1 /*
2 * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "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 "stubGenerator_x86_64.hpp"
43 #ifdef COMPILER2
44 #include "opto/runtime.hpp"
45 #include "opto/c2_globals.hpp"
46 #endif
47 #if INCLUDE_JVMCI
48 #include "jvmci/jvmci_globals.hpp"
49 #endif
50
51 // For a more detailed description of the stub routine structure
52 // see the comment in stubRoutines.hpp
53
54 #define __ _masm->
55 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
56
57 #ifdef PRODUCT
58 #define BLOCK_COMMENT(str) /* nothing */
59 #else
60 #define BLOCK_COMMENT(str) __ block_comment(str)
61 #endif // PRODUCT
62
63 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
64
65 //
66 // Linux Arguments:
67 // c_rarg0: call wrapper address address
68 // c_rarg1: result address
69 // c_rarg2: result type BasicType
70 // c_rarg3: method Method*
71 // c_rarg4: (interpreter) entry point address
72 // c_rarg5: parameters intptr_t*
73 // 16(rbp): parameter size (in words) int
74 // 24(rbp): thread Thread*
75 //
76 // [ return_from_Java ] <--- rsp
77 // [ argument word n ]
78 // ...
79 // -12 [ argument word 1 ]
80 // -11 [ saved r15 ] <--- rsp_after_call
81 // -10 [ saved r14 ]
82 // -9 [ saved r13 ]
83 // -8 [ saved r12 ]
84 // -7 [ saved rbx ]
85 // -6 [ call wrapper ]
86 // -5 [ result ]
87 // -4 [ result type ]
88 // -3 [ method ]
89 // -2 [ entry point ]
90 // -1 [ parameters ]
91 // 0 [ saved rbp ] <--- rbp
92 // 1 [ return address ]
93 // 2 [ parameter size ]
94 // 3 [ thread ]
95 //
96 // Windows Arguments:
97 // c_rarg0: call wrapper address address
98 // c_rarg1: result address
99 // c_rarg2: result type BasicType
100 // c_rarg3: method Method*
101 // 48(rbp): (interpreter) entry point address
102 // 56(rbp): parameters intptr_t*
103 // 64(rbp): parameter size (in words) int
104 // 72(rbp): thread Thread*
105 //
106 // [ return_from_Java ] <--- rsp
107 // [ argument word n ]
108 // ...
109 // -28 [ argument word 1 ]
110 // -27 [ saved xmm15 ] <--- rsp after_call
111 // [ saved xmm7-xmm14 ]
112 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
113 // -7 [ saved r15 ]
114 // -6 [ saved r14 ]
115 // -5 [ saved r13 ]
116 // -4 [ saved r12 ]
117 // -3 [ saved rdi ]
118 // -2 [ saved rsi ]
119 // -1 [ saved rbx ]
120 // 0 [ saved rbp ] <--- rbp
121 // 1 [ return address ]
122 // 2 [ call wrapper ]
123 // 3 [ result ]
124 // 4 [ result type ]
125 // 5 [ method ]
126 // 6 [ entry point ]
127 // 7 [ parameters ]
128 // 8 [ parameter size ]
129 // 9 [ thread ]
130 //
131 // Windows reserves the callers stack space for arguments 1-4.
132 // We spill c_rarg0-c_rarg3 to this space.
133
134 // Call stub stack layout word offsets from rbp
135 #ifdef _WIN64
136 enum call_stub_layout {
137 xmm_save_first = 6, // save from xmm6
138 xmm_save_last = 15, // to xmm15
139 xmm_save_base = -9,
140 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
141 r15_off = -7,
142 r14_off = -6,
143 r13_off = -5,
144 r12_off = -4,
145 rdi_off = -3,
146 rsi_off = -2,
147 rbx_off = -1,
148 rbp_off = 0,
149 retaddr_off = 1,
150 call_wrapper_off = 2,
151 result_off = 3,
152 result_type_off = 4,
153 method_off = 5,
154 entry_point_off = 6,
155 parameters_off = 7,
156 parameter_size_off = 8,
157 thread_off = 9
158 };
159
160 static Address xmm_save(int reg) {
161 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
162 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
163 }
164 #else // !_WIN64
165 enum call_stub_layout {
166 rsp_after_call_off = -12,
167 mxcsr_off = rsp_after_call_off,
168 r15_off = -11,
169 r14_off = -10,
170 r13_off = -9,
171 r12_off = -8,
172 rbx_off = -7,
173 call_wrapper_off = -6,
174 result_off = -5,
175 result_type_off = -4,
176 method_off = -3,
177 entry_point_off = -2,
178 parameters_off = -1,
179 rbp_off = 0,
180 retaddr_off = 1,
181 parameter_size_off = 2,
182 thread_off = 3
183 };
184 #endif // _WIN64
185
186 address StubGenerator::generate_call_stub(address& return_address) {
187
188 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
189 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
190 "adjust this code");
191 StubCodeMark mark(this, "StubRoutines", "call_stub");
192 address start = __ pc();
193
194 // same as in generate_catch_exception()!
195 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
196
197 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
198 const Address result (rbp, result_off * wordSize);
199 const Address result_type (rbp, result_type_off * wordSize);
200 const Address method (rbp, method_off * wordSize);
201 const Address entry_point (rbp, entry_point_off * wordSize);
202 const Address parameters (rbp, parameters_off * wordSize);
203 const Address parameter_size(rbp, parameter_size_off * wordSize);
204
205 // same as in generate_catch_exception()!
206 const Address thread (rbp, thread_off * wordSize);
207
208 const Address r15_save(rbp, r15_off * wordSize);
209 const Address r14_save(rbp, r14_off * wordSize);
210 const Address r13_save(rbp, r13_off * wordSize);
211 const Address r12_save(rbp, r12_off * wordSize);
212 const Address rbx_save(rbp, rbx_off * wordSize);
213
214 // stub code
215 __ enter();
216 __ subptr(rsp, -rsp_after_call_off * wordSize);
217
218 // save register parameters
219 #ifndef _WIN64
220 __ movptr(parameters, c_rarg5); // parameters
221 __ movptr(entry_point, c_rarg4); // entry_point
222 #endif
223
224 __ movptr(method, c_rarg3); // method
225 __ movl(result_type, c_rarg2); // result type
226 __ movptr(result, c_rarg1); // result
227 __ movptr(call_wrapper, c_rarg0); // call wrapper
228
229 // save regs belonging to calling function
230 __ movptr(rbx_save, rbx);
231 __ movptr(r12_save, r12);
232 __ movptr(r13_save, r13);
233 __ movptr(r14_save, r14);
234 __ movptr(r15_save, r15);
235
236 #ifdef _WIN64
237 int last_reg = 15;
238 for (int i = xmm_save_first; i <= last_reg; i++) {
239 __ movdqu(xmm_save(i), as_XMMRegister(i));
240 }
241
242 const Address rdi_save(rbp, rdi_off * wordSize);
243 const Address rsi_save(rbp, rsi_off * wordSize);
244
245 __ movptr(rsi_save, rsi);
246 __ movptr(rdi_save, rdi);
247 #else
248 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
249 {
250 Label skip_ldmx;
251 __ stmxcsr(mxcsr_save);
252 __ movl(rax, mxcsr_save);
253 __ andl(rax, 0xFFC0); // Mask out any pending exceptions (only check control and mask bits)
254 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
255 __ cmp32(rax, mxcsr_std, rscratch1);
256 __ jcc(Assembler::equal, skip_ldmx);
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(c_rarg0, result);
308 Label is_long, is_float, is_double, exit;
309 __ movl(c_rarg1, result_type);
310 __ cmpl(c_rarg1, T_OBJECT);
311 __ jcc(Assembler::equal, is_long);
312 __ cmpl(c_rarg1, T_LONG);
313 __ jcc(Assembler::equal, is_long);
314 __ cmpl(c_rarg1, T_FLOAT);
315 __ jcc(Assembler::equal, is_float);
316 __ cmpl(c_rarg1, T_DOUBLE);
317 __ jcc(Assembler::equal, is_double);
318 #ifdef ASSERT
319 // make sure the type is INT
320 {
321 Label L;
322 __ cmpl(c_rarg1, 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(c_rarg0, 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(is_long);
389 __ movq(Address(c_rarg0, 0), rax);
390 __ jmp(exit);
391
392 __ BIND(is_float);
393 __ movflt(Address(c_rarg0, 0), xmm0);
394 __ jmp(exit);
395
396 __ BIND(is_double);
397 __ movdbl(Address(c_rarg0, 0), xmm0);
398 __ jmp(exit);
399
400 return start;
401 }
402
403 // Return point for a Java call if there's an exception thrown in
404 // Java code. The exception is caught and transformed into a
405 // pending exception stored in JavaThread that can be tested from
406 // within the VM.
407 //
408 // Note: Usually the parameters are removed by the callee. In case
409 // of an exception crossing an activation frame boundary, that is
410 // not the case if the callee is compiled code => need to setup the
411 // rsp.
412 //
413 // rax: exception oop
414
415 address StubGenerator::generate_catch_exception() {
416 StubCodeMark mark(this, "StubRoutines", "catch_exception");
417 address start = __ pc();
418
419 // same as in generate_call_stub():
420 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
421 const Address thread (rbp, thread_off * wordSize);
422
423 #ifdef ASSERT
424 // verify that threads correspond
425 {
426 Label L1, L2, L3;
427 __ cmpptr(r15_thread, thread);
428 __ jcc(Assembler::equal, L1);
429 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
430 __ bind(L1);
431 __ get_thread(rbx);
432 __ cmpptr(r15_thread, thread);
433 __ jcc(Assembler::equal, L2);
434 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
435 __ bind(L2);
436 __ cmpptr(r15_thread, rbx);
437 __ jcc(Assembler::equal, L3);
438 __ stop("StubRoutines::catch_exception: threads must correspond");
439 __ bind(L3);
440 }
441 #endif
442
443 // set pending exception
444 __ verify_oop(rax);
445
446 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
447 __ lea(rscratch1, ExternalAddress((address)__FILE__));
448 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
449 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
450
451 // complete return to VM
452 assert(StubRoutines::_call_stub_return_address != nullptr,
453 "_call_stub_return_address must have been generated before");
454 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
455
456 return start;
457 }
458
459 // Continuation point for runtime calls returning with a pending
460 // exception. The pending exception check happened in the runtime
461 // or native call stub. The pending exception in Thread is
462 // converted into a Java-level exception.
463 //
464 // Contract with Java-level exception handlers:
465 // rax: exception
466 // rdx: throwing pc
467 //
468 // NOTE: At entry of this stub, exception-pc must be on stack !!
469
470 address StubGenerator::generate_forward_exception() {
471 StubCodeMark mark(this, "StubRoutines", "forward exception");
472 address start = __ pc();
473
474 // Upon entry, the sp points to the return address returning into
475 // Java (interpreted or compiled) code; i.e., the return address
476 // becomes the throwing pc.
477 //
478 // Arguments pushed before the runtime call are still on the stack
479 // but the exception handler will reset the stack pointer ->
480 // ignore them. A potential result in registers can be ignored as
481 // well.
482
483 #ifdef ASSERT
484 // make sure this code is only executed if there is a pending exception
485 {
486 Label L;
487 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
488 __ jcc(Assembler::notEqual, L);
489 __ stop("StubRoutines::forward exception: no pending exception (1)");
490 __ bind(L);
491 }
492 #endif
493
494 // compute exception handler into rbx
495 __ movptr(c_rarg0, Address(rsp, 0));
496 BLOCK_COMMENT("call exception_handler_for_return_address");
497 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
498 SharedRuntime::exception_handler_for_return_address),
499 r15_thread, c_rarg0);
500 __ mov(rbx, rax);
501
502 // setup rax & rdx, remove return address & clear pending exception
503 __ pop(rdx);
504 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
505 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
506
507 #ifdef ASSERT
508 // make sure exception is set
509 {
510 Label L;
511 __ testptr(rax, rax);
512 __ jcc(Assembler::notEqual, L);
513 __ stop("StubRoutines::forward exception: no pending exception (2)");
514 __ bind(L);
515 }
516 #endif
517
518 // continue at exception handler (return address removed)
519 // rax: exception
520 // rbx: exception handler
521 // rdx: throwing pc
522 __ verify_oop(rax);
523 __ jmp(rbx);
524
525 return start;
526 }
527
528 // Support for intptr_t OrderAccess::fence()
529 //
530 // Arguments :
531 //
532 // Result:
533 address StubGenerator::generate_orderaccess_fence() {
534 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
535 address start = __ pc();
536
537 __ membar(Assembler::StoreLoad);
538 __ ret(0);
539
540 return start;
541 }
542
543
544 // Support for intptr_t get_previous_sp()
545 //
546 // This routine is used to find the previous stack pointer for the
547 // caller.
548 address StubGenerator::generate_get_previous_sp() {
549 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
550 address start = __ pc();
551
552 __ movptr(rax, rsp);
553 __ addptr(rax, 8); // return address is at the top of the stack.
554 __ ret(0);
555
556 return start;
557 }
558
559 //----------------------------------------------------------------------------------------------------
560 // Support for void verify_mxcsr()
561 //
562 // This routine is used with -Xcheck:jni to verify that native
563 // JNI code does not return to Java code without restoring the
564 // MXCSR register to our expected state.
565
566 address StubGenerator::generate_verify_mxcsr() {
567 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
568 address start = __ pc();
569
570 const Address mxcsr_save(rsp, 0);
571
572 if (CheckJNICalls) {
573 Label ok_ret;
574 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
575 __ push(rax);
576 __ subptr(rsp, wordSize); // allocate a temp location
577 __ stmxcsr(mxcsr_save);
578 __ movl(rax, mxcsr_save);
579 __ andl(rax, 0xFFC0); // Mask out any pending exceptions (only check control and mask bits)
580 __ cmp32(rax, mxcsr_std, rscratch1);
581 __ jcc(Assembler::equal, ok_ret);
582
583 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
584
585 __ ldmxcsr(mxcsr_std, rscratch1);
586
587 __ bind(ok_ret);
588 __ addptr(rsp, wordSize);
589 __ pop(rax);
590 }
591
592 __ ret(0);
593
594 return start;
595 }
596
597 address StubGenerator::generate_f2i_fixup() {
598 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
599 Address inout(rsp, 5 * wordSize); // return address + 4 saves
600
601 address start = __ pc();
602
603 Label L;
604
605 __ push(rax);
606 __ push(c_rarg3);
607 __ push(c_rarg2);
608 __ push(c_rarg1);
609
610 __ movl(rax, 0x7f800000);
611 __ xorl(c_rarg3, c_rarg3);
612 __ movl(c_rarg2, inout);
613 __ movl(c_rarg1, c_rarg2);
614 __ andl(c_rarg1, 0x7fffffff);
615 __ cmpl(rax, c_rarg1); // NaN? -> 0
616 __ jcc(Assembler::negative, L);
617 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
618 __ movl(c_rarg3, 0x80000000);
619 __ movl(rax, 0x7fffffff);
620 __ cmovl(Assembler::positive, c_rarg3, rax);
621
622 __ bind(L);
623 __ movptr(inout, c_rarg3);
624
625 __ pop(c_rarg1);
626 __ pop(c_rarg2);
627 __ pop(c_rarg3);
628 __ pop(rax);
629
630 __ ret(0);
631
632 return start;
633 }
634
635 address StubGenerator::generate_f2l_fixup() {
636 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
637 Address inout(rsp, 5 * wordSize); // return address + 4 saves
638 address start = __ pc();
639
640 Label L;
641
642 __ push(rax);
643 __ push(c_rarg3);
644 __ push(c_rarg2);
645 __ push(c_rarg1);
646
647 __ movl(rax, 0x7f800000);
648 __ xorl(c_rarg3, c_rarg3);
649 __ movl(c_rarg2, inout);
650 __ movl(c_rarg1, c_rarg2);
651 __ andl(c_rarg1, 0x7fffffff);
652 __ cmpl(rax, c_rarg1); // NaN? -> 0
653 __ jcc(Assembler::negative, L);
654 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
655 __ mov64(c_rarg3, 0x8000000000000000);
656 __ mov64(rax, 0x7fffffffffffffff);
657 __ cmov(Assembler::positive, c_rarg3, rax);
658
659 __ bind(L);
660 __ movptr(inout, c_rarg3);
661
662 __ pop(c_rarg1);
663 __ pop(c_rarg2);
664 __ pop(c_rarg3);
665 __ pop(rax);
666
667 __ ret(0);
668
669 return start;
670 }
671
672 address StubGenerator::generate_d2i_fixup() {
673 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
674 Address inout(rsp, 6 * wordSize); // return address + 5 saves
675
676 address start = __ pc();
677
678 Label L;
679
680 __ push(rax);
681 __ push(c_rarg3);
682 __ push(c_rarg2);
683 __ push(c_rarg1);
684 __ push(c_rarg0);
685
686 __ movl(rax, 0x7ff00000);
687 __ movq(c_rarg2, inout);
688 __ movl(c_rarg3, c_rarg2);
689 __ mov(c_rarg1, c_rarg2);
690 __ mov(c_rarg0, c_rarg2);
691 __ negl(c_rarg3);
692 __ shrptr(c_rarg1, 0x20);
693 __ orl(c_rarg3, c_rarg2);
694 __ andl(c_rarg1, 0x7fffffff);
695 __ xorl(c_rarg2, c_rarg2);
696 __ shrl(c_rarg3, 0x1f);
697 __ orl(c_rarg1, c_rarg3);
698 __ cmpl(rax, c_rarg1);
699 __ jcc(Assembler::negative, L); // NaN -> 0
700 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
701 __ movl(c_rarg2, 0x80000000);
702 __ movl(rax, 0x7fffffff);
703 __ cmov(Assembler::positive, c_rarg2, rax);
704
705 __ bind(L);
706 __ movptr(inout, c_rarg2);
707
708 __ pop(c_rarg0);
709 __ pop(c_rarg1);
710 __ pop(c_rarg2);
711 __ pop(c_rarg3);
712 __ pop(rax);
713
714 __ ret(0);
715
716 return start;
717 }
718
719 address StubGenerator::generate_d2l_fixup() {
720 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
721 Address inout(rsp, 6 * wordSize); // return address + 5 saves
722
723 address start = __ pc();
724
725 Label L;
726
727 __ push(rax);
728 __ push(c_rarg3);
729 __ push(c_rarg2);
730 __ push(c_rarg1);
731 __ push(c_rarg0);
732
733 __ movl(rax, 0x7ff00000);
734 __ movq(c_rarg2, inout);
735 __ movl(c_rarg3, c_rarg2);
736 __ mov(c_rarg1, c_rarg2);
737 __ mov(c_rarg0, c_rarg2);
738 __ negl(c_rarg3);
739 __ shrptr(c_rarg1, 0x20);
740 __ orl(c_rarg3, c_rarg2);
741 __ andl(c_rarg1, 0x7fffffff);
742 __ xorl(c_rarg2, c_rarg2);
743 __ shrl(c_rarg3, 0x1f);
744 __ orl(c_rarg1, c_rarg3);
745 __ cmpl(rax, c_rarg1);
746 __ jcc(Assembler::negative, L); // NaN -> 0
747 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
748 __ mov64(c_rarg2, 0x8000000000000000);
749 __ mov64(rax, 0x7fffffffffffffff);
750 __ cmovq(Assembler::positive, c_rarg2, rax);
751
752 __ bind(L);
753 __ movq(inout, c_rarg2);
754
755 __ pop(c_rarg0);
756 __ pop(c_rarg1);
757 __ pop(c_rarg2);
758 __ pop(c_rarg3);
759 __ pop(rax);
760
761 __ ret(0);
762
763 return start;
764 }
765
766 address StubGenerator::generate_count_leading_zeros_lut(const char *stub_name) {
767 __ align64();
768 StubCodeMark mark(this, "StubRoutines", stub_name);
769 address start = __ pc();
770
771 __ emit_data64(0x0101010102020304, relocInfo::none);
772 __ emit_data64(0x0000000000000000, relocInfo::none);
