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/assembler.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "classfile/javaClasses.hpp"
28 #include "classfile/vmIntrinsics.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 "utilities/macros.hpp"
43 #include "vmreg_x86.inline.hpp"
44 #include "stubGenerator_x86_64.hpp"
45 #ifdef COMPILER2
46 #include "opto/runtime.hpp"
47 #include "opto/c2_globals.hpp"
48 #endif
49 #if INCLUDE_JVMCI
50 #include "jvmci/jvmci_globals.hpp"
51 #endif
52
53 // For a more detailed description of the stub routine structure
54 // see the comment in stubRoutines.hpp
55
56 #define __ _masm->
57 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
58
59 #ifdef PRODUCT
60 #define BLOCK_COMMENT(str) /* nothing */
61 #else
62 #define BLOCK_COMMENT(str) __ block_comment(str)
63 #endif // PRODUCT
64
65 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
66
67 //
68 // Linux Arguments:
69 // c_rarg0: call wrapper address address
70 // c_rarg1: result address
71 // c_rarg2: result type BasicType
72 // c_rarg3: method Method*
73 // c_rarg4: (interpreter) entry point address
74 // c_rarg5: parameters intptr_t*
75 // 16(rbp): parameter size (in words) int
76 // 24(rbp): thread Thread*
77 //
78 // [ return_from_Java ] <--- rsp
79 // [ argument word n ]
80 // ...
81 // -12 [ argument word 1 ]
82 // -11 [ saved r15 ] <--- rsp_after_call
83 // -10 [ saved r14 ]
84 // -9 [ saved r13 ]
85 // -8 [ saved r12 ]
86 // -7 [ saved rbx ]
87 // -6 [ call wrapper ]
88 // -5 [ result ]
89 // -4 [ result type ]
90 // -3 [ method ]
91 // -2 [ entry point ]
92 // -1 [ parameters ]
93 // 0 [ saved rbp ] <--- rbp
94 // 1 [ return address ]
95 // 2 [ parameter size ]
96 // 3 [ thread ]
97 //
98 // Windows Arguments:
99 // c_rarg0: call wrapper address address
100 // c_rarg1: result address
101 // c_rarg2: result type BasicType
102 // c_rarg3: method Method*
103 // 48(rbp): (interpreter) entry point address
104 // 56(rbp): parameters intptr_t*
105 // 64(rbp): parameter size (in words) int
106 // 72(rbp): thread Thread*
107 //
108 // [ return_from_Java ] <--- rsp
109 // [ argument word n ]
110 // ...
111 // -28 [ argument word 1 ]
112 // -27 [ saved xmm15 ] <--- rsp after_call
113 // [ saved xmm7-xmm14 ]
114 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
115 // -7 [ saved r15 ]
116 // -6 [ saved r14 ]
117 // -5 [ saved r13 ]
118 // -4 [ saved r12 ]
119 // -3 [ saved rdi ]
120 // -2 [ saved rsi ]
121 // -1 [ saved rbx ]
122 // 0 [ saved rbp ] <--- rbp
123 // 1 [ return address ]
124 // 2 [ call wrapper ]
125 // 3 [ result ]
126 // 4 [ result type ]
127 // 5 [ method ]
128 // 6 [ entry point ]
129 // 7 [ parameters ]
130 // 8 [ parameter size ]
131 // 9 [ thread ]
132 //
133 // Windows reserves the callers stack space for arguments 1-4.
134 // We spill c_rarg0-c_rarg3 to this space.
135
136 // Call stub stack layout word offsets from rbp
137 #ifdef _WIN64
138 enum call_stub_layout {
139 xmm_save_first = 6, // save from xmm6
140 xmm_save_last = 15, // to xmm15
141 xmm_save_base = -9,
142 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
143 r15_off = -7,
144 r14_off = -6,
145 r13_off = -5,
146 r12_off = -4,
147 rdi_off = -3,
148 rsi_off = -2,
149 rbx_off = -1,
150 rbp_off = 0,
151 retaddr_off = 1,
152 call_wrapper_off = 2,
153 result_off = 3,
154 result_type_off = 4,
155 method_off = 5,
156 entry_point_off = 6,
157 parameters_off = 7,
158 parameter_size_off = 8,
159 thread_off = 9
160 };
161
162 static Address xmm_save(int reg) {
163 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
164 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
165 }
166 #else // !_WIN64
167 enum call_stub_layout {
168 rsp_after_call_off = -12,
169 mxcsr_off = rsp_after_call_off,
170 r15_off = -11,
171 r14_off = -10,
172 r13_off = -9,
173 r12_off = -8,
174 rbx_off = -7,
175 call_wrapper_off = -6,
176 result_off = -5,
177 result_type_off = -4,
178 method_off = -3,
179 entry_point_off = -2,
180 parameters_off = -1,
181 rbp_off = 0,
182 retaddr_off = 1,
183 parameter_size_off = 2,
184 thread_off = 3
185 };
186 #endif // _WIN64
187
188 address StubGenerator::generate_call_stub(address& return_address) {
189
190 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
191 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
192 "adjust this code");
193 StubGenStubId stub_id = StubGenStubId::call_stub_id;
194 StubCodeMark mark(this, stub_id);
195 address start = __ pc();
196
197 // same as in generate_catch_exception()!
198 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
199
200 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
201 const Address result (rbp, result_off * wordSize);
202 const Address result_type (rbp, result_type_off * wordSize);
203 const Address method (rbp, method_off * wordSize);
204 const Address entry_point (rbp, entry_point_off * wordSize);
205 const Address parameters (rbp, parameters_off * wordSize);
206 const Address parameter_size(rbp, parameter_size_off * wordSize);
207
208 // same as in generate_catch_exception()!
209 const Address thread (rbp, thread_off * wordSize);
210
211 const Address r15_save(rbp, r15_off * wordSize);
212 const Address r14_save(rbp, r14_off * wordSize);
213 const Address r13_save(rbp, r13_off * wordSize);
214 const Address r12_save(rbp, r12_off * wordSize);
215 const Address rbx_save(rbp, rbx_off * wordSize);
216
217 // stub code
218 __ enter();
219 __ subptr(rsp, -rsp_after_call_off * wordSize);
220
221 // save register parameters
222 #ifndef _WIN64
223 __ movptr(parameters, c_rarg5); // parameters
224 __ movptr(entry_point, c_rarg4); // entry_point
225 #endif
226
227 __ movptr(method, c_rarg3); // method
228 __ movl(result_type, c_rarg2); // result type
229 __ movptr(result, c_rarg1); // result
230 __ movptr(call_wrapper, c_rarg0); // call wrapper
231
232 // save regs belonging to calling function
233 __ movptr(rbx_save, rbx);
234 __ movptr(r12_save, r12);
235 __ movptr(r13_save, r13);
236 __ movptr(r14_save, r14);
237 __ movptr(r15_save, r15);
238
239 #ifdef _WIN64
240 int last_reg = 15;
241 for (int i = xmm_save_first; i <= last_reg; i++) {
242 __ movdqu(xmm_save(i), as_XMMRegister(i));
243 }
244
245 const Address rdi_save(rbp, rdi_off * wordSize);
246 const Address rsi_save(rbp, rsi_off * wordSize);
247
248 __ movptr(rsi_save, rsi);
249 __ movptr(rdi_save, rdi);
250 #else
251 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
252 {
253 Label skip_ldmx;
254 __ cmp32_mxcsr_std(mxcsr_save, rax, rscratch1);
255 __ jcc(Assembler::equal, skip_ldmx);
256 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
257 __ ldmxcsr(mxcsr_std, rscratch1);
258 __ bind(skip_ldmx);
259 }
260 #endif
261
262 // Load up thread register
263 __ movptr(r15_thread, thread);
264 __ reinit_heapbase();
265
266 #ifdef ASSERT
267 // make sure we have no pending exceptions
268 {
269 Label L;
270 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
271 __ jcc(Assembler::equal, L);
272 __ stop("StubRoutines::call_stub: entered with pending exception");
273 __ bind(L);
274 }
275 #endif
276
277 // pass parameters if any
278 BLOCK_COMMENT("pass parameters if any");
279 Label parameters_done;
280 __ movl(c_rarg3, parameter_size);
281 __ testl(c_rarg3, c_rarg3);
282 __ jcc(Assembler::zero, parameters_done);
283
284 Label loop;
285 __ movptr(c_rarg2, parameters); // parameter pointer
286 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
287 __ BIND(loop);
288 __ movptr(rax, Address(c_rarg2, 0));// get parameter
289 __ addptr(c_rarg2, wordSize); // advance to next parameter
290 __ decrementl(c_rarg1); // decrement counter
291 __ push(rax); // pass parameter
292 __ jcc(Assembler::notZero, loop);
293
294 // call Java function
295 __ BIND(parameters_done);
296 __ movptr(rbx, method); // get Method*
297 __ movptr(c_rarg1, entry_point); // get entry_point
298 __ mov(r13, rsp); // set sender sp
299 BLOCK_COMMENT("call Java function");
300 __ call(c_rarg1);
301
302 BLOCK_COMMENT("call_stub_return_address:");
303 return_address = __ pc();
304
305 // store result depending on type (everything that is not
306 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
307 __ movptr(r13, result);
308 Label is_long, is_float, is_double, check_prim, exit;
309 __ movl(rbx, result_type);
310 __ cmpl(rbx, T_OBJECT);
311 __ jcc(Assembler::equal, check_prim);
312 __ cmpl(rbx, T_LONG);
313 __ jcc(Assembler::equal, is_long);
314 __ cmpl(rbx, T_FLOAT);
315 __ jcc(Assembler::equal, is_float);
316 __ cmpl(rbx, T_DOUBLE);
317 __ jcc(Assembler::equal, is_double);
318 #ifdef ASSERT
319 // make sure the type is INT
320 {
321 Label L;
322 __ cmpl(rbx, T_INT);
323 __ jcc(Assembler::equal, L);
324 __ stop("StubRoutines::call_stub: unexpected result type");
325 __ bind(L);
326 }
327 #endif
328
329 // handle T_INT case
330 __ movl(Address(r13, 0), rax);
331
332 __ BIND(exit);
333
334 // pop parameters
335 __ lea(rsp, rsp_after_call);
336
337 #ifdef ASSERT
338 // verify that threads correspond
339 {
340 Label L1, L2, L3;
341 __ cmpptr(r15_thread, thread);
342 __ jcc(Assembler::equal, L1);
343 __ stop("StubRoutines::call_stub: r15_thread is corrupted");
344 __ bind(L1);
345 __ get_thread(rbx);
346 __ cmpptr(r15_thread, thread);
347 __ jcc(Assembler::equal, L2);
348 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
349 __ bind(L2);
350 __ cmpptr(r15_thread, rbx);
351 __ jcc(Assembler::equal, L3);
352 __ stop("StubRoutines::call_stub: threads must correspond");
353 __ bind(L3);
354 }
355 #endif
356
357 __ pop_cont_fastpath();
358
359 // restore regs belonging to calling function
360 #ifdef _WIN64
361 // emit the restores for xmm regs
362 for (int i = xmm_save_first; i <= last_reg; i++) {
363 __ movdqu(as_XMMRegister(i), xmm_save(i));
364 }
365 #endif
366 __ movptr(r15, r15_save);
367 __ movptr(r14, r14_save);
368 __ movptr(r13, r13_save);
369 __ movptr(r12, r12_save);
370 __ movptr(rbx, rbx_save);
371
372 #ifdef _WIN64
373 __ movptr(rdi, rdi_save);
374 __ movptr(rsi, rsi_save);
375 #else
376 __ ldmxcsr(mxcsr_save);
377 #endif
378
379 // restore rsp
380 __ addptr(rsp, -rsp_after_call_off * wordSize);
381
382 // return
383 __ vzeroupper();
384 __ pop(rbp);
385 __ ret(0);
386
387 // handle return types different from T_INT
388 __ BIND(check_prim);
389 if (InlineTypeReturnedAsFields) {
390 // Check for scalarized return value
391 __ testptr(rax, 1);
392 __ jcc(Assembler::zero, is_long);
393 // Load pack handler address
394 __ andptr(rax, -2);
395 __ movptr(rax, Address(rax, InstanceKlass::adr_inlineklass_fixed_block_offset()));
396 __ movptr(rbx, Address(rax, InlineKlass::pack_handler_jobject_offset()));
397 // Call pack handler to initialize the buffer
398 __ call(rbx);
399 __ jmp(exit);
400 }
401 __ BIND(is_long);
402 __ movq(Address(r13, 0), rax);
403 __ jmp(exit);
404
405 __ BIND(is_float);
406 __ movflt(Address(r13, 0), xmm0);
407 __ jmp(exit);
408
409 __ BIND(is_double);
410 __ movdbl(Address(r13, 0), xmm0);
411 __ jmp(exit);
412
413 return start;
414 }
415
416 // Return point for a Java call if there's an exception thrown in
417 // Java code. The exception is caught and transformed into a
418 // pending exception stored in JavaThread that can be tested from
419 // within the VM.
420 //
421 // Note: Usually the parameters are removed by the callee. In case
422 // of an exception crossing an activation frame boundary, that is
423 // not the case if the callee is compiled code => need to setup the
424 // rsp.
425 //
426 // rax: exception oop
427
428 address StubGenerator::generate_catch_exception() {
429 StubGenStubId stub_id = StubGenStubId::catch_exception_id;
430 StubCodeMark mark(this, stub_id);
431 address start = __ pc();
432
433 // same as in generate_call_stub():
434 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
435 const Address thread (rbp, thread_off * wordSize);
436
437 #ifdef ASSERT
438 // verify that threads correspond
439 {
440 Label L1, L2, L3;
441 __ cmpptr(r15_thread, thread);
442 __ jcc(Assembler::equal, L1);
443 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
444 __ bind(L1);
445 __ get_thread(rbx);
446 __ cmpptr(r15_thread, thread);
447 __ jcc(Assembler::equal, L2);
448 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
449 __ bind(L2);
450 __ cmpptr(r15_thread, rbx);
451 __ jcc(Assembler::equal, L3);
452 __ stop("StubRoutines::catch_exception: threads must correspond");
453 __ bind(L3);
454 }
455 #endif
456
457 // set pending exception
458 __ verify_oop(rax);
459
460 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
461 __ lea(rscratch1, ExternalAddress((address)__FILE__));
462 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
463 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
464
465 // complete return to VM
466 assert(StubRoutines::_call_stub_return_address != nullptr,
467 "_call_stub_return_address must have been generated before");
468 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
469
470 return start;
471 }
472
473 // Continuation point for runtime calls returning with a pending
474 // exception. The pending exception check happened in the runtime
475 // or native call stub. The pending exception in Thread is
476 // converted into a Java-level exception.
477 //
478 // Contract with Java-level exception handlers:
479 // rax: exception
480 // rdx: throwing pc
481 //
482 // NOTE: At entry of this stub, exception-pc must be on stack !!
483
484 address StubGenerator::generate_forward_exception() {
485 StubGenStubId stub_id = StubGenStubId::forward_exception_id;
486 StubCodeMark mark(this, stub_id);
487 address start = __ pc();
488
489 // Upon entry, the sp points to the return address returning into
490 // Java (interpreted or compiled) code; i.e., the return address
491 // becomes the throwing pc.
492 //
493 // Arguments pushed before the runtime call are still on the stack
494 // but the exception handler will reset the stack pointer ->
495 // ignore them. A potential result in registers can be ignored as
496 // well.
497
498 #ifdef ASSERT
499 // make sure this code is only executed if there is a pending exception
500 {
501 Label L;
502 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
503 __ jcc(Assembler::notEqual, L);
504 __ stop("StubRoutines::forward exception: no pending exception (1)");
505 __ bind(L);
506 }
507 #endif
508
509 // compute exception handler into rbx
510 __ movptr(c_rarg0, Address(rsp, 0));
511 BLOCK_COMMENT("call exception_handler_for_return_address");
512 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
513 SharedRuntime::exception_handler_for_return_address),
514 r15_thread, c_rarg0);
515 __ mov(rbx, rax);
516
517 // setup rax & rdx, remove return address & clear pending exception
518 __ pop(rdx);
519 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
520 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
521
522 #ifdef ASSERT
523 // make sure exception is set
524 {
525 Label L;
526 __ testptr(rax, rax);
527 __ jcc(Assembler::notEqual, L);
528 __ stop("StubRoutines::forward exception: no pending exception (2)");
529 __ bind(L);
530 }
531 #endif
532
533 // continue at exception handler (return address removed)
534 // rax: exception
535 // rbx: exception handler
536 // rdx: throwing pc
537 __ verify_oop(rax);
538 __ jmp(rbx);
539
540 return start;
541 }
542
543 // Support for intptr_t OrderAccess::fence()
544 //
545 // Arguments :
546 //
547 // Result:
548 address StubGenerator::generate_orderaccess_fence() {
549 StubGenStubId stub_id = StubGenStubId::fence_id;
550 StubCodeMark mark(this, stub_id);
551 address start = __ pc();
552
553 __ membar(Assembler::StoreLoad);
554 __ ret(0);
555
556 return start;
557 }
558
559
560 // Support for intptr_t get_previous_sp()
561 //
562 // This routine is used to find the previous stack pointer for the
563 // caller.
564 address StubGenerator::generate_get_previous_sp() {
565 StubGenStubId stub_id = StubGenStubId::get_previous_sp_id;
566 StubCodeMark mark(this, stub_id);
567 address start = __ pc();
568
569 __ movptr(rax, rsp);
570 __ addptr(rax, 8); // return address is at the top of the stack.
571 __ ret(0);
572
573 return start;
574 }
575
576 //----------------------------------------------------------------------------------------------------
577 // Support for void verify_mxcsr()
578 //
579 // This routine is used with -Xcheck:jni to verify that native
580 // JNI code does not return to Java code without restoring the
581 // MXCSR register to our expected state.
