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/assembler.hpp"
27 #include "asm/macroAssembler.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/javaThread.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubRoutines.hpp"
41 #include "utilities/macros.hpp"
42 #include "vmreg_x86.inline.hpp"
43 #include "stubGenerator_x86_64.hpp"
44 #ifdef COMPILER2
45 #include "opto/runtime.hpp"
46 #include "opto/c2_globals.hpp"
47 #endif
48 #if INCLUDE_JVMCI
49 #include "jvmci/jvmci_globals.hpp"
50 #endif
51 #if INCLUDE_JFR
52 #include "jfr/support/jfrIntrinsics.hpp"
53 #endif
54
55 // For a more detailed description of the stub routine structure
56 // see the comment in stubRoutines.hpp
57
58 #define __ _masm->
59 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
60
61 #ifdef PRODUCT
62 #define BLOCK_COMMENT(str) /* nothing */
63 #else
64 #define BLOCK_COMMENT(str) __ block_comment(str)
65 #endif // PRODUCT
66
67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
68
69 //
70 // Linux Arguments:
71 // c_rarg0: call wrapper address address
72 // c_rarg1: result address
73 // c_rarg2: result type BasicType
74 // c_rarg3: method Method*
75 // c_rarg4: (interpreter) entry point address
76 // c_rarg5: parameters intptr_t*
77 // 16(rbp): parameter size (in words) int
78 // 24(rbp): thread Thread*
79 //
80 // [ return_from_Java ] <--- rsp
81 // [ argument word n ]
82 // ...
83 // -12 [ argument word 1 ]
84 // -11 [ saved r15 ] <--- rsp_after_call
85 // -10 [ saved r14 ]
86 // -9 [ saved r13 ]
87 // -8 [ saved r12 ]
88 // -7 [ saved rbx ]
89 // -6 [ call wrapper ]
90 // -5 [ result ]
91 // -4 [ result type ]
92 // -3 [ method ]
93 // -2 [ entry point ]
94 // -1 [ parameters ]
95 // 0 [ saved rbp ] <--- rbp
96 // 1 [ return address ]
97 // 2 [ parameter size ]
98 // 3 [ thread ]
99 //
100 // Windows Arguments:
101 // c_rarg0: call wrapper address address
102 // c_rarg1: result address
103 // c_rarg2: result type BasicType
104 // c_rarg3: method Method*
105 // 48(rbp): (interpreter) entry point address
106 // 56(rbp): parameters intptr_t*
107 // 64(rbp): parameter size (in words) int
108 // 72(rbp): thread Thread*
109 //
110 // [ return_from_Java ] <--- rsp
111 // [ argument word n ]
112 // ...
113 // -28 [ argument word 1 ]
114 // -27 [ saved xmm15 ] <--- rsp after_call
115 // [ saved xmm7-xmm14 ]
116 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
117 // -7 [ saved r15 ]
118 // -6 [ saved r14 ]
119 // -5 [ saved r13 ]
120 // -4 [ saved r12 ]
121 // -3 [ saved rdi ]
122 // -2 [ saved rsi ]
123 // -1 [ saved rbx ]
124 // 0 [ saved rbp ] <--- rbp
125 // 1 [ return address ]
126 // 2 [ call wrapper ]
127 // 3 [ result ]
128 // 4 [ result type ]
129 // 5 [ method ]
130 // 6 [ entry point ]
131 // 7 [ parameters ]
132 // 8 [ parameter size ]
133 // 9 [ thread ]
134 //
135 // Windows reserves the callers stack space for arguments 1-4.
136 // We spill c_rarg0-c_rarg3 to this space.
137
138 // Call stub stack layout word offsets from rbp
139 #ifdef _WIN64
140 enum call_stub_layout {
141 xmm_save_first = 6, // save from xmm6
142 xmm_save_last = 15, // to xmm15
143 xmm_save_base = -9,
144 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
145 r15_off = -7,
146 r14_off = -6,
147 r13_off = -5,
148 r12_off = -4,
149 rdi_off = -3,
150 rsi_off = -2,
151 rbx_off = -1,
152 rbp_off = 0,
153 retaddr_off = 1,
154 call_wrapper_off = 2,
155 result_off = 3,
156 result_type_off = 4,
157 method_off = 5,
158 entry_point_off = 6,
159 parameters_off = 7,
160 parameter_size_off = 8,
161 thread_off = 9
162 };
163
164 static Address xmm_save(int reg) {
165 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
166 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
167 }
168 #else // !_WIN64
169 enum call_stub_layout {
170 rsp_after_call_off = -12,
171 mxcsr_off = rsp_after_call_off,
172 r15_off = -11,
173 r14_off = -10,
174 r13_off = -9,
175 r12_off = -8,
176 rbx_off = -7,
177 call_wrapper_off = -6,
178 result_off = -5,
179 result_type_off = -4,
180 method_off = -3,
181 entry_point_off = -2,
182 parameters_off = -1,
183 rbp_off = 0,
184 retaddr_off = 1,
185 parameter_size_off = 2,
186 thread_off = 3
187 };
188 #endif // _WIN64
189
190 address StubGenerator::generate_call_stub(address& return_address) {
191
192 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
193 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
194 "adjust this code");
195 StubCodeMark mark(this, "StubRoutines", "call_stub");
196 address start = __ pc();
197
198 // same as in generate_catch_exception()!
199 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
200
201 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
202 const Address result (rbp, result_off * wordSize);
203 const Address result_type (rbp, result_type_off * wordSize);
204 const Address method (rbp, method_off * wordSize);
205 const Address entry_point (rbp, entry_point_off * wordSize);
206 const Address parameters (rbp, parameters_off * wordSize);
207 const Address parameter_size(rbp, parameter_size_off * wordSize);
208
209 // same as in generate_catch_exception()!