773 __ emit_data64(0x0101010102020304, relocInfo::none);
774 __ emit_data64(0x0000000000000000, relocInfo::none);
775 __ emit_data64(0x0101010102020304, relocInfo::none);
776 __ emit_data64(0x0000000000000000, relocInfo::none);
777 __ emit_data64(0x0101010102020304, relocInfo::none);
778 __ emit_data64(0x0000000000000000, relocInfo::none);
779
780 return start;
781 }
782
783 address StubGenerator::generate_popcount_avx_lut(const char *stub_name) {
784 __ align64();
785 StubCodeMark mark(this, "StubRoutines", stub_name);
786 address start = __ pc();
787
788 __ emit_data64(0x0302020102010100, relocInfo::none);
789 __ emit_data64(0x0403030203020201, relocInfo::none);
790 __ emit_data64(0x0302020102010100, relocInfo::none);
791 __ emit_data64(0x0403030203020201, relocInfo::none);
792 __ emit_data64(0x0302020102010100, relocInfo::none);
793 __ emit_data64(0x0403030203020201, relocInfo::none);
794 __ emit_data64(0x0302020102010100, relocInfo::none);
795 __ emit_data64(0x0403030203020201, relocInfo::none);
796
797 return start;
798 }
799
800 address StubGenerator::generate_iota_indices(const char *stub_name) {
801 __ align(CodeEntryAlignment);
802 StubCodeMark mark(this, "StubRoutines", stub_name);
803 address start = __ pc();
804 // B
805 __ emit_data64(0x0706050403020100, relocInfo::none);
806 __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
807 __ emit_data64(0x1716151413121110, relocInfo::none);
808 __ emit_data64(0x1F1E1D1C1B1A1918, relocInfo::none);
809 __ emit_data64(0x2726252423222120, relocInfo::none);
810 __ emit_data64(0x2F2E2D2C2B2A2928, relocInfo::none);
811 __ emit_data64(0x3736353433323130, relocInfo::none);
812 __ emit_data64(0x3F3E3D3C3B3A3938, relocInfo::none);
813 // W
814 __ emit_data64(0x0003000200010000, relocInfo::none);
815 __ emit_data64(0x0007000600050004, relocInfo::none);
816 __ emit_data64(0x000B000A00090008, relocInfo::none);
817 __ emit_data64(0x000F000E000D000C, relocInfo::none);
818 __ emit_data64(0x0013001200110010, relocInfo::none);
819 __ emit_data64(0x0017001600150014, relocInfo::none);
820 __ emit_data64(0x001B001A00190018, relocInfo::none);
821 __ emit_data64(0x001F001E001D001C, relocInfo::none);
822 // D
823 __ emit_data64(0x0000000100000000, relocInfo::none);
824 __ emit_data64(0x0000000300000002, relocInfo::none);
825 __ emit_data64(0x0000000500000004, relocInfo::none);
826 __ emit_data64(0x0000000700000006, relocInfo::none);
827 __ emit_data64(0x0000000900000008, relocInfo::none);
828 __ emit_data64(0x0000000B0000000A, relocInfo::none);
829 __ emit_data64(0x0000000D0000000C, relocInfo::none);
830 __ emit_data64(0x0000000F0000000E, relocInfo::none);
831 // Q
832 __ emit_data64(0x0000000000000000, relocInfo::none);
833 __ emit_data64(0x0000000000000001, relocInfo::none);
834 __ emit_data64(0x0000000000000002, relocInfo::none);
835 __ emit_data64(0x0000000000000003, relocInfo::none);
836 __ emit_data64(0x0000000000000004, relocInfo::none);
837 __ emit_data64(0x0000000000000005, relocInfo::none);
838 __ emit_data64(0x0000000000000006, relocInfo::none);
839 __ emit_data64(0x0000000000000007, relocInfo::none);
840 // D - FP
841 __ emit_data64(0x3F80000000000000, relocInfo::none); // 0.0f, 1.0f
842 __ emit_data64(0x4040000040000000, relocInfo::none); // 2.0f, 3.0f
843 __ emit_data64(0x40A0000040800000, relocInfo::none); // 4.0f, 5.0f
844 __ emit_data64(0x40E0000040C00000, relocInfo::none); // 6.0f, 7.0f
845 __ emit_data64(0x4110000041000000, relocInfo::none); // 8.0f, 9.0f
846 __ emit_data64(0x4130000041200000, relocInfo::none); // 10.0f, 11.0f
847 __ emit_data64(0x4150000041400000, relocInfo::none); // 12.0f, 13.0f
848 __ emit_data64(0x4170000041600000, relocInfo::none); // 14.0f, 15.0f
849 // Q - FP
850 __ emit_data64(0x0000000000000000, relocInfo::none); // 0.0d
851 __ emit_data64(0x3FF0000000000000, relocInfo::none); // 1.0d
852 __ emit_data64(0x4000000000000000, relocInfo::none); // 2.0d
853 __ emit_data64(0x4008000000000000, relocInfo::none); // 3.0d
854 __ emit_data64(0x4010000000000000, relocInfo::none); // 4.0d
855 __ emit_data64(0x4014000000000000, relocInfo::none); // 5.0d
856 __ emit_data64(0x4018000000000000, relocInfo::none); // 6.0d
857 __ emit_data64(0x401c000000000000, relocInfo::none); // 7.0d
858 return start;
859 }
860
861 address StubGenerator::generate_vector_reverse_bit_lut(const char *stub_name) {
862 __ align(CodeEntryAlignment);
863 StubCodeMark mark(this, "StubRoutines", stub_name);
864 address start = __ pc();
865
866 __ emit_data64(0x0E060A020C040800, relocInfo::none);
867 __ emit_data64(0x0F070B030D050901, relocInfo::none);
868 __ emit_data64(0x0E060A020C040800, relocInfo::none);
869 __ emit_data64(0x0F070B030D050901, relocInfo::none);
870 __ emit_data64(0x0E060A020C040800, relocInfo::none);
871 __ emit_data64(0x0F070B030D050901, relocInfo::none);
872 __ emit_data64(0x0E060A020C040800, relocInfo::none);
873 __ emit_data64(0x0F070B030D050901, relocInfo::none);
874
875 return start;
876 }
877
878 address StubGenerator::generate_vector_reverse_byte_perm_mask_long(const char *stub_name) {
879 __ align(CodeEntryAlignment);
880 StubCodeMark mark(this, "StubRoutines", stub_name);
881 address start = __ pc();
882
883 __ emit_data64(0x0001020304050607, relocInfo::none);
884 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
885 __ emit_data64(0x0001020304050607, relocInfo::none);
886 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
887 __ emit_data64(0x0001020304050607, relocInfo::none);
888 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
889 __ emit_data64(0x0001020304050607, relocInfo::none);
890 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
891
892 return start;
893 }
894
895 address StubGenerator::generate_vector_reverse_byte_perm_mask_int(const char *stub_name) {
896 __ align(CodeEntryAlignment);
897 StubCodeMark mark(this, "StubRoutines", stub_name);
898 address start = __ pc();
899
900 __ emit_data64(0x0405060700010203, relocInfo::none);
901 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
902 __ emit_data64(0x0405060700010203, relocInfo::none);
903 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
904 __ emit_data64(0x0405060700010203, relocInfo::none);
905 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
906 __ emit_data64(0x0405060700010203, relocInfo::none);
907 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
908
909 return start;
910 }
911
912 address StubGenerator::generate_vector_reverse_byte_perm_mask_short(const char *stub_name) {
913 __ align(CodeEntryAlignment);
914 StubCodeMark mark(this, "StubRoutines", stub_name);
915 address start = __ pc();
916
917 __ emit_data64(0x0607040502030001, relocInfo::none);
918 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
919 __ emit_data64(0x0607040502030001, relocInfo::none);
920 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
921 __ emit_data64(0x0607040502030001, relocInfo::none);
922 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
923 __ emit_data64(0x0607040502030001, relocInfo::none);
924 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
925
926 return start;
927 }
928
929 address StubGenerator::generate_vector_byte_shuffle_mask(const char *stub_name) {
930 __ align(CodeEntryAlignment);
931 StubCodeMark mark(this, "StubRoutines", stub_name);
932 address start = __ pc();
933
934 __ emit_data64(0x7070707070707070, relocInfo::none);
935 __ emit_data64(0x7070707070707070, relocInfo::none);
936 __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
937 __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
938
939 return start;
940 }
941
942 address StubGenerator::generate_fp_mask(const char *stub_name, int64_t mask) {
943 __ align(CodeEntryAlignment);
944 StubCodeMark mark(this, "StubRoutines", stub_name);
945 address start = __ pc();
946
947 __ emit_data64( mask, relocInfo::none );
948 __ emit_data64( mask, relocInfo::none );
949
950 return start;
951 }
952
953 address StubGenerator::generate_compress_perm_table(const char *stub_name, int32_t esize) {
954 __ align(CodeEntryAlignment);
955 StubCodeMark mark(this, "StubRoutines", stub_name);
956 address start = __ pc();
957 if (esize == 32) {
958 // Loop to generate 256 x 8 int compression permute index table. A row is
959 // accessed using 8 bit index computed using vector mask. An entry in
960 // a row holds either a valid permute index corresponding to set bit position
961 // or a -1 (default) value.
962 for (int mask = 0; mask < 256; mask++) {
963 int ctr = 0;
964 for (int j = 0; j < 8; j++) {
965 if (mask & (1 << j)) {
966 __ emit_data(j, relocInfo::none);
967 ctr++;
968 }
969 }
970 for (; ctr < 8; ctr++) {
971 __ emit_data(-1, relocInfo::none);
972 }
973 }
974 } else {
975 assert(esize == 64, "");
976 // Loop to generate 16 x 4 long compression permute index table. A row is
977 // accessed using 4 bit index computed using vector mask. An entry in
978 // a row holds either a valid permute index pair for a quadword corresponding
979 // to set bit position or a -1 (default) value.
980 for (int mask = 0; mask < 16; mask++) {
981 int ctr = 0;
982 for (int j = 0; j < 4; j++) {
983 if (mask & (1 << j)) {
984 __ emit_data(2 * j, relocInfo::none);
985 __ emit_data(2 * j + 1, relocInfo::none);
986 ctr++;
987 }
988 }
989 for (; ctr < 4; ctr++) {
990 __ emit_data64(-1L, relocInfo::none);
991 }
992 }
993 }
994 return start;
995 }
996
997 address StubGenerator::generate_expand_perm_table(const char *stub_name, int32_t esize) {
998 __ align(CodeEntryAlignment);
999 StubCodeMark mark(this, "StubRoutines", stub_name);
1000 address start = __ pc();
1001 if (esize == 32) {
1002 // Loop to generate 256 x 8 int expand permute index table. A row is accessed
1003 // using 8 bit index computed using vector mask. An entry in a row holds either
1004 // a valid permute index (starting from least significant lane) placed at poisition
1005 // corresponding to set bit position or a -1 (default) value.
1006 for (int mask = 0; mask < 256; mask++) {
1007 int ctr = 0;
1008 for (int j = 0; j < 8; j++) {
1009 if (mask & (1 << j)) {
1010 __ emit_data(ctr++, relocInfo::none);
1011 } else {
1012 __ emit_data(-1, relocInfo::none);
1013 }
1014 }
1015 }
1016 } else {
1017 assert(esize == 64, "");
1018 // Loop to generate 16 x 4 long expand permute index table. A row is accessed
1019 // using 4 bit index computed using vector mask. An entry in a row holds either
1020 // a valid doubleword permute index pair representing a quadword index (starting
1021 // from least significant lane) placed at poisition corresponding to set bit
1022 // position or a -1 (default) value.
1023 for (int mask = 0; mask < 16; mask++) {
1024 int ctr = 0;
1025 for (int j = 0; j < 4; j++) {
1026 if (mask & (1 << j)) {
1027 __ emit_data(2 * ctr, relocInfo::none);
1028 __ emit_data(2 * ctr + 1, relocInfo::none);
1029 ctr++;
1030 } else {
1031 __ emit_data64(-1L, relocInfo::none);
1032 }
1033 }
1034 }
1035 }
1036 return start;
1037 }
1038
1039 address StubGenerator::generate_vector_mask(const char *stub_name, int64_t mask) {
1040 __ align(CodeEntryAlignment);
1041 StubCodeMark mark(this, "StubRoutines", stub_name);
1042 address start = __ pc();
1043
1044 __ emit_data64(mask, relocInfo::none);
1045 __ emit_data64(mask, relocInfo::none);
1046 __ emit_data64(mask, relocInfo::none);
1047 __ emit_data64(mask, relocInfo::none);
1048 __ emit_data64(mask, relocInfo::none);
1049 __ emit_data64(mask, relocInfo::none);
1050 __ emit_data64(mask, relocInfo::none);
1051 __ emit_data64(mask, relocInfo::none);
1052
1053 return start;
1054 }
1055
1056 address StubGenerator::generate_vector_byte_perm_mask(const char *stub_name) {
1057 __ align(CodeEntryAlignment);
1058 StubCodeMark mark(this, "StubRoutines", stub_name);
1059 address start = __ pc();
1060
1061 __ emit_data64(0x0000000000000001, relocInfo::none);
1062 __ emit_data64(0x0000000000000003, relocInfo::none);
1063 __ emit_data64(0x0000000000000005, relocInfo::none);
1064 __ emit_data64(0x0000000000000007, relocInfo::none);
1065 __ emit_data64(0x0000000000000000, relocInfo::none);
1066 __ emit_data64(0x0000000000000002, relocInfo::none);
1067 __ emit_data64(0x0000000000000004, relocInfo::none);
1068 __ emit_data64(0x0000000000000006, relocInfo::none);
1069
1070 return start;
1071 }
1072
1073 address StubGenerator::generate_vector_fp_mask(const char *stub_name, int64_t mask) {
1074 __ align(CodeEntryAlignment);
1075 StubCodeMark mark(this, "StubRoutines", stub_name);
1076 address start = __ pc();
1077
1078 __ emit_data64(mask, relocInfo::none);
1079 __ emit_data64(mask, relocInfo::none);
1080 __ emit_data64(mask, relocInfo::none);
1081 __ emit_data64(mask, relocInfo::none);
1082 __ emit_data64(mask, relocInfo::none);
1083 __ emit_data64(mask, relocInfo::none);
1084 __ emit_data64(mask, relocInfo::none);
1085 __ emit_data64(mask, relocInfo::none);
1086
1087 return start;
1088 }
1089
1090 address StubGenerator::generate_vector_custom_i32(const char *stub_name, Assembler::AvxVectorLen len,
1091 int32_t val0, int32_t val1, int32_t val2, int32_t val3,
1092 int32_t val4, int32_t val5, int32_t val6, int32_t val7,
1093 int32_t val8, int32_t val9, int32_t val10, int32_t val11,
1094 int32_t val12, int32_t val13, int32_t val14, int32_t val15) {
1095 __ align(CodeEntryAlignment);
1096 StubCodeMark mark(this, "StubRoutines", stub_name);
1097 address start = __ pc();
1098
1099 assert(len != Assembler::AVX_NoVec, "vector len must be specified");
1100 __ emit_data(val0, relocInfo::none, 0);
1101 __ emit_data(val1, relocInfo::none, 0);
1102 __ emit_data(val2, relocInfo::none, 0);
1103 __ emit_data(val3, relocInfo::none, 0);
1104 if (len >= Assembler::AVX_256bit) {
1105 __ emit_data(val4, relocInfo::none, 0);
1106 __ emit_data(val5, relocInfo::none, 0);
1107 __ emit_data(val6, relocInfo::none, 0);
1108 __ emit_data(val7, relocInfo::none, 0);
1109 if (len >= Assembler::AVX_512bit) {
1110 __ emit_data(val8, relocInfo::none, 0);
1111 __ emit_data(val9, relocInfo::none, 0);
1112 __ emit_data(val10, relocInfo::none, 0);
1113 __ emit_data(val11, relocInfo::none, 0);
1114 __ emit_data(val12, relocInfo::none, 0);
1115 __ emit_data(val13, relocInfo::none, 0);
1116 __ emit_data(val14, relocInfo::none, 0);
1117 __ emit_data(val15, relocInfo::none, 0);
1118 }
1119 }
1120 return start;
1121 }
1122
1123 // Non-destructive plausibility checks for oops
1124 //
1125 // Arguments:
1126 // all args on stack!
1127 //
1128 // Stack after saving c_rarg3:
1129 // [tos + 0]: saved c_rarg3
1130 // [tos + 1]: saved c_rarg2
1131 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1132 // [tos + 3]: saved flags
1133 // [tos + 4]: return address
1134 // * [tos + 5]: error message (char*)
1135 // * [tos + 6]: object to verify (oop)
1136 // * [tos + 7]: saved rax - saved by caller and bashed
1137 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1138 // * = popped on exit
1139 address StubGenerator::generate_verify_oop() {
1140 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1141 address start = __ pc();
1142
1143 Label exit, error;
1144
1145 __ pushf();
1146 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()), rscratch1);
1147
1148 __ push(r12);
1149
1150 // save c_rarg2 and c_rarg3
1151 __ push(c_rarg2);
1152 __ push(c_rarg3);
1153
1154 enum {
1155 // After previous pushes.
1156 oop_to_verify = 6 * wordSize,
1157 saved_rax = 7 * wordSize,
1158 saved_r10 = 8 * wordSize,
1159
1160 // Before the call to MacroAssembler::debug(), see below.
1161 return_addr = 16 * wordSize,
1162 error_msg = 17 * wordSize
1163 };
1164
1165 // get object
1166 __ movptr(rax, Address(rsp, oop_to_verify));
1167
1168 // make sure object is 'reasonable'
1169 __ testptr(rax, rax);
1170 __ jcc(Assembler::zero, exit); // if obj is null it is OK
1171
1172 BarrierSetAssembler* bs_asm = BarrierSet::barrier_set()->barrier_set_assembler();
1173 bs_asm->check_oop(_masm, rax, c_rarg2, c_rarg3, error);
1174
1175 // return if everything seems ok
1176 __ bind(exit);
1177 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1178 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1179 __ pop(c_rarg3); // restore c_rarg3
1180 __ pop(c_rarg2); // restore c_rarg2
1181 __ pop(r12); // restore r12
1182 __ popf(); // restore flags
1183 __ ret(4 * wordSize); // pop caller saved stuff
1184
1185 // handle errors
1186 __ bind(error);
1187 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1188 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1189 __ pop(c_rarg3); // get saved c_rarg3 back
1190 __ pop(c_rarg2); // get saved c_rarg2 back
1191 __ pop(r12); // get saved r12 back
1192 __ popf(); // get saved flags off stack --
1193 // will be ignored
1194
1195 __ pusha(); // push registers
1196 // (rip is already
1197 // already pushed)
1198 // debug(char* msg, int64_t pc, int64_t regs[])
1199 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1200 // pushed all the registers, so now the stack looks like:
1201 // [tos + 0] 16 saved registers
1202 // [tos + 16] return address
1203 // * [tos + 17] error message (char*)
1204 // * [tos + 18] object to verify (oop)
1205 // * [tos + 19] saved rax - saved by caller and bashed
1206 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1207 // * = popped on exit
1208
1209 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1210 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1211 __ movq(c_rarg2, rsp); // pass address of regs on stack
1212 __ mov(r12, rsp); // remember rsp
1213 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1214 __ andptr(rsp, -16); // align stack as required by ABI
1215 BLOCK_COMMENT("call MacroAssembler::debug");
1216 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1217 __ hlt();
1218
1219 return start;
1220 }
1221
1222
1223 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1224 //
1225 // Outputs:
1226 // rdi - rcx
1227 // rsi - rdx
1228 // rdx - r8
1229 // rcx - r9
1230 //
1231 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1232 // are non-volatile. r9 and r10 should not be used by the caller.