582
583 address StubGenerator::generate_verify_mxcsr() {
584 StubGenStubId stub_id = StubGenStubId::verify_mxcsr_id;
585 StubCodeMark mark(this, stub_id);
586 address start = __ pc();
587
588 const Address mxcsr_save(rsp, 0);
589
590 if (CheckJNICalls) {
591 Label ok_ret;
592 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
593 __ push(rax);
594 __ subptr(rsp, wordSize); // allocate a temp location
595 __ cmp32_mxcsr_std(mxcsr_save, rax, rscratch1);
596 __ jcc(Assembler::equal, ok_ret);
597
598 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
599
600 __ ldmxcsr(mxcsr_std, rscratch1);
601
602 __ bind(ok_ret);
603 __ addptr(rsp, wordSize);
604 __ pop(rax);
605 }
606
607 __ ret(0);
608
609 return start;
610 }
611
612 address StubGenerator::generate_f2i_fixup() {
613 StubGenStubId stub_id = StubGenStubId::f2i_fixup_id;
614 StubCodeMark mark(this, stub_id);
615 Address inout(rsp, 5 * wordSize); // return address + 4 saves
616
617 address start = __ pc();
618
619 Label L;
620
621 __ push(rax);
622 __ push(c_rarg3);
623 __ push(c_rarg2);
624 __ push(c_rarg1);
625
626 __ movl(rax, 0x7f800000);
627 __ xorl(c_rarg3, c_rarg3);
628 __ movl(c_rarg2, inout);
629 __ movl(c_rarg1, c_rarg2);
630 __ andl(c_rarg1, 0x7fffffff);
631 __ cmpl(rax, c_rarg1); // NaN? -> 0
632 __ jcc(Assembler::negative, L);
633 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
634 __ movl(c_rarg3, 0x80000000);
635 __ movl(rax, 0x7fffffff);
636 __ cmovl(Assembler::positive, c_rarg3, rax);
637
638 __ bind(L);
639 __ movptr(inout, c_rarg3);
640
641 __ pop(c_rarg1);
642 __ pop(c_rarg2);
643 __ pop(c_rarg3);
644 __ pop(rax);
645
646 __ ret(0);
647
648 return start;
649 }
650
651 address StubGenerator::generate_f2l_fixup() {
652 StubGenStubId stub_id = StubGenStubId::f2l_fixup_id;
653 StubCodeMark mark(this, stub_id);
654 Address inout(rsp, 5 * wordSize); // return address + 4 saves
655 address start = __ pc();
656
657 Label L;
658
659 __ push(rax);
660 __ push(c_rarg3);
661 __ push(c_rarg2);
662 __ push(c_rarg1);
663
664 __ movl(rax, 0x7f800000);
665 __ xorl(c_rarg3, c_rarg3);
666 __ movl(c_rarg2, inout);
667 __ movl(c_rarg1, c_rarg2);
668 __ andl(c_rarg1, 0x7fffffff);
669 __ cmpl(rax, c_rarg1); // NaN? -> 0
670 __ jcc(Assembler::negative, L);
671 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
672 __ mov64(c_rarg3, 0x8000000000000000);
673 __ mov64(rax, 0x7fffffffffffffff);
674 __ cmov(Assembler::positive, c_rarg3, rax);
675
676 __ bind(L);
677 __ movptr(inout, c_rarg3);
678
679 __ pop(c_rarg1);
680 __ pop(c_rarg2);
681 __ pop(c_rarg3);
682 __ pop(rax);
683
684 __ ret(0);
685
686 return start;
687 }
688
689 address StubGenerator::generate_d2i_fixup() {
690 StubGenStubId stub_id = StubGenStubId::d2i_fixup_id;
691 StubCodeMark mark(this, stub_id);
692 Address inout(rsp, 6 * wordSize); // return address + 5 saves
693
694 address start = __ pc();
695
696 Label L;
697
698 __ push(rax);
699 __ push(c_rarg3);
700 __ push(c_rarg2);
701 __ push(c_rarg1);
702 __ push(c_rarg0);
703
704 __ movl(rax, 0x7ff00000);
705 __ movq(c_rarg2, inout);
706 __ movl(c_rarg3, c_rarg2);
707 __ mov(c_rarg1, c_rarg2);
708 __ mov(c_rarg0, c_rarg2);
709 __ negl(c_rarg3);
710 __ shrptr(c_rarg1, 0x20);
711 __ orl(c_rarg3, c_rarg2);
712 __ andl(c_rarg1, 0x7fffffff);
713 __ xorl(c_rarg2, c_rarg2);
714 __ shrl(c_rarg3, 0x1f);
715 __ orl(c_rarg1, c_rarg3);
716 __ cmpl(rax, c_rarg1);
717 __ jcc(Assembler::negative, L); // NaN -> 0
718 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
719 __ movl(c_rarg2, 0x80000000);
720 __ movl(rax, 0x7fffffff);
721 __ cmov(Assembler::positive, c_rarg2, rax);
722
723 __ bind(L);
724 __ movptr(inout, c_rarg2);
725
726 __ pop(c_rarg0);
727 __ pop(c_rarg1);
728 __ pop(c_rarg2);
729 __ pop(c_rarg3);
730 __ pop(rax);
731
732 __ ret(0);
733
734 return start;
735 }
736
737 address StubGenerator::generate_d2l_fixup() {
738 StubGenStubId stub_id = StubGenStubId::d2l_fixup_id;
739 StubCodeMark mark(this, stub_id);
740 Address inout(rsp, 6 * wordSize); // return address + 5 saves
741
742 address start = __ pc();
743
744 Label L;
745
746 __ push(rax);
747 __ push(c_rarg3);
748 __ push(c_rarg2);
749 __ push(c_rarg1);
750 __ push(c_rarg0);
751
752 __ movl(rax, 0x7ff00000);
753 __ movq(c_rarg2, inout);
754 __ movl(c_rarg3, c_rarg2);
755 __ mov(c_rarg1, c_rarg2);
756 __ mov(c_rarg0, c_rarg2);
757 __ negl(c_rarg3);
758 __ shrptr(c_rarg1, 0x20);
759 __ orl(c_rarg3, c_rarg2);
760 __ andl(c_rarg1, 0x7fffffff);
761 __ xorl(c_rarg2, c_rarg2);
762 __ shrl(c_rarg3, 0x1f);
763 __ orl(c_rarg1, c_rarg3);
764 __ cmpl(rax, c_rarg1);
765 __ jcc(Assembler::negative, L); // NaN -> 0
766 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
767 __ mov64(c_rarg2, 0x8000000000000000);
768 __ mov64(rax, 0x7fffffffffffffff);
769 __ cmovq(Assembler::positive, c_rarg2, rax);
770
771 __ bind(L);
772 __ movq(inout, c_rarg2);
773
774 __ pop(c_rarg0);
775 __ pop(c_rarg1);
776 __ pop(c_rarg2);
777 __ pop(c_rarg3);
778 __ pop(rax);
779
780 __ ret(0);
781
782 return start;
783 }
784
785 address StubGenerator::generate_count_leading_zeros_lut() {
786 __ align64();
787 StubGenStubId stub_id = StubGenStubId::vector_count_leading_zeros_lut_id;
788 StubCodeMark mark(this, stub_id);
789 address start = __ pc();
790
791 __ emit_data64(0x0101010102020304, relocInfo::none);
792 __ emit_data64(0x0000000000000000, relocInfo::none);
793 __ emit_data64(0x0101010102020304, relocInfo::none);
794 __ emit_data64(0x0000000000000000, relocInfo::none);
795 __ emit_data64(0x0101010102020304, relocInfo::none);
796 __ emit_data64(0x0000000000000000, relocInfo::none);
797 __ emit_data64(0x0101010102020304, relocInfo::none);
798 __ emit_data64(0x0000000000000000, relocInfo::none);
799
800 return start;
801 }
802
803 address StubGenerator::generate_popcount_avx_lut() {
804 __ align64();
805 StubGenStubId stub_id = StubGenStubId::vector_popcount_lut_id;
806 StubCodeMark mark(this, stub_id);
807 address start = __ pc();
808
809 __ emit_data64(0x0302020102010100, relocInfo::none);
810 __ emit_data64(0x0403030203020201, relocInfo::none);
811 __ emit_data64(0x0302020102010100, relocInfo::none);
812 __ emit_data64(0x0403030203020201, relocInfo::none);
813 __ emit_data64(0x0302020102010100, relocInfo::none);
814 __ emit_data64(0x0403030203020201, relocInfo::none);
815 __ emit_data64(0x0302020102010100, relocInfo::none);
816 __ emit_data64(0x0403030203020201, relocInfo::none);
817
818 return start;
819 }
820
821 address StubGenerator::generate_iota_indices() {
822 __ align(CodeEntryAlignment);
823 StubGenStubId stub_id = StubGenStubId::vector_iota_indices_id;
824 StubCodeMark mark(this, stub_id);
825 address start = __ pc();
826 // B
827 __ emit_data64(0x0706050403020100, relocInfo::none);
828 __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
829 __ emit_data64(0x1716151413121110, relocInfo::none);
830 __ emit_data64(0x1F1E1D1C1B1A1918, relocInfo::none);
831 __ emit_data64(0x2726252423222120, relocInfo::none);
832 __ emit_data64(0x2F2E2D2C2B2A2928, relocInfo::none);
833 __ emit_data64(0x3736353433323130, relocInfo::none);
834 __ emit_data64(0x3F3E3D3C3B3A3938, relocInfo::none);
835 // W
836 __ emit_data64(0x0003000200010000, relocInfo::none);
837 __ emit_data64(0x0007000600050004, relocInfo::none);
838 __ emit_data64(0x000B000A00090008, relocInfo::none);
839 __ emit_data64(0x000F000E000D000C, relocInfo::none);
840 __ emit_data64(0x0013001200110010, relocInfo::none);
841 __ emit_data64(0x0017001600150014, relocInfo::none);
842 __ emit_data64(0x001B001A00190018, relocInfo::none);
843 __ emit_data64(0x001F001E001D001C, relocInfo::none);
844 // D
845 __ emit_data64(0x0000000100000000, relocInfo::none);
846 __ emit_data64(0x0000000300000002, relocInfo::none);
847 __ emit_data64(0x0000000500000004, relocInfo::none);
848 __ emit_data64(0x0000000700000006, relocInfo::none);
849 __ emit_data64(0x0000000900000008, relocInfo::none);
850 __ emit_data64(0x0000000B0000000A, relocInfo::none);
851 __ emit_data64(0x0000000D0000000C, relocInfo::none);
852 __ emit_data64(0x0000000F0000000E, relocInfo::none);
853 // Q
854 __ emit_data64(0x0000000000000000, relocInfo::none);
855 __ emit_data64(0x0000000000000001, relocInfo::none);
856 __ emit_data64(0x0000000000000002, relocInfo::none);
857 __ emit_data64(0x0000000000000003, relocInfo::none);
858 __ emit_data64(0x0000000000000004, relocInfo::none);
859 __ emit_data64(0x0000000000000005, relocInfo::none);
860 __ emit_data64(0x0000000000000006, relocInfo::none);
861 __ emit_data64(0x0000000000000007, relocInfo::none);
862 // D - FP
863 __ emit_data64(0x3F80000000000000, relocInfo::none); // 0.0f, 1.0f
864 __ emit_data64(0x4040000040000000, relocInfo::none); // 2.0f, 3.0f
865 __ emit_data64(0x40A0000040800000, relocInfo::none); // 4.0f, 5.0f
866 __ emit_data64(0x40E0000040C00000, relocInfo::none); // 6.0f, 7.0f
867 __ emit_data64(0x4110000041000000, relocInfo::none); // 8.0f, 9.0f
868 __ emit_data64(0x4130000041200000, relocInfo::none); // 10.0f, 11.0f
869 __ emit_data64(0x4150000041400000, relocInfo::none); // 12.0f, 13.0f
870 __ emit_data64(0x4170000041600000, relocInfo::none); // 14.0f, 15.0f
871 // Q - FP
872 __ emit_data64(0x0000000000000000, relocInfo::none); // 0.0d
873 __ emit_data64(0x3FF0000000000000, relocInfo::none); // 1.0d
874 __ emit_data64(0x4000000000000000, relocInfo::none); // 2.0d
875 __ emit_data64(0x4008000000000000, relocInfo::none); // 3.0d
876 __ emit_data64(0x4010000000000000, relocInfo::none); // 4.0d
877 __ emit_data64(0x4014000000000000, relocInfo::none); // 5.0d
878 __ emit_data64(0x4018000000000000, relocInfo::none); // 6.0d
879 __ emit_data64(0x401c000000000000, relocInfo::none); // 7.0d
880 return start;
881 }
882
883 address StubGenerator::generate_vector_reverse_bit_lut() {
884 __ align(CodeEntryAlignment);
885 StubGenStubId stub_id = StubGenStubId::vector_reverse_bit_lut_id;
886 StubCodeMark mark(this, stub_id);
887 address start = __ pc();
888
889 __ emit_data64(0x0E060A020C040800, relocInfo::none);
890 __ emit_data64(0x0F070B030D050901, relocInfo::none);
891 __ emit_data64(0x0E060A020C040800, relocInfo::none);
892 __ emit_data64(0x0F070B030D050901, relocInfo::none);
893 __ emit_data64(0x0E060A020C040800, relocInfo::none);
894 __ emit_data64(0x0F070B030D050901, relocInfo::none);
895 __ emit_data64(0x0E060A020C040800, relocInfo::none);
896 __ emit_data64(0x0F070B030D050901, relocInfo::none);
897
898 return start;
899 }
900
901 address StubGenerator::generate_vector_reverse_byte_perm_mask_long() {
902 __ align(CodeEntryAlignment);
903 StubGenStubId stub_id = StubGenStubId::vector_reverse_byte_perm_mask_long_id;
904 StubCodeMark mark(this, stub_id);
905 address start = __ pc();
906
907 __ emit_data64(0x0001020304050607, relocInfo::none);
908 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
909 __ emit_data64(0x0001020304050607, relocInfo::none);
910 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
911 __ emit_data64(0x0001020304050607, relocInfo::none);
912 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
913 __ emit_data64(0x0001020304050607, relocInfo::none);
914 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
915
916 return start;
917 }
918
919 address StubGenerator::generate_vector_reverse_byte_perm_mask_int() {
920 __ align(CodeEntryAlignment);
921 StubGenStubId stub_id = StubGenStubId::vector_reverse_byte_perm_mask_int_id;
922 StubCodeMark mark(this, stub_id);
923 address start = __ pc();
924
925 __ emit_data64(0x0405060700010203, relocInfo::none);
926 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
927 __ emit_data64(0x0405060700010203, relocInfo::none);
928 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
929 __ emit_data64(0x0405060700010203, relocInfo::none);
930 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
931 __ emit_data64(0x0405060700010203, relocInfo::none);
932 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
933
934 return start;
935 }
936
937 address StubGenerator::generate_vector_reverse_byte_perm_mask_short() {
938 __ align(CodeEntryAlignment);
939 StubGenStubId stub_id = StubGenStubId::vector_reverse_byte_perm_mask_short_id;
940 StubCodeMark mark(this, stub_id);
941 address start = __ pc();
942
943 __ emit_data64(0x0607040502030001, relocInfo::none);
944 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
945 __ emit_data64(0x0607040502030001, relocInfo::none);
946 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
947 __ emit_data64(0x0607040502030001, relocInfo::none);
948 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
949 __ emit_data64(0x0607040502030001, relocInfo::none);
950 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
951
952 return start;
953 }
954
955 address StubGenerator::generate_vector_byte_shuffle_mask() {
956 __ align(CodeEntryAlignment);
957 StubGenStubId stub_id = StubGenStubId::vector_byte_shuffle_mask_id;
958 StubCodeMark mark(this, stub_id);
959 address start = __ pc();
960
961 __ emit_data64(0x7070707070707070, relocInfo::none);
962 __ emit_data64(0x7070707070707070, relocInfo::none);
963 __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
964 __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
965
966 return start;
967 }
968
969 address StubGenerator::generate_fp_mask(StubGenStubId stub_id, int64_t mask) {
970 __ align(CodeEntryAlignment);
971 StubCodeMark mark(this, stub_id);
972 address start = __ pc();
973
974 __ emit_data64( mask, relocInfo::none );
975 __ emit_data64( mask, relocInfo::none );
976
977 return start;
978 }
979
980 address StubGenerator::generate_compress_perm_table(StubGenStubId stub_id) {
981 int esize;
982 switch (stub_id) {
983 case compress_perm_table32_id:
984 esize = 32;
985 break;
986 case compress_perm_table64_id:
987 esize = 64;
988 break;
989 default:
990 ShouldNotReachHere();
991 }
992 __ align(CodeEntryAlignment);
993 StubCodeMark mark(this, stub_id);
994 address start = __ pc();
995 if (esize == 32) {
996 // Loop to generate 256 x 8 int compression permute index table. A row is
997 // accessed using 8 bit index computed using vector mask. An entry in
998 // a row holds either a valid permute index corresponding to set bit position
999 // or a -1 (default) value.
1000 for (int mask = 0; mask < 256; mask++) {
1001 int ctr = 0;
1002 for (int j = 0; j < 8; j++) {
1003 if (mask & (1 << j)) {
1004 __ emit_data(j, relocInfo::none);
1005 ctr++;
1006 }
1007 }
1008 for (; ctr < 8; ctr++) {
1009 __ emit_data(-1, relocInfo::none);
1010 }
1011 }
1012 } else {
1013 assert(esize == 64, "");
1014 // Loop to generate 16 x 4 long compression permute index table. A row is
1015 // accessed using 4 bit index computed using vector mask. An entry in
1016 // a row holds either a valid permute index pair for a quadword corresponding
1017 // to set bit position or a -1 (default) value.
1018 for (int mask = 0; mask < 16; mask++) {
1019 int ctr = 0;
1020 for (int j = 0; j < 4; j++) {
1021 if (mask & (1 << j)) {
1022 __ emit_data(2 * j, relocInfo::none);
1023 __ emit_data(2 * j + 1, relocInfo::none);
1024 ctr++;
1025 }
1026 }
1027 for (; ctr < 4; ctr++) {
1028 __ emit_data64(-1L, relocInfo::none);
1029 }
1030 }
1031 }
1032 return start;
1033 }
1034
1035 address StubGenerator::generate_expand_perm_table(StubGenStubId stub_id) {
1036 int esize;
1037 switch (stub_id) {
1038 case expand_perm_table32_id:
1039 esize = 32;
1040 break;
1041 case expand_perm_table64_id:
1042 esize = 64;
1043 break;
1044 default:
1045 ShouldNotReachHere();
1046 }
1047 __ align(CodeEntryAlignment);
1048 StubCodeMark mark(this, stub_id);
1049 address start = __ pc();
1050 if (esize == 32) {
1051 // Loop to generate 256 x 8 int expand permute index table. A row is accessed
1052 // using 8 bit index computed using vector mask. An entry in a row holds either
1053 // a valid permute index (starting from least significant lane) placed at poisition
1054 // corresponding to set bit position or a -1 (default) value.
1055 for (int mask = 0; mask < 256; mask++) {
1056 int ctr = 0;
1057 for (int j = 0; j < 8; j++) {
1058 if (mask & (1 << j)) {
1059 __ emit_data(ctr++, relocInfo::none);
1060 } else {
1061 __ emit_data(-1, relocInfo::none);
1062 }
1063 }
1064 }
1065 } else {
1066 assert(esize == 64, "");
1067 // Loop to generate 16 x 4 long expand permute index table. A row is accessed
1068 // using 4 bit index computed using vector mask. An entry in a row holds either
1069 // a valid doubleword permute index pair representing a quadword index (starting
1070 // from least significant lane) placed at poisition corresponding to set bit
1071 // position or a -1 (default) value.