210 const Address thread (rbp, thread_off * wordSize);
211
212 const Address r15_save(rbp, r15_off * wordSize);
213 const Address r14_save(rbp, r14_off * wordSize);
214 const Address r13_save(rbp, r13_off * wordSize);
215 const Address r12_save(rbp, r12_off * wordSize);
216 const Address rbx_save(rbp, rbx_off * wordSize);
217
218 // stub code
219 __ enter();
220 __ subptr(rsp, -rsp_after_call_off * wordSize);
221
222 // save register parameters
223 #ifndef _WIN64
224 __ movptr(parameters, c_rarg5); // parameters
225 __ movptr(entry_point, c_rarg4); // entry_point
226 #endif
227
228 __ movptr(method, c_rarg3); // method
229 __ movl(result_type, c_rarg2); // result type
230 __ movptr(result, c_rarg1); // result
231 __ movptr(call_wrapper, c_rarg0); // call wrapper
232
233 // save regs belonging to calling function
234 __ movptr(rbx_save, rbx);
235 __ movptr(r12_save, r12);
236 __ movptr(r13_save, r13);
237 __ movptr(r14_save, r14);
238 __ movptr(r15_save, r15);
239
240 #ifdef _WIN64
241 int last_reg = 15;
242 for (int i = xmm_save_first; i <= last_reg; i++) {
243 __ movdqu(xmm_save(i), as_XMMRegister(i));
244 }
245
246 const Address rdi_save(rbp, rdi_off * wordSize);
247 const Address rsi_save(rbp, rsi_off * wordSize);
248
249 __ movptr(rsi_save, rsi);
250 __ movptr(rdi_save, rdi);
251 #else
252 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
253 {
254 Label skip_ldmx;
255 __ stmxcsr(mxcsr_save);
256 __ movl(rax, mxcsr_save);
257 __ andl(rax, 0xFFC0); // Mask out any pending exceptions (only check control and mask bits)
258 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
259 __ cmp32(rax, mxcsr_std, rscratch1);
260 __ jcc(Assembler::equal, skip_ldmx);
261 __ ldmxcsr(mxcsr_std, rscratch1);
262 __ bind(skip_ldmx);
263 }
264 #endif
265
266 // Load up thread register
267 __ movptr(r15_thread, thread);
268 __ reinit_heapbase();
269
270 #ifdef ASSERT
271 // make sure we have no pending exceptions
272 {
273 Label L;
274 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
275 __ jcc(Assembler::equal, L);
276 __ stop("StubRoutines::call_stub: entered with pending exception");
277 __ bind(L);
278 }
279 #endif
280
281 // pass parameters if any
282 BLOCK_COMMENT("pass parameters if any");
283 Label parameters_done;
284 __ movl(c_rarg3, parameter_size);
285 __ testl(c_rarg3, c_rarg3);
286 __ jcc(Assembler::zero, parameters_done);
287
288 Label loop;
289 __ movptr(c_rarg2, parameters); // parameter pointer
290 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
291 __ BIND(loop);
292 __ movptr(rax, Address(c_rarg2, 0));// get parameter
293 __ addptr(c_rarg2, wordSize); // advance to next parameter
294 __ decrementl(c_rarg1); // decrement counter
295 __ push(rax); // pass parameter
296 __ jcc(Assembler::notZero, loop);
297
298 // call Java function
299 __ BIND(parameters_done);
300 __ movptr(rbx, method); // get Method*
301 __ movptr(c_rarg1, entry_point); // get entry_point
302 __ mov(r13, rsp); // set sender sp
303 BLOCK_COMMENT("call Java function");
304 __ call(c_rarg1);
305
306 BLOCK_COMMENT("call_stub_return_address:");
307 return_address = __ pc();
308
309 // store result depending on type (everything that is not
310 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
311 __ movptr(r13, result);
312 Label is_long, is_float, is_double, check_prim, exit;
313 __ movl(rbx, result_type);
314 __ cmpl(rbx, T_OBJECT);
315 __ jcc(Assembler::equal, check_prim);
316 __ cmpl(rbx, T_LONG);
317 __ jcc(Assembler::equal, is_long);
318 __ cmpl(rbx, T_FLOAT);
319 __ jcc(Assembler::equal, is_float);
320 __ cmpl(rbx, T_DOUBLE);
321 __ jcc(Assembler::equal, is_double);
322 #ifdef ASSERT
323 // make sure the type is INT
324 {
325 Label L;
326 __ cmpl(rbx, T_INT);
327 __ jcc(Assembler::equal, L);
328 __ stop("StubRoutines::call_stub: unexpected result type");
329 __ bind(L);
330 }
331 #endif
332
333 // handle T_INT case
334 __ movl(Address(r13, 0), rax);
335
336 __ BIND(exit);
337
338 // pop parameters
339 __ lea(rsp, rsp_after_call);
340
341 #ifdef ASSERT
342 // verify that threads correspond
343 {
344 Label L1, L2, L3;
345 __ cmpptr(r15_thread, thread);
346 __ jcc(Assembler::equal, L1);
347 __ stop("StubRoutines::call_stub: r15_thread is corrupted");
348 __ bind(L1);
349 __ get_thread(rbx);
350 __ cmpptr(r15_thread, thread);
351 __ jcc(Assembler::equal, L2);
352 __ stop("StubRoutines::call_stub: r15_thread is modified by call");
353 __ bind(L2);
354 __ cmpptr(r15_thread, rbx);
355 __ jcc(Assembler::equal, L3);
356 __ stop("StubRoutines::call_stub: threads must correspond");
357 __ bind(L3);
358 }
359 #endif
360
361 __ pop_cont_fastpath();
362
363 // restore regs belonging to calling function
364 #ifdef _WIN64
365 // emit the restores for xmm regs
366 for (int i = xmm_save_first; i <= last_reg; i++) {
367 __ movdqu(as_XMMRegister(i), xmm_save(i));
368 }
369 #endif
370 __ movptr(r15, r15_save);
371 __ movptr(r14, r14_save);
372 __ movptr(r13, r13_save);
373 __ movptr(r12, r12_save);
374 __ movptr(rbx, rbx_save);
375
376 #ifdef _WIN64
377 __ movptr(rdi, rdi_save);
378 __ movptr(rsi, rsi_save);
379 #else
380 __ ldmxcsr(mxcsr_save);
381 #endif
382
383 // restore rsp
384 __ addptr(rsp, -rsp_after_call_off * wordSize);
385
386 // return
387 __ vzeroupper();
388 __ pop(rbp);
389 __ ret(0);
390
391 // handle return types different from T_INT
392 __ BIND(check_prim);
393 if (InlineTypeReturnedAsFields) {
394 // Check for scalarized return value
395 __ testptr(rax, 1);
396 __ jcc(Assembler::zero, is_long);
397 // Load pack handler address
398 __ andptr(rax, -2);
399 __ movptr(rax, Address(rax, InstanceKlass::adr_inlineklass_fixed_block_offset()));
400 __ movptr(rbx, Address(rax, InlineKlass::pack_handler_jobject_offset()));
401 // Call pack handler to initialize the buffer
402 __ call(rbx);
403 __ jmp(exit);
404 }
405 __ BIND(is_long);
406 __ movq(Address(r13, 0), rax);
407 __ jmp(exit);
408
409 __ BIND(is_float);
410 __ movflt(Address(r13, 0), xmm0);
411 __ jmp(exit);
412
413 __ BIND(is_double);
414 __ movdbl(Address(r13, 0), xmm0);
415 __ jmp(exit);
416
417 return start;
418 }
419
420 // Return point for a Java call if there's an exception thrown in
421 // Java code. The exception is caught and transformed into a
422 // pending exception stored in JavaThread that can be tested from
423 // within the VM.
424 //
425 // Note: Usually the parameters are removed by the callee. In case
426 // of an exception crossing an activation frame boundary, that is
427 // not the case if the callee is compiled code => need to setup the
428 // rsp.