1233 //
1234 void StubGenerator::setup_arg_regs(int nargs) {
1235 const Register saved_rdi = r9;
1236 const Register saved_rsi = r10;
1237 assert(nargs == 3 || nargs == 4, "else fix");
1238 #ifdef _WIN64
1239 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1240 "unexpected argument registers");
1241 if (nargs == 4) {
1242 __ mov(rax, r9); // r9 is also saved_rdi
1243 }
1244 __ movptr(saved_rdi, rdi);
1245 __ movptr(saved_rsi, rsi);
1246 __ mov(rdi, rcx); // c_rarg0
1247 __ mov(rsi, rdx); // c_rarg1
1248 __ mov(rdx, r8); // c_rarg2
1249 if (nargs == 4) {
1250 __ mov(rcx, rax); // c_rarg3 (via rax)
1251 }
1252 #else
1253 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1254 "unexpected argument registers");
1255 #endif
1256 DEBUG_ONLY(_regs_in_thread = false;)
1257 }
1258
1259
1260 void StubGenerator::restore_arg_regs() {
1261 assert(!_regs_in_thread, "wrong call to restore_arg_regs");
1262 const Register saved_rdi = r9;
1263 const Register saved_rsi = r10;
1264 #ifdef _WIN64
1265 __ movptr(rdi, saved_rdi);
1266 __ movptr(rsi, saved_rsi);
1267 #endif
1268 }
1269
1270
1271 // This is used in places where r10 is a scratch register, and can
1272 // be adapted if r9 is needed also.
1273 void StubGenerator::setup_arg_regs_using_thread(int nargs) {
1274 const Register saved_r15 = r9;
1275 assert(nargs == 3 || nargs == 4, "else fix");
1276 #ifdef _WIN64
1277 if (nargs == 4) {
1278 __ mov(rax, r9); // r9 is also saved_r15
1279 }
1280 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1281 __ get_thread(r15_thread);
1282 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1283 "unexpected argument registers");
1284 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1285 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1286
1287 __ mov(rdi, rcx); // c_rarg0
1288 __ mov(rsi, rdx); // c_rarg1
1289 __ mov(rdx, r8); // c_rarg2
1290 if (nargs == 4) {
1291 __ mov(rcx, rax); // c_rarg3 (via rax)
1292 }
1293 #else
1294 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1295 "unexpected argument registers");
1296 #endif
1297 DEBUG_ONLY(_regs_in_thread = true;)
1298 }
1299
1300
1301 void StubGenerator::restore_arg_regs_using_thread() {
1302 assert(_regs_in_thread, "wrong call to restore_arg_regs");
1303 const Register saved_r15 = r9;
1304 #ifdef _WIN64
1305 __ get_thread(r15_thread);
1306 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1307 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1308 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1309 #endif
1310 }
1311
1312
1313 void StubGenerator::setup_argument_regs(BasicType type) {
1314 if (type == T_BYTE || type == T_SHORT) {
1315 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1316 // r9 and r10 may be used to save non-volatile registers
1317 } else {
1318 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1319 // r9 is used to save r15_thread
1320 }
1321 }
1322
1323
1324 void StubGenerator::restore_argument_regs(BasicType type) {
1325 if (type == T_BYTE || type == T_SHORT) {
1326 restore_arg_regs();
1327 } else {
1328 restore_arg_regs_using_thread();
1329 }
1330 }
1331
1332 address StubGenerator::generate_data_cache_writeback() {
1333 const Register src = c_rarg0; // source address
1334
1335 __ align(CodeEntryAlignment);
1336
1337 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback");
1338
1339 address start = __ pc();
1340
1341 __ enter();
1342 __ cache_wb(Address(src, 0));
1343 __ leave();
1344 __ ret(0);
1345
1346 return start;
1347 }
1348
1349 address StubGenerator::generate_data_cache_writeback_sync() {
1350 const Register is_pre = c_rarg0; // pre or post sync
1351
1352 __ align(CodeEntryAlignment);
1353
1354 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback_sync");
1355
1356 // pre wbsync is a no-op
1357 // post wbsync translates to an sfence
1358
1359 Label skip;
1360 address start = __ pc();
1361
1362 __ enter();
1363 __ cmpl(is_pre, 0);
1364 __ jcc(Assembler::notEqual, skip);
1365 __ cache_wbsync(false);
1366 __ bind(skip);
1367 __ leave();
1368 __ ret(0);
1369
1370 return start;
1371 }
1372
1373 // ofs and limit are use for multi-block byte array.
1374 // int com.sun.security.provider.MD5.implCompress(byte[] b, int ofs)
1375 address StubGenerator::generate_md5_implCompress(bool multi_block, const char *name) {
1376 __ align(CodeEntryAlignment);
1377 StubCodeMark mark(this, "StubRoutines", name);
1378 address start = __ pc();
1379
1380 const Register buf_param = r15;
1381 const Address state_param(rsp, 0 * wordSize);
1382 const Address ofs_param (rsp, 1 * wordSize );
1383 const Address limit_param(rsp, 1 * wordSize + 4);
1384
1385 __ enter();
1386 __ push(rbx);
1387 __ push(rdi);
1388 __ push(rsi);
1389 __ push(r15);
1390 __ subptr(rsp, 2 * wordSize);
1391
1392 __ movptr(buf_param, c_rarg0);
1393 __ movptr(state_param, c_rarg1);
1394 if (multi_block) {
1395 __ movl(ofs_param, c_rarg2);
1396 __ movl(limit_param, c_rarg3);
1397 }
1398 __ fast_md5(buf_param, state_param, ofs_param, limit_param, multi_block);
1399
1400 __ addptr(rsp, 2 * wordSize);
1401 __ pop(r15);
1402 __ pop(rsi);
1403 __ pop(rdi);
1404 __ pop(rbx);
1405 __ leave();
1406 __ ret(0);
1407
1408 return start;
1409 }
1410
1411 address StubGenerator::generate_upper_word_mask() {
1412 __ align64();
1413 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
1414 address start = __ pc();
1415
1416 __ emit_data64(0x0000000000000000, relocInfo::none);
1417 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
1418
1419 return start;
1420 }
1421
1422 address StubGenerator::generate_shuffle_byte_flip_mask() {
1423 __ align64();
1424 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
1425 address start = __ pc();
1426
1427 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1428 __ emit_data64(0x0001020304050607, relocInfo::none);
1429
1430 return start;
1431 }
1432
1433 // ofs and limit are use for multi-block byte array.
1434 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1435 address StubGenerator::generate_sha1_implCompress(bool multi_block, const char *name) {
1436 __ align(CodeEntryAlignment);
1437 StubCodeMark mark(this, "StubRoutines", name);
1438 address start = __ pc();
1439
1440 Register buf = c_rarg0;
1441 Register state = c_rarg1;
1442 Register ofs = c_rarg2;
1443 Register limit = c_rarg3;
1444
1445 const XMMRegister abcd = xmm0;
1446 const XMMRegister e0 = xmm1;
1447 const XMMRegister e1 = xmm2;
1448 const XMMRegister msg0 = xmm3;
1449
1450 const XMMRegister msg1 = xmm4;
1451 const XMMRegister msg2 = xmm5;
1452 const XMMRegister msg3 = xmm6;
1453 const XMMRegister shuf_mask = xmm7;
1454
1455 __ enter();
1456
1457 __ subptr(rsp, 4 * wordSize);
1458
1459 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
1460 buf, state, ofs, limit, rsp, multi_block);
1461
1462 __ addptr(rsp, 4 * wordSize);
1463
1464 __ leave();
1465 __ ret(0);
1466
1467 return start;
1468 }
1469
1470 address StubGenerator::generate_pshuffle_byte_flip_mask() {
1471 __ align64();
1472 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
1473 address start = __ pc();
1474
1475 __ emit_data64(0x0405060700010203, relocInfo::none);
1476 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1477
1478 if (VM_Version::supports_avx2()) {
1479 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
1480 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1481 // _SHUF_00BA
1482 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1483 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1484 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1485 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1486 // _SHUF_DC00
1487 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1488 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1489 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1490 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1491 }
1492
1493 return start;
1494 }
1495
1496 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
1497 address StubGenerator::generate_pshuffle_byte_flip_mask_sha512() {
1498 __ align32();
1499 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
1500 address start = __ pc();
1501
1502 if (VM_Version::supports_avx2()) {
1503 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
1504 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1505 __ emit_data64(0x1011121314151617, relocInfo::none);
1506 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
1507 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
1508 __ emit_data64(0x0000000000000000, relocInfo::none);
1509 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1510 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1511 }
1512
1513 return start;
1514 }
1515
1516 // ofs and limit are use for multi-block byte array.
1517 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1518 address StubGenerator::generate_sha256_implCompress(bool multi_block, const char *name) {
1519 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
1520 __ align(CodeEntryAlignment);
1521 StubCodeMark mark(this, "StubRoutines", name);
1522 address start = __ pc();
1523
1524 Register buf = c_rarg0;
1525 Register state = c_rarg1;
1526 Register ofs = c_rarg2;
1527 Register limit = c_rarg3;
1528
1529 const XMMRegister msg = xmm0;
1530 const XMMRegister state0 = xmm1;
1531 const XMMRegister state1 = xmm2;
1532 const XMMRegister msgtmp0 = xmm3;
1533
1534 const XMMRegister msgtmp1 = xmm4;
1535 const XMMRegister msgtmp2 = xmm5;
1536 const XMMRegister msgtmp3 = xmm6;
1537 const XMMRegister msgtmp4 = xmm7;
1538
1539 const XMMRegister shuf_mask = xmm8;
1540
1541 __ enter();
1542
1543 __ subptr(rsp, 4 * wordSize);
1544
1545 if (VM_Version::supports_sha()) {
1546 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1547 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1548 } else if (VM_Version::supports_avx2()) {
1549 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1550 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1551 }
1552 __ addptr(rsp, 4 * wordSize);
1553 __ vzeroupper();
1554 __ leave();
1555 __ ret(0);
1556
1557 return start;
1558 }
1559
1560 address StubGenerator::generate_sha512_implCompress(bool multi_block, const char *name) {
1561 assert(VM_Version::supports_avx2(), "");
1562 assert(VM_Version::supports_bmi2() || VM_Version::supports_sha512(), "");
1563 __ align(CodeEntryAlignment);
1564 StubCodeMark mark(this, "StubRoutines", name);
1565 address start = __ pc();
1566
1567 Register buf = c_rarg0;
1568 Register state = c_rarg1;
1569 Register ofs = c_rarg2;
1570 Register limit = c_rarg3;
1571
1572 __ enter();
1573
1574 if (VM_Version::supports_sha512()) {
1575 __ sha512_update_ni_x1(state, buf, ofs, limit, multi_block);
1576 } else {
1577 const XMMRegister msg = xmm0;
1578 const XMMRegister state0 = xmm1;
1579 const XMMRegister state1 = xmm2;
1580 const XMMRegister msgtmp0 = xmm3;
1581 const XMMRegister msgtmp1 = xmm4;
1582 const XMMRegister msgtmp2 = xmm5;
1583 const XMMRegister msgtmp3 = xmm6;
1584 const XMMRegister msgtmp4 = xmm7;
1585
1586 const XMMRegister shuf_mask = xmm8;
1587 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1588 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1589 }
1590 __ vzeroupper();
1591 __ leave();
1592 __ ret(0);
1593
1594 return start;
1595 }
1596
1597 address StubGenerator::base64_shuffle_addr() {
1598 __ align64();
1599 StubCodeMark mark(this, "StubRoutines", "shuffle_base64");
1600 address start = __ pc();
1601
1602 assert(((unsigned long long)start & 0x3f) == 0,
1603 "Alignment problem (0x%08llx)", (unsigned long long)start);
1604 __ emit_data64(0x0405030401020001, relocInfo::none);
1605 __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1606 __ emit_data64(0x10110f100d0e0c0d, relocInfo::none);
1607 __ emit_data64(0x1617151613141213, relocInfo::none);
1608 __ emit_data64(0x1c1d1b1c191a1819, relocInfo::none);
1609 __ emit_data64(0x222321221f201e1f, relocInfo::none);
1610 __ emit_data64(0x2829272825262425, relocInfo::none);
1611 __ emit_data64(0x2e2f2d2e2b2c2a2b, relocInfo::none);
1612
1613 return start;
1614 }
1615
1616 address StubGenerator::base64_avx2_shuffle_addr() {
1617 __ align32();
1618 StubCodeMark mark(this, "StubRoutines", "avx2_shuffle_base64");
1619 address start = __ pc();
1620
1621 __ emit_data64(0x0809070805060405, relocInfo::none);
1622 __ emit_data64(0x0e0f0d0e0b0c0a0b, relocInfo::none);
1623 __ emit_data64(0x0405030401020001, relocInfo::none);
1624 __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1625
1626 return start;
1627 }
1628
1629 address StubGenerator::base64_avx2_input_mask_addr() {
1630 __ align32();
1631 StubCodeMark mark(this, "StubRoutines", "avx2_input_mask_base64");
1632 address start = __ pc();
1633
1634 __ emit_data64(0x8000000000000000, relocInfo::none);
1635 __ emit_data64(0x8000000080000000, relocInfo::none);
1636 __ emit_data64(0x8000000080000000, relocInfo::none);
1637 __ emit_data64(0x8000000080000000, relocInfo::none);
1638
1639 return start;
1640 }
1641
1642 address StubGenerator::base64_avx2_lut_addr() {
1643 __ align32();
1644 StubCodeMark mark(this, "StubRoutines", "avx2_lut_base64");
1645 address start = __ pc();
1646
1647 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1648 __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1649 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1650 __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1651
1652 // URL LUT
1653 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1654 __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1655 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1656 __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1657
1658 return start;
1659 }
1660
1661 address StubGenerator::base64_encoding_table_addr() {
1662 __ align64();
1663 StubCodeMark mark(this, "StubRoutines", "encoding_table_base64");
1664 address start = __ pc();
1665
1666 assert(((unsigned long long)start & 0x3f) == 0, "Alignment problem (0x%08llx)", (unsigned long long)start);
1667 __ emit_data64(0x4847464544434241, relocInfo::none);
1668 __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1669 __ emit_data64(0x5857565554535251, relocInfo::none);
1670 __ emit_data64(0x6665646362615a59, relocInfo::none);
1671 __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1672 __ emit_data64(0x767574737271706f, relocInfo::none);
1673 __ emit_data64(0x333231307a797877, relocInfo::none);
1674 __ emit_data64(0x2f2b393837363534, relocInfo::none);
1675
1676 // URL table
1677 __ emit_data64(0x4847464544434241, relocInfo::none);
1678 __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1679 __ emit_data64(0x5857565554535251, relocInfo::none);
1680 __ emit_data64(0x6665646362615a59, relocInfo::none);
1681 __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1682 __ emit_data64(0x767574737271706f, relocInfo::none);
1683 __ emit_data64(0x333231307a797877, relocInfo::none);
1684 __ emit_data64(0x5f2d393837363534, relocInfo::none);
1685
1686 return start;
1687 }
1688
1689 // Code for generating Base64 encoding.
1690 // Intrinsic function prototype in Base64.java:
1691 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp,
1692 // boolean isURL) {
1693 address StubGenerator::generate_base64_encodeBlock()
1694 {
1695 __ align(CodeEntryAlignment);
1696 StubCodeMark mark(this, "StubRoutines", "implEncode");
1697 address start = __ pc();
1698
1699 __ enter();
1700
1701 // Save callee-saved registers before using them
1702 __ push(r12);
1703 __ push(r13);
1704 __ push(r14);
1705 __ push(r15);
1706
1707 // arguments
1708 const Register source = c_rarg0; // Source Array
1709 const Register start_offset = c_rarg1; // start offset
1710 const Register end_offset = c_rarg2; // end offset
1711 const Register dest = c_rarg3; // destination array
1712
1713 #ifndef _WIN64
1714 const Register dp = c_rarg4; // Position for writing to dest array
1715 const Register isURL = c_rarg5; // Base64 or URL character set
1716 #else
1717 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
1718 const Address isURL_mem(rbp, 7 * wordSize);
1719 const Register isURL = r10; // pick the volatile windows register
1720 const Register dp = r12;
1721 __ movl(dp, dp_mem);
1722 __ movl(isURL, isURL_mem);
1723 #endif
1724
1725 const Register length = r14;
1726 const Register encode_table = r13;
1727 Label L_process3, L_exit, L_processdata, L_vbmiLoop, L_not512, L_32byteLoop;
1728
1729 // calculate length from offsets
1730 __ movl(length, end_offset);
1731 __ subl(length, start_offset);
1732 __ jcc(Assembler::lessEqual, L_exit);
1733
1734 // Code for 512-bit VBMI encoding. Encodes 48 input bytes into 64
1735 // output bytes. We read 64 input bytes and ignore the last 16, so be
1736 // sure not to read past the end of the input buffer.
1737 if (VM_Version::supports_avx512_vbmi()) {
1738 __ cmpl(length, 64); // Do not overrun input buffer.
1739 __ jcc(Assembler::below, L_not512);
1740
1741 __ shll(isURL, 6); // index into decode table based on isURL
1742 __ lea(encode_table, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
1743 __ addptr(encode_table, isURL);
1744 __ shrl(isURL, 6); // restore isURL
1745
1746 __ mov64(rax, 0x3036242a1016040aull); // Shifts
1747 __ evmovdquq(xmm3, ExternalAddress(StubRoutines::x86::base64_shuffle_addr()), Assembler::AVX_512bit, r15);
1748 __ evmovdquq(xmm2, Address(encode_table, 0), Assembler::AVX_512bit);
1749 __ evpbroadcastq(xmm1, rax, Assembler::AVX_512bit);
1750
1751 __ align32();
1752 __ BIND(L_vbmiLoop);
1753
1754 __ vpermb(xmm0, xmm3, Address(source, start_offset), Assembler::AVX_512bit);
1755 __ subl(length, 48);
1756
1757 // Put the input bytes into the proper lanes for writing, then
1758 // encode them.
1759 __ evpmultishiftqb(xmm0, xmm1, xmm0, Assembler::AVX_512bit);
1760 __ vpermb(xmm0, xmm0, xmm2, Assembler::AVX_512bit);
1761
1762 // Write to destination
1763 __ evmovdquq(Address(dest, dp), xmm0, Assembler::AVX_512bit);
1764
1765 __ addptr(dest, 64);
1766 __ addptr(source, 48);
1767 __ cmpl(length, 64);
1768 __ jcc(Assembler::aboveEqual, L_vbmiLoop);
1769
1770 __ vzeroupper();
1771 }
1772
1773 __ BIND(L_not512);
1774 if (VM_Version::supports_avx2()) {
1775 /*
1776 ** This AVX2 encoder is based off the paper at:
1777 ** https://dl.acm.org/doi/10.1145/3132709
1778 **
1779 ** We use AVX2 SIMD instructions to encode 24 bytes into 32
1780 ** output bytes.
1781 **
1782 */
1783 // Lengths under 32 bytes are done with scalar routine
1784 __ cmpl(length, 31);
1785 __ jcc(Assembler::belowEqual, L_process3);
1786
1787 // Set up supporting constant table data
1788 __ vmovdqu(xmm9, ExternalAddress(StubRoutines::x86::base64_avx2_shuffle_addr()), rax);
1789 // 6-bit mask for 2nd and 4th (and multiples) 6-bit values
1790 __ movl(rax, 0x0fc0fc00);
1791 __ movdl(xmm8, rax);
1792 __ vmovdqu(xmm1, ExternalAddress(StubRoutines::x86::base64_avx2_input_mask_addr()), rax);
1793 __ vpbroadcastd(xmm8, xmm8, Assembler::AVX_256bit);
1794
1795 // Multiplication constant for "shifting" right by 6 and 10
1796 // bits
1797 __ movl(rax, 0x04000040);
1798
1799 __ subl(length, 24);
1800 __ movdl(xmm7, rax);
1801 __ vpbroadcastd(xmm7, xmm7, Assembler::AVX_256bit);
1802
1803 // For the first load, we mask off reading of the first 4
1804 // bytes into the register. This is so we can get 4 3-byte
1805 // chunks into each lane of the register, avoiding having to
1806 // handle end conditions. We then shuffle these bytes into a
1807 // specific order so that manipulation is easier.
