdef COMPILER2
4233 if ((UseAVX == 2) && EnableX86ECoreOpts) {
4234 generate_string_indexof(StubRoutines::_string_indexof_array);
4235 }
4236 #endif
4237
4238 if (UseAdler32Intrinsics) {
4239 StubRoutines::_updateBytesAdler32 = generate_updateBytesAdler32();
4240 }
4241
4242 if (UsePoly1305Intrinsics) {
4243 StubRoutines::_poly1305_processBlocks = generate_poly1305_processBlocks();
4244 }
4245
4246 if (UseIntPolyIntrinsics) {
4247 StubRoutines::_intpoly_montgomeryMult_P256 = generate_intpoly_montgomeryMult_P256();
4248 StubRoutines::_intpoly_assign = generate_intpoly_assign();
4249 }
4250
4251 if (UseMD5Intrinsics) {
4252 StubRoutines::_md5_implCompress = generate_md5_implCompress(StubGenStubId::md5_implCompress_id);
4253 StubRoutines::_md5_implCompressMB = generate_md5_implCompress(StubGenStubId::md5_implCompressMB_id);
4254 }
4255
4256 if (UseSHA1Intrinsics) {
4257 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
4258 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
4259 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(StubGenStubId::sha1_implCompress_id);
4260 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(StubGenStubId::sha1_implCompressMB_id);
4261 }
4262
4263 if (UseSHA256Intrinsics) {
4264 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
4265 char* dst = (char*)StubRoutines::x86::_k256_W;
4266 char* src = (char*)StubRoutines::x86::_k256;
4267 for (int ii = 0; ii < 16; ++ii) {
4268 memcpy(dst + 32 * ii, src + 16 * ii, 16);
4269 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
4270 }
4271 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
4272 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
4273 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(StubGenStubId::sha256_implCompress_id);
4274 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(StubGenStubId::sha256_implCompressMB_id);
4275 }
4276
4277 if (UseSHA512Intrinsics) {
4278 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
4279 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
4280 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(StubGenStubId::sha512_implCompress_id);
4281 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(StubGenStubId::sha512_implCompressMB_id);
4282 }
4283
4284 if (UseBASE64Intrinsics) {
4285 if(VM_Version::supports_avx2()) {
4286 StubRoutines::x86::_avx2_shuffle_base64 = base64_avx2_shuffle_addr();
4287 StubRoutines::x86::_avx2_input_mask_base64 = base64_avx2_input_mask_addr();
4288 StubRoutines::x86::_avx2_lut_base64 = base64_avx2_lut_addr();
4289 StubRoutines::x86::_avx2_decode_tables_base64 = base64_AVX2_decode_tables_addr();
4290 StubRoutines::x86::_avx2_decode_lut_tables_base64 = base64_AVX2_decode_LUT_tables_addr();
4291 }
4292 StubRoutines::x86::_encoding_table_base64 = base64_encoding_table_addr();
4293 if (VM_Version::supports_avx512_vbmi()) {
4294 StubRoutines::x86::_shuffle_base64 = base64_shuffle_addr();
4295 StubRoutines::x86::_lookup_lo_base64 = base64_vbmi_lookup_lo_addr();
4296 StubRoutines::x86::_lookup_hi_base64 = base64_vbmi_lookup_hi_addr();
4297 StubRoutines::x86::_lookup_lo_base64url = base64_vbmi_lookup_lo_url_addr();
4298 StubRoutines::x86::_lookup_hi_base64url = base64_vbmi_lookup_hi_url_addr();
4299 StubRoutines::x86::_pack_vec_base64 = base64_vbmi_pack_vec_addr();
4300 StubRoutines::x86::_join_0_1_base64 = base64_vbmi_join_0_1_addr();
4301 StubRoutines::x86::_join_1_2_base64 = base64_vbmi_join_1_2_addr();
4302 StubRoutines::x86::_join_2_3_base64 = base64_vbmi_join_2_3_addr();
4303 }
4304 StubRoutines::x86::_decoding_table_base64 = base64_decoding_table_addr();
4305 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
4306 StubRoutines::_base64_decodeBlock = generate_base64_decodeBlock();
4307 }
4308
4309 #ifdef COMPILER2
4310 if (UseMultiplyToLenIntrinsic) {
4311 StubRoutines::_multiplyToLen = generate_multiplyToLen();
4312 }
4313 if (UseSquareToLenIntrinsic) {
4314 StubRoutines::_squareToLen = generate_squareToLen();
4315 }
4316 if (UseMulAddIntrinsic) {
4317 StubRoutines::_mulAdd = generate_mulAdd();
4318 }
4319 if (VM_Version::supports_avx512_vbmi2()) {
4320 StubRoutines::_bigIntegerRightShiftWorker = generate_bigIntegerRightShift();
4321 StubRoutines::_bigIntegerLeftShiftWorker = generate_bigIntegerLeftShift();
4322 }
4323 if (UseMontgomeryMultiplyIntrinsic) {
4324 StubRoutines::_montgomeryMultiply
4325 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
4326 }
4327 if (UseMontgomerySquareIntrinsic) {
4328 StubRoutines::_montgomerySquare
4329 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
4330 }
4331
4332 // Load x86_64_sort library on supported hardware to enable SIMD sort and partition intrinsics
4333