429 //
430 // rax: exception oop
431
432 address StubGenerator::generate_catch_exception() {
433 StubCodeMark mark(this, "StubRoutines", "catch_exception");
434 address start = __ pc();
435
436 // same as in generate_call_stub():
437 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
438 const Address thread (rbp, thread_off * wordSize);
439
440 #ifdef ASSERT
441 // verify that threads correspond
442 {
443 Label L1, L2, L3;
444 __ cmpptr(r15_thread, thread);
445 __ jcc(Assembler::equal, L1);
446 __ stop("StubRoutines::catch_exception: r15_thread is corrupted");
447 __ bind(L1);
448 __ get_thread(rbx);
449 __ cmpptr(r15_thread, thread);
450 __ jcc(Assembler::equal, L2);
451 __ stop("StubRoutines::catch_exception: r15_thread is modified by call");
452 __ bind(L2);
453 __ cmpptr(r15_thread, rbx);
454 __ jcc(Assembler::equal, L3);
455 __ stop("StubRoutines::catch_exception: threads must correspond");
456 __ bind(L3);
457 }
458 #endif
459
460 // set pending exception
461 __ verify_oop(rax);
462
463 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
464 __ lea(rscratch1, ExternalAddress((address)__FILE__));
465 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
466 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
467
468 // complete return to VM
469 assert(StubRoutines::_call_stub_return_address != nullptr,
470 "_call_stub_return_address must have been generated before");
471 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
472
473 return start;
474 }
475
476 // Continuation point for runtime calls returning with a pending
477 // exception. The pending exception check happened in the runtime
478 // or native call stub. The pending exception in Thread is
479 // converted into a Java-level exception.
480 //
481 // Contract with Java-level exception handlers:
482 // rax: exception
483 // rdx: throwing pc
484 //
485 // NOTE: At entry of this stub, exception-pc must be on stack !!
486
487 address StubGenerator::generate_forward_exception() {
488 StubCodeMark mark(this, "StubRoutines", "forward exception");
489 address start = __ pc();
490
491 // Upon entry, the sp points to the return address returning into
492 // Java (interpreted or compiled) code; i.e., the return address
493 // becomes the throwing pc.
494 //
495 // Arguments pushed before the runtime call are still on the stack
496 // but the exception handler will reset the stack pointer ->
497 // ignore them. A potential result in registers can be ignored as
498 // well.
499
500 #ifdef ASSERT
501 // make sure this code is only executed if there is a pending exception
502 {
503 Label L;
504 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
505 __ jcc(Assembler::notEqual, L);
506 __ stop("StubRoutines::forward exception: no pending exception (1)");
507 __ bind(L);
508 }
509 #endif
510
511 // compute exception handler into rbx
512 __ movptr(c_rarg0, Address(rsp, 0));
513 BLOCK_COMMENT("call exception_handler_for_return_address");
514 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
515 SharedRuntime::exception_handler_for_return_address),
516 r15_thread, c_rarg0);
517 __ mov(rbx, rax);
518
519 // setup rax & rdx, remove return address & clear pending exception
520 __ pop(rdx);
521 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
522 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
523
524 #ifdef ASSERT
525 // make sure exception is set
526 {
527 Label L;
528 __ testptr(rax, rax);
529 __ jcc(Assembler::notEqual, L);
530 __ stop("StubRoutines::forward exception: no pending exception (2)");
531 __ bind(L);
532 }
533 #endif
534
535 // continue at exception handler (return address removed)
536 // rax: exception
537 // rbx: exception handler
538 // rdx: throwing pc
539 __ verify_oop(rax);
540 __ jmp(rbx);
541
542 return start;
543 }
544
545 // Support for intptr_t OrderAccess::fence()
546 //
547 // Arguments :
548 //
549 // Result:
550 address StubGenerator::generate_orderaccess_fence() {
551 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
552 address start = __ pc();
553
554 __ membar(Assembler::StoreLoad);
555 __ ret(0);
556
557 return start;
558 }
559
560
561 // Support for intptr_t get_previous_sp()
562 //
563 // This routine is used to find the previous stack pointer for the
564 // caller.
565 address StubGenerator::generate_get_previous_sp() {
566 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
567 address start = __ pc();
568
569 __ movptr(rax, rsp);
570 __ addptr(rax, 8); // return address is at the top of the stack.
571 __ ret(0);
572
573 return start;
574 }
575
576 //----------------------------------------------------------------------------------------------------
577 // Support for void verify_mxcsr()
578 //
579 // This routine is used with -Xcheck:jni to verify that native
580 // JNI code does not return to Java code without restoring the
581 // MXCSR register to our expected state.
582
583 address StubGenerator::generate_verify_mxcsr() {
584 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
585 address start = __ pc();
586
587 const Address mxcsr_save(rsp, 0);
588
589 if (CheckJNICalls) {
590 Label ok_ret;
591 ExternalAddress mxcsr_std(StubRoutines::x86::addr_mxcsr_std());
592 __ push(rax);
593 __ subptr(rsp, wordSize); // allocate a temp location
594 __ stmxcsr(mxcsr_save);
595 __ movl(rax, mxcsr_save);
596 __ andl(rax, 0xFFC0); // Mask out any pending exceptions (only check control and mask bits)
597 __ cmp32(rax, mxcsr_std, rscratch1);
598 __ jcc(Assembler::equal, ok_ret);
599
600 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
601
602 __ ldmxcsr(mxcsr_std, rscratch1);
603
604 __ bind(ok_ret);
605 __ addptr(rsp, wordSize);
606 __ pop(rax);
607 }
608
609 __ ret(0);
610
611 return start;
612 }
613
614 address StubGenerator::generate_f2i_fixup() {
615 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
616 Address inout(rsp, 5 * wordSize); // return address + 4 saves
617
618 address start = __ pc();
619
620 Label L;
621
622 __ push(rax);
623 __ push(c_rarg3);
624 __ push(c_rarg2);
625 __ push(c_rarg1);
626
627 __ movl(rax, 0x7f800000);
628 __ xorl(c_rarg3, c_rarg3);
629 __ movl(c_rarg2, inout);
630 __ movl(c_rarg1, c_rarg2);
631 __ andl(c_rarg1, 0x7fffffff);