1808 //
1809 // The initial read loads the XMM register like this:
1810 //
1811 // Lower 128-bit lane:
1812 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1813 // | XX | XX | XX | XX | A0 | A1 | A2 | B0 | B1 | B2 | C0 | C1
1814 // | C2 | D0 | D1 | D2 |
1815 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1816 //
1817 // Upper 128-bit lane:
1818 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1819 // | E0 | E1 | E2 | F0 | F1 | F2 | G0 | G1 | G2 | H0 | H1 | H2
1820 // | XX | XX | XX | XX |
1821 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1822 //
1823 // Where A0 is the first input byte, B0 is the fourth, etc.
1824 // The alphabetical significance denotes the 3 bytes to be
1825 // consumed and encoded into 4 bytes.
1826 //
1827 // We then shuffle the register so each 32-bit word contains
1828 // the sequence:
1829 // A1 A0 A2 A1, B1, B0, B2, B1, etc.
1830 // Each of these byte sequences are then manipulated into 4
1831 // 6-bit values ready for encoding.
1832 //
1833 // If we focus on one set of 3-byte chunks, changing the
1834 // nomenclature such that A0 => a, A1 => b, and A2 => c, we
1835 // shuffle such that each 24-bit chunk contains:
1836 //
1837 // b7 b6 b5 b4 b3 b2 b1 b0 | a7 a6 a5 a4 a3 a2 a1 a0 | c7 c6
1838 // c5 c4 c3 c2 c1 c0 | b7 b6 b5 b4 b3 b2 b1 b0
1839 // Explain this step.
1840 // b3 b2 b1 b0 c5 c4 c3 c2 | c1 c0 d5 d4 d3 d2 d1 d0 | a5 a4
1841 // a3 a2 a1 a0 b5 b4 | b3 b2 b1 b0 c5 c4 c3 c2
1842 //
1843 // W first and off all but bits 4-9 and 16-21 (c5..c0 and
1844 // a5..a0) and shift them using a vector multiplication
1845 // operation (vpmulhuw) which effectively shifts c right by 6
1846 // bits and a right by 10 bits. We similarly mask bits 10-15
1847 // (d5..d0) and 22-27 (b5..b0) and shift them left by 8 and 4
1848 // bits respectively. This is done using vpmullw. We end up
1849 // with 4 6-bit values, thus splitting the 3 input bytes,
1850 // ready for encoding:
1851 // 0 0 d5..d0 0 0 c5..c0 0 0 b5..b0 0 0 a5..a0
1852 //
1853 // For translation, we recognize that there are 5 distinct
1854 // ranges of legal Base64 characters as below:
1855 //
1856 // +-------------+-------------+------------+
1857 // | 6-bit value | ASCII range | offset |
1858 // +-------------+-------------+------------+
1859 // | 0..25 | A..Z | 65 |
1860 // | 26..51 | a..z | 71 |
1861 // | 52..61 | 0..9 | -4 |
1862 // | 62 | + or - | -19 or -17 |
1863 // | 63 | / or _ | -16 or 32 |
1864 // +-------------+-------------+------------+
1865 //
1866 // We note that vpshufb does a parallel lookup in a
1867 // destination register using the lower 4 bits of bytes from a
1868 // source register. If we use a saturated subtraction and
1869 // subtract 51 from each 6-bit value, bytes from [0,51]
1870 // saturate to 0, and [52,63] map to a range of [1,12]. We
1871 // distinguish the [0,25] and [26,51] ranges by assigning a
1872 // value of 13 for all 6-bit values less than 26. We end up
1873 // with:
1874 //
1875 // +-------------+-------------+------------+
1876 // | 6-bit value | Reduced | offset |
1877 // +-------------+-------------+------------+
1878 // | 0..25 | 13 | 65 |
1879 // | 26..51 | 0 | 71 |
1880 // | 52..61 | 0..9 | -4 |
1881 // | 62 | 11 | -19 or -17 |
1882 // | 63 | 12 | -16 or 32 |
1883 // +-------------+-------------+------------+
1884 //
1885 // We then use a final vpshufb to add the appropriate offset,
1886 // translating the bytes.
1887 //
1888 // Load input bytes - only 28 bytes. Mask the first load to
1889 // not load into the full register.
1890 __ vpmaskmovd(xmm1, xmm1, Address(source, start_offset, Address::times_1, -4), Assembler::AVX_256bit);
1891
1892 // Move 3-byte chunks of input (12 bytes) into 16 bytes,
1893 // ordering by:
1894 // 1, 0, 2, 1; 4, 3, 5, 4; etc. This groups 6-bit chunks
1895 // for easy masking
1896 __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1897
1898 __ addl(start_offset, 24);
1899
1900 // Load masking register for first and third (and multiples)
1901 // 6-bit values.
1902 __ movl(rax, 0x003f03f0);
1903 __ movdl(xmm6, rax);
1904 __ vpbroadcastd(xmm6, xmm6, Assembler::AVX_256bit);
1905 // Multiplication constant for "shifting" left by 4 and 8 bits
1906 __ movl(rax, 0x01000010);
1907 __ movdl(xmm5, rax);
1908 __ vpbroadcastd(xmm5, xmm5, Assembler::AVX_256bit);
1909
1910 // Isolate 6-bit chunks of interest
1911 __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
1912
1913 // Load constants for encoding
1914 __ movl(rax, 0x19191919);
1915 __ movdl(xmm3, rax);
1916 __ vpbroadcastd(xmm3, xmm3, Assembler::AVX_256bit);
1917 __ movl(rax, 0x33333333);
1918 __ movdl(xmm4, rax);
1919 __ vpbroadcastd(xmm4, xmm4, Assembler::AVX_256bit);
1920
1921 // Shift output bytes 0 and 2 into proper lanes
1922 __ vpmulhuw(xmm2, xmm0, xmm7, Assembler::AVX_256bit);
1923
1924 // Mask and shift output bytes 1 and 3 into proper lanes and
1925 // combine
1926 __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
1927 __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
1928 __ vpor(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
1929
1930 // Find out which are 0..25. This indicates which input
1931 // values fall in the range of 'A'-'Z', which require an
1932 // additional offset (see comments above)
1933 __ vpcmpgtb(xmm2, xmm0, xmm3, Assembler::AVX_256bit);
1934 __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
1935 __ vpsubb(xmm1, xmm1, xmm2, Assembler::AVX_256bit);
1936
1937 // Load the proper lookup table
1938 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_avx2_lut_addr()));
1939 __ movl(r15, isURL);
1940 __ shll(r15, 5);
1941 __ vmovdqu(xmm2, Address(r11, r15));
1942
1943 // Shuffle the offsets based on the range calculation done
1944 // above. This allows us to add the correct offset to the
1945 // 6-bit value corresponding to the range documented above.
1946 __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
1947 __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
1948
1949 // Store the encoded bytes
1950 __ vmovdqu(Address(dest, dp), xmm0);
1951 __ addl(dp, 32);
1952
1953 __ cmpl(length, 31);
1954 __ jcc(Assembler::belowEqual, L_process3);
1955
1956 __ align32();
1957 __ BIND(L_32byteLoop);
1958
1959 // Get next 32 bytes
1960 __ vmovdqu(xmm1, Address(source, start_offset, Address::times_1, -4));
1961
1962 __ subl(length, 24);
1963 __ addl(start_offset, 24);
1964
1965 // This logic is identical to the above, with only constant
1966 // register loads removed. Shuffle the input, mask off 6-bit
1967 // chunks, shift them into place, then add the offset to
1968 // encode.
1969 __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1970
1971 __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
1972 __ vpmulhuw(xmm10, xmm0, xmm7, Assembler::AVX_256bit);
1973 __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
1974 __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
1975 __ vpor(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
1976 __ vpcmpgtb(xmm10, xmm0, xmm3, Assembler::AVX_256bit);
1977 __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
1978 __ vpsubb(xmm1, xmm1, xmm10, Assembler::AVX_256bit);
1979 __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
1980 __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
1981
1982 // Store the encoded bytes
1983 __ vmovdqu(Address(dest, dp), xmm0);
1984 __ addl(dp, 32);
1985
1986 __ cmpl(length, 31);
1987 __ jcc(Assembler::above, L_32byteLoop);
1988
1989 __ BIND(L_process3);
1990 __ vzeroupper();
1991 } else {
1992 __ BIND(L_process3);
1993 }
1994
1995 __ cmpl(length, 3);
1996 __ jcc(Assembler::below, L_exit);
1997
1998 // Load the encoding table based on isURL
1999 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
2000 __ movl(r15, isURL);
2001 __ shll(r15, 6);
2002 __ addptr(r11, r15);
2003
2004 __ BIND(L_processdata);
2005
2006 // Load 3 bytes
2007 __ load_unsigned_byte(r15, Address(source, start_offset));
2008 __ load_unsigned_byte(r10, Address(source, start_offset, Address::times_1, 1));
2009 __ load_unsigned_byte(r13, Address(source, start_offset, Address::times_1, 2));
2010
2011 // Build a 32-bit word with bytes 1, 2, 0, 1
2012 __ movl(rax, r10);
2013 __ shll(r10, 24);
2014 __ orl(rax, r10);
2015
2016 __ subl(length, 3);
2017
2018 __ shll(r15, 8);
2019 __ shll(r13, 16);
2020 __ orl(rax, r15);
2021
2022 __ addl(start_offset, 3);
2023
2024 __ orl(rax, r13);
2025 // At this point, rax contains | byte1 | byte2 | byte0 | byte1
2026 // r13 has byte2 << 16 - need low-order 6 bits to translate.
2027 // This translated byte is the fourth output byte.
2028 __ shrl(r13, 16);
2029 __ andl(r13, 0x3f);
2030
2031 // The high-order 6 bits of r15 (byte0) is translated.
2032 // The translated byte is the first output byte.
2033 __ shrl(r15, 10);
2034
2035 __ load_unsigned_byte(r13, Address(r11, r13));
2036 __ load_unsigned_byte(r15, Address(r11, r15));
2037
2038 __ movb(Address(dest, dp, Address::times_1, 3), r13);
2039
2040 // Extract high-order 4 bits of byte1 and low-order 2 bits of byte0.
2041 // This translated byte is the second output byte.
2042 __ shrl(rax, 4);
2043 __ movl(r10, rax);
2044 __ andl(rax, 0x3f);
2045
2046 __ movb(Address(dest, dp, Address::times_1, 0), r15);
2047
2048 __ load_unsigned_byte(rax, Address(r11, rax));
2049
2050 // Extract low-order 2 bits of byte1 and high-order 4 bits of byte2.
2051 // This translated byte is the third output byte.
2052 __ shrl(r10, 18);
2053 __ andl(r10, 0x3f);
2054
2055 __ load_unsigned_byte(r10, Address(r11, r10));
2056
2057 __ movb(Address(dest, dp, Address::times_1, 1), rax);
2058 __ movb(Address(dest, dp, Address::times_1, 2), r10);
2059
2060 __ addl(dp, 4);
2061 __ cmpl(length, 3);
2062 __ jcc(Assembler::aboveEqual, L_processdata);
2063
2064 __ BIND(L_exit);
2065 __ pop(r15);
2066 __ pop(r14);
2067 __ pop(r13);
2068 __ pop(r12);
2069 __ leave();
2070 __ ret(0);
2071
2072 return start;
2073 }
2074
2075 // base64 AVX512vbmi tables
2076 address StubGenerator::base64_vbmi_lookup_lo_addr() {
2077 __ align64();
2078 StubCodeMark mark(this, "StubRoutines", "lookup_lo_base64");
2079 address start = __ pc();
2080
2081 assert(((unsigned long long)start & 0x3f) == 0,
2082 "Alignment problem (0x%08llx)", (unsigned long long)start);
2083 __ emit_data64(0x8080808080808080, relocInfo::none);
2084 __ emit_data64(0x8080808080808080, relocInfo::none);
2085 __ emit_data64(0x8080808080808080, relocInfo::none);
2086 __ emit_data64(0x8080808080808080, relocInfo::none);
2087 __ emit_data64(0x8080808080808080, relocInfo::none);
2088 __ emit_data64(0x3f8080803e808080, relocInfo::none);
2089 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2090 __ emit_data64(0x8080808080803d3c, relocInfo::none);
2091
2092 return start;
2093 }
2094
2095 address StubGenerator::base64_vbmi_lookup_hi_addr() {
2096 __ align64();
2097 StubCodeMark mark(this, "StubRoutines", "lookup_hi_base64");
2098 address start = __ pc();
2099
2100 assert(((unsigned long long)start & 0x3f) == 0,
2101 "Alignment problem (0x%08llx)", (unsigned long long)start);
2102 __ emit_data64(0x0605040302010080, relocInfo::none);
2103 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2104 __ emit_data64(0x161514131211100f, relocInfo::none);
2105 __ emit_data64(0x8080808080191817, relocInfo::none);
2106 __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2107 __ emit_data64(0x2827262524232221, relocInfo::none);
2108 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2109 __ emit_data64(0x8080808080333231, relocInfo::none);
2110
2111 return start;
2112 }
2113 address StubGenerator::base64_vbmi_lookup_lo_url_addr() {
2114 __ align64();
2115 StubCodeMark mark(this, "StubRoutines", "lookup_lo_base64url");
2116 address start = __ pc();
2117
2118 assert(((unsigned long long)start & 0x3f) == 0,
2119 "Alignment problem (0x%08llx)", (unsigned long long)start);
2120 __ emit_data64(0x8080808080808080, relocInfo::none);
2121 __ emit_data64(0x8080808080808080, relocInfo::none);
2122 __ emit_data64(0x8080808080808080, relocInfo::none);
2123 __ emit_data64(0x8080808080808080, relocInfo::none);
2124 __ emit_data64(0x8080808080808080, relocInfo::none);
2125 __ emit_data64(0x80803e8080808080, relocInfo::none);
2126 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2127 __ emit_data64(0x8080808080803d3c, relocInfo::none);
2128
2129 return start;
2130 }
2131
2132 address StubGenerator::base64_vbmi_lookup_hi_url_addr() {
2133 __ align64();
2134 StubCodeMark mark(this, "StubRoutines", "lookup_hi_base64url");
2135 address start = __ pc();
2136
2137 assert(((unsigned long long)start & 0x3f) == 0,
2138 "Alignment problem (0x%08llx)", (unsigned long long)start);
2139 __ emit_data64(0x0605040302010080, relocInfo::none);
2140 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2141 __ emit_data64(0x161514131211100f, relocInfo::none);
2142 __ emit_data64(0x3f80808080191817, relocInfo::none);
2143 __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2144 __ emit_data64(0x2827262524232221, relocInfo::none);
2145 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2146 __ emit_data64(0x8080808080333231, relocInfo::none);
2147
2148 return start;
2149 }
2150
2151 address StubGenerator::base64_vbmi_pack_vec_addr() {
2152 __ align64();
2153 StubCodeMark mark(this, "StubRoutines", "pack_vec_base64");
2154 address start = __ pc();
2155
2156 assert(((unsigned long long)start & 0x3f) == 0,
2157 "Alignment problem (0x%08llx)", (unsigned long long)start);
2158 __ emit_data64(0x090a040506000102, relocInfo::none);
2159 __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2160 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2161 __ emit_data64(0x292a242526202122, relocInfo::none);
2162 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2163 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2164 __ emit_data64(0x0000000000000000, relocInfo::none);
2165 __ emit_data64(0x0000000000000000, relocInfo::none);
2166
2167 return start;
2168 }
2169
2170 address StubGenerator::base64_vbmi_join_0_1_addr() {
2171 __ align64();
2172 StubCodeMark mark(this, "StubRoutines", "join_0_1_base64");
2173 address start = __ pc();
2174
2175 assert(((unsigned long long)start & 0x3f) == 0,
2176 "Alignment problem (0x%08llx)", (unsigned long long)start);
2177 __ emit_data64(0x090a040506000102, relocInfo::none);
2178 __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2179 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2180 __ emit_data64(0x292a242526202122, relocInfo::none);
2181 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2182 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2183 __ emit_data64(0x494a444546404142, relocInfo::none);
2184 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2185
2186 return start;
2187 }
2188
2189 address StubGenerator::base64_vbmi_join_1_2_addr() {
2190 __ align64();
2191 StubCodeMark mark(this, "StubRoutines", "join_1_2_base64");
2192 address start = __ pc();
2193
2194 assert(((unsigned long long)start & 0x3f) == 0,
2195 "Alignment problem (0x%08llx)", (unsigned long long)start);
2196 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2197 __ emit_data64(0x292a242526202122, relocInfo::none);
2198 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2199 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2200 __ emit_data64(0x494a444546404142, relocInfo::none);
2201 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2202 __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2203 __ emit_data64(0x696a646566606162, relocInfo::none);
2204
2205 return start;
2206 }
2207
2208 address StubGenerator::base64_vbmi_join_2_3_addr() {
2209 __ align64();
2210 StubCodeMark mark(this, "StubRoutines", "join_2_3_base64");
2211 address start = __ pc();
2212
2213 assert(((unsigned long long)start & 0x3f) == 0,
2214 "Alignment problem (0x%08llx)", (unsigned long long)start);
2215 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2216 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2217 __ emit_data64(0x494a444546404142, relocInfo::none);
2218 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2219 __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2220 __ emit_data64(0x696a646566606162, relocInfo::none);
2221 __ emit_data64(0x767071726c6d6e68, relocInfo::none);
2222 __ emit_data64(0x7c7d7e78797a7475, relocInfo::none);
2223
2224 return start;
2225 }
2226
2227 address StubGenerator::base64_AVX2_decode_tables_addr() {
2228 __ align64();
2229 StubCodeMark mark(this, "StubRoutines", "AVX2_tables_base64");
2230 address start = __ pc();
2231
2232 assert(((unsigned long long)start & 0x3f) == 0,
2233 "Alignment problem (0x%08llx)", (unsigned long long)start);
2234 __ emit_data(0x2f2f2f2f, relocInfo::none, 0);
2235 __ emit_data(0x5f5f5f5f, relocInfo::none, 0); // for URL
2236
2237 __ emit_data(0xffffffff, relocInfo::none, 0);
2238 __ emit_data(0xfcfcfcfc, relocInfo::none, 0); // for URL
2239
2240 // Permute table
2241 __ emit_data64(0x0000000100000000, relocInfo::none);
2242 __ emit_data64(0x0000000400000002, relocInfo::none);
2243 __ emit_data64(0x0000000600000005, relocInfo::none);
2244 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2245
2246 // Shuffle table
2247 __ emit_data64(0x090a040506000102, relocInfo::none);
2248 __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2249 __ emit_data64(0x090a040506000102, relocInfo::none);
2250 __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2251
2252 // merge table
2253 __ emit_data(0x01400140, relocInfo::none, 0);
2254
2255 // merge multiplier
2256 __ emit_data(0x00011000, relocInfo::none, 0);
2257
2258 return start;
2259 }
2260
2261 address StubGenerator::base64_AVX2_decode_LUT_tables_addr() {
2262 __ align64();
2263 StubCodeMark mark(this, "StubRoutines", "AVX2_tables_URL_base64");
2264 address start = __ pc();
2265
2266 assert(((unsigned long long)start & 0x3f) == 0,
2267 "Alignment problem (0x%08llx)", (unsigned long long)start);
2268 // lut_lo
2269 __ emit_data64(0x1111111111111115, relocInfo::none);
2270 __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2271 __ emit_data64(0x1111111111111115, relocInfo::none);
2272 __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2273
2274 // lut_roll
2275 __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2276 __ emit_data64(0x0000000000000000, relocInfo::none);
2277 __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2278 __ emit_data64(0x0000000000000000, relocInfo::none);
2279
2280 // lut_lo URL
2281 __ emit_data64(0x1111111111111115, relocInfo::none);
2282 __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2283 __ emit_data64(0x1111111111111115, relocInfo::none);
2284 __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2285
2286 // lut_roll URL
2287 __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2288 __ emit_data64(0x0000000000000000, relocInfo::none);
2289 __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2290 __ emit_data64(0x0000000000000000, relocInfo::none);
2291
2292 // lut_hi
2293 __ emit_data64(0x0804080402011010, relocInfo::none);
2294 __ emit_data64(0x1010101010101010, relocInfo::none);
2295 __ emit_data64(0x0804080402011010, relocInfo::none);
2296 __ emit_data64(0x1010101010101010, relocInfo::none);
2297
2298 return start;
2299 }
2300
2301 address StubGenerator::base64_decoding_table_addr() {
2302 StubCodeMark mark(this, "StubRoutines", "decoding_table_base64");
2303 address start = __ pc();
2304
2305 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2306 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2307 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2308 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2309 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2310 __ emit_data64(0x3fffffff3effffff, relocInfo::none);
2311 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2312 __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2313 __ emit_data64(0x06050403020100ff, relocInfo::none);
2314 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2315 __ emit_data64(0x161514131211100f, relocInfo::none);
2316 __ emit_data64(0xffffffffff191817, relocInfo::none);
2317 __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2318 __ emit_data64(0x2827262524232221, relocInfo::none);
2319 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2320 __ emit_data64(0xffffffffff333231, relocInfo::none);
2321 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2322 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2323 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2324 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2325 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2326 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2327 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2328 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2329 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2330 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2331 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2332 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2333 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2334 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2335 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2336 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2337
2338 // URL table
2339 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2340 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2341 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2342 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2343 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2344 __ emit_data64(0xffff3effffffffff, relocInfo::none);
2345 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2346 __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2347 __ emit_data64(0x06050403020100ff, relocInfo::none);
2348 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2349 __ emit_data64(0x161514131211100f, relocInfo::none);
2350 __ emit_data64(0x3fffffffff191817, relocInfo::none);
2351 __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2352 __ emit_data64(0x2827262524232221, relocInfo::none);
2353 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2354 __ emit_data64(0xffffffffff333231, relocInfo::none);
2355 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2356 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2357 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2358 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2359 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2360 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2361 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2362 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2363 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2364 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2365 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2366 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2367 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2368 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2369 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2370 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2371
2372 return start;
2373 }
2374
2375
2376 // Code for generating Base64 decoding.