1072 for (int mask = 0; mask < 16; mask++) {
1073 int ctr = 0;
1074 for (int j = 0; j < 4; j++) {
1075 if (mask & (1 << j)) {
1076 __ emit_data(2 * ctr, relocInfo::none);
1077 __ emit_data(2 * ctr + 1, relocInfo::none);
1078 ctr++;
1079 } else {
1080 __ emit_data64(-1L, relocInfo::none);
1081 }
1082 }
1083 }
1084 }
1085 return start;
1086 }
1087
1088 address StubGenerator::generate_vector_mask(StubGenStubId stub_id, int64_t mask) {
1089 __ align(CodeEntryAlignment);
1090 StubCodeMark mark(this, stub_id);
1091 address start = __ pc();
1092
1093 __ emit_data64(mask, relocInfo::none);
1094 __ emit_data64(mask, relocInfo::none);
1095 __ emit_data64(mask, relocInfo::none);
1096 __ emit_data64(mask, relocInfo::none);
1097 __ emit_data64(mask, relocInfo::none);
1098 __ emit_data64(mask, relocInfo::none);
1099 __ emit_data64(mask, relocInfo::none);
1100 __ emit_data64(mask, relocInfo::none);
1101
1102 return start;
1103 }
1104
1105 address StubGenerator::generate_vector_byte_perm_mask() {
1106 __ align(CodeEntryAlignment);
1107 StubGenStubId stub_id = StubGenStubId::vector_byte_perm_mask_id;
1108 StubCodeMark mark(this, stub_id);
1109 address start = __ pc();
1110
1111 __ emit_data64(0x0000000000000001, relocInfo::none);
1112 __ emit_data64(0x0000000000000003, relocInfo::none);
1113 __ emit_data64(0x0000000000000005, relocInfo::none);
1114 __ emit_data64(0x0000000000000007, relocInfo::none);
1115 __ emit_data64(0x0000000000000000, relocInfo::none);
1116 __ emit_data64(0x0000000000000002, relocInfo::none);
1117 __ emit_data64(0x0000000000000004, relocInfo::none);
1118 __ emit_data64(0x0000000000000006, relocInfo::none);
1119
1120 return start;
1121 }
1122
1123 address StubGenerator::generate_vector_fp_mask(StubGenStubId stub_id, int64_t mask) {
1124 __ align(CodeEntryAlignment);
1125 StubCodeMark mark(this, stub_id);
1126 address start = __ pc();
1127
1128 __ emit_data64(mask, relocInfo::none);
1129 __ emit_data64(mask, relocInfo::none);
1130 __ emit_data64(mask, relocInfo::none);
1131 __ emit_data64(mask, relocInfo::none);
1132 __ emit_data64(mask, relocInfo::none);
1133 __ emit_data64(mask, relocInfo::none);
1134 __ emit_data64(mask, relocInfo::none);
1135 __ emit_data64(mask, relocInfo::none);
1136
1137 return start;
1138 }
1139
1140 address StubGenerator::generate_vector_custom_i32(StubGenStubId stub_id, Assembler::AvxVectorLen len,
1141 int32_t val0, int32_t val1, int32_t val2, int32_t val3,
1142 int32_t val4, int32_t val5, int32_t val6, int32_t val7,
1143 int32_t val8, int32_t val9, int32_t val10, int32_t val11,
1144 int32_t val12, int32_t val13, int32_t val14, int32_t val15) {
1145 __ align(CodeEntryAlignment);
1146 StubCodeMark mark(this, stub_id);
1147 address start = __ pc();
1148
1149 assert(len != Assembler::AVX_NoVec, "vector len must be specified");
1150 __ emit_data(val0, relocInfo::none, 0);
1151 __ emit_data(val1, relocInfo::none, 0);
1152 __ emit_data(val2, relocInfo::none, 0);
1153 __ emit_data(val3, relocInfo::none, 0);
1154 if (len >= Assembler::AVX_256bit) {
1155 __ emit_data(val4, relocInfo::none, 0);
1156 __ emit_data(val5, relocInfo::none, 0);
1157 __ emit_data(val6, relocInfo::none, 0);
1158 __ emit_data(val7, relocInfo::none, 0);
1159 if (len >= Assembler::AVX_512bit) {
1160 __ emit_data(val8, relocInfo::none, 0);
1161 __ emit_data(val9, relocInfo::none, 0);
1162 __ emit_data(val10, relocInfo::none, 0);
1163 __ emit_data(val11, relocInfo::none, 0);
1164 __ emit_data(val12, relocInfo::none, 0);
1165 __ emit_data(val13, relocInfo::none, 0);
1166 __ emit_data(val14, relocInfo::none, 0);
1167 __ emit_data(val15, relocInfo::none, 0);
1168 }
1169 }
1170 return start;
1171 }
1172
1173 // Non-destructive plausibility checks for oops
1174 //
1175 // Arguments:
1176 // all args on stack!
1177 //
1178 // Stack after saving c_rarg3:
1179 // [tos + 0]: saved c_rarg3
1180 // [tos + 1]: saved c_rarg2
1181 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1182 // [tos + 3]: saved flags
1183 // [tos + 4]: return address
1184 // * [tos + 5]: error message (char*)
1185 // * [tos + 6]: object to verify (oop)
1186 // * [tos + 7]: saved rax - saved by caller and bashed
1187 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1188 // * = popped on exit
1189 address StubGenerator::generate_verify_oop() {
1190 StubGenStubId stub_id = StubGenStubId::verify_oop_id;
1191 StubCodeMark mark(this, stub_id);
1192 address start = __ pc();
1193
1194 Label exit, error;
1195
1196 __ pushf();
1197 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()), rscratch1);
1198
1199 __ push(r12);
1200
1201 // save c_rarg2 and c_rarg3
1202 __ push(c_rarg2);
1203 __ push(c_rarg3);
1204
1205 enum {
1206 // After previous pushes.
1207 oop_to_verify = 6 * wordSize,
1208 saved_rax = 7 * wordSize,
1209 saved_r10 = 8 * wordSize,
1210
1211 // Before the call to MacroAssembler::debug(), see below.
1212 return_addr = 16 * wordSize,
1213 error_msg = 17 * wordSize
1214 };
1215
1216 // get object
1217 __ movptr(rax, Address(rsp, oop_to_verify));
1218
1219 // make sure object is 'reasonable'
1220 __ testptr(rax, rax);
1221 __ jcc(Assembler::zero, exit); // if obj is null it is OK
1222
1223 BarrierSetAssembler* bs_asm = BarrierSet::barrier_set()->barrier_set_assembler();
1224 bs_asm->check_oop(_masm, rax, c_rarg2, c_rarg3, error);
1225
1226 // return if everything seems ok
1227 __ bind(exit);
1228 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1229 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1230 __ pop(c_rarg3); // restore c_rarg3
1231 __ pop(c_rarg2); // restore c_rarg2
1232 __ pop(r12); // restore r12
1233 __ popf(); // restore flags
1234 __ ret(4 * wordSize); // pop caller saved stuff
1235
1236 // handle errors
1237 __ bind(error);
1238 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1239 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1240 __ pop(c_rarg3); // get saved c_rarg3 back
1241 __ pop(c_rarg2); // get saved c_rarg2 back
1242 __ pop(r12); // get saved r12 back
1243 __ popf(); // get saved flags off stack --
1244 // will be ignored
1245
1246 __ pusha(); // push registers
1247 // (rip is already
1248 // already pushed)
1249 // debug(char* msg, int64_t pc, int64_t regs[])
1250 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1251 // pushed all the registers, so now the stack looks like:
1252 // [tos + 0] 16 saved registers
1253 // [tos + 16] return address
1254 // * [tos + 17] error message (char*)
1255 // * [tos + 18] object to verify (oop)
1256 // * [tos + 19] saved rax - saved by caller and bashed
1257 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1258 // * = popped on exit
1259
1260 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1261 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1262 __ movq(c_rarg2, rsp); // pass address of regs on stack
1263 __ mov(r12, rsp); // remember rsp
1264 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1265 __ andptr(rsp, -16); // align stack as required by ABI
1266 BLOCK_COMMENT("call MacroAssembler::debug");
1267 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1268 __ hlt();
1269
1270 return start;
1271 }
1272
1273
1274 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1275 //
1276 // Outputs:
1277 // rdi - rcx
1278 // rsi - rdx
1279 // rdx - r8
1280 // rcx - r9
1281 //
1282 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1283 // are non-volatile. r9 and r10 should not be used by the caller.
1284 //
1285 void StubGenerator::setup_arg_regs(int nargs) {
1286 const Register saved_rdi = r9;
1287 const Register saved_rsi = r10;
1288 assert(nargs == 3 || nargs == 4, "else fix");
1289 #ifdef _WIN64
1290 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1291 "unexpected argument registers");
1292 if (nargs == 4) {
1293 __ mov(rax, r9); // r9 is also saved_rdi
1294 }
1295 __ movptr(saved_rdi, rdi);
1296 __ movptr(saved_rsi, rsi);
1297 __ mov(rdi, rcx); // c_rarg0
1298 __ mov(rsi, rdx); // c_rarg1
1299 __ mov(rdx, r8); // c_rarg2
1300 if (nargs == 4) {
1301 __ mov(rcx, rax); // c_rarg3 (via rax)
1302 }
1303 #else
1304 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1305 "unexpected argument registers");
1306 #endif
1307 DEBUG_ONLY(_regs_in_thread = false;)
1308 }
1309
1310
1311 void StubGenerator::restore_arg_regs() {
1312 assert(!_regs_in_thread, "wrong call to restore_arg_regs");
1313 const Register saved_rdi = r9;
1314 const Register saved_rsi = r10;
1315 #ifdef _WIN64
1316 __ movptr(rdi, saved_rdi);
1317 __ movptr(rsi, saved_rsi);
1318 #endif
1319 }
1320
1321
1322 // This is used in places where r10 is a scratch register, and can
1323 // be adapted if r9 is needed also.
1324 void StubGenerator::setup_arg_regs_using_thread(int nargs) {
1325 const Register saved_r15 = r9;
1326 assert(nargs == 3 || nargs == 4, "else fix");
1327 #ifdef _WIN64
1328 if (nargs == 4) {
1329 __ mov(rax, r9); // r9 is also saved_r15
1330 }
1331 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1332 __ get_thread(r15_thread);
1333 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1334 "unexpected argument registers");
1335 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1336 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1337
1338 __ mov(rdi, rcx); // c_rarg0
1339 __ mov(rsi, rdx); // c_rarg1
1340 __ mov(rdx, r8); // c_rarg2
1341 if (nargs == 4) {
1342 __ mov(rcx, rax); // c_rarg3 (via rax)
1343 }
1344 #else
1345 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1346 "unexpected argument registers");
1347 #endif
1348 DEBUG_ONLY(_regs_in_thread = true;)
1349 }
1350
1351
1352 void StubGenerator::restore_arg_regs_using_thread() {
1353 assert(_regs_in_thread, "wrong call to restore_arg_regs");
1354 const Register saved_r15 = r9;
1355 #ifdef _WIN64
1356 __ get_thread(r15_thread);
1357 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1358 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1359 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1360 #endif
1361 }
1362
1363
1364 void StubGenerator::setup_argument_regs(BasicType type) {
1365 if (type == T_BYTE || type == T_SHORT) {
1366 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1367 // r9 and r10 may be used to save non-volatile registers
1368 } else {
1369 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1370 // r9 is used to save r15_thread
1371 }
1372 }
1373
1374
1375 void StubGenerator::restore_argument_regs(BasicType type) {
1376 if (type == T_BYTE || type == T_SHORT) {
1377 restore_arg_regs();
1378 } else {
1379 restore_arg_regs_using_thread();
1380 }
1381 }
1382
1383 address StubGenerator::generate_data_cache_writeback() {
1384 const Register src = c_rarg0; // source address
1385
1386 __ align(CodeEntryAlignment);
1387
1388 StubGenStubId stub_id = StubGenStubId::data_cache_writeback_id;
1389 StubCodeMark mark(this, stub_id);
1390
1391 address start = __ pc();
1392
1393 __ enter();
1394 __ cache_wb(Address(src, 0));
1395 __ leave();
1396 __ ret(0);
1397
1398 return start;
1399 }
1400
1401 address StubGenerator::generate_data_cache_writeback_sync() {
1402 const Register is_pre = c_rarg0; // pre or post sync
1403
1404 __ align(CodeEntryAlignment);
1405
1406 StubGenStubId stub_id = StubGenStubId::data_cache_writeback_sync_id;
1407 StubCodeMark mark(this, stub_id);
1408
1409 // pre wbsync is a no-op
1410 // post wbsync translates to an sfence
1411
1412 Label skip;
1413 address start = __ pc();
1414
1415 __ enter();
1416 __ cmpl(is_pre, 0);
1417 __ jcc(Assembler::notEqual, skip);
1418 __ cache_wbsync(false);
1419 __ bind(skip);
1420 __ leave();
1421 __ ret(0);
1422
1423 return start;
1424 }
1425
1426 // ofs and limit are use for multi-block byte array.
1427 // int com.sun.security.provider.MD5.implCompress(byte[] b, int ofs)
1428 address StubGenerator::generate_md5_implCompress(StubGenStubId stub_id) {
1429 bool multi_block;
1430 switch (stub_id) {
1431 case md5_implCompress_id:
1432 multi_block = false;
1433 break;
1434 case md5_implCompressMB_id:
1435 multi_block = true;
1436 break;
1437 default:
1438 ShouldNotReachHere();
1439 }
1440 __ align(CodeEntryAlignment);
1441 StubCodeMark mark(this, stub_id);
1442 address start = __ pc();
1443
1444 const Register buf_param = r15;
1445 const Address state_param(rsp, 0 * wordSize);
1446 const Address ofs_param (rsp, 1 * wordSize );
1447 const Address limit_param(rsp, 1 * wordSize + 4);
1448
1449 __ enter();
1450 __ push(rbx);
1451 __ push(rdi);
1452 __ push(rsi);
1453 __ push(r15);
1454 __ subptr(rsp, 2 * wordSize);
1455
1456 __ movptr(buf_param, c_rarg0);
1457 __ movptr(state_param, c_rarg1);
1458 if (multi_block) {
1459 __ movl(ofs_param, c_rarg2);
1460 __ movl(limit_param, c_rarg3);
1461 }
1462 __ fast_md5(buf_param, state_param, ofs_param, limit_param, multi_block);
1463
1464 __ addptr(rsp, 2 * wordSize);
1465 __ pop(r15);
1466 __ pop(rsi);
1467 __ pop(rdi);
1468 __ pop(rbx);
1469 __ leave();
1470 __ ret(0);
1471
1472 return start;
1473 }
1474
1475 address StubGenerator::generate_upper_word_mask() {
1476 __ align64();
1477 StubGenStubId stub_id = StubGenStubId::upper_word_mask_id;
1478 StubCodeMark mark(this, stub_id);
1479 address start = __ pc();
1480
1481 __ emit_data64(0x0000000000000000, relocInfo::none);
1482 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
1483
1484 return start;
1485 }
1486
1487 address StubGenerator::generate_shuffle_byte_flip_mask() {
1488 __ align64();
1489 StubGenStubId stub_id = StubGenStubId::shuffle_byte_flip_mask_id;
1490 StubCodeMark mark(this, stub_id);
1491 address start = __ pc();
1492
1493 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1494 __ emit_data64(0x0001020304050607, relocInfo::none);
1495
1496 return start;
1497 }
1498
1499 // ofs and limit are use for multi-block byte array.
1500 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1501 address StubGenerator::generate_sha1_implCompress(StubGenStubId stub_id) {
1502 bool multi_block;
1503 switch (stub_id) {
1504 case sha1_implCompress_id:
1505 multi_block = false;
1506 break;
1507 case sha1_implCompressMB_id:
1508 multi_block = true;
1509 break;
1510 default:
1511 ShouldNotReachHere();
1512 }
1513 __ align(CodeEntryAlignment);
1514 StubCodeMark mark(this, stub_id);
1515 address start = __ pc();
1516
1517 Register buf = c_rarg0;
1518 Register state = c_rarg1;
1519 Register ofs = c_rarg2;
1520 Register limit = c_rarg3;
1521
1522 const XMMRegister abcd = xmm0;
1523 const XMMRegister e0 = xmm1;
1524 const XMMRegister e1 = xmm2;
1525 const XMMRegister msg0 = xmm3;
1526
1527 const XMMRegister msg1 = xmm4;
1528 const XMMRegister msg2 = xmm5;
1529 const XMMRegister msg3 = xmm6;
1530 const XMMRegister shuf_mask = xmm7;
1531
1532 __ enter();
1533
1534 __ subptr(rsp, 4 * wordSize);
1535
1536 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
1537 buf, state, ofs, limit, rsp, multi_block);
1538
1539 __ addptr(rsp, 4 * wordSize);
1540
1541 __ leave();
1542 __ ret(0);
1543
1544 return start;
1545 }
1546
1547 address StubGenerator::generate_pshuffle_byte_flip_mask() {
1548 __ align64();
1549 StubGenStubId stub_id = StubGenStubId::pshuffle_byte_flip_mask_id;
1550 StubCodeMark mark(this, stub_id);
1551 address start = __ pc();
1552
1553 __ emit_data64(0x0405060700010203, relocInfo::none);
1554 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1555
1556 if (VM_Version::supports_avx2()) {
1557 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
1558 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1559 // _SHUF_00BA
1560 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1561 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1562 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1563 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1564 // _SHUF_DC00
1565 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1566 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1567 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1568 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1569 }
1570
1571 return start;
1572 }
1573
1574 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
1575 address StubGenerator::generate_pshuffle_byte_flip_mask_sha512() {
1576 __ align32();
1577 StubGenStubId stub_id = StubGenStubId::pshuffle_byte_flip_mask_sha512_id;
1578 StubCodeMark mark(this, stub_id);
1579 address start = __ pc();
1580
1581 if (VM_Version::supports_avx2()) {
1582 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
1583 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1584 __ emit_data64(0x1011121314151617, relocInfo::none);
1585 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
1586 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
1587 __ emit_data64(0x0000000000000000, relocInfo::none);
1588 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1589 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1590 }
1591
1592 return start;
1593 }
1594
1595 // ofs and limit are use for multi-block byte array.