632 __ cmpl(rax, c_rarg1); // NaN? -> 0
633 __ jcc(Assembler::negative, L);
634 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
635 __ movl(c_rarg3, 0x80000000);
636 __ movl(rax, 0x7fffffff);
637 __ cmovl(Assembler::positive, c_rarg3, rax);
638
639 __ bind(L);
640 __ movptr(inout, c_rarg3);
641
642 __ pop(c_rarg1);
643 __ pop(c_rarg2);
644 __ pop(c_rarg3);
645 __ pop(rax);
646
647 __ ret(0);
648
649 return start;
650 }
651
652 address StubGenerator::generate_f2l_fixup() {
653 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
654 Address inout(rsp, 5 * wordSize); // return address + 4 saves
655 address start = __ pc();
656
657 Label L;
658
659 __ push(rax);
660 __ push(c_rarg3);
661 __ push(c_rarg2);
662 __ push(c_rarg1);
663
664 __ movl(rax, 0x7f800000);
665 __ xorl(c_rarg3, c_rarg3);
666 __ movl(c_rarg2, inout);
667 __ movl(c_rarg1, c_rarg2);
668 __ andl(c_rarg1, 0x7fffffff);
669 __ cmpl(rax, c_rarg1); // NaN? -> 0
670 __ jcc(Assembler::negative, L);
671 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
672 __ mov64(c_rarg3, 0x8000000000000000);
673 __ mov64(rax, 0x7fffffffffffffff);
674 __ cmov(Assembler::positive, c_rarg3, rax);
675
676 __ bind(L);
677 __ movptr(inout, c_rarg3);
678
679 __ pop(c_rarg1);
680 __ pop(c_rarg2);
681 __ pop(c_rarg3);
682 __ pop(rax);
683
684 __ ret(0);
685
686 return start;
687 }
688
689 address StubGenerator::generate_d2i_fixup() {
690 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
691 Address inout(rsp, 6 * wordSize); // return address + 5 saves
692
693 address start = __ pc();
694
695 Label L;
696
697 __ push(rax);
698 __ push(c_rarg3);
699 __ push(c_rarg2);
700 __ push(c_rarg1);
701 __ push(c_rarg0);
702
703 __ movl(rax, 0x7ff00000);
704 __ movq(c_rarg2, inout);
705 __ movl(c_rarg3, c_rarg2);
706 __ mov(c_rarg1, c_rarg2);
707 __ mov(c_rarg0, c_rarg2);
708 __ negl(c_rarg3);
709 __ shrptr(c_rarg1, 0x20);
710 __ orl(c_rarg3, c_rarg2);
711 __ andl(c_rarg1, 0x7fffffff);
712 __ xorl(c_rarg2, c_rarg2);
713 __ shrl(c_rarg3, 0x1f);
714 __ orl(c_rarg1, c_rarg3);
715 __ cmpl(rax, c_rarg1);
716 __ jcc(Assembler::negative, L); // NaN -> 0
717 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
718 __ movl(c_rarg2, 0x80000000);
719 __ movl(rax, 0x7fffffff);
720 __ cmov(Assembler::positive, c_rarg2, rax);
721
722 __ bind(L);
723 __ movptr(inout, c_rarg2);
724
725 __ pop(c_rarg0);
726 __ pop(c_rarg1);
727 __ pop(c_rarg2);
728 __ pop(c_rarg3);
729 __ pop(rax);
730
731 __ ret(0);
732
733 return start;
734 }
735
736 address StubGenerator::generate_d2l_fixup() {
737 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
738 Address inout(rsp, 6 * wordSize); // return address + 5 saves
739
740 address start = __ pc();
741
742 Label L;
743
744 __ push(rax);
745 __ push(c_rarg3);
746 __ push(c_rarg2);
747 __ push(c_rarg1);
748 __ push(c_rarg0);
749
750 __ movl(rax, 0x7ff00000);
751 __ movq(c_rarg2, inout);
752 __ movl(c_rarg3, c_rarg2);
753 __ mov(c_rarg1, c_rarg2);
754 __ mov(c_rarg0, c_rarg2);
755 __ negl(c_rarg3);
756 __ shrptr(c_rarg1, 0x20);
757 __ orl(c_rarg3, c_rarg2);
758 __ andl(c_rarg1, 0x7fffffff);
759 __ xorl(c_rarg2, c_rarg2);
760 __ shrl(c_rarg3, 0x1f);
761 __ orl(c_rarg1, c_rarg3);
762 __ cmpl(rax, c_rarg1);
763 __ jcc(Assembler::negative, L); // NaN -> 0
764 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
765 __ mov64(c_rarg2, 0x8000000000000000);
766 __ mov64(rax, 0x7fffffffffffffff);
767 __ cmovq(Assembler::positive, c_rarg2, rax);
768
769 __ bind(L);
770 __ movq(inout, c_rarg2);
771
772 __ pop(c_rarg0);
773 __ pop(c_rarg1);
774 __ pop(c_rarg2);
775 __ pop(c_rarg3);
776 __ pop(rax);
777
778 __ ret(0);
779
780 return start;
781 }
782
783 address StubGenerator::generate_count_leading_zeros_lut(const char *stub_name) {
784 __ align64();
785 StubCodeMark mark(this, "StubRoutines", stub_name);
786 address start = __ pc();
787
788 __ emit_data64(0x0101010102020304, relocInfo::none);
789 __ emit_data64(0x0000000000000000, relocInfo::none);
790 __ emit_data64(0x0101010102020304, relocInfo::none);
791 __ emit_data64(0x0000000000000000, relocInfo::none);
792 __ emit_data64(0x0101010102020304, relocInfo::none);
793 __ emit_data64(0x0000000000000000, relocInfo::none);
794 __ emit_data64(0x0101010102020304, relocInfo::none);
795 __ emit_data64(0x0000000000000000, relocInfo::none);
796
797 return start;
798 }
799
800 address StubGenerator::generate_popcount_avx_lut(const char *stub_name) {
801 __ align64();
802 StubCodeMark mark(this, "StubRoutines", stub_name);
803 address start = __ pc();
804
805 __ emit_data64(0x0302020102010100, relocInfo::none);
806 __ emit_data64(0x0403030203020201, relocInfo::none);
807 __ emit_data64(0x0302020102010100, relocInfo::none);
808 __ emit_data64(0x0403030203020201, relocInfo::none);
809 __ emit_data64(0x0302020102010100, relocInfo::none);
810 __ emit_data64(0x0403030203020201, relocInfo::none);
811 __ emit_data64(0x0302020102010100, relocInfo::none);
812 __ emit_data64(0x0403030203020201, relocInfo::none);
813
814 return start;
815 }
816
817 address StubGenerator::generate_iota_indices(const char *stub_name) {
818 __ align(CodeEntryAlignment);
819 StubCodeMark mark(this, "StubRoutines", stub_name);
820 address start = __ pc();
821 // B
822 __ emit_data64(0x0706050403020100, relocInfo::none);
823 __ emit_data64(0x0F0E0D0C0B0A0908, relocInfo::none);
824 __ emit_data64(0x1716151413121110, relocInfo::none);
825 __ emit_data64(0x1F1E1D1C1B1A1918, relocInfo::none);
826 __ emit_data64(0x2726252423222120, relocInfo::none);
827 __ emit_data64(0x2F2E2D2C2B2A2928, relocInfo::none);
828 __ emit_data64(0x3736353433323130, relocInfo::none);
829 __ emit_data64(0x3F3E3D3C3B3A3938, relocInfo::none);
830 // W
831 __ emit_data64(0x0003000200010000, relocInfo::none);
832 __ emit_data64(0x0007000600050004, relocInfo::none);
833 __ emit_data64(0x000B000A00090008, relocInfo::none);
834 __ emit_data64(0x000F000E000D000C, relocInfo::none);
835 __ emit_data64(0x0013001200110010, relocInfo::none);
836 __ emit_data64(0x0017001600150014, relocInfo::none);
837 __ emit_data64(0x001B001A00190018, relocInfo::none);
838 __ emit_data64(0x001F001E001D001C, relocInfo::none);
839 // D