2377 //
2378 // Based on the article (and associated code) from https://arxiv.org/abs/1910.05109.
2379 //
2380 // Intrinsic function prototype in Base64.java:
2381 // private void decodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL, isMIME) {
2382 address StubGenerator::generate_base64_decodeBlock() {
2383 __ align(CodeEntryAlignment);
2384 StubCodeMark mark(this, "StubRoutines", "implDecode");
2385 address start = __ pc();
2386
2387 __ enter();
2388
2389 // Save callee-saved registers before using them
2390 __ push(r12);
2391 __ push(r13);
2392 __ push(r14);
2393 __ push(r15);
2394 __ push(rbx);
2395
2396 // arguments
2397 const Register source = c_rarg0; // Source Array
2398 const Register start_offset = c_rarg1; // start offset
2399 const Register end_offset = c_rarg2; // end offset
2400 const Register dest = c_rarg3; // destination array
2401 const Register isMIME = rbx;
2402
2403 #ifndef _WIN64
2404 const Register dp = c_rarg4; // Position for writing to dest array
2405 const Register isURL = c_rarg5;// Base64 or URL character set
2406 __ movl(isMIME, Address(rbp, 2 * wordSize));
2407 #else
2408 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
2409 const Address isURL_mem(rbp, 7 * wordSize);
2410 const Register isURL = r10; // pick the volatile windows register
2411 const Register dp = r12;
2412 __ movl(dp, dp_mem);
2413 __ movl(isURL, isURL_mem);
2414 __ movl(isMIME, Address(rbp, 8 * wordSize));
2415 #endif
2416
2417 const XMMRegister lookup_lo = xmm5;
2418 const XMMRegister lookup_hi = xmm6;
2419 const XMMRegister errorvec = xmm7;
2420 const XMMRegister pack16_op = xmm9;
2421 const XMMRegister pack32_op = xmm8;
2422 const XMMRegister input0 = xmm3;
2423 const XMMRegister input1 = xmm20;
2424 const XMMRegister input2 = xmm21;
2425 const XMMRegister input3 = xmm19;
2426 const XMMRegister join01 = xmm12;
2427 const XMMRegister join12 = xmm11;
2428 const XMMRegister join23 = xmm10;
2429 const XMMRegister translated0 = xmm2;
2430 const XMMRegister translated1 = xmm1;
2431 const XMMRegister translated2 = xmm0;
2432 const XMMRegister translated3 = xmm4;
2433
2434 const XMMRegister merged0 = xmm2;
2435 const XMMRegister merged1 = xmm1;
2436 const XMMRegister merged2 = xmm0;
2437 const XMMRegister merged3 = xmm4;
2438 const XMMRegister merge_ab_bc0 = xmm2;
2439 const XMMRegister merge_ab_bc1 = xmm1;
2440 const XMMRegister merge_ab_bc2 = xmm0;
2441 const XMMRegister merge_ab_bc3 = xmm4;
2442
2443 const XMMRegister pack24bits = xmm4;
2444
2445 const Register length = r14;
2446 const Register output_size = r13;
2447 const Register output_mask = r15;
2448 const KRegister input_mask = k1;
2449
2450 const XMMRegister input_initial_valid_b64 = xmm0;
2451 const XMMRegister tmp = xmm10;
2452 const XMMRegister mask = xmm0;
2453 const XMMRegister invalid_b64 = xmm1;
2454
2455 Label L_process256, L_process64, L_process64Loop, L_exit, L_processdata, L_loadURL;
2456 Label L_continue, L_finalBit, L_padding, L_donePadding, L_bruteForce;
2457 Label L_forceLoop, L_bottomLoop, L_checkMIME, L_exit_no_vzero, L_lastChunk;
2458
2459 // calculate length from offsets
2460 __ movl(length, end_offset);
2461 __ subl(length, start_offset);
2462 __ push(dest); // Save for return value calc
2463
2464 // If AVX512 VBMI not supported, just compile non-AVX code
2465 if(VM_Version::supports_avx512_vbmi() &&
2466 VM_Version::supports_avx512bw()) {
2467 __ cmpl(length, 31); // 32-bytes is break-even for AVX-512
2468 __ jcc(Assembler::lessEqual, L_lastChunk);
2469
2470 __ cmpl(isMIME, 0);
2471 __ jcc(Assembler::notEqual, L_lastChunk);
2472
2473 // Load lookup tables based on isURL
2474 __ cmpl(isURL, 0);
2475 __ jcc(Assembler::notZero, L_loadURL);
2476
2477 __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_addr()), Assembler::AVX_512bit, r13);
2478 __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_addr()), Assembler::AVX_512bit, r13);
2479
2480 __ BIND(L_continue);
2481
2482 __ movl(r15, 0x01400140);
2483 __ evpbroadcastd(pack16_op, r15, Assembler::AVX_512bit);
2484
2485 __ movl(r15, 0x00011000);
2486 __ evpbroadcastd(pack32_op, r15, Assembler::AVX_512bit);
2487
2488 __ cmpl(length, 0xff);
2489 __ jcc(Assembler::lessEqual, L_process64);
2490
2491 // load masks required for decoding data
2492 __ BIND(L_processdata);
2493 __ evmovdquq(join01, ExternalAddress(StubRoutines::x86::base64_vbmi_join_0_1_addr()), Assembler::AVX_512bit,r13);
2494 __ evmovdquq(join12, ExternalAddress(StubRoutines::x86::base64_vbmi_join_1_2_addr()), Assembler::AVX_512bit, r13);
2495 __ evmovdquq(join23, ExternalAddress(StubRoutines::x86::base64_vbmi_join_2_3_addr()), Assembler::AVX_512bit, r13);
2496
2497 __ align32();
2498 __ BIND(L_process256);
2499 // Grab input data
2500 __ evmovdquq(input0, Address(source, start_offset, Address::times_1, 0x00), Assembler::AVX_512bit);
2501 __ evmovdquq(input1, Address(source, start_offset, Address::times_1, 0x40), Assembler::AVX_512bit);
2502 __ evmovdquq(input2, Address(source, start_offset, Address::times_1, 0x80), Assembler::AVX_512bit);
2503 __ evmovdquq(input3, Address(source, start_offset, Address::times_1, 0xc0), Assembler::AVX_512bit);
2504
2505 // Copy the low part of the lookup table into the destination of the permutation
2506 __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2507 __ evmovdquq(translated1, lookup_lo, Assembler::AVX_512bit);
2508 __ evmovdquq(translated2, lookup_lo, Assembler::AVX_512bit);
2509 __ evmovdquq(translated3, lookup_lo, Assembler::AVX_512bit);
2510
2511 // Translate the base64 input into "decoded" bytes
2512 __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2513 __ evpermt2b(translated1, input1, lookup_hi, Assembler::AVX_512bit);
2514 __ evpermt2b(translated2, input2, lookup_hi, Assembler::AVX_512bit);
2515 __ evpermt2b(translated3, input3, lookup_hi, Assembler::AVX_512bit);
2516
2517 // OR all of the translations together to check for errors (high-order bit of byte set)
2518 __ vpternlogd(input0, 0xfe, input1, input2, Assembler::AVX_512bit);
2519
2520 __ vpternlogd(input3, 0xfe, translated0, translated1, Assembler::AVX_512bit);
2521 __ vpternlogd(input0, 0xfe, translated2, translated3, Assembler::AVX_512bit);
2522 __ vpor(errorvec, input3, input0, Assembler::AVX_512bit);
2523
2524 // Check if there was an error - if so, try 64-byte chunks
2525 __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2526 __ kortestql(k3, k3);
2527 __ jcc(Assembler::notZero, L_process64);
2528
2529 // The merging and shuffling happens here
2530 // We multiply each byte pair [00dddddd | 00cccccc | 00bbbbbb | 00aaaaaa]
2531 // Multiply [00cccccc] by 2^6 added to [00dddddd] to get [0000cccc | ccdddddd]
2532 // The pack16_op is a vector of 0x01400140, so multiply D by 1 and C by 0x40
2533 __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2534 __ vpmaddubsw(merge_ab_bc1, translated1, pack16_op, Assembler::AVX_512bit);
2535 __ vpmaddubsw(merge_ab_bc2, translated2, pack16_op, Assembler::AVX_512bit);
2536 __ vpmaddubsw(merge_ab_bc3, translated3, pack16_op, Assembler::AVX_512bit);
2537
2538 // Now do the same with packed 16-bit values.
2539 // We start with [0000cccc | ccdddddd | 0000aaaa | aabbbbbb]
2540 // pack32_op is 0x00011000 (2^12, 1), so this multiplies [0000aaaa | aabbbbbb] by 2^12
2541 // and adds [0000cccc | ccdddddd] to yield [00000000 | aaaaaabb | bbbbcccc | ccdddddd]
2542 __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2543 __ vpmaddwd(merged1, merge_ab_bc1, pack32_op, Assembler::AVX_512bit);
2544 __ vpmaddwd(merged2, merge_ab_bc2, pack32_op, Assembler::AVX_512bit);
2545 __ vpmaddwd(merged3, merge_ab_bc3, pack32_op, Assembler::AVX_512bit);
2546
2547 // The join vectors specify which byte from which vector goes into the outputs
2548 // One of every 4 bytes in the extended vector is zero, so we pack them into their
2549 // final positions in the register for storing (256 bytes in, 192 bytes out)
2550 __ evpermt2b(merged0, join01, merged1, Assembler::AVX_512bit);
2551 __ evpermt2b(merged1, join12, merged2, Assembler::AVX_512bit);
2552 __ evpermt2b(merged2, join23, merged3, Assembler::AVX_512bit);
2553
2554 // Store result
2555 __ evmovdquq(Address(dest, dp, Address::times_1, 0x00), merged0, Assembler::AVX_512bit);
2556 __ evmovdquq(Address(dest, dp, Address::times_1, 0x40), merged1, Assembler::AVX_512bit);
2557 __ evmovdquq(Address(dest, dp, Address::times_1, 0x80), merged2, Assembler::AVX_512bit);
2558
2559 __ addptr(source, 0x100);
2560 __ addptr(dest, 0xc0);
2561 __ subl(length, 0x100);
2562 __ cmpl(length, 64 * 4);
2563 __ jcc(Assembler::greaterEqual, L_process256);
2564
2565 // At this point, we've decoded 64 * 4 * n bytes.
2566 // The remaining length will be <= 64 * 4 - 1.
2567 // UNLESS there was an error decoding the first 256-byte chunk. In this
2568 // case, the length will be arbitrarily long.
2569 //
2570 // Note that this will be the path for MIME-encoded strings.
2571
2572 __ BIND(L_process64);
2573
2574 __ evmovdquq(pack24bits, ExternalAddress(StubRoutines::x86::base64_vbmi_pack_vec_addr()), Assembler::AVX_512bit, r13);
2575
2576 __ cmpl(length, 63);
2577 __ jcc(Assembler::lessEqual, L_finalBit);
2578
2579 __ mov64(rax, 0x0000ffffffffffff);
2580 __ kmovql(k2, rax);
2581
2582 __ align32();
2583 __ BIND(L_process64Loop);
2584
2585 // Handle first 64-byte block
2586
2587 __ evmovdquq(input0, Address(source, start_offset), Assembler::AVX_512bit);
2588 __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2589 __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2590
2591 __ vpor(errorvec, translated0, input0, Assembler::AVX_512bit);
2592
2593 // Check for error and bomb out before updating dest
2594 __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2595 __ kortestql(k3, k3);
2596 __ jcc(Assembler::notZero, L_exit);
2597
2598 // Pack output register, selecting correct byte ordering
2599 __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2600 __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2601 __ vpermb(merged0, pack24bits, merged0, Assembler::AVX_512bit);
2602
2603 __ evmovdqub(Address(dest, dp), k2, merged0, true, Assembler::AVX_512bit);
2604
2605 __ subl(length, 64);
2606 __ addptr(source, 64);
2607 __ addptr(dest, 48);
2608
2609 __ cmpl(length, 64);
2610 __ jcc(Assembler::greaterEqual, L_process64Loop);
2611
2612 __ cmpl(length, 0);
2613 __ jcc(Assembler::lessEqual, L_exit);
2614
2615 __ BIND(L_finalBit);
2616 // Now have 1 to 63 bytes left to decode
2617
2618 // I was going to let Java take care of the final fragment
2619 // however it will repeatedly call this routine for every 4 bytes
2620 // of input data, so handle the rest here.
2621 __ movq(rax, -1);
2622 __ bzhiq(rax, rax, length); // Input mask in rax
2623
2624 __ movl(output_size, length);
2625 __ shrl(output_size, 2); // Find (len / 4) * 3 (output length)
2626 __ lea(output_size, Address(output_size, output_size, Address::times_2, 0));
2627 // output_size in r13
2628
2629 // Strip pad characters, if any, and adjust length and mask
2630 __ addq(length, start_offset);
2631 __ cmpb(Address(source, length, Address::times_1, -1), '=');
2632 __ jcc(Assembler::equal, L_padding);
2633
2634 __ BIND(L_donePadding);
2635 __ subq(length, start_offset);
2636
2637 // Output size is (64 - output_size), output mask is (all 1s >> output_size).
2638 __ kmovql(input_mask, rax);
2639 __ movq(output_mask, -1);
2640 __ bzhiq(output_mask, output_mask, output_size);
2641
2642 // Load initial input with all valid base64 characters. Will be used
2643 // in merging source bytes to avoid masking when determining if an error occurred.
2644 __ movl(rax, 0x61616161);
2645 __ evpbroadcastd(input_initial_valid_b64, rax, Assembler::AVX_512bit);
2646
2647 // A register containing all invalid base64 decoded values
2648 __ movl(rax, 0x80808080);
2649 __ evpbroadcastd(invalid_b64, rax, Assembler::AVX_512bit);
2650
2651 // input_mask is in k1
2652 // output_size is in r13
2653 // output_mask is in r15
2654 // zmm0 - free
2655 // zmm1 - 0x00011000
2656 // zmm2 - 0x01400140
2657 // zmm3 - errorvec
2658 // zmm4 - pack vector
2659 // zmm5 - lookup_lo
2660 // zmm6 - lookup_hi
2661 // zmm7 - errorvec
2662 // zmm8 - 0x61616161
2663 // zmm9 - 0x80808080
2664
2665 // Load only the bytes from source, merging into our "fully-valid" register
2666 __ evmovdqub(input_initial_valid_b64, input_mask, Address(source, start_offset, Address::times_1, 0x0), true, Assembler::AVX_512bit);
2667
2668 // Decode all bytes within our merged input
2669 __ evmovdquq(tmp, lookup_lo, Assembler::AVX_512bit);
2670 __ evpermt2b(tmp, input_initial_valid_b64, lookup_hi, Assembler::AVX_512bit);
2671 __ evporq(mask, tmp, input_initial_valid_b64, Assembler::AVX_512bit);
2672
2673 // Check for error. Compare (decoded | initial) to all invalid.
2674 // If any bytes have their high-order bit set, then we have an error.
2675 __ evptestmb(k2, mask, invalid_b64, Assembler::AVX_512bit);
2676 __ kortestql(k2, k2);
2677
2678 // If we have an error, use the brute force loop to decode what we can (4-byte chunks).
2679 __ jcc(Assembler::notZero, L_bruteForce);
2680
2681 // Shuffle output bytes
2682 __ vpmaddubsw(tmp, tmp, pack16_op, Assembler::AVX_512bit);
2683 __ vpmaddwd(tmp, tmp, pack32_op, Assembler::AVX_512bit);
2684
2685 __ vpermb(tmp, pack24bits, tmp, Assembler::AVX_512bit);
2686 __ kmovql(k1, output_mask);
2687 __ evmovdqub(Address(dest, dp), k1, tmp, true, Assembler::AVX_512bit);
2688
2689 __ addptr(dest, output_size);
2690
2691 __ BIND(L_exit);
2692 __ vzeroupper();
2693 __ pop(rax); // Get original dest value
2694 __ subptr(dest, rax); // Number of bytes converted
2695 __ movptr(rax, dest);
2696 __ pop(rbx);
2697 __ pop(r15);
2698 __ pop(r14);
2699 __ pop(r13);
2700 __ pop(r12);
2701 __ leave();
2702 __ ret(0);
2703
2704 __ BIND(L_loadURL);
2705 __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_url_addr()), Assembler::AVX_512bit, r13);
2706 __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_url_addr()), Assembler::AVX_512bit, r13);
2707 __ jmp(L_continue);
2708
2709 __ BIND(L_padding);
2710 __ decrementq(output_size, 1);
2711 __ shrq(rax, 1);
2712
2713 __ cmpb(Address(source, length, Address::times_1, -2), '=');
2714 __ jcc(Assembler::notEqual, L_donePadding);
2715
2716 __ decrementq(output_size, 1);
2717 __ shrq(rax, 1);
2718 __ jmp(L_donePadding);
2719
2720 __ align32();
2721 __ BIND(L_bruteForce);
2722 } // End of if(avx512_vbmi)
2723
2724 if (VM_Version::supports_avx2()) {
2725 Label L_tailProc, L_topLoop, L_enterLoop;
2726
2727 __ cmpl(isMIME, 0);
2728 __ jcc(Assembler::notEqual, L_lastChunk);
2729
2730 // Check for buffer too small (for algorithm)
2731 __ subl(length, 0x2c);
2732 __ jcc(Assembler::less, L_tailProc);
2733
2734 __ shll(isURL, 2);
2735
2736 // Algorithm adapted from https://arxiv.org/abs/1704.00605, "Faster Base64
2737 // Encoding and Decoding using AVX2 Instructions". URL modifications added.