1596 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1597 address StubGenerator::generate_sha256_implCompress(StubGenStubId stub_id) {
1598 bool multi_block;
1599 switch (stub_id) {
1600 case sha256_implCompress_id:
1601 multi_block = false;
1602 break;
1603 case sha256_implCompressMB_id:
1604 multi_block = true;
1605 break;
1606 default:
1607 ShouldNotReachHere();
1608 }
1609 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
1610 __ align(CodeEntryAlignment);
1611 StubCodeMark mark(this, stub_id);
1612 address start = __ pc();
1613
1614 Register buf = c_rarg0;
1615 Register state = c_rarg1;
1616 Register ofs = c_rarg2;
1617 Register limit = c_rarg3;
1618
1619 const XMMRegister msg = xmm0;
1620 const XMMRegister state0 = xmm1;
1621 const XMMRegister state1 = xmm2;
1622 const XMMRegister msgtmp0 = xmm3;
1623
1624 const XMMRegister msgtmp1 = xmm4;
1625 const XMMRegister msgtmp2 = xmm5;
1626 const XMMRegister msgtmp3 = xmm6;
1627 const XMMRegister msgtmp4 = xmm7;
1628
1629 const XMMRegister shuf_mask = xmm8;
1630
1631 __ enter();
1632
1633 __ subptr(rsp, 4 * wordSize);
1634
1635 if (VM_Version::supports_sha()) {
1636 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1637 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1638 } else if (VM_Version::supports_avx2()) {
1639 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1640 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1641 }
1642 __ addptr(rsp, 4 * wordSize);
1643 __ vzeroupper();
1644 __ leave();
1645 __ ret(0);
1646
1647 return start;
1648 }
1649
1650 address StubGenerator::generate_sha512_implCompress(StubGenStubId stub_id) {
1651 bool multi_block;
1652 switch (stub_id) {
1653 case sha512_implCompress_id:
1654 multi_block = false;
1655 break;
1656 case sha512_implCompressMB_id:
1657 multi_block = true;
1658 break;
1659 default:
1660 ShouldNotReachHere();
1661 }
1662 assert(VM_Version::supports_avx2(), "");
1663 assert(VM_Version::supports_bmi2() || VM_Version::supports_sha512(), "");
1664 __ align(CodeEntryAlignment);
1665 StubCodeMark mark(this, stub_id);
1666 address start = __ pc();
1667
1668 Register buf = c_rarg0;
1669 Register state = c_rarg1;
1670 Register ofs = c_rarg2;
1671 Register limit = c_rarg3;
1672
1673 __ enter();
1674
1675 if (VM_Version::supports_sha512()) {
1676 __ sha512_update_ni_x1(state, buf, ofs, limit, multi_block);
1677 } else {
1678 const XMMRegister msg = xmm0;
1679 const XMMRegister state0 = xmm1;
1680 const XMMRegister state1 = xmm2;
1681 const XMMRegister msgtmp0 = xmm3;
1682 const XMMRegister msgtmp1 = xmm4;
1683 const XMMRegister msgtmp2 = xmm5;
1684 const XMMRegister msgtmp3 = xmm6;
1685 const XMMRegister msgtmp4 = xmm7;
1686
1687 const XMMRegister shuf_mask = xmm8;
1688 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1689 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1690 }
1691 __ vzeroupper();
1692 __ leave();
1693 __ ret(0);
1694
1695 return start;
1696 }
1697
1698 address StubGenerator::base64_shuffle_addr() {
1699 __ align64();
1700 StubGenStubId stub_id = StubGenStubId::shuffle_base64_id;
1701 StubCodeMark mark(this, stub_id);
1702 address start = __ pc();
1703
1704 assert(((unsigned long long)start & 0x3f) == 0,
1705 "Alignment problem (0x%08llx)", (unsigned long long)start);
1706 __ emit_data64(0x0405030401020001, relocInfo::none);
1707 __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1708 __ emit_data64(0x10110f100d0e0c0d, relocInfo::none);
1709 __ emit_data64(0x1617151613141213, relocInfo::none);
1710 __ emit_data64(0x1c1d1b1c191a1819, relocInfo::none);
1711 __ emit_data64(0x222321221f201e1f, relocInfo::none);
1712 __ emit_data64(0x2829272825262425, relocInfo::none);
1713 __ emit_data64(0x2e2f2d2e2b2c2a2b, relocInfo::none);
1714
1715 return start;
1716 }
1717
1718 address StubGenerator::base64_avx2_shuffle_addr() {
1719 __ align32();
1720 StubGenStubId stub_id = StubGenStubId::avx2_shuffle_base64_id;
1721 StubCodeMark mark(this, stub_id);
1722 address start = __ pc();
1723
1724 __ emit_data64(0x0809070805060405, relocInfo::none);
1725 __ emit_data64(0x0e0f0d0e0b0c0a0b, relocInfo::none);
1726 __ emit_data64(0x0405030401020001, relocInfo::none);
1727 __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1728
1729 return start;
1730 }
1731
1732 address StubGenerator::base64_avx2_input_mask_addr() {
1733 __ align32();
1734 StubGenStubId stub_id = StubGenStubId::avx2_input_mask_base64_id;
1735 StubCodeMark mark(this, stub_id);
1736 address start = __ pc();
1737
1738 __ emit_data64(0x8000000000000000, relocInfo::none);
1739 __ emit_data64(0x8000000080000000, relocInfo::none);
1740 __ emit_data64(0x8000000080000000, relocInfo::none);
1741 __ emit_data64(0x8000000080000000, relocInfo::none);
1742
1743 return start;
1744 }
1745
1746 address StubGenerator::base64_avx2_lut_addr() {
1747 __ align32();
1748 StubGenStubId stub_id = StubGenStubId::avx2_lut_base64_id;
1749 StubCodeMark mark(this, stub_id);
1750 address start = __ pc();
1751
1752 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1753 __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1754 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1755 __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1756
1757 // URL LUT
1758 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1759 __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1760 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1761 __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1762
1763 return start;
1764 }
1765
1766 address StubGenerator::base64_encoding_table_addr() {
1767 __ align64();
1768 StubGenStubId stub_id = StubGenStubId::encoding_table_base64_id;
1769 StubCodeMark mark(this, stub_id);
1770 address start = __ pc();
1771
1772 assert(((unsigned long long)start & 0x3f) == 0, "Alignment problem (0x%08llx)", (unsigned long long)start);
1773 __ emit_data64(0x4847464544434241, relocInfo::none);
1774 __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1775 __ emit_data64(0x5857565554535251, relocInfo::none);
1776 __ emit_data64(0x6665646362615a59, relocInfo::none);
1777 __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1778 __ emit_data64(0x767574737271706f, relocInfo::none);
1779 __ emit_data64(0x333231307a797877, relocInfo::none);
1780 __ emit_data64(0x2f2b393837363534, relocInfo::none);
1781
1782 // URL table
1783 __ emit_data64(0x4847464544434241, relocInfo::none);
1784 __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1785 __ emit_data64(0x5857565554535251, relocInfo::none);
1786 __ emit_data64(0x6665646362615a59, relocInfo::none);
1787 __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1788 __ emit_data64(0x767574737271706f, relocInfo::none);
1789 __ emit_data64(0x333231307a797877, relocInfo::none);
1790 __ emit_data64(0x5f2d393837363534, relocInfo::none);
1791
1792 return start;
1793 }
1794
1795 // Code for generating Base64 encoding.
1796 // Intrinsic function prototype in Base64.java:
1797 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp,
1798 // boolean isURL) {
1799 address StubGenerator::generate_base64_encodeBlock()
1800 {
1801 __ align(CodeEntryAlignment);
1802 StubGenStubId stub_id = StubGenStubId::base64_encodeBlock_id;
1803 StubCodeMark mark(this, stub_id);
1804 address start = __ pc();
1805
1806 __ enter();
1807
1808 // Save callee-saved registers before using them
1809 __ push(r12);
1810 __ push(r13);
1811 __ push(r14);
1812 __ push(r15);
1813
1814 // arguments
1815 const Register source = c_rarg0; // Source Array
1816 const Register start_offset = c_rarg1; // start offset
1817 const Register end_offset = c_rarg2; // end offset
1818 const Register dest = c_rarg3; // destination array
1819
1820 #ifndef _WIN64
1821 const Register dp = c_rarg4; // Position for writing to dest array
1822 const Register isURL = c_rarg5; // Base64 or URL character set
1823 #else
1824 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
1825 const Address isURL_mem(rbp, 7 * wordSize);
1826 const Register isURL = r10; // pick the volatile windows register
1827 const Register dp = r12;
1828 __ movl(dp, dp_mem);
1829 __ movl(isURL, isURL_mem);
1830 #endif
1831
1832 const Register length = r14;
1833 const Register encode_table = r13;
1834 Label L_process3, L_exit, L_processdata, L_vbmiLoop, L_not512, L_32byteLoop;
1835
1836 // calculate length from offsets
1837 __ movl(length, end_offset);
1838 __ subl(length, start_offset);
1839 __ jcc(Assembler::lessEqual, L_exit);
1840
1841 // Code for 512-bit VBMI encoding. Encodes 48 input bytes into 64
1842 // output bytes. We read 64 input bytes and ignore the last 16, so be
1843 // sure not to read past the end of the input buffer.
1844 if (VM_Version::supports_avx512_vbmi()) {
1845 __ cmpl(length, 64); // Do not overrun input buffer.
1846 __ jcc(Assembler::below, L_not512);
1847
1848 __ shll(isURL, 6); // index into decode table based on isURL
1849 __ lea(encode_table, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
1850 __ addptr(encode_table, isURL);
1851 __ shrl(isURL, 6); // restore isURL
1852
1853 __ mov64(rax, 0x3036242a1016040aull); // Shifts
1854 __ evmovdquq(xmm3, ExternalAddress(StubRoutines::x86::base64_shuffle_addr()), Assembler::AVX_512bit, r15);
1855 __ evmovdquq(xmm2, Address(encode_table, 0), Assembler::AVX_512bit);
1856 __ evpbroadcastq(xmm1, rax, Assembler::AVX_512bit);
1857
1858 __ align32();
1859 __ BIND(L_vbmiLoop);
1860
1861 __ vpermb(xmm0, xmm3, Address(source, start_offset), Assembler::AVX_512bit);
1862 __ subl(length, 48);
1863
1864 // Put the input bytes into the proper lanes for writing, then
1865 // encode them.
1866 __ evpmultishiftqb(xmm0, xmm1, xmm0, Assembler::AVX_512bit);
1867 __ vpermb(xmm0, xmm0, xmm2, Assembler::AVX_512bit);
1868
1869 // Write to destination
1870 __ evmovdquq(Address(dest, dp), xmm0, Assembler::AVX_512bit);
1871
1872 __ addptr(dest, 64);
1873 __ addptr(source, 48);
1874 __ cmpl(length, 64);
1875 __ jcc(Assembler::aboveEqual, L_vbmiLoop);
1876
1877 __ vzeroupper();
1878 }
1879
1880 __ BIND(L_not512);
1881 if (VM_Version::supports_avx2()) {
1882 /*
1883 ** This AVX2 encoder is based off the paper at:
1884 ** https://dl.acm.org/doi/10.1145/3132709
1885 **
1886 ** We use AVX2 SIMD instructions to encode 24 bytes into 32
1887 ** output bytes.
1888 **
1889 */
1890 // Lengths under 32 bytes are done with scalar routine
1891 __ cmpl(length, 31);
1892 __ jcc(Assembler::belowEqual, L_process3);
1893
1894 // Set up supporting constant table data
1895 __ vmovdqu(xmm9, ExternalAddress(StubRoutines::x86::base64_avx2_shuffle_addr()), rax);
1896 // 6-bit mask for 2nd and 4th (and multiples) 6-bit values
1897 __ movl(rax, 0x0fc0fc00);
1898 __ movdl(xmm8, rax);
1899 __ vmovdqu(xmm1, ExternalAddress(StubRoutines::x86::base64_avx2_input_mask_addr()), rax);
1900 __ vpbroadcastd(xmm8, xmm8, Assembler::AVX_256bit);
1901
1902 // Multiplication constant for "shifting" right by 6 and 10
1903 // bits
1904 __ movl(rax, 0x04000040);
1905
1906 __ subl(length, 24);
1907 __ movdl(xmm7, rax);
1908 __ vpbroadcastd(xmm7, xmm7, Assembler::AVX_256bit);
1909
1910 // For the first load, we mask off reading of the first 4
1911 // bytes into the register. This is so we can get 4 3-byte
1912 // chunks into each lane of the register, avoiding having to
1913 // handle end conditions. We then shuffle these bytes into a
1914 // specific order so that manipulation is easier.
1915 //
1916 // The initial read loads the XMM register like this:
1917 //
1918 // Lower 128-bit lane:
1919 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1920 // | XX | XX | XX | XX | A0 | A1 | A2 | B0 | B1 | B2 | C0 | C1
1921 // | C2 | D0 | D1 | D2 |
1922 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1923 //
1924 // Upper 128-bit lane:
1925 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1926 // | E0 | E1 | E2 | F0 | F1 | F2 | G0 | G1 | G2 | H0 | H1 | H2
1927 // | XX | XX | XX | XX |
1928 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1929 //
1930 // Where A0 is the first input byte, B0 is the fourth, etc.
1931 // The alphabetical significance denotes the 3 bytes to be
1932 // consumed and encoded into 4 bytes.
1933 //
1934 // We then shuffle the register so each 32-bit word contains
1935 // the sequence:
1936 // A1 A0 A2 A1, B1, B0, B2, B1, etc.
1937 // Each of these byte sequences are then manipulated into 4
1938 // 6-bit values ready for encoding.
1939 //
1940 // If we focus on one set of 3-byte chunks, changing the
1941 // nomenclature such that A0 => a, A1 => b, and A2 => c, we
1942 // shuffle such that each 24-bit chunk contains:
1943 //
1944 // b7 b6 b5 b4 b3 b2 b1 b0 | a7 a6 a5 a4 a3 a2 a1 a0 | c7 c6
1945 // c5 c4 c3 c2 c1 c0 | b7 b6 b5 b4 b3 b2 b1 b0
1946 // Explain this step.
1947 // b3 b2 b1 b0 c5 c4 c3 c2 | c1 c0 d5 d4 d3 d2 d1 d0 | a5 a4
1948 // a3 a2 a1 a0 b5 b4 | b3 b2 b1 b0 c5 c4 c3 c2
1949 //
1950 // W first and off all but bits 4-9 and 16-21 (c5..c0 and
1951 // a5..a0) and shift them using a vector multiplication
1952 // operation (vpmulhuw) which effectively shifts c right by 6
1953 // bits and a right by 10 bits. We similarly mask bits 10-15
1954 // (d5..d0) and 22-27 (b5..b0) and shift them left by 8 and 4
1955 // bits respectively. This is done using vpmullw. We end up
1956 // with 4 6-bit values, thus splitting the 3 input bytes,
1957 // ready for encoding:
1958 // 0 0 d5..d0 0 0 c5..c0 0 0 b5..b0 0 0 a5..a0
1959 //
1960 // For translation, we recognize that there are 5 distinct
1961 // ranges of legal Base64 characters as below:
1962 //
1963 // +-------------+-------------+------------+
1964 // | 6-bit value | ASCII range | offset |
1965 // +-------------+-------------+------------+
1966 // | 0..25 | A..Z | 65 |
1967 // | 26..51 | a..z | 71 |
1968 // | 52..61 | 0..9 | -4 |
1969 // | 62 | + or - | -19 or -17 |
1970 // | 63 | / or _ | -16 or 32 |
1971 // +-------------+-------------+------------+
1972 //
1973 // We note that vpshufb does a parallel lookup in a
1974 // destination register using the lower 4 bits of bytes from a
1975 // source register. If we use a saturated subtraction and
1976 // subtract 51 from each 6-bit value, bytes from [0,51]
1977 // saturate to 0, and [52,63] map to a range of [1,12]. We
1978 // distinguish the [0,25] and [26,51] ranges by assigning a
1979 // value of 13 for all 6-bit values less than 26. We end up
1980 // with:
1981 //
1982 // +-------------+-------------+------------+
1983 // | 6-bit value | Reduced | offset |
1984 // +-------------+-------------+------------+
1985 // | 0..25 | 13 | 65 |
1986 // | 26..51 | 0 | 71 |
1987 // | 52..61 | 0..9 | -4 |
1988 // | 62 | 11 | -19 or -17 |
1989 // | 63 | 12 | -16 or 32 |
1990 // +-------------+-------------+------------+
1991 //
1992 // We then use a final vpshufb to add the appropriate offset,
1993 // translating the bytes.
1994 //
1995 // Load input bytes - only 28 bytes. Mask the first load to
1996 // not load into the full register.
1997 __ vpmaskmovd(xmm1, xmm1, Address(source, start_offset, Address::times_1, -4), Assembler::AVX_256bit);
1998
1999 // Move 3-byte chunks of input (12 bytes) into 16 bytes,
2000 // ordering by:
2001 // 1, 0, 2, 1; 4, 3, 5, 4; etc. This groups 6-bit chunks
2002 // for easy masking
2003 __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
2004
2005 __ addl(start_offset, 24);
2006
2007 // Load masking register for first and third (and multiples)
2008 // 6-bit values.
2009 __ movl(rax, 0x003f03f0);
2010 __ movdl(xmm6, rax);
2011 __ vpbroadcastd(xmm6, xmm6, Assembler::AVX_256bit);
2012 // Multiplication constant for "shifting" left by 4 and 8 bits
2013 __ movl(rax, 0x01000010);
2014 __ movdl(xmm5, rax);
2015 __ vpbroadcastd(xmm5, xmm5, Assembler::AVX_256bit);
2016
2017 // Isolate 6-bit chunks of interest
2018 __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
2019
2020 // Load constants for encoding
2021 __ movl(rax, 0x19191919);
2022 __ movdl(xmm3, rax);
2023 __ vpbroadcastd(xmm3, xmm3, Assembler::AVX_256bit);
2024 __ movl(rax, 0x33333333);
2025 __ movdl(xmm4, rax);
2026 __ vpbroadcastd(xmm4, xmm4, Assembler::AVX_256bit);
2027
2028 // Shift output bytes 0 and 2 into proper lanes
2029 __ vpmulhuw(xmm2, xmm0, xmm7, Assembler::AVX_256bit);
2030
2031 // Mask and shift output bytes 1 and 3 into proper lanes and
2032 // combine
2033 __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
2034 __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
2035 __ vpor(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
2036
2037 // Find out which are 0..25. This indicates which input
2038 // values fall in the range of 'A'-'Z', which require an
2039 // additional offset (see comments above)
2040 __ vpcmpgtb(xmm2, xmm0, xmm3, Assembler::AVX_256bit);
2041 __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
2042 __ vpsubb(xmm1, xmm1, xmm2, Assembler::AVX_256bit);
2043
2044 // Load the proper lookup table
2045 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_avx2_lut_addr()));
2046 __ movl(r15, isURL);
2047 __ shll(r15, 5);
2048 __ vmovdqu(xmm2, Address(r11, r15));
2049
2050 // Shuffle the offsets based on the range calculation done
2051 // above. This allows us to add the correct offset to the
2052 // 6-bit value corresponding to the range documented above.
2053 __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
2054 __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
2055
2056 // Store the encoded bytes
2057 __ vmovdqu(Address(dest, dp), xmm0);
2058 __ addl(dp, 32);
2059
2060 __ cmpl(length, 31);
2061 __ jcc(Assembler::belowEqual, L_process3);
2062
2063 __ align32();
2064 __ BIND(L_32byteLoop);
2065
2066 // Get next 32 bytes
2067 __ vmovdqu(xmm1, Address(source, start_offset, Address::times_1, -4));
2068
2069 __ subl(length, 24);
2070 __ addl(start_offset, 24);
2071
2072 // This logic is identical to the above, with only constant
2073 // register loads removed. Shuffle the input, mask off 6-bit
2074 // chunks, shift them into place, then add the offset to
2075 // encode.
2076 __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
2077
2078 __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
2079 __ vpmulhuw(xmm10, xmm0, xmm7, Assembler::AVX_256bit);
2080 __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
2081 __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
2082 __ vpor(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
2083 __ vpcmpgtb(xmm10, xmm0, xmm3, Assembler::AVX_256bit);
2084 __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
2085 __ vpsubb(xmm1, xmm1, xmm10, Assembler::AVX_256bit);
2086 __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
2087 __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
2088
2089 // Store the encoded bytes
2090 __ vmovdqu(Address(dest, dp), xmm0);
2091 __ addl(dp, 32);
2092
2093 __ cmpl(length, 31);
2094 __ jcc(Assembler::above, L_32byteLoop);
2095
2096 __ BIND(L_process3);
2097 __ vzeroupper();
2098 } else {
2099 __ BIND(L_process3);
2100 }
2101
2102 __ cmpl(length, 3);
2103 __ jcc(Assembler::below, L_exit);
2104
2105 // Load the encoding table based on isURL
2106 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
2107 __ movl(r15, isURL);
2108 __ shll(r15, 6);
2109 __ addptr(r11, r15);
2110
2111 __ BIND(L_processdata);
2112
2113 // Load 3 bytes
2114 __ load_unsigned_byte(r15, Address(source, start_offset));
2115 __ load_unsigned_byte(r10, Address(source, start_offset, Address::times_1, 1));
2116 __ load_unsigned_byte(r13, Address(source, start_offset, Address::times_1, 2));
2117
2118 // Build a 32-bit word with bytes 1, 2, 0, 1
2119 __ movl(rax, r10);
2120 __ shll(r10, 24);
2121 __ orl(rax, r10);
2122
2123 __ subl(length, 3);
2124
2125 __ shll(r15, 8);
2126 __ shll(r13, 16);
2127 __ orl(rax, r15);
2128
2129 __ addl(start_offset, 3);
2130
2131 __ orl(rax, r13);
2132 // At this point, rax contains | byte1 | byte2 | byte0 | byte1
2133 // r13 has byte2 << 16 - need low-order 6 bits to translate.
2134 // This translated byte is the fourth output byte.
2135 __ shrl(r13, 16);
2136 __ andl(r13, 0x3f);
2137
2138 // The high-order 6 bits of r15 (byte0) is translated.
2139 // The translated byte is the first output byte.