840 __ emit_data64(0x0000000100000000, relocInfo::none);
841 __ emit_data64(0x0000000300000002, relocInfo::none);
842 __ emit_data64(0x0000000500000004, relocInfo::none);
843 __ emit_data64(0x0000000700000006, relocInfo::none);
844 __ emit_data64(0x0000000900000008, relocInfo::none);
845 __ emit_data64(0x0000000B0000000A, relocInfo::none);
846 __ emit_data64(0x0000000D0000000C, relocInfo::none);
847 __ emit_data64(0x0000000F0000000E, relocInfo::none);
848 // Q
849 __ emit_data64(0x0000000000000000, relocInfo::none);
850 __ emit_data64(0x0000000000000001, relocInfo::none);
851 __ emit_data64(0x0000000000000002, relocInfo::none);
852 __ emit_data64(0x0000000000000003, relocInfo::none);
853 __ emit_data64(0x0000000000000004, relocInfo::none);
854 __ emit_data64(0x0000000000000005, relocInfo::none);
855 __ emit_data64(0x0000000000000006, relocInfo::none);
856 __ emit_data64(0x0000000000000007, relocInfo::none);
857 // D - FP
858 __ emit_data64(0x3F80000000000000, relocInfo::none); // 0.0f, 1.0f
859 __ emit_data64(0x4040000040000000, relocInfo::none); // 2.0f, 3.0f
860 __ emit_data64(0x40A0000040800000, relocInfo::none); // 4.0f, 5.0f
861 __ emit_data64(0x40E0000040C00000, relocInfo::none); // 6.0f, 7.0f
862 __ emit_data64(0x4110000041000000, relocInfo::none); // 8.0f, 9.0f
863 __ emit_data64(0x4130000041200000, relocInfo::none); // 10.0f, 11.0f
864 __ emit_data64(0x4150000041400000, relocInfo::none); // 12.0f, 13.0f
865 __ emit_data64(0x4170000041600000, relocInfo::none); // 14.0f, 15.0f
866 // Q - FP
867 __ emit_data64(0x0000000000000000, relocInfo::none); // 0.0d
868 __ emit_data64(0x3FF0000000000000, relocInfo::none); // 1.0d
869 __ emit_data64(0x4000000000000000, relocInfo::none); // 2.0d
870 __ emit_data64(0x4008000000000000, relocInfo::none); // 3.0d
871 __ emit_data64(0x4010000000000000, relocInfo::none); // 4.0d
872 __ emit_data64(0x4014000000000000, relocInfo::none); // 5.0d
873 __ emit_data64(0x4018000000000000, relocInfo::none); // 6.0d
874 __ emit_data64(0x401c000000000000, relocInfo::none); // 7.0d
875 return start;
876 }
877
878 address StubGenerator::generate_vector_reverse_bit_lut(const char *stub_name) {
879 __ align(CodeEntryAlignment);
880 StubCodeMark mark(this, "StubRoutines", stub_name);
881 address start = __ pc();
882
883 __ emit_data64(0x0E060A020C040800, relocInfo::none);
884 __ emit_data64(0x0F070B030D050901, relocInfo::none);
885 __ emit_data64(0x0E060A020C040800, relocInfo::none);
886 __ emit_data64(0x0F070B030D050901, relocInfo::none);
887 __ emit_data64(0x0E060A020C040800, relocInfo::none);
888 __ emit_data64(0x0F070B030D050901, relocInfo::none);
889 __ emit_data64(0x0E060A020C040800, relocInfo::none);
890 __ emit_data64(0x0F070B030D050901, relocInfo::none);
891
892 return start;
893 }
894
895 address StubGenerator::generate_vector_reverse_byte_perm_mask_long(const char *stub_name) {
896 __ align(CodeEntryAlignment);
897 StubCodeMark mark(this, "StubRoutines", stub_name);
898 address start = __ pc();
899
900 __ emit_data64(0x0001020304050607, relocInfo::none);
901 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
902 __ emit_data64(0x0001020304050607, relocInfo::none);
903 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
904 __ emit_data64(0x0001020304050607, relocInfo::none);
905 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
906 __ emit_data64(0x0001020304050607, relocInfo::none);
907 __ emit_data64(0x08090A0B0C0D0E0F, relocInfo::none);
908
909 return start;
910 }
911
912 address StubGenerator::generate_vector_reverse_byte_perm_mask_int(const char *stub_name) {
913 __ align(CodeEntryAlignment);
914 StubCodeMark mark(this, "StubRoutines", stub_name);
915 address start = __ pc();
916
917 __ emit_data64(0x0405060700010203, relocInfo::none);
918 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
919 __ emit_data64(0x0405060700010203, relocInfo::none);
920 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
921 __ emit_data64(0x0405060700010203, relocInfo::none);
922 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
923 __ emit_data64(0x0405060700010203, relocInfo::none);
924 __ emit_data64(0x0C0D0E0F08090A0B, relocInfo::none);
925
926 return start;
927 }
928
929 address StubGenerator::generate_vector_reverse_byte_perm_mask_short(const char *stub_name) {
930 __ align(CodeEntryAlignment);
931 StubCodeMark mark(this, "StubRoutines", stub_name);
932 address start = __ pc();
933
934 __ emit_data64(0x0607040502030001, relocInfo::none);
935 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
936 __ emit_data64(0x0607040502030001, relocInfo::none);
937 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
938 __ emit_data64(0x0607040502030001, relocInfo::none);
939 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
940 __ emit_data64(0x0607040502030001, relocInfo::none);
941 __ emit_data64(0x0E0F0C0D0A0B0809, relocInfo::none);
942
943 return start;
944 }
945
946 address StubGenerator::generate_vector_byte_shuffle_mask(const char *stub_name) {
947 __ align(CodeEntryAlignment);
948 StubCodeMark mark(this, "StubRoutines", stub_name);
949 address start = __ pc();
950
951 __ emit_data64(0x7070707070707070, relocInfo::none);
952 __ emit_data64(0x7070707070707070, relocInfo::none);
953 __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
954 __ emit_data64(0xF0F0F0F0F0F0F0F0, relocInfo::none);
955
956 return start;
957 }
958
959 address StubGenerator::generate_fp_mask(const char *stub_name, int64_t mask) {
960 __ align(CodeEntryAlignment);
961 StubCodeMark mark(this, "StubRoutines", stub_name);
962 address start = __ pc();
963
964 __ emit_data64( mask, relocInfo::none );
965 __ emit_data64( mask, relocInfo::none );
966
967 return start;
968 }
969
970 address StubGenerator::generate_compress_perm_table(const char *stub_name, int32_t esize) {
971 __ align(CodeEntryAlignment);
972 StubCodeMark mark(this, "StubRoutines", stub_name);
973 address start = __ pc();
974 if (esize == 32) {
975 // Loop to generate 256 x 8 int compression permute index table. A row is
976 // accessed using 8 bit index computed using vector mask. An entry in
977 // a row holds either a valid permute index corresponding to set bit position
978 // or a -1 (default) value.