2738
2739 // Set up constants
2740 __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_tables_addr()));
2741 __ vpbroadcastd(xmm4, Address(r13, isURL, Address::times_1), Assembler::AVX_256bit); // 2F or 5F
2742 __ vpbroadcastd(xmm10, Address(r13, isURL, Address::times_1, 0x08), Assembler::AVX_256bit); // -1 or -4
2743 __ vmovdqu(xmm12, Address(r13, 0x10)); // permute
2744 __ vmovdqu(xmm13, Address(r13, 0x30)); // shuffle
2745 __ vpbroadcastd(xmm7, Address(r13, 0x50), Assembler::AVX_256bit); // merge
2746 __ vpbroadcastd(xmm6, Address(r13, 0x54), Assembler::AVX_256bit); // merge mult
2747
2748 __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_LUT_tables_addr()));
2749 __ shll(isURL, 4);
2750 __ vmovdqu(xmm11, Address(r13, isURL, Address::times_1, 0x00)); // lut_lo
2751 __ vmovdqu(xmm8, Address(r13, isURL, Address::times_1, 0x20)); // lut_roll
2752 __ shrl(isURL, 6); // restore isURL
2753 __ vmovdqu(xmm9, Address(r13, 0x80)); // lut_hi
2754 __ jmp(L_enterLoop);
2755
2756 __ align32();
2757 __ bind(L_topLoop);
2758 // Add in the offset value (roll) to get 6-bit out values
2759 __ vpaddb(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
2760 // Merge and permute the output bits into appropriate output byte lanes
2761 __ vpmaddubsw(xmm0, xmm0, xmm7, Assembler::AVX_256bit);
2762 __ vpmaddwd(xmm0, xmm0, xmm6, Assembler::AVX_256bit);
2763 __ vpshufb(xmm0, xmm0, xmm13, Assembler::AVX_256bit);
2764 __ vpermd(xmm0, xmm12, xmm0, Assembler::AVX_256bit);
2765 // Store the output bytes
2766 __ vmovdqu(Address(dest, dp, Address::times_1, 0), xmm0);
2767 __ addptr(source, 0x20);
2768 __ addptr(dest, 0x18);
2769 __ subl(length, 0x20);
2770 __ jcc(Assembler::less, L_tailProc);
2771
2772 __ bind(L_enterLoop);
2773
2774 // Load in encoded string (32 bytes)
2775 __ vmovdqu(xmm2, Address(source, start_offset, Address::times_1, 0x0));
2776 // Extract the high nibble for indexing into the lut tables. High 4 bits are don't care.
2777 __ vpsrld(xmm1, xmm2, 0x4, Assembler::AVX_256bit);
2778 __ vpand(xmm1, xmm4, xmm1, Assembler::AVX_256bit);
2779 // Extract the low nibble. 5F/2F will isolate the low-order 4 bits. High 4 bits are don't care.
2780 __ vpand(xmm3, xmm2, xmm4, Assembler::AVX_256bit);
2781 // Check for special-case (0x2F or 0x5F (URL))
2782 __ vpcmpeqb(xmm0, xmm4, xmm2, Assembler::AVX_256bit);
2783 // Get the bitset based on the low nibble. vpshufb uses low-order 4 bits only.
2784 __ vpshufb(xmm3, xmm11, xmm3, Assembler::AVX_256bit);
2785 // Get the bit value of the high nibble
2786 __ vpshufb(xmm5, xmm9, xmm1, Assembler::AVX_256bit);
2787 // Make sure 2F / 5F shows as valid
2788 __ vpandn(xmm3, xmm0, xmm3, Assembler::AVX_256bit);
2789 // Make adjustment for roll index. For non-URL, this is a no-op,
2790 // for URL, this adjusts by -4. This is to properly index the
2791 // roll value for 2F / 5F.
2792 __ vpand(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
2793 // If the and of the two is non-zero, we have an invalid input character
2794 __ vptest(xmm3, xmm5);
2795 // Extract the "roll" value - value to add to the input to get 6-bit out value
2796 __ vpaddb(xmm0, xmm0, xmm1, Assembler::AVX_256bit); // Handle 2F / 5F
2797 __ vpshufb(xmm0, xmm8, xmm0, Assembler::AVX_256bit);
2798 __ jcc(Assembler::equal, L_topLoop); // Fall through on error
2799
2800 __ bind(L_tailProc);
2801
2802 __ addl(length, 0x2c);
2803
2804 __ vzeroupper();
2805 }
2806
2807 // Use non-AVX code to decode 4-byte chunks into 3 bytes of output
2808
2809 // Register state (Linux):
2810 // r12-15 - saved on stack
2811 // rdi - src
2812 // rsi - sp
2813 // rdx - sl
2814 // rcx - dst
2815 // r8 - dp
2816 // r9 - isURL
2817
2818 // Register state (Windows):
2819 // r12-15 - saved on stack
2820 // rcx - src
2821 // rdx - sp
2822 // r8 - sl
2823 // r9 - dst
2824 // r12 - dp
2825 // r10 - isURL
2826
2827 // Registers (common):
2828 // length (r14) - bytes in src
2829
2830 const Register decode_table = r11;
2831 const Register out_byte_count = rbx;
2832 const Register byte1 = r13;
2833 const Register byte2 = r15;
2834 const Register byte3 = WIN64_ONLY(r8) NOT_WIN64(rdx);
2835 const Register byte4 = WIN64_ONLY(r10) NOT_WIN64(r9);
2836
2837 __ bind(L_lastChunk);
2838
2839 __ shrl(length, 2); // Multiple of 4 bytes only - length is # 4-byte chunks
2840 __ cmpl(length, 0);
2841 __ jcc(Assembler::lessEqual, L_exit_no_vzero);
2842
2843 __ shll(isURL, 8); // index into decode table based on isURL
2844 __ lea(decode_table, ExternalAddress(StubRoutines::x86::base64_decoding_table_addr()));
2845 __ addptr(decode_table, isURL);
2846
2847 __ jmp(L_bottomLoop);
2848
2849 __ align32();
2850 __ BIND(L_forceLoop);
2851 __ shll(byte1, 18);
2852 __ shll(byte2, 12);
2853 __ shll(byte3, 6);
2854 __ orl(byte1, byte2);
2855 __ orl(byte1, byte3);
2856 __ orl(byte1, byte4);
2857
2858 __ addptr(source, 4);
2859
2860 __ movb(Address(dest, dp, Address::times_1, 2), byte1);
2861 __ shrl(byte1, 8);
2862 __ movb(Address(dest, dp, Address::times_1, 1), byte1);
2863 __ shrl(byte1, 8);
2864 __ movb(Address(dest, dp, Address::times_1, 0), byte1);
2865
2866 __ addptr(dest, 3);
2867 __ decrementl(length, 1);
2868 __ jcc(Assembler::zero, L_exit_no_vzero);
2869
2870 __ BIND(L_bottomLoop);
2871 __ load_unsigned_byte(byte1, Address(source, start_offset, Address::times_1, 0x00));
2872 __ load_unsigned_byte(byte2, Address(source, start_offset, Address::times_1, 0x01));
2873 __ load_signed_byte(byte1, Address(decode_table, byte1));
2874 __ load_signed_byte(byte2, Address(decode_table, byte2));
2875 __ load_unsigned_byte(byte3, Address(source, start_offset, Address::times_1, 0x02));
2876 __ load_unsigned_byte(byte4, Address(source, start_offset, Address::times_1, 0x03));
2877 __ load_signed_byte(byte3, Address(decode_table, byte3));
2878 __ load_signed_byte(byte4, Address(decode_table, byte4));
2879
2880 __ mov(rax, byte1);
2881 __ orl(rax, byte2);
2882 __ orl(rax, byte3);
2883 __ orl(rax, byte4);
2884 __ jcc(Assembler::positive, L_forceLoop);
2885
2886 __ BIND(L_exit_no_vzero);
2887 __ pop(rax); // Get original dest value
2888 __ subptr(dest, rax); // Number of bytes converted
2889 __ movptr(rax, dest);
2890 __ pop(rbx);
2891 __ pop(r15);
2892 __ pop(r14);
2893 __ pop(r13);
2894 __ pop(r12);
2895 __ leave();
2896 __ ret(0);
2897
2898 return start;
2899 }
2900
2901
2902 /**
2903 * Arguments:
2904 *
2905 * Inputs:
2906 * c_rarg0 - int crc
2907 * c_rarg1 - byte* buf
2908 * c_rarg2 - int length
2909 *
2910 * Output:
2911 * rax - int crc result
2912 */
2913 address StubGenerator::generate_updateBytesCRC32() {
2914 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2915
2916 __ align(CodeEntryAlignment);
2917 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2918
2919 address start = __ pc();
2920
2921 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
2922 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
2923 // rscratch1: r10
2924 const Register crc = c_rarg0; // crc
2925 const Register buf = c_rarg1; // source java byte array address
2926 const Register len = c_rarg2; // length
2927 const Register table = c_rarg3; // crc_table address (reuse register)
2928 const Register tmp1 = r11;
2929 const Register tmp2 = r10;
2930 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax);
2931
2932 BLOCK_COMMENT("Entry:");
2933 __ enter(); // required for proper stackwalking of RuntimeStub frame
2934
2935 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
2936 VM_Version::supports_avx512bw() &&
2937 VM_Version::supports_avx512vl()) {
2938 // The constants used in the CRC32 algorithm requires the 1's compliment of the initial crc value.
2939 // However, the constant table for CRC32-C assumes the original crc value. Account for this
2940 // difference before calling and after returning.
2941 __ lea(table, ExternalAddress(StubRoutines::x86::crc_table_avx512_addr()));
2942 __ notl(crc);
2943 __ kernel_crc32_avx512(crc, buf, len, table, tmp1, tmp2);
2944 __ notl(crc);
2945 } else {
2946 __ kernel_crc32(crc, buf, len, table, tmp1);
2947 }
2948
2949 __ movl(rax, crc);
2950 __ vzeroupper();
2951 __ leave(); // required for proper stackwalking of RuntimeStub frame
2952 __ ret(0);
2953
2954 return start;
2955 }
2956
2957 /**
2958 * Arguments:
2959 *
2960 * Inputs:
2961 * c_rarg0 - int crc
2962 * c_rarg1 - byte* buf
2963 * c_rarg2 - long length
2964 * c_rarg3 - table_start - optional (present only when doing a library_call,
2965 * not used by x86 algorithm)
2966 *
2967 * Output:
2968 * rax - int crc result
2969 */
2970 address StubGenerator::generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
2971 assert(UseCRC32CIntrinsics, "need SSE4_2");
2972 __ align(CodeEntryAlignment);
2973 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
2974 address start = __ pc();
2975
2976 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
2977 //Windows RCX RDX R8 R9 none none XMM0..XMM3
2978 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
2979 const Register crc = c_rarg0; // crc
2980 const Register buf = c_rarg1; // source java byte array address
2981 const Register len = c_rarg2; // length
2982 const Register a = rax;
2983 const Register j = r9;
2984 const Register k = r10;
2985 const Register l = r11;
2986 #ifdef _WIN64
2987 const Register y = rdi;
2988 const Register z = rsi;
2989 #else
2990 const Register y = rcx;
2991 const Register z = r8;
2992 #endif
2993 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
2994
2995 BLOCK_COMMENT("Entry:");
2996 __ enter(); // required for proper stackwalking of RuntimeStub frame
2997 Label L_continue;
2998
2999 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
3000 VM_Version::supports_avx512bw() &&
3001 VM_Version::supports_avx512vl()) {
3002 Label L_doSmall;
3003
3004 __ cmpl(len, 384);
3005 __ jcc(Assembler::lessEqual, L_doSmall);
3006
3007 __ lea(j, ExternalAddress(StubRoutines::x86::crc32c_table_avx512_addr()));
3008 __ kernel_crc32_avx512(crc, buf, len, j, l, k);
3009
3010 __ jmp(L_continue);
3011
3012 __ bind(L_doSmall);
3013 }
3014 #ifdef _WIN64
3015 __ push(y);
3016 __ push(z);
3017 #endif
3018 __ crc32c_ipl_alg2_alt2(crc, buf, len,
3019 a, j, k,
3020 l, y, z,
3021 c_farg0, c_farg1, c_farg2,
3022 is_pclmulqdq_supported);
3023 #ifdef _WIN64
3024 __ pop(z);
3025 __ pop(y);
3026 #endif
3027
3028 __ bind(L_continue);
3029 __ movl(rax, crc);
3030 __ vzeroupper();
3031 __ leave(); // required for proper stackwalking of RuntimeStub frame
3032 __ ret(0);
3033
3034 return start;
3035 }
3036
3037
3038 /**
3039 * Arguments:
3040 *
3041 * Input:
3042 * c_rarg0 - x address
3043 * c_rarg1 - x length
3044 * c_rarg2 - y address
3045 * c_rarg3 - y length
3046 * not Win64
3047 * c_rarg4 - z address
3048 * Win64
3049 * rsp+40 - z address
3050 */
3051 address StubGenerator::generate_multiplyToLen() {
3052 __ align(CodeEntryAlignment);
3053 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
3054 address start = __ pc();
3055
3056 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3057 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3058 const Register x = rdi;
3059 const Register xlen = rax;
3060 const Register y = rsi;
3061 const Register ylen = rcx;
3062 const Register z = r8;
3063
3064 // Next registers will be saved on stack in multiply_to_len().
3065 const Register tmp0 = r11;
3066 const Register tmp1 = r12;
3067 const Register tmp2 = r13;
3068 const Register tmp3 = r14;
3069 const Register tmp4 = r15;
3070 const Register tmp5 = rbx;
3071
3072 BLOCK_COMMENT("Entry:");
3073 __ enter(); // required for proper stackwalking of RuntimeStub frame
3074
3075 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
3076 // ylen => rcx, z => r8
3077 // r9 and r10 may be used to save non-volatile registers
3078 #ifdef _WIN64
3079 // last argument (#4) is on stack on Win64
3080 __ movptr(z, Address(rsp, 6 * wordSize));
3081 #endif
3082
3083 __ movptr(xlen, rsi);
3084 __ movptr(y, rdx);
3085 __ multiply_to_len(x, xlen, y, ylen, z, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
3086
3087 restore_arg_regs();
3088
3089 __ leave(); // required for proper stackwalking of RuntimeStub frame
3090 __ ret(0);
3091
3092 return start;
3093 }
3094
3095 /**
3096 * Arguments:
3097 *
3098 * Input:
3099 * c_rarg0 - obja address
3100 * c_rarg1 - objb address
3101 * c_rarg3 - length length
3102 * c_rarg4 - scale log2_array_indxscale
3103 *
3104 * Output:
3105 * rax - int >= mismatched index, < 0 bitwise complement of tail
3106 */
3107 address StubGenerator::generate_vectorizedMismatch() {
3108 __ align(CodeEntryAlignment);
3109 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
3110 address start = __ pc();
3111
3112 BLOCK_COMMENT("Entry:");
3113 __ enter();
3114
3115 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3116 const Register scale = c_rarg0; //rcx, will exchange with r9
3117 const Register objb = c_rarg1; //rdx
3118 const Register length = c_rarg2; //r8
3119 const Register obja = c_rarg3; //r9
3120 __ xchgq(obja, scale); //now obja and scale contains the correct contents
3121
3122 const Register tmp1 = r10;
3123 const Register tmp2 = r11;
3124 #endif
3125 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3126 const Register obja = c_rarg0; //U:rdi
3127 const Register objb = c_rarg1; //U:rsi
3128 const Register length = c_rarg2; //U:rdx
3129 const Register scale = c_rarg3; //U:rcx
3130 const Register tmp1 = r8;
3131 const Register tmp2 = r9;
3132 #endif
3133 const Register result = rax; //return value
3134 const XMMRegister vec0 = xmm0;
3135 const XMMRegister vec1 = xmm1;
3136 const XMMRegister vec2 = xmm2;
3137
3138 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
3139
3140 __ vzeroupper();
3141 __ leave();
3142 __ ret(0);
3143
3144 return start;
3145 }
3146
3147 /**
3148 * Arguments:
3149 *
3150 // Input:
3151 // c_rarg0 - x address
3152 // c_rarg1 - x length
3153 // c_rarg2 - z address
3154 // c_rarg3 - z length
3155 *
3156 */
3157 address StubGenerator::generate_squareToLen() {
3158
3159 __ align(CodeEntryAlignment);
3160 StubCodeMark mark(this, "StubRoutines", "squareToLen");
3161 address start = __ pc();
3162
3163 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3164 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
3165 const Register x = rdi;
3166 const Register len = rsi;
3167 const Register z = r8;
3168 const Register zlen = rcx;
3169
3170 const Register tmp1 = r12;
3171 const Register tmp2 = r13;
3172 const Register tmp3 = r14;
3173 const Register tmp4 = r15;
3174 const Register tmp5 = rbx;
3175
3176 BLOCK_COMMENT("Entry:");
3177 __ enter(); // required for proper stackwalking of RuntimeStub frame
3178
3179 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
3180 // zlen => rcx
3181 // r9 and r10 may be used to save non-volatile registers
3182 __ movptr(r8, rdx);
3183 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3184
3185 restore_arg_regs();
3186
3187 __ leave(); // required for proper stackwalking of RuntimeStub frame
3188 __ ret(0);
3189
3190 return start;
3191 }
3192
3193 address StubGenerator::generate_method_entry_barrier() {
3194 __ align(CodeEntryAlignment);
3195 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
3196 address start = __ pc();
3197
3198 Label deoptimize_label;
3199
3200 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
3201
3202 BLOCK_COMMENT("Entry:");
3203 __ enter(); // save rbp
3204
3205 // save c_rarg0, because we want to use that value.
3206 // We could do without it but then we depend on the number of slots used by pusha
3207 __ push(c_rarg0);
3208
3209 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
3210
3211 __ pusha();
3212
3213 // The method may have floats as arguments, and we must spill them before calling
3214 // the VM runtime.
3215 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
3216 const int xmm_size = wordSize * 2;
3217 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
3218 __ subptr(rsp, xmm_spill_size);
3219 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
3220 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
3221 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
3222 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
3223 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
3224 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
3225 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
3226 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
3227
3228 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
3229
3230 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
3231 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
3232 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
3233 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
3234 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
3235 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
3236 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
3237 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
3238 __ addptr(rsp, xmm_spill_size);
3239
3240 __ cmpl(rax, 1); // 1 means deoptimize
3241 __ jcc(Assembler::equal, deoptimize_label);
3242
3243 __ popa();
3244 __ pop(c_rarg0);
3245
3246 __ leave();
3247
3248 __ addptr(rsp, 1 * wordSize); // cookie
3249 __ ret(0);
3250
3251
3252 __ BIND(deoptimize_label);
3253
3254 __ popa();
3255 __ pop(c_rarg0);
3256
3257 __ leave();
3258
3259 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
3260 // here while still having a correct stack is valuable
3261 __ testptr(rsp, Address(rsp, 0));
3262
3263 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
3264 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
3265
3266 return start;
3267 }
3268
3269 /**
3270 * Arguments:
3271 *
3272 * Input:
3273 * c_rarg0 - out address
3274 * c_rarg1 - in address
3275 * c_rarg2 - offset
3276 * c_rarg3 - len
3277 * not Win64
3278 * c_rarg4 - k
3279 * Win64
3280 * rsp+40 - k
3281 */
3282 address StubGenerator::generate_mulAdd() {
3283 __ align(CodeEntryAlignment);
3284 StubCodeMark mark(this, "StubRoutines", "mulAdd");
3285 address start = __ pc();
3286
3287 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3288 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3289 const Register out = rdi;
3290 const Register in = rsi;
3291 const Register offset = r11;
3292 const Register len = rcx;
3293 const Register k = r8;
3294
3295 // Next registers will be saved on stack in mul_add().