2140 __ shrl(r15, 10);
2141
2142 __ load_unsigned_byte(r13, Address(r11, r13));
2143 __ load_unsigned_byte(r15, Address(r11, r15));
2144
2145 __ movb(Address(dest, dp, Address::times_1, 3), r13);
2146
2147 // Extract high-order 4 bits of byte1 and low-order 2 bits of byte0.
2148 // This translated byte is the second output byte.
2149 __ shrl(rax, 4);
2150 __ movl(r10, rax);
2151 __ andl(rax, 0x3f);
2152
2153 __ movb(Address(dest, dp, Address::times_1, 0), r15);
2154
2155 __ load_unsigned_byte(rax, Address(r11, rax));
2156
2157 // Extract low-order 2 bits of byte1 and high-order 4 bits of byte2.
2158 // This translated byte is the third output byte.
2159 __ shrl(r10, 18);
2160 __ andl(r10, 0x3f);
2161
2162 __ load_unsigned_byte(r10, Address(r11, r10));
2163
2164 __ movb(Address(dest, dp, Address::times_1, 1), rax);
2165 __ movb(Address(dest, dp, Address::times_1, 2), r10);
2166
2167 __ addl(dp, 4);
2168 __ cmpl(length, 3);
2169 __ jcc(Assembler::aboveEqual, L_processdata);
2170
2171 __ BIND(L_exit);
2172 __ pop(r15);
2173 __ pop(r14);
2174 __ pop(r13);
2175 __ pop(r12);
2176 __ leave();
2177 __ ret(0);
2178
2179 return start;
2180 }
2181
2182 // base64 AVX512vbmi tables
2183 address StubGenerator::base64_vbmi_lookup_lo_addr() {
2184 __ align64();
2185 StubGenStubId stub_id = StubGenStubId::lookup_lo_base64_id;
2186 StubCodeMark mark(this, stub_id);
2187 address start = __ pc();
2188
2189 assert(((unsigned long long)start & 0x3f) == 0,
2190 "Alignment problem (0x%08llx)", (unsigned long long)start);
2191 __ emit_data64(0x8080808080808080, relocInfo::none);
2192 __ emit_data64(0x8080808080808080, relocInfo::none);
2193 __ emit_data64(0x8080808080808080, relocInfo::none);
2194 __ emit_data64(0x8080808080808080, relocInfo::none);
2195 __ emit_data64(0x8080808080808080, relocInfo::none);
2196 __ emit_data64(0x3f8080803e808080, relocInfo::none);
2197 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2198 __ emit_data64(0x8080808080803d3c, relocInfo::none);
2199
2200 return start;
2201 }
2202
2203 address StubGenerator::base64_vbmi_lookup_hi_addr() {
2204 __ align64();
2205 StubGenStubId stub_id = StubGenStubId::lookup_hi_base64_id;
2206 StubCodeMark mark(this, stub_id);
2207 address start = __ pc();
2208
2209 assert(((unsigned long long)start & 0x3f) == 0,
2210 "Alignment problem (0x%08llx)", (unsigned long long)start);
2211 __ emit_data64(0x0605040302010080, relocInfo::none);
2212 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2213 __ emit_data64(0x161514131211100f, relocInfo::none);
2214 __ emit_data64(0x8080808080191817, relocInfo::none);
2215 __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2216 __ emit_data64(0x2827262524232221, relocInfo::none);
2217 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2218 __ emit_data64(0x8080808080333231, relocInfo::none);
2219
2220 return start;
2221 }
2222 address StubGenerator::base64_vbmi_lookup_lo_url_addr() {
2223 __ align64();
2224 StubGenStubId stub_id = StubGenStubId::lookup_lo_base64url_id;
2225 StubCodeMark mark(this, stub_id);
2226 address start = __ pc();
2227
2228 assert(((unsigned long long)start & 0x3f) == 0,
2229 "Alignment problem (0x%08llx)", (unsigned long long)start);
2230 __ emit_data64(0x8080808080808080, relocInfo::none);
2231 __ emit_data64(0x8080808080808080, relocInfo::none);
2232 __ emit_data64(0x8080808080808080, relocInfo::none);
2233 __ emit_data64(0x8080808080808080, relocInfo::none);
2234 __ emit_data64(0x8080808080808080, relocInfo::none);
2235 __ emit_data64(0x80803e8080808080, relocInfo::none);
2236 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2237 __ emit_data64(0x8080808080803d3c, relocInfo::none);
2238
2239 return start;
2240 }
2241
2242 address StubGenerator::base64_vbmi_lookup_hi_url_addr() {
2243 __ align64();
2244 StubGenStubId stub_id = StubGenStubId::lookup_hi_base64url_id;
2245 StubCodeMark mark(this, stub_id);
2246 address start = __ pc();
2247
2248 assert(((unsigned long long)start & 0x3f) == 0,
2249 "Alignment problem (0x%08llx)", (unsigned long long)start);
2250 __ emit_data64(0x0605040302010080, relocInfo::none);
2251 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2252 __ emit_data64(0x161514131211100f, relocInfo::none);
2253 __ emit_data64(0x3f80808080191817, relocInfo::none);
2254 __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2255 __ emit_data64(0x2827262524232221, relocInfo::none);
2256 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2257 __ emit_data64(0x8080808080333231, relocInfo::none);
2258
2259 return start;
2260 }
2261
2262 address StubGenerator::base64_vbmi_pack_vec_addr() {
2263 __ align64();
2264 StubGenStubId stub_id = StubGenStubId::pack_vec_base64_id;
2265 StubCodeMark mark(this, stub_id);
2266 address start = __ pc();
2267
2268 assert(((unsigned long long)start & 0x3f) == 0,
2269 "Alignment problem (0x%08llx)", (unsigned long long)start);
2270 __ emit_data64(0x090a040506000102, relocInfo::none);
2271 __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2272 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2273 __ emit_data64(0x292a242526202122, relocInfo::none);
2274 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2275 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2276 __ emit_data64(0x0000000000000000, relocInfo::none);
2277 __ emit_data64(0x0000000000000000, relocInfo::none);
2278
2279 return start;
2280 }
2281
2282 address StubGenerator::base64_vbmi_join_0_1_addr() {
2283 __ align64();
2284 StubGenStubId stub_id = StubGenStubId::join_0_1_base64_id;
2285 StubCodeMark mark(this, stub_id);
2286 address start = __ pc();
2287
2288 assert(((unsigned long long)start & 0x3f) == 0,
2289 "Alignment problem (0x%08llx)", (unsigned long long)start);
2290 __ emit_data64(0x090a040506000102, relocInfo::none);
2291 __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2292 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2293 __ emit_data64(0x292a242526202122, relocInfo::none);
2294 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2295 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2296 __ emit_data64(0x494a444546404142, relocInfo::none);
2297 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2298
2299 return start;
2300 }
2301
2302 address StubGenerator::base64_vbmi_join_1_2_addr() {
2303 __ align64();
2304 StubGenStubId stub_id = StubGenStubId::join_1_2_base64_id;
2305 StubCodeMark mark(this, stub_id);
2306 address start = __ pc();
2307
2308 assert(((unsigned long long)start & 0x3f) == 0,
2309 "Alignment problem (0x%08llx)", (unsigned long long)start);
2310 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2311 __ emit_data64(0x292a242526202122, relocInfo::none);
2312 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2313 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2314 __ emit_data64(0x494a444546404142, relocInfo::none);
2315 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2316 __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2317 __ emit_data64(0x696a646566606162, relocInfo::none);
2318
2319 return start;
2320 }
2321
2322 address StubGenerator::base64_vbmi_join_2_3_addr() {
2323 __ align64();
2324 StubGenStubId stub_id = StubGenStubId::join_2_3_base64_id;
2325 StubCodeMark mark(this, stub_id);
2326 address start = __ pc();
2327
2328 assert(((unsigned long long)start & 0x3f) == 0,
2329 "Alignment problem (0x%08llx)", (unsigned long long)start);
2330 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2331 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2332 __ emit_data64(0x494a444546404142, relocInfo::none);
2333 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2334 __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2335 __ emit_data64(0x696a646566606162, relocInfo::none);
2336 __ emit_data64(0x767071726c6d6e68, relocInfo::none);
2337 __ emit_data64(0x7c7d7e78797a7475, relocInfo::none);
2338
2339 return start;
2340 }
2341
2342 address StubGenerator::base64_AVX2_decode_tables_addr() {
2343 __ align64();
2344 StubGenStubId stub_id = StubGenStubId::avx2_decode_tables_base64_id;
2345 StubCodeMark mark(this, stub_id);
2346 address start = __ pc();
2347
2348 assert(((unsigned long long)start & 0x3f) == 0,
2349 "Alignment problem (0x%08llx)", (unsigned long long)start);
2350 __ emit_data(0x2f2f2f2f, relocInfo::none, 0);
2351 __ emit_data(0x5f5f5f5f, relocInfo::none, 0); // for URL
2352
2353 __ emit_data(0xffffffff, relocInfo::none, 0);
2354 __ emit_data(0xfcfcfcfc, relocInfo::none, 0); // for URL
2355
2356 // Permute table
2357 __ emit_data64(0x0000000100000000, relocInfo::none);
2358 __ emit_data64(0x0000000400000002, relocInfo::none);
2359 __ emit_data64(0x0000000600000005, relocInfo::none);
2360 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2361
2362 // Shuffle table
2363 __ emit_data64(0x090a040506000102, relocInfo::none);
2364 __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2365 __ emit_data64(0x090a040506000102, relocInfo::none);
2366 __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2367
2368 // merge table
2369 __ emit_data(0x01400140, relocInfo::none, 0);
2370
2371 // merge multiplier
2372 __ emit_data(0x00011000, relocInfo::none, 0);
2373
2374 return start;
2375 }
2376
2377 address StubGenerator::base64_AVX2_decode_LUT_tables_addr() {
2378 __ align64();
2379 StubGenStubId stub_id = StubGenStubId::avx2_decode_lut_tables_base64_id;
2380 StubCodeMark mark(this, stub_id);
2381 address start = __ pc();
2382
2383 assert(((unsigned long long)start & 0x3f) == 0,
2384 "Alignment problem (0x%08llx)", (unsigned long long)start);
2385 // lut_lo
2386 __ emit_data64(0x1111111111111115, relocInfo::none);
2387 __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2388 __ emit_data64(0x1111111111111115, relocInfo::none);
2389 __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2390
2391 // lut_roll
2392 __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2393 __ emit_data64(0x0000000000000000, relocInfo::none);
2394 __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2395 __ emit_data64(0x0000000000000000, relocInfo::none);
2396
2397 // lut_lo URL
2398 __ emit_data64(0x1111111111111115, relocInfo::none);
2399 __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2400 __ emit_data64(0x1111111111111115, relocInfo::none);
2401 __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2402
2403 // lut_roll URL
2404 __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2405 __ emit_data64(0x0000000000000000, relocInfo::none);
2406 __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2407 __ emit_data64(0x0000000000000000, relocInfo::none);
2408
2409 // lut_hi
2410 __ emit_data64(0x0804080402011010, relocInfo::none);
2411 __ emit_data64(0x1010101010101010, relocInfo::none);
2412 __ emit_data64(0x0804080402011010, relocInfo::none);
2413 __ emit_data64(0x1010101010101010, relocInfo::none);
2414
2415 return start;
2416 }
2417
2418 address StubGenerator::base64_decoding_table_addr() {
2419 StubGenStubId stub_id = StubGenStubId::decoding_table_base64_id;
2420 StubCodeMark mark(this, stub_id);
2421 address start = __ pc();
2422
2423 __ emit_data64(0xffffffffffffffff, 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(0x3fffffff3effffff, relocInfo::none);
2429 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2430 __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2431 __ emit_data64(0x06050403020100ff, relocInfo::none);
2432 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2433 __ emit_data64(0x161514131211100f, relocInfo::none);
2434 __ emit_data64(0xffffffffff191817, relocInfo::none);
2435 __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2436 __ emit_data64(0x2827262524232221, relocInfo::none);
2437 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2438 __ emit_data64(0xffffffffff333231, relocInfo::none);
2439 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2440 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2441 __ emit_data64(0xffffffffffffffff, relocInfo::none);
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(0xffffffffffffffff, relocInfo::none);
2448 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2449 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2450 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2451 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2452 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2453 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2454 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2455
2456 // URL table
2457 __ emit_data64(0xffffffffffffffff, 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(0xffff3effffffffff, relocInfo::none);
2463 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2464 __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2465 __ emit_data64(0x06050403020100ff, relocInfo::none);
2466 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2467 __ emit_data64(0x161514131211100f, relocInfo::none);
2468 __ emit_data64(0x3fffffffff191817, relocInfo::none);
2469 __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2470 __ emit_data64(0x2827262524232221, relocInfo::none);
2471 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2472 __ emit_data64(0xffffffffff333231, relocInfo::none);
2473 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2474 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2475 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2476 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2477 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2478 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2479 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2480 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2481 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2482 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2483 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2484 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2485 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2486 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2487 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2488 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2489
2490 return start;
2491 }
2492
2493
2494 // Code for generating Base64 decoding.
2495 //
2496 // Based on the article (and associated code) from https://arxiv.org/abs/1910.05109.
2497 //
2498 // Intrinsic function prototype in Base64.java:
2499 // private void decodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL, isMIME) {
2500 address StubGenerator::generate_base64_decodeBlock() {
2501 __ align(CodeEntryAlignment);
2502 StubGenStubId stub_id = StubGenStubId::base64_decodeBlock_id;
2503 StubCodeMark mark(this, stub_id);
2504 address start = __ pc();
2505
2506 __ enter();
2507
2508 // Save callee-saved registers before using them
2509 __ push(r12);
2510 __ push(r13);
2511 __ push(r14);
2512 __ push(r15);
2513 __ push(rbx);
2514
2515 // arguments
2516 const Register source = c_rarg0; // Source Array
2517 const Register start_offset = c_rarg1; // start offset
2518 const Register end_offset = c_rarg2; // end offset
2519 const Register dest = c_rarg3; // destination array
2520 const Register isMIME = rbx;
2521
2522 #ifndef _WIN64
2523 const Register dp = c_rarg4; // Position for writing to dest array
2524 const Register isURL = c_rarg5;// Base64 or URL character set
2525 __ movl(isMIME, Address(rbp, 2 * wordSize));
2526 #else
2527 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
2528 const Address isURL_mem(rbp, 7 * wordSize);
2529 const Register isURL = r10; // pick the volatile windows register
2530 const Register dp = r12;
2531 __ movl(dp, dp_mem);
2532 __ movl(isURL, isURL_mem);
2533 __ movl(isMIME, Address(rbp, 8 * wordSize));
2534 #endif
2535
2536 const XMMRegister lookup_lo = xmm5;
2537 const XMMRegister lookup_hi = xmm6;
2538 const XMMRegister errorvec = xmm7;
2539 const XMMRegister pack16_op = xmm9;
2540 const XMMRegister pack32_op = xmm8;
2541 const XMMRegister input0 = xmm3;
2542 const XMMRegister input1 = xmm20;
2543 const XMMRegister input2 = xmm21;
2544 const XMMRegister input3 = xmm19;
2545 const XMMRegister join01 = xmm12;
2546 const XMMRegister join12 = xmm11;
2547 const XMMRegister join23 = xmm10;
2548 const XMMRegister translated0 = xmm2;
2549 const XMMRegister translated1 = xmm1;
2550 const XMMRegister translated2 = xmm0;
2551 const XMMRegister translated3 = xmm4;
2552
2553 const XMMRegister merged0 = xmm2;
2554 const XMMRegister merged1 = xmm1;
2555 const XMMRegister merged2 = xmm0;
2556 const XMMRegister merged3 = xmm4;
2557 const XMMRegister merge_ab_bc0 = xmm2;
2558 const XMMRegister merge_ab_bc1 = xmm1;
2559 const XMMRegister merge_ab_bc2 = xmm0;
2560 const XMMRegister merge_ab_bc3 = xmm4;
2561
2562 const XMMRegister pack24bits = xmm4;
2563
2564 const Register length = r14;
2565 const Register output_size = r13;
2566 const Register output_mask = r15;
2567 const KRegister input_mask = k1;
2568
2569 const XMMRegister input_initial_valid_b64 = xmm0;
2570 const XMMRegister tmp = xmm10;
2571 const XMMRegister mask = xmm0;
2572 const XMMRegister invalid_b64 = xmm1;
2573
2574 Label L_process256, L_process64, L_process64Loop, L_exit, L_processdata, L_loadURL;
2575 Label L_continue, L_finalBit, L_padding, L_donePadding, L_bruteForce;
2576 Label L_forceLoop, L_bottomLoop, L_checkMIME, L_exit_no_vzero, L_lastChunk;
2577
2578 // calculate length from offsets
2579 __ movl(length, end_offset);
2580 __ subl(length, start_offset);
2581 __ push(dest); // Save for return value calc
2582
2583 // If AVX512 VBMI not supported, just compile non-AVX code
2584 if(VM_Version::supports_avx512_vbmi() &&
2585 VM_Version::supports_avx512bw()) {
2586 __ cmpl(length, 31); // 32-bytes is break-even for AVX-512
2587 __ jcc(Assembler::lessEqual, L_lastChunk);
2588
2589 __ cmpl(isMIME, 0);
2590 __ jcc(Assembler::notEqual, L_lastChunk);
2591
2592 // Load lookup tables based on isURL
2593 __ cmpl(isURL, 0);
2594 __ jcc(Assembler::notZero, L_loadURL);
2595
2596 __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_addr()), Assembler::AVX_512bit, r13);
2597 __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_addr()), Assembler::AVX_512bit, r13);
2598
2599 __ BIND(L_continue);
2600
2601 __ movl(r15, 0x01400140);
2602 __ evpbroadcastd(pack16_op, r15, Assembler::AVX_512bit);
2603
2604 __ movl(r15, 0x00011000);
2605 __ evpbroadcastd(pack32_op, r15, Assembler::AVX_512bit);
2606
2607 __ cmpl(length, 0xff);
2608 __ jcc(Assembler::lessEqual, L_process64);
2609
2610 // load masks required for decoding data
2611 __ BIND(L_processdata);
2612 __ evmovdquq(join01, ExternalAddress(StubRoutines::x86::base64_vbmi_join_0_1_addr()), Assembler::AVX_512bit,r13);
2613 __ evmovdquq(join12, ExternalAddress(StubRoutines::x86::base64_vbmi_join_1_2_addr()), Assembler::AVX_512bit, r13);
2614 __ evmovdquq(join23, ExternalAddress(StubRoutines::x86::base64_vbmi_join_2_3_addr()), Assembler::AVX_512bit, r13);
2615
2616 __ align32();
2617 __ BIND(L_process256);
2618 // Grab input data
2619 __ evmovdquq(input0, Address(source, start_offset, Address::times_1, 0x00), Assembler::AVX_512bit);
2620 __ evmovdquq(input1, Address(source, start_offset, Address::times_1, 0x40), Assembler::AVX_512bit);
2621 __ evmovdquq(input2, Address(source, start_offset, Address::times_1, 0x80), Assembler::AVX_512bit);
2622 __ evmovdquq(input3, Address(source, start_offset, Address::times_1, 0xc0), Assembler::AVX_512bit);
2623
2624 // Copy the low part of the lookup table into the destination of the permutation
2625 __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2626 __ evmovdquq(translated1, lookup_lo, Assembler::AVX_512bit);
2627 __ evmovdquq(translated2, lookup_lo, Assembler::AVX_512bit);
2628 __ evmovdquq(translated3, lookup_lo, Assembler::AVX_512bit);
2629
2630 // Translate the base64 input into "decoded" bytes
2631 __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2632 __ evpermt2b(translated1, input1, lookup_hi, Assembler::AVX_512bit);
2633 __ evpermt2b(translated2, input2, lookup_hi, Assembler::AVX_512bit);
2634 __ evpermt2b(translated3, input3, lookup_hi, Assembler::AVX_512bit);
2635
2636 // OR all of the translations together to check for errors (high-order bit of byte set)
2637 __ vpternlogd(input0, 0xfe, input1, input2, Assembler::AVX_512bit);
2638
2639 __ vpternlogd(input3, 0xfe, translated0, translated1, Assembler::AVX_512bit);
2640 __ vpternlogd(input0, 0xfe, translated2, translated3, Assembler::AVX_512bit);
2641 __ vpor(errorvec, input3, input0, Assembler::AVX_512bit);
2642
2643 // Check if there was an error - if so, try 64-byte chunks
2644 __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2645 __ kortestql(k3, k3);
2646 __ jcc(Assembler::notZero, L_process64);
2647
2648 // The merging and shuffling happens here
2649 // We multiply each byte pair [00dddddd | 00cccccc | 00bbbbbb | 00aaaaaa]
2650 // Multiply [00cccccc] by 2^6 added to [00dddddd] to get [0000cccc | ccdddddd]
2651 // The pack16_op is a vector of 0x01400140, so multiply D by 1 and C by 0x40
2652 __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2653 __ vpmaddubsw(merge_ab_bc1, translated1, pack16_op, Assembler::AVX_512bit);
2654 __ vpmaddubsw(merge_ab_bc2, translated2, pack16_op, Assembler::AVX_512bit);
2655 __ vpmaddubsw(merge_ab_bc3, translated3, pack16_op, Assembler::AVX_512bit);
2656
2657 // Now do the same with packed 16-bit values.