4334 if (VM_Version::is_intel() && (VM_Version::supports_avx512dq() || VM_Version::supports_avx2())) {
4335 void *libsimdsort = nullptr;
4336 char ebuf_[1024];
4337 char dll_name_simd_sort[JVM_MAXPATHLEN];
4338 if (os::dll_locate_lib(dll_name_simd_sort, sizeof(dll_name_simd_sort), Arguments::get_dll_dir(), "simdsort")) {
4339 libsimdsort = os::dll_load(dll_name_simd_sort, ebuf_, sizeof ebuf_);
4340 }
4341 // Get addresses for SIMD sort and partition routines
4342 if (libsimdsort != nullptr) {
4343 log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "simdsort" JNI_LIB_SUFFIX, p2i(libsimdsort));
4344
4345 snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512dq() ? "avx512_sort" : "avx2_sort");
4346 StubRoutines::_array_sort = (address)os::dll_lookup(libsimdsort, ebuf_);
4347
4348 snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512dq() ? "avx512_partition" : "avx2_partition");
4349 StubRoutines::_array_partition = (address)os::dll_lookup(libsimdsort, ebuf_);
4350 }
4351 }
4352
4353 // Get svml stub routine addresses
4354 void *libjsvml = nullptr;
4355 char ebuf[1024];
4356 char dll_name[JVM_MAXPATHLEN];
4357 if (os::dll_locate_lib(dll_name, sizeof(dll_name), Arguments::get_dll_dir(), "jsvml")) {
4358 libjsvml = os::dll_load(dll_name, ebuf, sizeof ebuf);
4359 }
4360 if (libjsvml != nullptr) {
4361 // SVML method naming convention
4362 // All the methods are named as __jsvml_op<T><N>_ha_<VV>
4363 // Where:
4364 // ha stands for high accuracy
4365 // <T> is optional to indicate float/double
4366 // Set to f for vector float operation
4367 // Omitted for vector double operation
4368 // <N> is the number of elements in the vector
4369 // 1, 2, 4, 8, 16
4370 // e.g. 128 bit float vector has 4 float elements
4371 // <VV> indicates the avx/sse level:
4372 // z0 is AVX512, l9 is AVX2, e9 is AVX1 and ex is for SSE2
4373 // e.g. __jsvml_expf16_ha_z0 is the method for computing 16 element vector float exp using AVX 512 insns
4374 // __jsvml_exp8_ha_z0 is the method for computing 8 element vector double exp using AVX 512 insns
4375
4376 log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "jsvml" JNI_LIB_SUFFIX, p2i(libjsvml));
4377 if (UseAVX > 2) {
4378 for (int op = 0; op < VectorSupport::NUM_VECTOR_OP_MATH; op++) {
4379 int vop = VectorSupport::VECTOR_OP_MATH_START + op;
4380 if ((!VM_Version::supports_avx512dq()) &&
4381 (vop == VectorSupport::VECTOR_OP_LOG || vop == VectorSupport::VECTOR_OP_LOG10 || vop == VectorSupport::VECTOR_OP_POW)) {
4382 continue;
4383 }
4384 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf16_ha_z0", VectorSupport::mathname[op]);
4385 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4386
4387 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s8_ha_z0", VectorSupport::mathname[op]);
4388 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4389 }
4390 }
4391 const char* avx_sse_str = (UseAVX >= 2) ? "l9" : ((UseAVX == 1) ? "e9" : "ex");
4392 for (int op = 0; op < VectorSupport::NUM_VECTOR_OP_MATH; op++) {
4393 int vop = VectorSupport::VECTOR_OP_MATH_START + op;
4394 if (vop == VectorSupport::VECTOR_OP_POW) {
4395 continue;
4396 }
4397 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4398 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4399
4400 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4401 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4402
4403 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf8_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4404 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4405
4406 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s1_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4407 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4408
4409 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s2_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4410 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4411
4412 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4413 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4414 }
4415 }
4416
4417 #endif // COMPILER2
4418 #endif // COMPILER2_OR_JVMCI
4419 }
4420
4421 StubGenerator::StubGenerator(CodeBuffer* code, StubGenBlobId blob_id) : StubCodeGenerator(code, blob_id) {
4422 switch(blob_id) {
4423 case initial_id:
4424 generate_initial_stubs();
4425 break;
4426 case continuation_id:
4427 generate_continuation_stubs();
4428 break;
4429 case compiler_id:
4430 generate_compiler_stubs();
4431 break;
4432 case final_id:
4433 generate_final_stubs();
4434 break;
4435 default:
4436 fatal("unexpected blob id: %d", blob_id);
4437 break;
4438 };
4439 }
4440
4441 void StubGenerator_generate(CodeBuffer* code, StubGenBlobId blob_id) {
4442 StubGenerator g(code, blob_id);
4443 }
4444
4445 #undef __