3296 const Register tmp1 = r12;
3297 const Register tmp2 = r13;
3298 const Register tmp3 = r14;
3299 const Register tmp4 = r15;
3300 const Register tmp5 = rbx;
3301
3302 BLOCK_COMMENT("Entry:");
3303 __ enter(); // required for proper stackwalking of RuntimeStub frame
3304
3305 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
3306 // len => rcx, k => r8
3307 // r9 and r10 may be used to save non-volatile registers
3308 #ifdef _WIN64
3309 // last argument is on stack on Win64
3310 __ movl(k, Address(rsp, 6 * wordSize));
3311 #endif
3312 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
3313 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3314
3315 restore_arg_regs();
3316
3317 __ leave(); // required for proper stackwalking of RuntimeStub frame
3318 __ ret(0);
3319
3320 return start;
3321 }
3322
3323 address StubGenerator::generate_bigIntegerRightShift() {
3324 __ align(CodeEntryAlignment);
3325 StubCodeMark mark(this, "StubRoutines", "bigIntegerRightShiftWorker");
3326 address start = __ pc();
3327
3328 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3329 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3330 const Register newArr = rdi;
3331 const Register oldArr = rsi;
3332 const Register newIdx = rdx;
3333 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3334 const Register totalNumIter = r8;
3335
3336 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3337 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3338 const Register tmp1 = r11; // Caller save.
3339 const Register tmp2 = rax; // Caller save.
3340 const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9); // Windows: Callee save. Linux: Caller save.
3341 const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10); // Windows: Callee save. Linux: Caller save.
3342 const Register tmp5 = r14; // Callee save.
3343 const Register tmp6 = r15;
3344
3345 const XMMRegister x0 = xmm0;
3346 const XMMRegister x1 = xmm1;
3347 const XMMRegister x2 = xmm2;
3348
3349 BLOCK_COMMENT("Entry:");
3350 __ enter(); // required for proper stackwalking of RuntimeStub frame
3351
3352 #ifdef _WIN64
3353 setup_arg_regs(4);
3354 // For windows, since last argument is on stack, we need to move it to the appropriate register.
3355 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3356 // Save callee save registers.
3357 __ push(tmp3);
3358 __ push(tmp4);
3359 #endif
3360 __ push(tmp5);
3361
3362 // Rename temps used throughout the code.
3363 const Register idx = tmp1;
3364 const Register nIdx = tmp2;
3365
3366 __ xorl(idx, idx);
3367
3368 // Start right shift from end of the array.
3369 // For example, if #iteration = 4 and newIdx = 1
3370 // then dest[4] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
3371 // if #iteration = 4 and newIdx = 0
3372 // then dest[3] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
3373 __ movl(idx, totalNumIter);
3374 __ movl(nIdx, idx);
3375 __ addl(nIdx, newIdx);
3376
3377 // If vectorization is enabled, check if the number of iterations is at least 64
3378 // If not, then go to ShifTwo processing 2 iterations
3379 if (VM_Version::supports_avx512_vbmi2()) {
3380 __ cmpptr(totalNumIter, (AVX3Threshold/64));
3381 __ jcc(Assembler::less, ShiftTwo);
3382
3383 if (AVX3Threshold < 16 * 64) {
3384 __ cmpl(totalNumIter, 16);
3385 __ jcc(Assembler::less, ShiftTwo);
3386 }
3387 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3388 __ subl(idx, 16);
3389 __ subl(nIdx, 16);
3390 __ BIND(Shift512Loop);
3391 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 4), Assembler::AVX_512bit);
3392 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3393 __ vpshrdvd(x2, x1, x0, Assembler::AVX_512bit);
3394 __ evmovdqul(Address(newArr, nIdx, Address::times_4), x2, Assembler::AVX_512bit);
3395 __ subl(nIdx, 16);
3396 __ subl(idx, 16);
3397 __ jcc(Assembler::greaterEqual, Shift512Loop);
3398 __ addl(idx, 16);
3399 __ addl(nIdx, 16);
3400 }
3401 __ BIND(ShiftTwo);
3402 __ cmpl(idx, 2);
3403 __ jcc(Assembler::less, ShiftOne);
3404 __ subl(idx, 2);
3405 __ subl(nIdx, 2);
3406 __ BIND(ShiftTwoLoop);
3407 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 8));
3408 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3409 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3410 __ shrdl(tmp5, tmp4);
3411 __ shrdl(tmp4, tmp3);
3412 __ movl(Address(newArr, nIdx, Address::times_4, 4), tmp5);
3413 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3414 __ subl(nIdx, 2);
3415 __ subl(idx, 2);
3416 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3417 __ addl(idx, 2);
3418 __ addl(nIdx, 2);
3419
3420 // Do the last iteration
3421 __ BIND(ShiftOne);
3422 __ cmpl(idx, 1);
3423 __ jcc(Assembler::less, Exit);
3424 __ subl(idx, 1);
3425 __ subl(nIdx, 1);
3426 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3427 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3428 __ shrdl(tmp4, tmp3);
3429 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3430 __ BIND(Exit);
3431 __ vzeroupper();
3432 // Restore callee save registers.
3433 __ pop(tmp5);
3434 #ifdef _WIN64
3435 __ pop(tmp4);
3436 __ pop(tmp3);
3437 restore_arg_regs();
3438 #endif
3439 __ leave(); // required for proper stackwalking of RuntimeStub frame
3440 __ ret(0);
3441
3442 return start;
3443 }
3444
3445 /**
3446 * Arguments:
3447 *
3448 * Input:
3449 * c_rarg0 - newArr address
3450 * c_rarg1 - oldArr address
3451 * c_rarg2 - newIdx
3452 * c_rarg3 - shiftCount
3453 * not Win64
3454 * c_rarg4 - numIter
3455 * Win64
3456 * rsp40 - numIter
3457 */
3458 address StubGenerator::generate_bigIntegerLeftShift() {
3459 __ align(CodeEntryAlignment);
3460 StubCodeMark mark(this, "StubRoutines", "bigIntegerLeftShiftWorker");
3461 address start = __ pc();
3462
3463 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3464 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3465 const Register newArr = rdi;
3466 const Register oldArr = rsi;
3467 const Register newIdx = rdx;
3468 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3469 const Register totalNumIter = r8;
3470 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3471 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3472 const Register tmp1 = r11; // Caller save.
3473 const Register tmp2 = rax; // Caller save.
3474 const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9); // Windows: Callee save. Linux: Caller save.
3475 const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10); // Windows: Callee save. Linux: Caller save.
3476 const Register tmp5 = r14; // Callee save.
3477
3478 const XMMRegister x0 = xmm0;
3479 const XMMRegister x1 = xmm1;
3480 const XMMRegister x2 = xmm2;
3481 BLOCK_COMMENT("Entry:");
3482 __ enter(); // required for proper stackwalking of RuntimeStub frame
3483
3484 #ifdef _WIN64
3485 setup_arg_regs(4);
3486 // For windows, since last argument is on stack, we need to move it to the appropriate register.
3487 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3488 // Save callee save registers.
3489 __ push(tmp3);
3490 __ push(tmp4);
3491 #endif
3492 __ push(tmp5);
3493
3494 // Rename temps used throughout the code
3495 const Register idx = tmp1;
3496 const Register numIterTmp = tmp2;
3497
3498 // Start idx from zero.
3499 __ xorl(idx, idx);
3500 // Compute interior pointer for new array. We do this so that we can use same index for both old and new arrays.
3501 __ lea(newArr, Address(newArr, newIdx, Address::times_4));
3502 __ movl(numIterTmp, totalNumIter);
3503
3504 // If vectorization is enabled, check if the number of iterations is at least 64
3505 // If not, then go to ShiftTwo shifting two numbers at a time
3506 if (VM_Version::supports_avx512_vbmi2()) {
3507 __ cmpl(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(numIterTmp, 16);
3516 __ BIND(Shift512Loop);
3517 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3518 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 0x4), Assembler::AVX_512bit);
3519 __ vpshldvd(x1, x2, x0, Assembler::AVX_512bit);
3520 __ evmovdqul(Address(newArr, idx, Address::times_4), x1, Assembler::AVX_512bit);
3521 __ addl(idx, 16);
3522 __ subl(numIterTmp, 16);
3523 __ jcc(Assembler::greaterEqual, Shift512Loop);
3524 __ addl(numIterTmp, 16);
3525 }
3526 __ BIND(ShiftTwo);
3527 __ cmpl(totalNumIter, 1);
3528 __ jcc(Assembler::less, Exit);
3529 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3530 __ subl(numIterTmp, 2);
3531 __ jcc(Assembler::less, ShiftOne);
3532
3533 __ BIND(ShiftTwoLoop);
3534 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3535 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 0x8));
3536 __ shldl(tmp3, tmp4);
3537 __ shldl(tmp4, tmp5);
3538 __ movl(Address(newArr, idx, Address::times_4), tmp3);
3539 __ movl(Address(newArr, idx, Address::times_4, 0x4), tmp4);
3540 __ movl(tmp3, tmp5);
3541 __ addl(idx, 2);
3542 __ subl(numIterTmp, 2);
3543 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3544
3545 // Do the last iteration
3546 __ BIND(ShiftOne);
3547 __ addl(numIterTmp, 2);
3548 __ cmpl(numIterTmp, 1);
3549 __ jcc(Assembler::less, Exit);
3550 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3551 __ shldl(tmp3, tmp4);
3552 __ movl(Address(newArr, idx, Address::times_4), tmp3);
3553
3554 __ BIND(Exit);
3555 __ vzeroupper();
3556 // Restore callee save registers.
3557 __ pop(tmp5);
3558 #ifdef _WIN64
3559 __ pop(tmp4);
3560 __ pop(tmp3);
3561 restore_arg_regs();
3562 #endif
3563 __ leave(); // required for proper stackwalking of RuntimeStub frame
3564 __ ret(0);
3565
3566 return start;
3567 }
3568
3569 void StubGenerator::generate_libm_stubs() {
3570 if (UseLibmIntrinsic && InlineIntrinsics) {
3571 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3572 StubRoutines::_dsin = generate_libmSin(); // from stubGenerator_x86_64_sin.cpp
3573 }
3574 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3575 StubRoutines::_dcos = generate_libmCos(); // from stubGenerator_x86_64_cos.cpp
3576 }
3577 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3578 StubRoutines::_dtan = generate_libmTan(); // from stubGenerator_x86_64_tan.cpp
3579 }
3580 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtanh)) {
3581 StubRoutines::_dtanh = generate_libmTanh(); // from stubGenerator_x86_64_tanh.cpp
3582 }
3583 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3584 StubRoutines::_dexp = generate_libmExp(); // from stubGenerator_x86_64_exp.cpp
3585 }
3586 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3587 StubRoutines::_dpow = generate_libmPow(); // from stubGenerator_x86_64_pow.cpp
3588 }
3589 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3590 StubRoutines::_dlog = generate_libmLog(); // from stubGenerator_x86_64_log.cpp
3591 }
3592 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3593 StubRoutines::_dlog10 = generate_libmLog10(); // from stubGenerator_x86_64_log.cpp
3594 }
3595 }
3596 }
3597
3598 /**
3599 * Arguments:
3600 *
3601 * Input:
3602 * c_rarg0 - float16 jshort
3603 *
3604 * Output:
3605 * xmm0 - float
3606 */
3607 address StubGenerator::generate_float16ToFloat() {
3608 StubCodeMark mark(this, "StubRoutines", "float16ToFloat");
3609
3610 address start = __ pc();
3611
3612 BLOCK_COMMENT("Entry:");
3613 // No need for RuntimeStub frame since it is called only during JIT compilation
3614
3615 // Load value into xmm0 and convert
3616 __ flt16_to_flt(xmm0, c_rarg0);
3617
3618 __ ret(0);
3619
3620 return start;
3621 }
3622
3623 /**
3624 * Arguments:
3625 *
3626 * Input:
3627 * xmm0 - float
3628 *
3629 * Output:
3630 * rax - float16 jshort
3631 */
3632 address StubGenerator::generate_floatToFloat16() {
3633 StubCodeMark mark(this, "StubRoutines", "floatToFloat16");
3634
3635 address start = __ pc();
3636
3637 BLOCK_COMMENT("Entry:");
3638 // No need for RuntimeStub frame since it is called only during JIT compilation
3639
3640 // Convert and put result into rax
3641 __ flt_to_flt16(rax, xmm0, xmm1);
3642
3643 __ ret(0);
3644
3645 return start;
3646 }
3647
3648 address StubGenerator::generate_cont_thaw(const char* label, Continuation::thaw_kind kind) {
3649 if (!Continuations::enabled()) return nullptr;
3650
3651 bool return_barrier = Continuation::is_thaw_return_barrier(kind);
3652 bool return_barrier_exception = Continuation::is_thaw_return_barrier_exception(kind);
3653
3654 StubCodeMark mark(this, "StubRoutines", label);
3655 address start = __ pc();
3656
3657 // TODO: Handle Valhalla return types. May require generating different return barriers.
3658
3659 if (!return_barrier) {
3660 // Pop return address. If we don't do this, we get a drift,
3661 // where the bottom-most frozen frame continuously grows.
3662 __ pop(c_rarg3);
3663 } else {
3664 __ movptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3665 }
3666
3667 #ifdef ASSERT
3668 {
3669 Label L_good_sp;
3670 __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3671 __ jcc(Assembler::equal, L_good_sp);
3672 __ stop("Incorrect rsp at thaw entry");
3673 __ BIND(L_good_sp);
3674 }
3675 #endif // ASSERT
3676
3677 if (return_barrier) {
3678 // Preserve possible return value from a method returning to the return barrier.
3679 __ push(rax);
3680 __ push_d(xmm0);
3681 }
3682
3683 __ movptr(c_rarg0, r15_thread);
3684 __ movptr(c_rarg1, (return_barrier ? 1 : 0));
3685 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Continuation::prepare_thaw), 2);
3686 __ movptr(rbx, rax);
3687
3688 if (return_barrier) {
3689 // Restore return value from a method returning to the return barrier.
3690 // No safepoint in the call to thaw, so even an oop return value should be OK.
3691 __ pop_d(xmm0);
3692 __ pop(rax);
3693 }
3694
3695 #ifdef ASSERT
3696 {
3697 Label L_good_sp;
3698 __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3699 __ jcc(Assembler::equal, L_good_sp);
3700 __ stop("Incorrect rsp after prepare thaw");
3701 __ BIND(L_good_sp);
3702 }
3703 #endif // ASSERT
3704
3705 // rbx contains the size of the frames to thaw, 0 if overflow or no more frames
3706 Label L_thaw_success;
3707 __ testptr(rbx, rbx);
3708 __ jccb(Assembler::notZero, L_thaw_success);
3709 __ jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
3710 __ bind(L_thaw_success);
3711
3712 // Make room for the thawed frames and align the stack.
3713 __ subptr(rsp, rbx);
3714 __ andptr(rsp, -StackAlignmentInBytes);
3715
3716 if (return_barrier) {
3717 // Preserve possible return value from a method returning to the return barrier. (Again.)
3718 __ push(rax);
3719 __ push_d(xmm0);
3720 }
3721
3722 // If we want, we can templatize thaw by kind, and have three different entries.
3723 __ movptr(c_rarg0, r15_thread);
3724 __ movptr(c_rarg1, kind);
3725 __ call_VM_leaf(Continuation::thaw_entry(), 2);
3726 __ movptr(rbx, rax);
3727
3728 if (return_barrier) {
3729 // Restore return value from a method returning to the return barrier. (Again.)
3730 // No safepoint in the call to thaw, so even an oop return value should be OK.
3731 __ pop_d(xmm0);
3732 __ pop(rax);
3733 } else {
3734 // Return 0 (success) from doYield.
3735 __ xorptr(rax, rax);
3736 }
3737
3738 // After thawing, rbx is the SP of the yielding frame.
3739 // Move there, and then to saved RBP slot.
3740 __ movptr(rsp, rbx);
3741 __ subptr(rsp, 2*wordSize);
3742
3743 if (return_barrier_exception) {
3744 __ movptr(c_rarg0, r15_thread);
3745 __ movptr(c_rarg1, Address(rsp, wordSize)); // return address
3746
3747 // rax still holds the original exception oop, save it before the call
3748 __ push(rax);
3749
3750 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 2);
3751 __ movptr(rbx, rax);
3752
3753 // Continue at exception handler:
3754 // rax: exception oop
3755 // rbx: exception handler
3756 // rdx: exception pc
3757 __ pop(rax);
3758 __ verify_oop(rax);
3759 __ pop(rbp); // pop out RBP here too
3760 __ pop(rdx);
3761 __ jmp(rbx);
3762 } else {
3763 // We are "returning" into the topmost thawed frame; see Thaw::push_return_frame
3764 __ pop(rbp);
3765 __ ret(0);
3766 }
3767
3768 return start;
3769 }
3770
3771 address StubGenerator::generate_cont_thaw() {
3772 return generate_cont_thaw("Cont thaw", Continuation::thaw_top);
3773 }
3774
3775 // TODO: will probably need multiple return barriers depending on return type
3776
3777 address StubGenerator::generate_cont_returnBarrier() {
3778 return generate_cont_thaw("Cont thaw return barrier", Continuation::thaw_return_barrier);
3779 }
3780
3781 address StubGenerator::generate_cont_returnBarrier_exception() {
3782 return generate_cont_thaw("Cont thaw return barrier exception", Continuation::thaw_return_barrier_exception);
3783 }
3784
3785 address StubGenerator::generate_cont_preempt_stub() {
3786 if (!Continuations::enabled()) return nullptr;
3787 StubCodeMark mark(this, "StubRoutines","Continuation preempt stub");
3788 address start = __ pc();
3789
3790 __ reset_last_Java_frame(true);
3791
3792 // Set rsp to enterSpecial frame, i.e. remove all frames copied into the heap.
3793 __ movptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3794
3795 Label preemption_cancelled;
3796 __ movbool(rscratch1, Address(r15_thread, JavaThread::preemption_cancelled_offset()));
3797 __ testbool(rscratch1);
3798 __ jcc(Assembler::notZero, preemption_cancelled);
3799
3800 // Remove enterSpecial frame from the stack and return to Continuation.run() to unmount.
3801 SharedRuntime::continuation_enter_cleanup(_masm);
3802 __ pop(rbp);
3803 __ ret(0);
3804
3805 // We acquired the monitor after freezing the frames so call thaw to continue execution.
3806 __ bind(preemption_cancelled);
3807 __ movbool(Address(r15_thread, JavaThread::preemption_cancelled_offset()), false);
3808 __ lea(rbp, Address(rsp, checked_cast<int32_t>(ContinuationEntry::size())));
3809 __ movptr(rscratch1, ExternalAddress(ContinuationEntry::thaw_call_pc_address()));
3810 __ jmp(rscratch1);
3811
3812 return start;
3813 }
3814
3815 // exception handler for upcall stubs
3816 address StubGenerator::generate_upcall_stub_exception_handler() {
3817 StubCodeMark mark(this, "StubRoutines", "upcall stub exception handler");
3818 address start = __ pc();
3819
3820 // native caller has no idea how to handle exceptions
3821 // we just crash here. Up to callee to catch exceptions.