2658 // We start with [0000cccc | ccdddddd | 0000aaaa | aabbbbbb]
2659 // pack32_op is 0x00011000 (2^12, 1), so this multiplies [0000aaaa | aabbbbbb] by 2^12
2660 // and adds [0000cccc | ccdddddd] to yield [00000000 | aaaaaabb | bbbbcccc | ccdddddd]
2661 __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2662 __ vpmaddwd(merged1, merge_ab_bc1, pack32_op, Assembler::AVX_512bit);
2663 __ vpmaddwd(merged2, merge_ab_bc2, pack32_op, Assembler::AVX_512bit);
2664 __ vpmaddwd(merged3, merge_ab_bc3, pack32_op, Assembler::AVX_512bit);
2665
2666 // The join vectors specify which byte from which vector goes into the outputs
2667 // One of every 4 bytes in the extended vector is zero, so we pack them into their
2668 // final positions in the register for storing (256 bytes in, 192 bytes out)
2669 __ evpermt2b(merged0, join01, merged1, Assembler::AVX_512bit);
2670 __ evpermt2b(merged1, join12, merged2, Assembler::AVX_512bit);
2671 __ evpermt2b(merged2, join23, merged3, Assembler::AVX_512bit);
2672
2673 // Store result
2674 __ evmovdquq(Address(dest, dp, Address::times_1, 0x00), merged0, Assembler::AVX_512bit);
2675 __ evmovdquq(Address(dest, dp, Address::times_1, 0x40), merged1, Assembler::AVX_512bit);
2676 __ evmovdquq(Address(dest, dp, Address::times_1, 0x80), merged2, Assembler::AVX_512bit);
2677
2678 __ addptr(source, 0x100);
2679 __ addptr(dest, 0xc0);
2680 __ subl(length, 0x100);
2681 __ cmpl(length, 64 * 4);
2682 __ jcc(Assembler::greaterEqual, L_process256);
2683
2684 // At this point, we've decoded 64 * 4 * n bytes.
2685 // The remaining length will be <= 64 * 4 - 1.
2686 // UNLESS there was an error decoding the first 256-byte chunk. In this
2687 // case, the length will be arbitrarily long.
2688 //
2689 // Note that this will be the path for MIME-encoded strings.
2690
2691 __ BIND(L_process64);
2692
2693 __ evmovdquq(pack24bits, ExternalAddress(StubRoutines::x86::base64_vbmi_pack_vec_addr()), Assembler::AVX_512bit, r13);
2694
2695 __ cmpl(length, 63);
2696 __ jcc(Assembler::lessEqual, L_finalBit);
2697
2698 __ mov64(rax, 0x0000ffffffffffff);
2699 __ kmovql(k2, rax);
2700
2701 __ align32();
2702 __ BIND(L_process64Loop);
2703
2704 // Handle first 64-byte block
2705
2706 __ evmovdquq(input0, Address(source, start_offset), Assembler::AVX_512bit);
2707 __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2708 __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2709
2710 __ vpor(errorvec, translated0, input0, Assembler::AVX_512bit);
2711
2712 // Check for error and bomb out before updating dest
2713 __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2714 __ kortestql(k3, k3);
2715 __ jcc(Assembler::notZero, L_exit);
2716
2717 // Pack output register, selecting correct byte ordering
2718 __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2719 __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2720 __ vpermb(merged0, pack24bits, merged0, Assembler::AVX_512bit);
2721
2722 __ evmovdqub(Address(dest, dp), k2, merged0, true, Assembler::AVX_512bit);
2723
2724 __ subl(length, 64);
2725 __ addptr(source, 64);
2726 __ addptr(dest, 48);
2727
2728 __ cmpl(length, 64);
2729 __ jcc(Assembler::greaterEqual, L_process64Loop);
2730
2731 __ cmpl(length, 0);
2732 __ jcc(Assembler::lessEqual, L_exit);
2733
2734 __ BIND(L_finalBit);
2735 // Now have 1 to 63 bytes left to decode
2736
2737 // I was going to let Java take care of the final fragment
2738 // however it will repeatedly call this routine for every 4 bytes
2739 // of input data, so handle the rest here.
2740 __ movq(rax, -1);
2741 __ bzhiq(rax, rax, length); // Input mask in rax
2742
2743 __ movl(output_size, length);
2744 __ shrl(output_size, 2); // Find (len / 4) * 3 (output length)
2745 __ lea(output_size, Address(output_size, output_size, Address::times_2, 0));
2746 // output_size in r13
2747
2748 // Strip pad characters, if any, and adjust length and mask
2749 __ addq(length, start_offset);
2750 __ cmpb(Address(source, length, Address::times_1, -1), '=');
2751 __ jcc(Assembler::equal, L_padding);
2752
2753 __ BIND(L_donePadding);
2754 __ subq(length, start_offset);
2755
2756 // Output size is (64 - output_size), output mask is (all 1s >> output_size).
2757 __ kmovql(input_mask, rax);
2758 __ movq(output_mask, -1);
2759 __ bzhiq(output_mask, output_mask, output_size);
2760
2761 // Load initial input with all valid base64 characters. Will be used
2762 // in merging source bytes to avoid masking when determining if an error occurred.
2763 __ movl(rax, 0x61616161);
2764 __ evpbroadcastd(input_initial_valid_b64, rax, Assembler::AVX_512bit);
2765
2766 // A register containing all invalid base64 decoded values
2767 __ movl(rax, 0x80808080);
2768 __ evpbroadcastd(invalid_b64, rax, Assembler::AVX_512bit);
2769
2770 // input_mask is in k1
2771 // output_size is in r13
2772 // output_mask is in r15
2773 // zmm0 - free
2774 // zmm1 - 0x00011000
2775 // zmm2 - 0x01400140
2776 // zmm3 - errorvec
2777 // zmm4 - pack vector
2778 // zmm5 - lookup_lo
2779 // zmm6 - lookup_hi
2780 // zmm7 - errorvec
2781 // zmm8 - 0x61616161
2782 // zmm9 - 0x80808080
2783
2784 // Load only the bytes from source, merging into our "fully-valid" register
2785 __ evmovdqub(input_initial_valid_b64, input_mask, Address(source, start_offset, Address::times_1, 0x0), true, Assembler::AVX_512bit);
2786
2787 // Decode all bytes within our merged input
2788 __ evmovdquq(tmp, lookup_lo, Assembler::AVX_512bit);
2789 __ evpermt2b(tmp, input_initial_valid_b64, lookup_hi, Assembler::AVX_512bit);
2790 __ evporq(mask, tmp, input_initial_valid_b64, Assembler::AVX_512bit);
2791
2792 // Check for error. Compare (decoded | initial) to all invalid.
2793 // If any bytes have their high-order bit set, then we have an error.
2794 __ evptestmb(k2, mask, invalid_b64, Assembler::AVX_512bit);
2795 __ kortestql(k2, k2);
2796
2797 // If we have an error, use the brute force loop to decode what we can (4-byte chunks).
2798 __ jcc(Assembler::notZero, L_bruteForce);
2799
2800 // Shuffle output bytes
2801 __ vpmaddubsw(tmp, tmp, pack16_op, Assembler::AVX_512bit);
2802 __ vpmaddwd(tmp, tmp, pack32_op, Assembler::AVX_512bit);
2803
2804 __ vpermb(tmp, pack24bits, tmp, Assembler::AVX_512bit);
2805 __ kmovql(k1, output_mask);
2806 __ evmovdqub(Address(dest, dp), k1, tmp, true, Assembler::AVX_512bit);
2807
2808 __ addptr(dest, output_size);
2809
2810 __ BIND(L_exit);
2811 __ vzeroupper();
2812 __ pop(rax); // Get original dest value
2813 __ subptr(dest, rax); // Number of bytes converted
2814 __ movptr(rax, dest);
2815 __ pop(rbx);
2816 __ pop(r15);
2817 __ pop(r14);
2818 __ pop(r13);
2819 __ pop(r12);
2820 __ leave();
2821 __ ret(0);
2822
2823 __ BIND(L_loadURL);
2824 __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_url_addr()), Assembler::AVX_512bit, r13);
2825 __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_url_addr()), Assembler::AVX_512bit, r13);
2826 __ jmp(L_continue);
2827
2828 __ BIND(L_padding);
2829 __ decrementq(output_size, 1);
2830 __ shrq(rax, 1);
2831
2832 __ cmpb(Address(source, length, Address::times_1, -2), '=');
2833 __ jcc(Assembler::notEqual, L_donePadding);
2834
2835 __ decrementq(output_size, 1);
2836 __ shrq(rax, 1);
2837 __ jmp(L_donePadding);
2838
2839 __ align32();
2840 __ BIND(L_bruteForce);
2841 } // End of if(avx512_vbmi)
2842
2843 if (VM_Version::supports_avx2()) {
2844 Label L_tailProc, L_topLoop, L_enterLoop;
2845
2846 __ cmpl(isMIME, 0);
2847 __ jcc(Assembler::notEqual, L_lastChunk);
2848
2849 // Check for buffer too small (for algorithm)
2850 __ subl(length, 0x2c);
2851 __ jcc(Assembler::less, L_tailProc);
2852
2853 __ shll(isURL, 2);
2854
2855 // Algorithm adapted from https://arxiv.org/abs/1704.00605, "Faster Base64
2856 // Encoding and Decoding using AVX2 Instructions". URL modifications added.
2857
2858 // Set up constants
2859 __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_tables_addr()));
2860 __ vpbroadcastd(xmm4, Address(r13, isURL, Address::times_1), Assembler::AVX_256bit); // 2F or 5F
2861 __ vpbroadcastd(xmm10, Address(r13, isURL, Address::times_1, 0x08), Assembler::AVX_256bit); // -1 or -4
2862 __ vmovdqu(xmm12, Address(r13, 0x10)); // permute
2863 __ vmovdqu(xmm13, Address(r13, 0x30)); // shuffle
2864 __ vpbroadcastd(xmm7, Address(r13, 0x50), Assembler::AVX_256bit); // merge
2865 __ vpbroadcastd(xmm6, Address(r13, 0x54), Assembler::AVX_256bit); // merge mult
2866
2867 __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_LUT_tables_addr()));
2868 __ shll(isURL, 4);
2869 __ vmovdqu(xmm11, Address(r13, isURL, Address::times_1, 0x00)); // lut_lo
2870 __ vmovdqu(xmm8, Address(r13, isURL, Address::times_1, 0x20)); // lut_roll
2871 __ shrl(isURL, 6); // restore isURL
2872 __ vmovdqu(xmm9, Address(r13, 0x80)); // lut_hi
2873 __ jmp(L_enterLoop);
2874
2875 __ align32();
2876 __ bind(L_topLoop);
2877 // Add in the offset value (roll) to get 6-bit out values
2878 __ vpaddb(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
2879 // Merge and permute the output bits into appropriate output byte lanes
2880 __ vpmaddubsw(xmm0, xmm0, xmm7, Assembler::AVX_256bit);
2881 __ vpmaddwd(xmm0, xmm0, xmm6, Assembler::AVX_256bit);
2882 __ vpshufb(xmm0, xmm0, xmm13, Assembler::AVX_256bit);
2883 __ vpermd(xmm0, xmm12, xmm0, Assembler::AVX_256bit);
2884 // Store the output bytes
2885 __ vmovdqu(Address(dest, dp, Address::times_1, 0), xmm0);
2886 __ addptr(source, 0x20);
2887 __ addptr(dest, 0x18);
2888 __ subl(length, 0x20);
2889 __ jcc(Assembler::less, L_tailProc);
2890
2891 __ bind(L_enterLoop);
2892
2893 // Load in encoded string (32 bytes)
2894 __ vmovdqu(xmm2, Address(source, start_offset, Address::times_1, 0x0));
2895 // Extract the high nibble for indexing into the lut tables. High 4 bits are don't care.
2896 __ vpsrld(xmm1, xmm2, 0x4, Assembler::AVX_256bit);
2897 __ vpand(xmm1, xmm4, xmm1, Assembler::AVX_256bit);
2898 // Extract the low nibble. 5F/2F will isolate the low-order 4 bits. High 4 bits are don't care.
2899 __ vpand(xmm3, xmm2, xmm4, Assembler::AVX_256bit);
2900 // Check for special-case (0x2F or 0x5F (URL))
2901 __ vpcmpeqb(xmm0, xmm4, xmm2, Assembler::AVX_256bit);
2902 // Get the bitset based on the low nibble. vpshufb uses low-order 4 bits only.
2903 __ vpshufb(xmm3, xmm11, xmm3, Assembler::AVX_256bit);
2904 // Get the bit value of the high nibble
2905 __ vpshufb(xmm5, xmm9, xmm1, Assembler::AVX_256bit);
2906 // Make sure 2F / 5F shows as valid
2907 __ vpandn(xmm3, xmm0, xmm3, Assembler::AVX_256bit);
2908 // Make adjustment for roll index. For non-URL, this is a no-op,
2909 // for URL, this adjusts by -4. This is to properly index the
2910 // roll value for 2F / 5F.
2911 __ vpand(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
2912 // If the and of the two is non-zero, we have an invalid input character
2913 __ vptest(xmm3, xmm5);
2914 // Extract the "roll" value - value to add to the input to get 6-bit out value
2915 __ vpaddb(xmm0, xmm0, xmm1, Assembler::AVX_256bit); // Handle 2F / 5F
2916 __ vpshufb(xmm0, xmm8, xmm0, Assembler::AVX_256bit);
2917 __ jcc(Assembler::equal, L_topLoop); // Fall through on error
2918
2919 __ bind(L_tailProc);
2920
2921 __ addl(length, 0x2c);
2922
2923 __ vzeroupper();
2924 }
2925
2926 // Use non-AVX code to decode 4-byte chunks into 3 bytes of output
2927
2928 // Register state (Linux):
2929 // r12-15 - saved on stack
2930 // rdi - src
2931 // rsi - sp
2932 // rdx - sl
2933 // rcx - dst
2934 // r8 - dp
2935 // r9 - isURL
2936
2937 // Register state (Windows):
2938 // r12-15 - saved on stack
2939 // rcx - src
2940 // rdx - sp
2941 // r8 - sl
2942 // r9 - dst
2943 // r12 - dp
2944 // r10 - isURL
2945
2946 // Registers (common):
2947 // length (r14) - bytes in src
2948
2949 const Register decode_table = r11;
2950 const Register out_byte_count = rbx;
2951 const Register byte1 = r13;
2952 const Register byte2 = r15;
2953 const Register byte3 = WIN64_ONLY(r8) NOT_WIN64(rdx);
2954 const Register byte4 = WIN64_ONLY(r10) NOT_WIN64(r9);
2955
2956 __ bind(L_lastChunk);
2957
2958 __ shrl(length, 2); // Multiple of 4 bytes only - length is # 4-byte chunks
2959 __ cmpl(length, 0);
2960 __ jcc(Assembler::lessEqual, L_exit_no_vzero);
2961
2962 __ shll(isURL, 8); // index into decode table based on isURL
2963 __ lea(decode_table, ExternalAddress(StubRoutines::x86::base64_decoding_table_addr()));
2964 __ addptr(decode_table, isURL);
2965
2966 __ jmp(L_bottomLoop);
2967
2968 __ align32();
2969 __ BIND(L_forceLoop);
2970 __ shll(byte1, 18);
2971 __ shll(byte2, 12);
2972 __ shll(byte3, 6);
2973 __ orl(byte1, byte2);
2974 __ orl(byte1, byte3);
2975 __ orl(byte1, byte4);
2976
2977 __ addptr(source, 4);
2978
2979 __ movb(Address(dest, dp, Address::times_1, 2), byte1);
2980 __ shrl(byte1, 8);
2981 __ movb(Address(dest, dp, Address::times_1, 1), byte1);
2982 __ shrl(byte1, 8);
2983 __ movb(Address(dest, dp, Address::times_1, 0), byte1);
2984
2985 __ addptr(dest, 3);
2986 __ decrementl(length, 1);
2987 __ jcc(Assembler::zero, L_exit_no_vzero);
2988
2989 __ BIND(L_bottomLoop);
2990 __ load_unsigned_byte(byte1, Address(source, start_offset, Address::times_1, 0x00));
2991 __ load_unsigned_byte(byte2, Address(source, start_offset, Address::times_1, 0x01));
2992 __ load_signed_byte(byte1, Address(decode_table, byte1));
2993 __ load_signed_byte(byte2, Address(decode_table, byte2));
2994 __ load_unsigned_byte(byte3, Address(source, start_offset, Address::times_1, 0x02));
2995 __ load_unsigned_byte(byte4, Address(source, start_offset, Address::times_1, 0x03));
2996 __ load_signed_byte(byte3, Address(decode_table, byte3));
2997 __ load_signed_byte(byte4, Address(decode_table, byte4));
2998
2999 __ mov(rax, byte1);
3000 __ orl(rax, byte2);
3001 __ orl(rax, byte3);
3002 __ orl(rax, byte4);
3003 __ jcc(Assembler::positive, L_forceLoop);
3004
3005 __ BIND(L_exit_no_vzero);
3006 __ pop(rax); // Get original dest value
3007 __ subptr(dest, rax); // Number of bytes converted
3008 __ movptr(rax, dest);
3009 __ pop(rbx);
3010 __ pop(r15);
3011 __ pop(r14);
3012 __ pop(r13);
3013 __ pop(r12);
3014 __ leave();
3015 __ ret(0);
3016
3017 return start;
3018 }
3019
3020
3021 /**
3022 * Arguments:
3023 *
3024 * Inputs:
3025 * c_rarg0 - int crc
3026 * c_rarg1 - byte* buf
3027 * c_rarg2 - int length
3028 *
3029 * Output:
3030 * rax - int crc result
3031 */
3032 address StubGenerator::generate_updateBytesCRC32() {
3033 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3034
3035 __ align(CodeEntryAlignment);
3036 StubGenStubId stub_id = StubGenStubId::updateBytesCRC32_id;
3037 StubCodeMark mark(this, stub_id);
3038
3039 address start = __ pc();
3040
3041 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3042 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3043 // rscratch1: r10
3044 const Register crc = c_rarg0; // crc
3045 const Register buf = c_rarg1; // source java byte array address
3046 const Register len = c_rarg2; // length
3047 const Register table = c_rarg3; // crc_table address (reuse register)
3048 const Register tmp1 = r11;
3049 const Register tmp2 = r10;
3050 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax);
3051
3052 BLOCK_COMMENT("Entry:");
3053 __ enter(); // required for proper stackwalking of RuntimeStub frame
3054
3055 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
3056 VM_Version::supports_avx512bw() &&
3057 VM_Version::supports_avx512vl()) {
3058 // The constants used in the CRC32 algorithm requires the 1's compliment of the initial crc value.