979 for (int mask = 0; mask < 256; mask++) {
980 int ctr = 0;
981 for (int j = 0; j < 8; j++) {
982 if (mask & (1 << j)) {
983 __ emit_data(j, relocInfo::none);
984 ctr++;
985 }
986 }
987 for (; ctr < 8; ctr++) {
988 __ emit_data(-1, relocInfo::none);
989 }
990 }
991 } else {
992 assert(esize == 64, "");
993 // Loop to generate 16 x 4 long compression permute index table. A row is
994 // accessed using 4 bit index computed using vector mask. An entry in
995 // a row holds either a valid permute index pair for a quadword corresponding
996 // to set bit position or a -1 (default) value.
997 for (int mask = 0; mask < 16; mask++) {
998 int ctr = 0;
999 for (int j = 0; j < 4; j++) {
1000 if (mask & (1 << j)) {
1001 __ emit_data(2 * j, relocInfo::none);
1002 __ emit_data(2 * j + 1, relocInfo::none);
1003 ctr++;
1004 }
1005 }
1006 for (; ctr < 4; ctr++) {
1007 __ emit_data64(-1L, relocInfo::none);
1008 }
1009 }
1010 }
1011 return start;
1012 }
1013
1014 address StubGenerator::generate_expand_perm_table(const char *stub_name, int32_t esize) {
1015 __ align(CodeEntryAlignment);
1016 StubCodeMark mark(this, "StubRoutines", stub_name);
1017 address start = __ pc();
1018 if (esize == 32) {
1019 // Loop to generate 256 x 8 int expand permute index table. A row is accessed
1020 // using 8 bit index computed using vector mask. An entry in a row holds either
1021 // a valid permute index (starting from least significant lane) placed at poisition
1022 // corresponding to set bit position or a -1 (default) value.
1023 for (int mask = 0; mask < 256; mask++) {
1024 int ctr = 0;
1025 for (int j = 0; j < 8; j++) {
1026 if (mask & (1 << j)) {
1027 __ emit_data(ctr++, relocInfo::none);
1028 } else {
1029 __ emit_data(-1, relocInfo::none);
1030 }
1031 }
1032 }
1033 } else {
1034 assert(esize == 64, "");
1035 // Loop to generate 16 x 4 long expand permute index table. A row is accessed
1036 // using 4 bit index computed using vector mask. An entry in a row holds either
1037 // a valid doubleword permute index pair representing a quadword index (starting
1038 // from least significant lane) placed at poisition corresponding to set bit
1039 // position or a -1 (default) value.
1040 for (int mask = 0; mask < 16; mask++) {
1041 int ctr = 0;
1042 for (int j = 0; j < 4; j++) {
1043 if (mask & (1 << j)) {
1044 __ emit_data(2 * ctr, relocInfo::none);
1045 __ emit_data(2 * ctr + 1, relocInfo::none);
1046 ctr++;
1047 } else {
1048 __ emit_data64(-1L, relocInfo::none);
1049 }
1050 }
1051 }
1052 }
1053 return start;
1054 }
1055
1056 address StubGenerator::generate_vector_mask(const char *stub_name, int64_t mask) {
1057 __ align(CodeEntryAlignment);
1058 StubCodeMark mark(this, "StubRoutines", stub_name);
1059 address start = __ pc();
1060
1061 __ emit_data64(mask, relocInfo::none);
1062 __ emit_data64(mask, relocInfo::none);
1063 __ emit_data64(mask, relocInfo::none);
1064 __ emit_data64(mask, relocInfo::none);
1065 __ emit_data64(mask, relocInfo::none);
1066 __ emit_data64(mask, relocInfo::none);
1067 __ emit_data64(mask, relocInfo::none);
1068 __ emit_data64(mask, relocInfo::none);
1069
1070 return start;
1071 }
1072
1073 address StubGenerator::generate_vector_byte_perm_mask(const char *stub_name) {
1074 __ align(CodeEntryAlignment);
1075 StubCodeMark mark(this, "StubRoutines", stub_name);
1076 address start = __ pc();
1077
1078 __ emit_data64(0x0000000000000001, relocInfo::none);
1079 __ emit_data64(0x0000000000000003, relocInfo::none);
1080 __ emit_data64(0x0000000000000005, relocInfo::none);
1081 __ emit_data64(0x0000000000000007, relocInfo::none);
1082 __ emit_data64(0x0000000000000000, relocInfo::none);
1083 __ emit_data64(0x0000000000000002, relocInfo::none);
1084 __ emit_data64(0x0000000000000004, relocInfo::none);
1085 __ emit_data64(0x0000000000000006, relocInfo::none);
1086
1087 return start;
1088 }
1089
1090 address StubGenerator::generate_vector_fp_mask(const char *stub_name, int64_t mask) {
1091 __ align(CodeEntryAlignment);
1092 StubCodeMark mark(this, "StubRoutines", stub_name);
1093 address start = __ pc();
1094
1095 __ emit_data64(mask, relocInfo::none);
1096 __ emit_data64(mask, relocInfo::none);
1097 __ emit_data64(mask, relocInfo::none);
1098 __ emit_data64(mask, relocInfo::none);
1099 __ emit_data64(mask, relocInfo::none);
1100 __ emit_data64(mask, relocInfo::none);
1101 __ emit_data64(mask, relocInfo::none);
1102 __ emit_data64(mask, relocInfo::none);
1103
1104 return start;
1105 }
1106
1107 address StubGenerator::generate_vector_custom_i32(const char *stub_name, Assembler::AvxVectorLen len,
1108 int32_t val0, int32_t val1, int32_t val2, int32_t val3,
1109 int32_t val4, int32_t val5, int32_t val6, int32_t val7,
1110 int32_t val8, int32_t val9, int32_t val10, int32_t val11,
1111 int32_t val12, int32_t val13, int32_t val14, int32_t val15) {
1112 __ align(CodeEntryAlignment);
1113 StubCodeMark mark(this, "StubRoutines", stub_name);
1114 address start = __ pc();
1115
1116 assert(len != Assembler::AVX_NoVec, "vector len must be specified");
1117 __ emit_data(val0, relocInfo::none, 0);
1118 __ emit_data(val1, relocInfo::none, 0);
1119 __ emit_data(val2, relocInfo::none, 0);
1120 __ emit_data(val3, relocInfo::none, 0);
1121 if (len >= Assembler::AVX_256bit) {
1122 __ emit_data(val4, relocInfo::none, 0);
1123 __ emit_data(val5, relocInfo::none, 0);
1124 __ emit_data(val6, relocInfo::none, 0);
1125 __ emit_data(val7, relocInfo::none, 0);
1126 if (len >= Assembler::AVX_512bit) {
1127 __ emit_data(val8, relocInfo::none, 0);
1128 __ emit_data(val9, relocInfo::none, 0);
1129 __ emit_data(val10, relocInfo::none, 0);
1130 __ emit_data(val11, relocInfo::none, 0);
1131 __ emit_data(val12, relocInfo::none, 0);
1132 __ emit_data(val13, relocInfo::none, 0);
1133 __ emit_data(val14, relocInfo::none, 0);
1134 __ emit_data(val15, relocInfo::none, 0);
1135 }
1136 }
1137 return start;
1138 }
1139
1140 // Non-destructive plausibility checks for oops
1141 //
1142 // Arguments:
1143 // all args on stack!