3822 __ verify_oop(rax);
3823 __ vzeroupper();
3824 __ mov(c_rarg0, rax);
3825 __ andptr(rsp, -StackAlignmentInBytes); // align stack as required by ABI
3826 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
3827 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, UpcallLinker::handle_uncaught_exception)));
3828 __ should_not_reach_here();
3829
3830 return start;
3831 }
3832
3833 // load Method* target of MethodHandle
3834 // j_rarg0 = jobject receiver
3835 // rbx = result
3836 address StubGenerator::generate_upcall_stub_load_target() {
3837 StubCodeMark mark(this, "StubRoutines", "upcall_stub_load_target");
3838 address start = __ pc();
3839
3840 __ resolve_global_jobject(j_rarg0, r15_thread, rscratch1);
3841 // Load target method from receiver
3842 __ load_heap_oop(rbx, Address(j_rarg0, java_lang_invoke_MethodHandle::form_offset()), rscratch1);
3843 __ load_heap_oop(rbx, Address(rbx, java_lang_invoke_LambdaForm::vmentry_offset()), rscratch1);
3844 __ load_heap_oop(rbx, Address(rbx, java_lang_invoke_MemberName::method_offset()), rscratch1);
3845 __ access_load_at(T_ADDRESS, IN_HEAP, rbx,
3846 Address(rbx, java_lang_invoke_ResolvedMethodName::vmtarget_offset()),
3847 noreg, noreg);
3848 __ movptr(Address(r15_thread, JavaThread::callee_target_offset()), rbx); // just in case callee is deoptimized
3849
3850 __ ret(0);
3851
3852 return start;
3853 }
3854
3855 address StubGenerator::generate_lookup_secondary_supers_table_stub(u1 super_klass_index) {
3856 StubCodeMark mark(this, "StubRoutines", "lookup_secondary_supers_table");
3857
3858 address start = __ pc();
3859
3860 const Register
3861 r_super_klass = rax,
3862 r_sub_klass = rsi,
3863 result = rdi;
3864
3865 __ lookup_secondary_supers_table_const(r_sub_klass, r_super_klass,
3866 rdx, rcx, rbx, r11, // temps
3867 result,
3868 super_klass_index);
3869 __ ret(0);
3870
3871 return start;
3872 }
3873
3874 // Slow path implementation for UseSecondarySupersTable.
3875 address StubGenerator::generate_lookup_secondary_supers_table_slow_path_stub() {
3876 StubCodeMark mark(this, "StubRoutines", "lookup_secondary_supers_table");
3877
3878 address start = __ pc();
3879
3880 const Register
3881 r_super_klass = rax,
3882 r_array_base = rbx,
3883 r_array_index = rdx,
3884 r_sub_klass = rsi,
3885 r_bitmap = r11,
3886 result = rdi;
3887
3888 Label L_success;
3889 __ lookup_secondary_supers_table_slow_path(r_super_klass, r_array_base, r_array_index, r_bitmap,
3890 rcx, rdi, // temps
3891 &L_success);
3892 // bind(L_failure);
3893 __ movl(result, 1);
3894 __ ret(0);
3895
3896 __ bind(L_success);
3897 __ movl(result, 0);
3898 __ ret(0);
3899
3900 return start;
3901 }
3902
3903 void StubGenerator::create_control_words() {
3904 // Round to nearest, 64-bit mode, exceptions masked, flags specialized
3905 StubRoutines::x86::_mxcsr_std = EnableX86ECoreOpts ? 0x1FBF : 0x1F80;
3906 // Round to zero, 64-bit mode, exceptions masked, flags specialized
3907 StubRoutines::x86::_mxcsr_rz = EnableX86ECoreOpts ? 0x7FBF : 0x7F80;
3908 }
3909
3910 // Initialization
3911 void StubGenerator::generate_initial_stubs() {
3912 // Generates all stubs and initializes the entry points
3913
3914 // This platform-specific settings are needed by generate_call_stub()
3915 create_control_words();
3916
3917 // Initialize table for unsafe copy memeory check.
3918 if (UnsafeMemoryAccess::_table == nullptr) {
3919 UnsafeMemoryAccess::create_table(16 + 4); // 16 for copyMemory; 4 for setMemory
3920 }
3921
3922 // entry points that exist in all platforms Note: This is code
3923 // that could be shared among different platforms - however the
3924 // benefit seems to be smaller than the disadvantage of having a
3925 // much more complicated generator structure. See also comment in
3926 // stubRoutines.hpp.
3927
3928 StubRoutines::_forward_exception_entry = generate_forward_exception();
3929
3930 StubRoutines::_call_stub_entry =
3931 generate_call_stub(StubRoutines::_call_stub_return_address);
3932
3933 // is referenced by megamorphic call
3934 StubRoutines::_catch_exception_entry = generate_catch_exception();
3935
3936 // atomic calls
3937 StubRoutines::_fence_entry = generate_orderaccess_fence();
3938
3939 // platform dependent
3940 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
3941
3942 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3943
3944 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
3945 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
3946 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
3947 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
3948
3949 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
3950 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
3951 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3952 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3953
3954 if (UseCRC32Intrinsics) {
3955 // set table address before stub generation which use it
3956 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3957 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3958 }
3959
3960 if (UseCRC32CIntrinsics) {
3961 bool supports_clmul = VM_Version::supports_clmul();
3962 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
3963 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
3964 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
3965 }
3966
3967 if (VM_Version::supports_float16()) {
3968 // For results consistency both intrinsics should be enabled.
3969 // vmIntrinsics checks InlineIntrinsics flag, no need to check it here.
3970 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_float16ToFloat) &&
3971 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_floatToFloat16)) {
3972 StubRoutines::_hf2f = generate_float16ToFloat();
3973 StubRoutines::_f2hf = generate_floatToFloat16();
3974 }
3975 }
3976
3977 generate_libm_stubs();
3978
3979 StubRoutines::_fmod = generate_libmFmod(); // from stubGenerator_x86_64_fmod.cpp
3980 }
3981
3982 void StubGenerator::generate_continuation_stubs() {
3983 // Continuation stubs:
3984 StubRoutines::_cont_thaw = generate_cont_thaw();
3985 StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
3986 StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
3987 StubRoutines::_cont_preempt_stub = generate_cont_preempt_stub();
3988 }
3989
3990 void StubGenerator::generate_final_stubs() {
3991 // Generates the rest of stubs and initializes the entry points
3992
3993 // support for verify_oop (must happen after universe_init)
3994 if (VerifyOops) {
3995 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3996 }
3997
3998 // data cache line writeback
3999 StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
4000 StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
4001
4002 // arraycopy stubs used by compilers
4003 generate_arraycopy_stubs();
4004
4005 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
4006 if (bs_nm != nullptr) {
4007 StubRoutines::_method_entry_barrier = generate_method_entry_barrier();
4008 }
4009
4010 if (UseVectorizedMismatchIntrinsic) {
4011 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
4012 }
4013
4014 StubRoutines::_upcall_stub_exception_handler = generate_upcall_stub_exception_handler();
4015 StubRoutines::_upcall_stub_load_target = generate_upcall_stub_load_target();
4016 }
4017
4018 void StubGenerator::generate_compiler_stubs() {
4019 #if COMPILER2_OR_JVMCI
4020
4021 // Entry points that are C2 compiler specific.
4022
4023 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
4024 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
4025 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4026 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
4027 StubRoutines::x86::_vector_all_bits_set = generate_vector_mask("vector_all_bits_set", 0xFFFFFFFFFFFFFFFF);
4028 StubRoutines::x86::_vector_int_mask_cmp_bits = generate_vector_mask("vector_int_mask_cmp_bits", 0x0000000100000001);
4029 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
4030 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
4031 StubRoutines::x86::_vector_int_to_byte_mask = generate_vector_mask("vector_int_to_byte_mask", 0x000000ff000000ff);
4032 StubRoutines::x86::_vector_int_to_short_mask = generate_vector_mask("vector_int_to_short_mask", 0x0000ffff0000ffff);
4033 StubRoutines::x86::_vector_32_bit_mask = generate_vector_custom_i32("vector_32_bit_mask", Assembler::AVX_512bit,
4034 0xFFFFFFFF, 0, 0, 0);
4035 StubRoutines::x86::_vector_64_bit_mask = generate_vector_custom_i32("vector_64_bit_mask", Assembler::AVX_512bit,
4036 0xFFFFFFFF, 0xFFFFFFFF, 0, 0);
4037 StubRoutines::x86::_vector_int_shuffle_mask = generate_vector_mask("vector_int_shuffle_mask", 0x0302010003020100);
4038 StubRoutines::x86::_vector_byte_shuffle_mask = generate_vector_byte_shuffle_mask("vector_byte_shuffle_mask");
4039 StubRoutines::x86::_vector_short_shuffle_mask = generate_vector_mask("vector_short_shuffle_mask", 0x0100010001000100);
4040 StubRoutines::x86::_vector_long_shuffle_mask = generate_vector_mask("vector_long_shuffle_mask", 0x0000000100000000);
4041 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
4042 StubRoutines::x86::_vector_iota_indices = generate_iota_indices("iota_indices");
4043 StubRoutines::x86::_vector_count_leading_zeros_lut = generate_count_leading_zeros_lut("count_leading_zeros_lut");
4044 StubRoutines::x86::_vector_reverse_bit_lut = generate_vector_reverse_bit_lut("reverse_bit_lut");
4045 StubRoutines::x86::_vector_reverse_byte_perm_mask_long = generate_vector_reverse_byte_perm_mask_long("perm_mask_long");
4046 StubRoutines::x86::_vector_reverse_byte_perm_mask_int = generate_vector_reverse_byte_perm_mask_int("perm_mask_int");
4047 StubRoutines::x86::_vector_reverse_byte_perm_mask_short = generate_vector_reverse_byte_perm_mask_short("perm_mask_short");
4048
4049 if (VM_Version::supports_avx2() && !VM_Version::supports_avx512vl()) {
4050 StubRoutines::x86::_compress_perm_table32 = generate_compress_perm_table("compress_perm_table32", 32);
4051 StubRoutines::x86::_compress_perm_table64 = generate_compress_perm_table("compress_perm_table64", 64);
4052 StubRoutines::x86::_expand_perm_table32 = generate_expand_perm_table("expand_perm_table32", 32);
4053 StubRoutines::x86::_expand_perm_table64 = generate_expand_perm_table("expand_perm_table64", 64);
4054 }
4055
4056 if (VM_Version::supports_avx2() && !VM_Version::supports_avx512_vpopcntdq()) {
4057 // lut implementation influenced by counting 1s algorithm from section 5-1 of Hackers' Delight.
4058 StubRoutines::x86::_vector_popcount_lut = generate_popcount_avx_lut("popcount_lut");
4059 }
4060
4061 generate_aes_stubs();
4062
4063 generate_ghash_stubs();
4064
4065 generate_chacha_stubs();
4066
4067 generate_sha3_stubs();
4068
4069 #ifdef COMPILER2
4070 if ((UseAVX == 2) && EnableX86ECoreOpts) {
4071 generate_string_indexof(StubRoutines::_string_indexof_array);
4072 }
4073 #endif
4074
4075 if (UseAdler32Intrinsics) {
4076 StubRoutines::_updateBytesAdler32 = generate_updateBytesAdler32();
4077 }
4078
4079 if (UsePoly1305Intrinsics) {
4080 StubRoutines::_poly1305_processBlocks = generate_poly1305_processBlocks();
4081 }
4082
4083 if (UseIntPolyIntrinsics) {
4084 StubRoutines::_intpoly_montgomeryMult_P256 = generate_intpoly_montgomeryMult_P256();
4085 StubRoutines::_intpoly_assign = generate_intpoly_assign();
4086 }
4087
4088 if (UseMD5Intrinsics) {
4089 StubRoutines::_md5_implCompress = generate_md5_implCompress(false, "md5_implCompress");
4090 StubRoutines::_md5_implCompressMB = generate_md5_implCompress(true, "md5_implCompressMB");
4091 }
4092
4093 if (UseSHA1Intrinsics) {
4094 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
4095 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
4096 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
4097 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
4098 }
4099
4100 if (UseSHA256Intrinsics) {
4101 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
4102 char* dst = (char*)StubRoutines::x86::_k256_W;
4103 char* src = (char*)StubRoutines::x86::_k256;
4104 for (int ii = 0; ii < 16; ++ii) {
4105 memcpy(dst + 32 * ii, src + 16 * ii, 16);
4106 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
4107 }
4108 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
4109 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
4110 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
4111 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
4112 }
4113
4114 if (UseSHA512Intrinsics) {
4115 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
4116 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
4117 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
4118 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
4119 }
4120
4121 if (UseBASE64Intrinsics) {
4122 if(VM_Version::supports_avx2()) {
4123 StubRoutines::x86::_avx2_shuffle_base64 = base64_avx2_shuffle_addr();
4124 StubRoutines::x86::_avx2_input_mask_base64 = base64_avx2_input_mask_addr();
4125 StubRoutines::x86::_avx2_lut_base64 = base64_avx2_lut_addr();
4126 StubRoutines::x86::_avx2_decode_tables_base64 = base64_AVX2_decode_tables_addr();
4127 StubRoutines::x86::_avx2_decode_lut_tables_base64 = base64_AVX2_decode_LUT_tables_addr();
4128 }
4129 StubRoutines::x86::_encoding_table_base64 = base64_encoding_table_addr();
4130 if (VM_Version::supports_avx512_vbmi()) {
4131 StubRoutines::x86::_shuffle_base64 = base64_shuffle_addr();
4132 StubRoutines::x86::_lookup_lo_base64 = base64_vbmi_lookup_lo_addr();
4133 StubRoutines::x86::_lookup_hi_base64 = base64_vbmi_lookup_hi_addr();
4134 StubRoutines::x86::_lookup_lo_base64url = base64_vbmi_lookup_lo_url_addr();
4135 StubRoutines::x86::_lookup_hi_base64url = base64_vbmi_lookup_hi_url_addr();
4136 StubRoutines::x86::_pack_vec_base64 = base64_vbmi_pack_vec_addr();
4137 StubRoutines::x86::_join_0_1_base64 = base64_vbmi_join_0_1_addr();
4138 StubRoutines::x86::_join_1_2_base64 = base64_vbmi_join_1_2_addr();
4139 StubRoutines::x86::_join_2_3_base64 = base64_vbmi_join_2_3_addr();
4140 }
4141 StubRoutines::x86::_decoding_table_base64 = base64_decoding_table_addr();
4142 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
4143 StubRoutines::_base64_decodeBlock = generate_base64_decodeBlock();
4144 }
4145
4146 #ifdef COMPILER2
4147 if (UseMultiplyToLenIntrinsic) {
4148 StubRoutines::_multiplyToLen = generate_multiplyToLen();
4149 }
4150 if (UseSquareToLenIntrinsic) {
4151 StubRoutines::_squareToLen = generate_squareToLen();
4152 }
4153 if (UseMulAddIntrinsic) {
4154 StubRoutines::_mulAdd = generate_mulAdd();
4155 }
4156 if (VM_Version::supports_avx512_vbmi2()) {
4157 StubRoutines::_bigIntegerRightShiftWorker = generate_bigIntegerRightShift();
4158 StubRoutines::_bigIntegerLeftShiftWorker = generate_bigIntegerLeftShift();
4159 }
4160 if (UseSecondarySupersTable) {
4161 StubRoutines::_lookup_secondary_supers_table_slow_path_stub = generate_lookup_secondary_supers_table_slow_path_stub();
4162 if (! InlineSecondarySupersTest) {
4163 for (int slot = 0; slot < Klass::SECONDARY_SUPERS_TABLE_SIZE; slot++) {
4164 StubRoutines::_lookup_secondary_supers_table_stubs[slot] = generate_lookup_secondary_supers_table_stub(slot);
4165 }
4166 }
4167 }
4168 if (UseMontgomeryMultiplyIntrinsic) {
4169 StubRoutines::_montgomeryMultiply
4170 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
4171 }
4172 if (UseMontgomerySquareIntrinsic) {
4173 StubRoutines::_montgomerySquare
4174 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
4175 }
4176
4177 // Load x86_64_sort library on supported hardware to enable SIMD sort and partition intrinsics
4178
4179 if (VM_Version::is_intel() && (VM_Version::supports_avx512dq() || VM_Version::supports_avx2())) {
4180 void *libsimdsort = nullptr;
4181 char ebuf_[1024];
4182 char dll_name_simd_sort[JVM_MAXPATHLEN];
4183 if (os::dll_locate_lib(dll_name_simd_sort, sizeof(dll_name_simd_sort), Arguments::get_dll_dir(), "simdsort")) {
4184 libsimdsort = os::dll_load(dll_name_simd_sort, ebuf_, sizeof ebuf_);
4185 }
4186 // Get addresses for SIMD sort and partition routines
4187 if (libsimdsort != nullptr) {
4188 log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "simdsort" JNI_LIB_SUFFIX, p2i(libsimdsort));
4189
4190 snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512dq() ? "avx512_sort" : "avx2_sort");
4191 StubRoutines::_array_sort = (address)os::dll_lookup(libsimdsort, ebuf_);
4192
4193 snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512dq() ? "avx512_partition" : "avx2_partition");
4194 StubRoutines::_array_partition = (address)os::dll_lookup(libsimdsort, ebuf_);
4195 }
4196 }
4197
4198 // Get svml stub routine addresses
4199 void *libjsvml = nullptr;
4200 char ebuf[1024];
4201 char dll_name[JVM_MAXPATHLEN];
4202 if (os::dll_locate_lib(dll_name, sizeof(dll_name), Arguments::get_dll_dir(), "jsvml")) {
4203 libjsvml = os::dll_load(dll_name, ebuf, sizeof ebuf);
4204 }
4205 if (libjsvml != nullptr) {
4206 // SVML method naming convention
4207 // All the methods are named as __jsvml_op<T><N>_ha_<VV>
4208 // Where:
4209 // ha stands for high accuracy
4210 // <T> is optional to indicate float/double
4211 // Set to f for vector float operation
4212 // Omitted for vector double operation
4213 // <N> is the number of elements in the vector
4214 // 1, 2, 4, 8, 16
4215 // e.g. 128 bit float vector has 4 float elements
4216 // <VV> indicates the avx/sse level:
4217 // z0 is AVX512, l9 is AVX2, e9 is AVX1 and ex is for SSE2
4218 // e.g. __jsvml_expf16_ha_z0 is the method for computing 16 element vector float exp using AVX 512 insns
4219 // __jsvml_exp8_ha_z0 is the method for computing 8 element vector double exp using AVX 512 insns
4220
4221 log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "jsvml" JNI_LIB_SUFFIX, p2i(libjsvml));
4222 if (UseAVX > 2) {
4223 for (int op = 0; op < VectorSupport::NUM_VECTOR_OP_MATH; op++) {
4224 int vop = VectorSupport::VECTOR_OP_MATH_START + op;
4225 if ((!VM_Version::supports_avx512dq()) &&
4226 (vop == VectorSupport::VECTOR_OP_LOG || vop == VectorSupport::VECTOR_OP_LOG10 || vop == VectorSupport::VECTOR_OP_POW)) {
4227 continue;
4228 }
4229 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf16_ha_z0", VectorSupport::mathname[op]);
4230 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4231
4232 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s8_ha_z0", VectorSupport::mathname[op]);
4233 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4234 }
4235 }
4236 const char* avx_sse_str = (UseAVX >= 2) ? "l9" : ((UseAVX == 1) ? "e9" : "ex");
4237 for (int op = 0; op < VectorSupport::NUM_VECTOR_OP_MATH; op++) {
4238 int vop = VectorSupport::VECTOR_OP_MATH_START + op;
4239 if (vop == VectorSupport::VECTOR_OP_POW) {
4240 continue;
4241 }
4242 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4243 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4244
4245 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4246 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4247
4248 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf8_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4249 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4250
4251 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s1_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4252 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4253
4254 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s2_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4255 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4256
4257 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s4_ha_%s", VectorSupport::mathname[op], avx_sse_str);
4258 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4259 }
4260 }
4261
4262 #endif // COMPILER2
4263 #endif // COMPILER2_OR_JVMCI
4264 }
4265
4266 StubGenerator::StubGenerator(CodeBuffer* code, StubsKind kind) : StubCodeGenerator(code) {
4267 DEBUG_ONLY( _regs_in_thread = false; )
4268 switch(kind) {
4269 case Initial_stubs:
4270 generate_initial_stubs();
4271 break;
4272 case Continuation_stubs:
4273 generate_continuation_stubs();
4274 break;
4275 case Compiler_stubs:
4276 generate_compiler_stubs();
4277 break;
4278 case Final_stubs:
4279 generate_final_stubs();
4280 break;
4281 default:
4282 fatal("unexpected stubs kind: %d", kind);
4283 break;
4284 };
4285 }
4286
4287 void StubGenerator_generate(CodeBuffer* code, StubCodeGenerator::StubsKind kind) {
4288 StubGenerator g(code, kind);
4289 }
4290
4291 #undef __