3059 // However, the constant table for CRC32-C assumes the original crc value. Account for this
3060 // difference before calling and after returning.
3061 __ lea(table, ExternalAddress(StubRoutines::x86::crc_table_avx512_addr()));
3062 __ notl(crc);
3063 __ kernel_crc32_avx512(crc, buf, len, table, tmp1, tmp2);
3064 __ notl(crc);
3065 } else {
3066 __ kernel_crc32(crc, buf, len, table, tmp1);
3067 }
3068
3069 __ movl(rax, crc);
3070 __ vzeroupper();
3071 __ leave(); // required for proper stackwalking of RuntimeStub frame
3072 __ ret(0);
3073
3074 return start;
3075 }
3076
3077 /**
3078 * Arguments:
3079 *
3080 * Inputs:
3081 * c_rarg0 - int crc
3082 * c_rarg1 - byte* buf
3083 * c_rarg2 - long length
3084 * c_rarg3 - table_start - optional (present only when doing a library_call,
3085 * not used by x86 algorithm)
3086 *
3087 * Output:
3088 * rax - int crc result
3089 */
3090 address StubGenerator::generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
3091 assert(UseCRC32CIntrinsics, "need SSE4_2");
3092 __ align(CodeEntryAlignment);
3093 StubGenStubId stub_id = StubGenStubId::updateBytesCRC32C_id;
3094 StubCodeMark mark(this, stub_id);
3095 address start = __ pc();
3096
3097 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
3098 //Windows RCX RDX R8 R9 none none XMM0..XMM3
3099 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
3100 const Register crc = c_rarg0; // crc
3101 const Register buf = c_rarg1; // source java byte array address
3102 const Register len = c_rarg2; // length
3103 const Register a = rax;
3104 const Register j = r9;
3105 const Register k = r10;
3106 const Register l = r11;
3107 #ifdef _WIN64
3108 const Register y = rdi;
3109 const Register z = rsi;
3110 #else
3111 const Register y = rcx;
3112 const Register z = r8;
3113 #endif
3114 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
3115
3116 BLOCK_COMMENT("Entry:");
3117 __ enter(); // required for proper stackwalking of RuntimeStub frame
3118 Label L_continue;
3119
3120 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
3121 VM_Version::supports_avx512bw() &&
3122 VM_Version::supports_avx512vl()) {
3123 Label L_doSmall;
3124
3125 __ cmpl(len, 384);
3126 __ jcc(Assembler::lessEqual, L_doSmall);
3127
3128 __ lea(j, ExternalAddress(StubRoutines::x86::crc32c_table_avx512_addr()));
3129 __ kernel_crc32_avx512(crc, buf, len, j, l, k);
3130
3131 __ jmp(L_continue);
3132
3133 __ bind(L_doSmall);
3134 }
3135 #ifdef _WIN64
3136 __ push(y);
3137 __ push(z);
3138 #endif
3139 __ crc32c_ipl_alg2_alt2(crc, buf, len,
3140 a, j, k,
3141 l, y, z,
3142 c_farg0, c_farg1, c_farg2,
3143 is_pclmulqdq_supported);
3144 #ifdef _WIN64
3145 __ pop(z);
3146 __ pop(y);
3147 #endif
3148
3149 __ bind(L_continue);
3150 __ movl(rax, crc);
3151 __ vzeroupper();
3152 __ leave(); // required for proper stackwalking of RuntimeStub frame
3153 __ ret(0);
3154
3155 return start;
3156 }
3157
3158
3159 /**
3160 * Arguments:
3161 *
3162 * Input:
3163 * c_rarg0 - x address
3164 * c_rarg1 - x length
3165 * c_rarg2 - y address
3166 * c_rarg3 - y length
3167 * not Win64
3168 * c_rarg4 - z address
3169 * Win64
3170 * rsp+40 - z address
3171 */
3172 address StubGenerator::generate_multiplyToLen() {
3173 __ align(CodeEntryAlignment);
3174 StubGenStubId stub_id = StubGenStubId::multiplyToLen_id;
3175 StubCodeMark mark(this, stub_id);
3176 address start = __ pc();
3177
3178 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3179 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3180 const Register x = rdi;
3181 const Register xlen = rax;
3182 const Register y = rsi;
3183 const Register ylen = rcx;
3184 const Register z = r8;
3185
3186 // Next registers will be saved on stack in multiply_to_len().
3187 const Register tmp0 = r11;
3188 const Register tmp1 = r12;
3189 const Register tmp2 = r13;
3190 const Register tmp3 = r14;
3191 const Register tmp4 = r15;
3192 const Register tmp5 = rbx;
3193
3194 BLOCK_COMMENT("Entry:");
3195 __ enter(); // required for proper stackwalking of RuntimeStub frame
3196
3197 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
3198 // ylen => rcx, z => r8
3199 // r9 and r10 may be used to save non-volatile registers
3200 #ifdef _WIN64
3201 // last argument (#4) is on stack on Win64
3202 __ movptr(z, Address(rsp, 6 * wordSize));
3203 #endif
3204
3205 __ movptr(xlen, rsi);
3206 __ movptr(y, rdx);
3207 __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
3208
3209 restore_arg_regs();
3210
3211 __ leave(); // required for proper stackwalking of RuntimeStub frame
3212 __ ret(0);
3213
3214 return start;
3215 }
3216
3217 /**
3218 * Arguments:
3219 *
3220 * Input:
3221 * c_rarg0 - obja address
3222 * c_rarg1 - objb address
3223 * c_rarg3 - length length
3224 * c_rarg4 - scale log2_array_indxscale
3225 *
3226 * Output:
3227 * rax - int >= mismatched index, < 0 bitwise complement of tail
3228 */
3229 address StubGenerator::generate_vectorizedMismatch() {
3230 __ align(CodeEntryAlignment);
3231 StubGenStubId stub_id = StubGenStubId::vectorizedMismatch_id;
3232 StubCodeMark mark(this, stub_id);
3233 address start = __ pc();
3234
3235 BLOCK_COMMENT("Entry:");
3236 __ enter();
3237
3238 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3239 const Register scale = c_rarg0; //rcx, will exchange with r9
3240 const Register objb = c_rarg1; //rdx
3241 const Register length = c_rarg2; //r8
3242 const Register obja = c_rarg3; //r9
3243 __ xchgq(obja, scale); //now obja and scale contains the correct contents
3244
3245 const Register tmp1 = r10;
3246 const Register tmp2 = r11;
3247 #endif
3248 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3249 const Register obja = c_rarg0; //U:rdi
3250 const Register objb = c_rarg1; //U:rsi
3251 const Register length = c_rarg2; //U:rdx
3252 const Register scale = c_rarg3; //U:rcx
3253 const Register tmp1 = r8;
3254 const Register tmp2 = r9;
3255 #endif
3256 const Register result = rax; //return value
3257 const XMMRegister vec0 = xmm0;
3258 const XMMRegister vec1 = xmm1;
3259 const XMMRegister vec2 = xmm2;
3260
3261 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
3262
3263 __ vzeroupper();
3264 __ leave();
3265 __ ret(0);
3266
3267 return start;
3268 }
3269
3270 /**
3271 * Arguments:
3272 *
3273 // Input:
3274 // c_rarg0 - x address
3275 // c_rarg1 - x length
3276 // c_rarg2 - z address
3277 // c_rarg3 - z length
3278 *
3279 */
3280 address StubGenerator::generate_squareToLen() {
3281
3282 __ align(CodeEntryAlignment);
3283 StubGenStubId stub_id = StubGenStubId::squareToLen_id;
3284 StubCodeMark mark(this, stub_id);
3285 address start = __ pc();
3286
3287 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3288 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
3289 const Register x = rdi;
3290 const Register len = rsi;
3291 const Register z = r8;
3292 const Register zlen = rcx;
3293
3294 const Register tmp1 = r12;
3295 const Register tmp2 = r13;
3296 const Register tmp3 = r14;
3297 const Register tmp4 = r15;
3298 const Register tmp5 = rbx;
3299
3300 BLOCK_COMMENT("Entry:");
3301 __ enter(); // required for proper stackwalking of RuntimeStub frame
3302
3303 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
3304 // zlen => rcx
3305 // r9 and r10 may be used to save non-volatile registers
3306 __ movptr(r8, rdx);
3307 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3308
3309 restore_arg_regs();
3310
3311 __ leave(); // required for proper stackwalking of RuntimeStub frame
3312 __ ret(0);
3313
3314 return start;
3315 }
3316
3317 address StubGenerator::generate_method_entry_barrier() {
3318 __ align(CodeEntryAlignment);
3319 StubGenStubId stub_id = StubGenStubId::method_entry_barrier_id;
3320 StubCodeMark mark(this, stub_id);
3321 address start = __ pc();
3322
3323 Label deoptimize_label;
3324
3325 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
3326
3327 BLOCK_COMMENT("Entry:");
3328 __ enter(); // save rbp
3329
3330 // save c_rarg0, because we want to use that value.
3331 // We could do without it but then we depend on the number of slots used by pusha
3332 __ push(c_rarg0);
3333
3334 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
3335
3336 __ pusha();
3337
3338 // The method may have floats as arguments, and we must spill them before calling
3339 // the VM runtime.
3340 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
3341 const int xmm_size = wordSize * 2;
3342 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
3343 __ subptr(rsp, xmm_spill_size);
3344 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
3345 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
3346 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
3347 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
3348 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
3349 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
3350 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
3351 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
3352
3353 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
3354
3355 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
3356 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
3357 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
3358 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
3359 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
3360 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
3361 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
3362 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
3363 __ addptr(rsp, xmm_spill_size);
3364
3365 __ cmpl(rax, 1); // 1 means deoptimize
3366 __ jcc(Assembler::equal, deoptimize_label);
3367
3368 __ popa();
3369 __ pop(c_rarg0);
3370
3371 __ leave();
3372
3373 __ addptr(rsp, 1 * wordSize); // cookie
3374 __ ret(0);
3375
3376
3377 __ BIND(deoptimize_label);
3378
3379 __ popa();
3380 __ pop(c_rarg0);
3381
3382 __ leave();
3383
3384 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
3385 // here while still having a correct stack is valuable
3386 __ testptr(rsp, Address(rsp, 0));
3387
3388 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
3389 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
3390
3391 return start;
3392 }
3393
3394 /**
3395 * Arguments:
3396 *
3397 * Input:
3398 * c_rarg0 - out address
3399 * c_rarg1 - in address
3400 * c_rarg2 - offset
3401 * c_rarg3 - len
3402 * not Win64
3403 * c_rarg4 - k
3404 * Win64
3405 * rsp+40 - k
3406 */
3407 address StubGenerator::generate_mulAdd() {
3408 __ align(CodeEntryAlignment);
3409 StubGenStubId stub_id = StubGenStubId::mulAdd_id;
3410 StubCodeMark mark(this, stub_id);
3411 address start = __ pc();
3412
3413 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3414 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3415 const Register out = rdi;
3416 const Register in = rsi;
3417 const Register offset = r11;
3418 const Register len = rcx;
3419 const Register k = r8;
3420
3421 // Next registers will be saved on stack in mul_add().
3422 const Register tmp1 = r12;
3423 const Register tmp2 = r13;
3424 const Register tmp3 = r14;
3425 const Register tmp4 = r15;
3426 const Register tmp5 = rbx;
3427
3428 BLOCK_COMMENT("Entry:");
3429 __ enter(); // required for proper stackwalking of RuntimeStub frame
3430
3431 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
3432 // len => rcx, k => r8
3433 // r9 and r10 may be used to save non-volatile registers
3434 #ifdef _WIN64
3435 // last argument is on stack on Win64
3436 __ movl(k, Address(rsp, 6 * wordSize));
3437 #endif
3438 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
3439 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3440
3441 restore_arg_regs();
3442
3443 __ leave(); // required for proper stackwalking of RuntimeStub frame
3444 __ ret(0);
3445
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 // Generate these first because they are called from other stubs
4085 if (InlineTypeReturnedAsFields) {
4086 StubRoutines::_load_inline_type_fields_in_regs =
4087 generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::load_inline_type_fields_in_regs),
4088 "load_inline_type_fields_in_regs", false);
4089 StubRoutines::_store_inline_type_fields_to_buf =
4090 generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::store_inline_type_fields_to_buf),
4091 "store_inline_type_fields_to_buf", true);
4092 }
4093
4094 StubRoutines::_call_stub_entry =
4095 generate_call_stub(StubRoutines::_call_stub_return_address);
4096
4097 // is referenced by megamorphic call
4098 StubRoutines::_catch_exception_entry = generate_catch_exception();
4099
4100 // atomic calls
4101 StubRoutines::_fence_entry = generate_orderaccess_fence();
4102
4103 // platform dependent
4104 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
4105
4106 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
4107
4108 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
4109 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
4110 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
4111 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
4112
4113 StubRoutines::x86::_float_sign_mask = generate_fp_mask(StubGenStubId::float_sign_mask_id, 0x7FFFFFFF7FFFFFFF);
4114 StubRoutines::x86::_float_sign_flip = generate_fp_mask(StubGenStubId::float_sign_flip_id, 0x8000000080000000);
4115 StubRoutines::x86::_double_sign_mask = generate_fp_mask(StubGenStubId::double_sign_mask_id, 0x7FFFFFFFFFFFFFFF);
4116 StubRoutines::x86::_double_sign_flip = generate_fp_mask(StubGenStubId::double_sign_flip_id, 0x8000000000000000);
4117
4118 if (UseCRC32Intrinsics) {
4119 // set table address before stub generation which use it
4120 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
4121 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
4122 }
4123
4124 if (UseCRC32CIntrinsics) {
4125 bool supports_clmul = VM_Version::supports_clmul();
4126 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
4127 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
4128 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
4129 }
4130
4131 if (VM_Version::supports_float16()) {
4132 // For results consistency both intrinsics should be enabled.
4133 // vmIntrinsics checks InlineIntrinsics flag, no need to check it here.
4134 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_float16ToFloat) &&
4135 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_floatToFloat16)) {
4136 StubRoutines::_hf2f = generate_float16ToFloat();
4137 StubRoutines::_f2hf = generate_floatToFloat16();
4138 }
4139 }
4140
4141 generate_libm_stubs();
4142
4143 StubRoutines::_fmod = generate_libmFmod(); // from stubGenerator_x86_64_fmod.cpp
4144 }
4145
4146 // Call here from the interpreter or compiled code to either load
4147 // multiple returned values from the inline type instance being
4148 // returned to registers or to store returned values to a newly
4149 // allocated inline type instance.
4150 // Register is a class, but it would be assigned numerical value.
4151 // "0" is assigned for xmm0. Thus we need to ignore -Wnonnull.
4152 PRAGMA_DIAG_PUSH
4153 PRAGMA_NONNULL_IGNORED
4154 address StubGenerator::generate_return_value_stub(address destination, const char* name, bool has_res) {
4155 // We need to save all registers the calling convention may use so
4156 // the runtime calls read or update those registers. This needs to
4157 // be in sync with SharedRuntime::java_return_convention().
4158 enum layout {
4159 pad_off = frame::arg_reg_save_area_bytes/BytesPerInt, pad_off_2,
4160 rax_off, rax_off_2,
4161 j_rarg5_off, j_rarg5_2,
4162 j_rarg4_off, j_rarg4_2,
4163 j_rarg3_off, j_rarg3_2,
4164 j_rarg2_off, j_rarg2_2,
4165 j_rarg1_off, j_rarg1_2,
4166 j_rarg0_off, j_rarg0_2,
4167 j_farg0_off, j_farg0_2,
4168 j_farg1_off, j_farg1_2,
4169 j_farg2_off, j_farg2_2,
4170 j_farg3_off, j_farg3_2,
4171 j_farg4_off, j_farg4_2,
4172 j_farg5_off, j_farg5_2,
4173 j_farg6_off, j_farg6_2,
4174 j_farg7_off, j_farg7_2,
4175 rbp_off, rbp_off_2,
4176 return_off, return_off_2,
4177
4178 framesize
4179 };
4180
4181 CodeBuffer buffer(name, 1000, 512);
4182 MacroAssembler* _masm = new MacroAssembler(&buffer);
4183
4184 int frame_size_in_bytes = align_up(framesize*BytesPerInt, 16);
4185 assert(frame_size_in_bytes == framesize*BytesPerInt, "misaligned");
4186 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
4187 int frame_size_in_words = frame_size_in_bytes / wordSize;
4188
4189 OopMapSet *oop_maps = new OopMapSet();
4190 OopMap* map = new OopMap(frame_size_in_slots, 0);
4191
4192 map->set_callee_saved(VMRegImpl::stack2reg(rax_off), rax->as_VMReg());
4193 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg5_off), j_rarg5->as_VMReg());
4194 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg4_off), j_rarg4->as_VMReg());
4195 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg3_off), j_rarg3->as_VMReg());
4196 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg2_off), j_rarg2->as_VMReg());
4197 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg1_off), j_rarg1->as_VMReg());
4198 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg0_off), j_rarg0->as_VMReg());
4199 map->set_callee_saved(VMRegImpl::stack2reg(j_farg0_off), j_farg0->as_VMReg());
4200 map->set_callee_saved(VMRegImpl::stack2reg(j_farg1_off), j_farg1->as_VMReg());
4201 map->set_callee_saved(VMRegImpl::stack2reg(j_farg2_off), j_farg2->as_VMReg());
4202 map->set_callee_saved(VMRegImpl::stack2reg(j_farg3_off), j_farg3->as_VMReg());
4203 map->set_callee_saved(VMRegImpl::stack2reg(j_farg4_off), j_farg4->as_VMReg());
4204 map->set_callee_saved(VMRegImpl::stack2reg(j_farg5_off), j_farg5->as_VMReg());
4205 map->set_callee_saved(VMRegImpl::stack2reg(j_farg6_off), j_farg6->as_VMReg());
4206 map->set_callee_saved(VMRegImpl::stack2reg(j_farg7_off), j_farg7->as_VMReg());
4207
4208 int start = __ offset();
4209
4210 __ subptr(rsp, frame_size_in_bytes - 8 /* return address*/);
4211
4212 __ movptr(Address(rsp, rbp_off * BytesPerInt), rbp);
4213 __ movdbl(Address(rsp, j_farg7_off * BytesPerInt), j_farg7);
4214 __ movdbl(Address(rsp, j_farg6_off * BytesPerInt), j_farg6);
4215 __ movdbl(Address(rsp, j_farg5_off * BytesPerInt), j_farg5);
4216 __ movdbl(Address(rsp, j_farg4_off * BytesPerInt), j_farg4);
4217 __ movdbl(Address(rsp, j_farg3_off * BytesPerInt), j_farg3);
4218 __ movdbl(Address(rsp, j_farg2_off * BytesPerInt), j_farg2);
4219 __ movdbl(Address(rsp, j_farg1_off * BytesPerInt), j_farg1);
4220 __ movdbl(Address(rsp, j_farg0_off * BytesPerInt), j_farg0);
4221
4222 __ movptr(Address(rsp, j_rarg0_off * BytesPerInt), j_rarg0);
4223 __ movptr(Address(rsp, j_rarg1_off * BytesPerInt), j_rarg1);
4224 __ movptr(Address(rsp, j_rarg2_off * BytesPerInt), j_rarg2);
4225 __ movptr(Address(rsp, j_rarg3_off * BytesPerInt), j_rarg3);
4226 __ movptr(Address(rsp, j_rarg4_off * BytesPerInt), j_rarg4);
4227 __ movptr(Address(rsp, j_rarg5_off * BytesPerInt), j_rarg5);
4228 __ movptr(Address(rsp, rax_off * BytesPerInt), rax);
4229
4230 int frame_complete = __ offset();
4231
4232 __ set_last_Java_frame(noreg, noreg, nullptr, rscratch1);
4233
4234 __ mov(c_rarg0, r15_thread);
4235 __ mov(c_rarg1, rax);
4236
4237 __ call(RuntimeAddress(destination));
4238
4239 // Set an oopmap for the call site.