1144 //
1145 // Stack after saving c_rarg3:
1146 // [tos + 0]: saved c_rarg3
1147 // [tos + 1]: saved c_rarg2
1148 // [tos + 2]: saved r12 (several TemplateTable methods use it)
1149 // [tos + 3]: saved flags
1150 // [tos + 4]: return address
1151 // * [tos + 5]: error message (char*)
1152 // * [tos + 6]: object to verify (oop)
1153 // * [tos + 7]: saved rax - saved by caller and bashed
1154 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
1155 // * = popped on exit
1156 address StubGenerator::generate_verify_oop() {
1157 StubCodeMark mark(this, "StubRoutines", "verify_oop");
1158 address start = __ pc();
1159
1160 Label exit, error;
1161
1162 __ pushf();
1163 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()), rscratch1);
1164
1165 __ push(r12);
1166
1167 // save c_rarg2 and c_rarg3
1168 __ push(c_rarg2);
1169 __ push(c_rarg3);
1170
1171 enum {
1172 // After previous pushes.
1173 oop_to_verify = 6 * wordSize,
1174 saved_rax = 7 * wordSize,
1175 saved_r10 = 8 * wordSize,
1176
1177 // Before the call to MacroAssembler::debug(), see below.
1178 return_addr = 16 * wordSize,
1179 error_msg = 17 * wordSize
1180 };
1181
1182 // get object
1183 __ movptr(rax, Address(rsp, oop_to_verify));
1184
1185 // make sure object is 'reasonable'
1186 __ testptr(rax, rax);
1187 __ jcc(Assembler::zero, exit); // if obj is null it is OK
1188
1189 BarrierSetAssembler* bs_asm = BarrierSet::barrier_set()->barrier_set_assembler();
1190 bs_asm->check_oop(_masm, rax, c_rarg2, c_rarg3, error);
1191
1192 // return if everything seems ok
1193 __ bind(exit);
1194 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1195 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1196 __ pop(c_rarg3); // restore c_rarg3
1197 __ pop(c_rarg2); // restore c_rarg2
1198 __ pop(r12); // restore r12
1199 __ popf(); // restore flags
1200 __ ret(4 * wordSize); // pop caller saved stuff
1201
1202 // handle errors
1203 __ bind(error);
1204 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1205 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1206 __ pop(c_rarg3); // get saved c_rarg3 back
1207 __ pop(c_rarg2); // get saved c_rarg2 back
1208 __ pop(r12); // get saved r12 back
1209 __ popf(); // get saved flags off stack --
1210 // will be ignored
1211
1212 __ pusha(); // push registers
1213 // (rip is already
1214 // already pushed)
1215 // debug(char* msg, int64_t pc, int64_t regs[])
1216 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1217 // pushed all the registers, so now the stack looks like:
1218 // [tos + 0] 16 saved registers
1219 // [tos + 16] return address
1220 // * [tos + 17] error message (char*)
1221 // * [tos + 18] object to verify (oop)
1222 // * [tos + 19] saved rax - saved by caller and bashed
1223 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1224 // * = popped on exit
1225
1226 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1227 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1228 __ movq(c_rarg2, rsp); // pass address of regs on stack
1229 __ mov(r12, rsp); // remember rsp
1230 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1231 __ andptr(rsp, -16); // align stack as required by ABI
1232 BLOCK_COMMENT("call MacroAssembler::debug");
1233 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1234 __ hlt();
1235
1236 return start;
1237 }
1238
1239
1240 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1241 //
1242 // Outputs:
1243 // rdi - rcx
1244 // rsi - rdx
1245 // rdx - r8
1246 // rcx - r9
1247 //
1248 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1249 // are non-volatile. r9 and r10 should not be used by the caller.
1250 //
1251 void StubGenerator::setup_arg_regs(int nargs) {
1252 const Register saved_rdi = r9;
1253 const Register saved_rsi = r10;
1254 assert(nargs == 3 || nargs == 4, "else fix");
1255 #ifdef _WIN64
1256 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1257 "unexpected argument registers");
1258 if (nargs == 4) {
1259 __ mov(rax, r9); // r9 is also saved_rdi
1260 }
1261 __ movptr(saved_rdi, rdi);
1262 __ movptr(saved_rsi, rsi);
1263 __ mov(rdi, rcx); // c_rarg0
1264 __ mov(rsi, rdx); // c_rarg1
1265 __ mov(rdx, r8); // c_rarg2
1266 if (nargs == 4) {
1267 __ mov(rcx, rax); // c_rarg3 (via rax)
1268 }
1269 #else
1270 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1271 "unexpected argument registers");
1272 #endif
1273 DEBUG_ONLY(_regs_in_thread = false;)
1274 }
1275
1276
1277 void StubGenerator::restore_arg_regs() {
1278 assert(!_regs_in_thread, "wrong call to restore_arg_regs");
1279 const Register saved_rdi = r9;
1280 const Register saved_rsi = r10;
1281 #ifdef _WIN64
1282 __ movptr(rdi, saved_rdi);
1283 __ movptr(rsi, saved_rsi);
1284 #endif
1285 }
1286
1287
1288 // This is used in places where r10 is a scratch register, and can
1289 // be adapted if r9 is needed also.