4240
4241 oop_maps->add_gc_map( __ offset() - start, map);
4242
4243 // clear last_Java_sp
4244 __ reset_last_Java_frame(false);
4245
4246 __ movptr(rbp, Address(rsp, rbp_off * BytesPerInt));
4247 __ movdbl(j_farg7, Address(rsp, j_farg7_off * BytesPerInt));
4248 __ movdbl(j_farg6, Address(rsp, j_farg6_off * BytesPerInt));
4249 __ movdbl(j_farg5, Address(rsp, j_farg5_off * BytesPerInt));
4250 __ movdbl(j_farg4, Address(rsp, j_farg4_off * BytesPerInt));
4251 __ movdbl(j_farg3, Address(rsp, j_farg3_off * BytesPerInt));
4252 __ movdbl(j_farg2, Address(rsp, j_farg2_off * BytesPerInt));
4253 __ movdbl(j_farg1, Address(rsp, j_farg1_off * BytesPerInt));
4254 __ movdbl(j_farg0, Address(rsp, j_farg0_off * BytesPerInt));
4255
4256 __ movptr(j_rarg0, Address(rsp, j_rarg0_off * BytesPerInt));
4257 __ movptr(j_rarg1, Address(rsp, j_rarg1_off * BytesPerInt));
4258 __ movptr(j_rarg2, Address(rsp, j_rarg2_off * BytesPerInt));
4259 __ movptr(j_rarg3, Address(rsp, j_rarg3_off * BytesPerInt));
4260 __ movptr(j_rarg4, Address(rsp, j_rarg4_off * BytesPerInt));
4261 __ movptr(j_rarg5, Address(rsp, j_rarg5_off * BytesPerInt));
4262 __ movptr(rax, Address(rsp, rax_off * BytesPerInt));
4263
4264 __ addptr(rsp, frame_size_in_bytes-8);
4265
4266 // check for pending exceptions
4267 Label pending;
4268 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
4269 __ jcc(Assembler::notEqual, pending);
4270
4271 if (has_res) {
4272 __ get_vm_result(rax, r15_thread);
4273 }
4274
4275 __ ret(0);
4276
4277 __ bind(pending);
4278
4279 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
4280 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
4281
4282 // -------------
4283 // make sure all code is generated
4284 _masm->flush();
4285
4286 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, false);
4287 return stub->entry_point();
4288 }
4289
4290 void StubGenerator::generate_continuation_stubs() {
4291 // Continuation stubs:
4292 StubRoutines::_cont_thaw = generate_cont_thaw();
4293 StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
4294 StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
4295 StubRoutines::_cont_preempt_stub = generate_cont_preempt_stub();
4296 }
4297
4298 void StubGenerator::generate_final_stubs() {
4299 // Generates the rest of stubs and initializes the entry points
4300
4301 // support for verify_oop (must happen after universe_init)
4302 if (VerifyOops) {
4303 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
4304 }
4305
4306 // arraycopy stubs used by compilers
4307 generate_arraycopy_stubs();
4308
4309 StubRoutines::_method_entry_barrier = generate_method_entry_barrier();
4310
4311 #ifdef COMPILER2
4312 if (UseSecondarySupersTable) {
4313 StubRoutines::_lookup_secondary_supers_table_slow_path_stub = generate_lookup_secondary_supers_table_slow_path_stub();
4314 if (! InlineSecondarySupersTest) {
4315 generate_lookup_secondary_supers_table_stub();
4316 }
4317 }
4318 #endif // COMPILER2
4319
4320 if (UseVectorizedMismatchIntrinsic) {
4321 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
4322 }
4323
4324 StubRoutines::_upcall_stub_exception_handler = generate_upcall_stub_exception_handler();
4325 StubRoutines::_upcall_stub_load_target = generate_upcall_stub_load_target();
4326 }
4327
4328 void StubGenerator::generate_compiler_stubs() {
4329 #if COMPILER2_OR_JVMCI
4330
4331 // Entry points that are C2 compiler specific.
4332
4333 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask(StubGenStubId::vector_float_sign_mask_id, 0x7FFFFFFF7FFFFFFF);
4334 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask(StubGenStubId::vector_float_sign_flip_id, 0x8000000080000000);
4335 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask(StubGenStubId::vector_double_sign_mask_id, 0x7FFFFFFFFFFFFFFF);
4336 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask(StubGenStubId::vector_double_sign_flip_id, 0x8000000000000000);
4337 StubRoutines::x86::_vector_all_bits_set = generate_vector_mask(StubGenStubId::vector_all_bits_set_id, 0xFFFFFFFFFFFFFFFF);
4338 StubRoutines::x86::_vector_int_mask_cmp_bits = generate_vector_mask(StubGenStubId::vector_int_mask_cmp_bits_id, 0x0000000100000001);
4339 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask(StubGenStubId::vector_short_to_byte_mask_id, 0x00ff00ff00ff00ff);
4340 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask();
4341 StubRoutines::x86::_vector_int_to_byte_mask = generate_vector_mask(StubGenStubId::vector_int_to_byte_mask_id, 0x000000ff000000ff);
4342 StubRoutines::x86::_vector_int_to_short_mask = generate_vector_mask(StubGenStubId::vector_int_to_short_mask_id, 0x0000ffff0000ffff);
4343 StubRoutines::x86::_vector_32_bit_mask = generate_vector_custom_i32(StubGenStubId::vector_32_bit_mask_id, Assembler::AVX_512bit,
4344 0xFFFFFFFF, 0, 0, 0);
4345 StubRoutines::x86::_vector_64_bit_mask = generate_vector_custom_i32(StubGenStubId::vector_64_bit_mask_id, Assembler::AVX_512bit,
4346 0xFFFFFFFF, 0xFFFFFFFF, 0, 0);
4347 StubRoutines::x86::_vector_int_shuffle_mask = generate_vector_mask(StubGenStubId::vector_int_shuffle_mask_id, 0x0302010003020100);
4348 StubRoutines::x86::_vector_byte_shuffle_mask = generate_vector_byte_shuffle_mask();
4349 StubRoutines::x86::_vector_short_shuffle_mask = generate_vector_mask(StubGenStubId::vector_short_shuffle_mask_id, 0x0100010001000100);
4350 StubRoutines::x86::_vector_long_shuffle_mask = generate_vector_mask(StubGenStubId::vector_long_shuffle_mask_id, 0x0000000100000000);
4351 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask(StubGenStubId::vector_long_sign_mask_id, 0x8000000000000000);
4352 StubRoutines::x86::_vector_iota_indices = generate_iota_indices();
4353 StubRoutines::x86::_vector_count_leading_zeros_lut = generate_count_leading_zeros_lut();
4354 StubRoutines::x86::_vector_reverse_bit_lut = generate_vector_reverse_bit_lut();
4355 StubRoutines::x86::_vector_reverse_byte_perm_mask_long = generate_vector_reverse_byte_perm_mask_long();
4356 StubRoutines::x86::_vector_reverse_byte_perm_mask_int = generate_vector_reverse_byte_perm_mask_int();
4357 StubRoutines::x86::_vector_reverse_byte_perm_mask_short = generate_vector_reverse_byte_perm_mask_short();
4358
4359 if (VM_Version::supports_avx2() && !VM_Version::supports_avx512vl()) {
4360 StubRoutines::x86::_compress_perm_table32 = generate_compress_perm_table(StubGenStubId::compress_perm_table32_id);
4361 StubRoutines::x86::_compress_perm_table64 = generate_compress_perm_table(StubGenStubId::compress_perm_table64_id);
4362 StubRoutines::x86::_expand_perm_table32 = generate_expand_perm_table(StubGenStubId::expand_perm_table32_id);
4363 StubRoutines::x86::_expand_perm_table64 = generate_expand_perm_table(StubGenStubId::expand_perm_table64_id);
4364 }
4365
4366 if (VM_Version::supports_avx2() && !VM_Version::supports_avx512_vpopcntdq()) {
4367 // lut implementation influenced by counting 1s algorithm from section 5-1 of Hackers' Delight.
4368 StubRoutines::x86::_vector_popcount_lut = generate_popcount_avx_lut();
4369 }
4370
4371 generate_aes_stubs();
4372
4373 generate_ghash_stubs();
4374
4375 generate_chacha_stubs();
4376
4377 generate_sha3_stubs();
4378
4379 // data cache line writeback
4380 StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
4381 StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
4382
4383 #ifdef COMPILER2
4384 if ((UseAVX == 2) && EnableX86ECoreOpts) {
4385 generate_string_indexof(StubRoutines::_string_indexof_array);
4386 }
4387 #endif
4388
4389 if (UseAdler32Intrinsics) {
4390 StubRoutines::_updateBytesAdler32 = generate_updateBytesAdler32();
4391 }
4392
4393 if (UsePoly1305Intrinsics) {
4394 StubRoutines::_poly1305_processBlocks = generate_poly1305_processBlocks();
4395 }
4396
4397 if (UseIntPolyIntrinsics) {
4398 StubRoutines::_intpoly_montgomeryMult_P256 = generate_intpoly_montgomeryMult_P256();
4399 StubRoutines::_intpoly_assign = generate_intpoly_assign();
4400 }
4401
4402 if (UseMD5Intrinsics) {
4403 StubRoutines::_md5_implCompress = generate_md5_implCompress(StubGenStubId::md5_implCompress_id);
4404 StubRoutines::_md5_implCompressMB = generate_md5_implCompress(StubGenStubId::md5_implCompressMB_id);
4405 }
4406
4407 if (UseSHA1Intrinsics) {
4408 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
4409 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
4410 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(StubGenStubId::sha1_implCompress_id);
4411 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(StubGenStubId::sha1_implCompressMB_id);
4412 }
4413
4414 if (UseSHA256Intrinsics) {
4415 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
4416 char* dst = (char*)StubRoutines::x86::_k256_W;
4417 char* src = (char*)StubRoutines::x86::_k256;
4418 for (int ii = 0; ii < 16; ++ii) {
4419 memcpy(dst + 32 * ii, src + 16 * ii, 16);
4420 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
4421 }
4422 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
4423 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
4424 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(StubGenStubId::sha256_implCompress_id);
4425 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(StubGenStubId::sha256_implCompressMB_id);
4426 }
4427
4428 if (UseSHA512Intrinsics) {
4429 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
4430 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
4431 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(StubGenStubId::sha512_implCompress_id);
4432 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(StubGenStubId::sha512_implCompressMB_id);
4433 }
4434
4435 if (UseBASE64Intrinsics) {
4436 if(VM_Version::supports_avx2()) {
4437 StubRoutines::x86::_avx2_shuffle_base64 = base64_avx2_shuffle_addr();
4438 StubRoutines::x86::_avx2_input_mask_base64 = base64_avx2_input_mask_addr();
4439 StubRoutines::x86::_avx2_lut_base64 = base64_avx2_lut_addr();
4440 StubRoutines::x86::_avx2_decode_tables_base64 = base64_AVX2_decode_tables_addr();
4441 StubRoutines::x86::_avx2_decode_lut_tables_base64 = base64_AVX2_decode_LUT_tables_addr();
4442 }
4443 StubRoutines::x86::_encoding_table_base64 = base64_encoding_table_addr();
4444 if (VM_Version::supports_avx512_vbmi()) {
4445 StubRoutines::x86::_shuffle_base64 = base64_shuffle_addr();
4446 StubRoutines::x86::_lookup_lo_base64 = base64_vbmi_lookup_lo_addr();
4447 StubRoutines::x86::_lookup_hi_base64 = base64_vbmi_lookup_hi_addr();
4448 StubRoutines::x86::_lookup_lo_base64url = base64_vbmi_lookup_lo_url_addr();
4449 StubRoutines::x86::_lookup_hi_base64url = base64_vbmi_lookup_hi_url_addr();
4450 StubRoutines::x86::_pack_vec_base64 = base64_vbmi_pack_vec_addr();
4451 StubRoutines::x86::_join_0_1_base64 = base64_vbmi_join_0_1_addr();
4452 StubRoutines::x86::_join_1_2_base64 = base64_vbmi_join_1_2_addr();
4453 StubRoutines::x86::_join_2_3_base64 = base64_vbmi_join_2_3_addr();
4454 }
4455 StubRoutines::x86::_decoding_table_base64 = base64_decoding_table_addr();
4456 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
4457 StubRoutines::_base64_decodeBlock = generate_base64_decodeBlock();
4458 }
4459
4460 #ifdef COMPILER2
4461 if (UseMultiplyToLenIntrinsic) {
4462 StubRoutines::_multiplyToLen = generate_multiplyToLen();
4463 }
4464 if (UseSquareToLenIntrinsic) {
4465 StubRoutines::_squareToLen = generate_squareToLen();
4466 }
4467 if (UseMulAddIntrinsic) {
4468 StubRoutines::_mulAdd = generate_mulAdd();
4469 }
4470 if (VM_Version::supports_avx512_vbmi2()) {
4471 StubRoutines::_bigIntegerRightShiftWorker = generate_bigIntegerRightShift();
4472 StubRoutines::_bigIntegerLeftShiftWorker = generate_bigIntegerLeftShift();
4473 }
4474 if (UseMontgomeryMultiplyIntrinsic) {
4475 StubRoutines::_montgomeryMultiply
4476 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
4477 }
4478 if (UseMontgomerySquareIntrinsic) {
4479 StubRoutines::_montgomerySquare
4480 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
4481 }
4482
4483 // Load x86_64_sort library on supported hardware to enable SIMD sort and partition intrinsics
4484
4485 if (VM_Version::supports_avx512dq() || VM_Version::supports_avx2()) {
4486 void *libsimdsort = nullptr;
4487 char ebuf_[1024];
4488 char dll_name_simd_sort[JVM_MAXPATHLEN];
4489 if (os::dll_locate_lib(dll_name_simd_sort, sizeof(dll_name_simd_sort), Arguments::get_dll_dir(), "simdsort")) {
4490 libsimdsort = os::dll_load(dll_name_simd_sort, ebuf_, sizeof ebuf_);
4491 }
4492 // Get addresses for SIMD sort and partition routines
4493 if (libsimdsort != nullptr) {
4494 log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "simdsort" JNI_LIB_SUFFIX, p2i(libsimdsort));
4495
4496 snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512_simd_sort() ? "avx512_sort" : "avx2_sort");
4497 StubRoutines::_array_sort = (address)os::dll_lookup(libsimdsort, ebuf_);
4498
4499 snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512_simd_sort() ? "avx512_partition" : "avx2_partition");
4500 StubRoutines::_array_partition = (address)os::dll_lookup(libsimdsort, ebuf_);
4501 }
4502 }
4503
4504 // Get svml stub routine addresses
4505 void *libjsvml = nullptr;
4506 char ebuf[1024];
4507 char dll_name[JVM_MAXPATHLEN];
4508 if (os::dll_locate_lib(dll_name, sizeof(dll_name), Arguments::get_dll_dir(), "jsvml")) {
4509 libjsvml = os::dll_load(dll_name, ebuf, sizeof ebuf);
4510 }
4511 if (libjsvml != nullptr) {
4512 // SVML method naming convention
4513 // All the methods are named as __jsvml_op<T><N>_ha_<VV>
4514 // Where:
4515 // ha stands for high accuracy
4516 // <T> is optional to indicate float/double
4517 // Set to f for vector float operation
4518 // Omitted for vector double operation
4519 // <N> is the number of elements in the vector
4520 // 1, 2, 4, 8, 16
4521 // e.g. 128 bit float vector has 4 float elements
4522 // <VV> indicates the avx/sse level:
4523 // z0 is AVX512, l9 is AVX2, e9 is AVX1 and ex is for SSE2
4524 // e.g. __jsvml_expf16_ha_z0 is the method for computing 16 element vector float exp using AVX 512 insns
4525 // __jsvml_exp8_ha_z0 is the method for computing 8 element vector double exp using AVX 512 insns
4526
4527 log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "jsvml" JNI_LIB_SUFFIX, p2i(libjsvml));
4528 if (UseAVX > 2) {
4529 for (int op = 0; op < VectorSupport::NUM_VECTOR_OP_MATH; op++) {
4530 int vop = VectorSupport::VECTOR_OP_MATH_START + op;
4531 if ((!VM_Version::supports_avx512dq()) &&
4532 (vop == VectorSupport::VECTOR_OP_LOG || vop == VectorSupport::VECTOR_OP_LOG10 || vop == VectorSupport::VECTOR_OP_POW)) {
4533 continue;
4534 }
4535 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf16_ha_z0", VectorSupport::mathname[op]);
4536 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4537
4538 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s8_ha_z0", VectorSupport::mathname[op]);
4539 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4540 }
4541 }
4542 const char* avx_sse_str = (UseAVX >= 2) ? "l9" : ((UseAVX == 1) ? "e9" : "ex");
4543 for (int op = 0; op < VectorSupport::NUM_VECTOR_OP_MATH; op++) {
4544 int vop = VectorSupport::VECTOR_OP_MATH_START + op;
4545 if (vop == VectorSupport::VECTOR_OP_POW) {
4546 continue;
4547 }
4548 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4549 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4550
4551 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4552 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4553
4554 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf8_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4555 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4556
4557 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s1_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4558 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4559
4560 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s2_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4561 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4562
4563 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4564 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4565 }
4566 }
4567
4568 #endif // COMPILER2
4569 #endif // COMPILER2_OR_JVMCI
4570 }
4571
4572 StubGenerator::StubGenerator(CodeBuffer* code, StubGenBlobId blob_id) : StubCodeGenerator(code, blob_id) {
4573 switch(blob_id) {
4574 case initial_id:
4575 generate_initial_stubs();
4576 break;
4577 case continuation_id:
4578 generate_continuation_stubs();
4579 break;
4580 case compiler_id:
4581 generate_compiler_stubs();
4582 break;
4583 case final_id:
4584 generate_final_stubs();
4585 break;
4586 default:
4587 fatal("unexpected blob id: %d", blob_id);
4588 break;
4589 };
4590 }
4591
4592 void StubGenerator_generate(CodeBuffer* code, StubGenBlobId blob_id) {
4593 StubGenerator g(code, blob_id);
4594 }
4595
4596 #undef __