1290 void StubGenerator::setup_arg_regs_using_thread(int nargs) {
1291 const Register saved_r15 = r9;
1292 assert(nargs == 3 || nargs == 4, "else fix");
1293 #ifdef _WIN64
1294 if (nargs == 4) {
1295 __ mov(rax, r9); // r9 is also saved_r15
1296 }
1297 __ mov(saved_r15, r15); // r15 is callee saved and needs to be restored
1298 __ get_thread(r15_thread);
1299 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1300 "unexpected argument registers");
1301 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())), rdi);
1302 __ movptr(Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())), rsi);
1303
1304 __ mov(rdi, rcx); // c_rarg0
1305 __ mov(rsi, rdx); // c_rarg1
1306 __ mov(rdx, r8); // c_rarg2
1307 if (nargs == 4) {
1308 __ mov(rcx, rax); // c_rarg3 (via rax)
1309 }
1310 #else
1311 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1312 "unexpected argument registers");
1313 #endif
1314 DEBUG_ONLY(_regs_in_thread = true;)
1315 }
1316
1317
1318 void StubGenerator::restore_arg_regs_using_thread() {
1319 assert(_regs_in_thread, "wrong call to restore_arg_regs");
1320 const Register saved_r15 = r9;
1321 #ifdef _WIN64
1322 __ get_thread(r15_thread);
1323 __ movptr(rsi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rsi_offset())));
1324 __ movptr(rdi, Address(r15_thread, in_bytes(JavaThread::windows_saved_rdi_offset())));
1325 __ mov(r15, saved_r15); // r15 is callee saved and needs to be restored
1326 #endif
1327 }
1328
1329
1330 void StubGenerator::setup_argument_regs(BasicType type) {
1331 if (type == T_BYTE || type == T_SHORT) {
1332 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1333 // r9 and r10 may be used to save non-volatile registers
1334 } else {
1335 setup_arg_regs_using_thread(); // from => rdi, to => rsi, count => rdx
1336 // r9 is used to save r15_thread
1337 }
1338 }
1339
1340
1341 void StubGenerator::restore_argument_regs(BasicType type) {
1342 if (type == T_BYTE || type == T_SHORT) {
1343 restore_arg_regs();
1344 } else {
1345 restore_arg_regs_using_thread();
1346 }
1347 }
1348
1349 address StubGenerator::generate_data_cache_writeback() {
1350 const Register src = c_rarg0; // source address
1351
1352 __ align(CodeEntryAlignment);
1353
1354 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback");
1355
1356 address start = __ pc();
1357
1358 __ enter();
1359 __ cache_wb(Address(src, 0));
1360 __ leave();
1361 __ ret(0);
1362
1363 return start;
1364 }
1365
1366 address StubGenerator::generate_data_cache_writeback_sync() {
1367 const Register is_pre = c_rarg0; // pre or post sync
1368
1369 __ align(CodeEntryAlignment);
1370
1371 StubCodeMark mark(this, "StubRoutines", "_data_cache_writeback_sync");
1372
1373 // pre wbsync is a no-op
1374 // post wbsync translates to an sfence
1375
1376 Label skip;
1377 address start = __ pc();
1378
1379 __ enter();
1380 __ cmpl(is_pre, 0);
1381 __ jcc(Assembler::notEqual, skip);
1382 __ cache_wbsync(false);
1383 __ bind(skip);
1384 __ leave();
1385 __ ret(0);
1386
1387 return start;
1388 }
1389
1390 // ofs and limit are use for multi-block byte array.
1391 // int com.sun.security.provider.MD5.implCompress(byte[] b, int ofs)
1392 address StubGenerator::generate_md5_implCompress(bool multi_block, const char *name) {
1393 __ align(CodeEntryAlignment);
1394 StubCodeMark mark(this, "StubRoutines", name);
1395 address start = __ pc();
1396
1397 const Register buf_param = r15;
1398 const Address state_param(rsp, 0 * wordSize);
1399 const Address ofs_param (rsp, 1 * wordSize );
1400 const Address limit_param(rsp, 1 * wordSize + 4);
1401
1402 __ enter();
1403 __ push(rbx);
1404 __ push(rdi);
1405 __ push(rsi);
1406 __ push(r15);
1407 __ subptr(rsp, 2 * wordSize);
1408
1409 __ movptr(buf_param, c_rarg0);
1410 __ movptr(state_param, c_rarg1);
1411 if (multi_block) {
1412 __ movl(ofs_param, c_rarg2);
1413 __ movl(limit_param, c_rarg3);
1414 }
1415 __ fast_md5(buf_param, state_param, ofs_param, limit_param, multi_block);
1416
1417 __ addptr(rsp, 2 * wordSize);
1418 __ pop(r15);
1419 __ pop(rsi);
1420 __ pop(rdi);
1421 __ pop(rbx);
1422 __ leave();
1423 __ ret(0);
1424
1425 return start;
1426 }
1427
1428 address StubGenerator::generate_upper_word_mask() {
1429 __ align64();
1430 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
1431 address start = __ pc();
1432
1433 __ emit_data64(0x0000000000000000, relocInfo::none);
1434 __ emit_data64(0xFFFFFFFF00000000, relocInfo::none);
1435
1436 return start;
1437 }
1438
1439 address StubGenerator::generate_shuffle_byte_flip_mask() {
1440 __ align64();
1441 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
1442 address start = __ pc();
1443
1444 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1445 __ emit_data64(0x0001020304050607, relocInfo::none);
1446
1447 return start;
1448 }
1449
1450 // ofs and limit are use for multi-block byte array.
1451 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1452 address StubGenerator::generate_sha1_implCompress(bool multi_block, const char *name) {
1453 __ align(CodeEntryAlignment);
1454 StubCodeMark mark(this, "StubRoutines", name);
1455 address start = __ pc();
1456
1457 Register buf = c_rarg0;
1458 Register state = c_rarg1;
1459 Register ofs = c_rarg2;
1460 Register limit = c_rarg3;
1461
1462 const XMMRegister abcd = xmm0;
1463 const XMMRegister e0 = xmm1;
1464 const XMMRegister e1 = xmm2;
1465 const XMMRegister msg0 = xmm3;
1466
1467 const XMMRegister msg1 = xmm4;
1468 const XMMRegister msg2 = xmm5;
1469 const XMMRegister msg3 = xmm6;
1470 const XMMRegister shuf_mask = xmm7;
1471
1472 __ enter();
1473
1474 __ subptr(rsp, 4 * wordSize);
1475
1476 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
1477 buf, state, ofs, limit, rsp, multi_block);
1478
1479 __ addptr(rsp, 4 * wordSize);
1480
1481 __ leave();
1482 __ ret(0);
1483
1484 return start;
1485 }
1486
1487 address StubGenerator::generate_pshuffle_byte_flip_mask() {
1488 __ align64();
1489 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
1490 address start = __ pc();
1491
1492 __ emit_data64(0x0405060700010203, relocInfo::none);
1493 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1494
1495 if (VM_Version::supports_avx2()) {
1496 __ emit_data64(0x0405060700010203, relocInfo::none); // second copy
1497 __ emit_data64(0x0c0d0e0f08090a0b, relocInfo::none);
1498 // _SHUF_00BA
1499 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1500 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1501 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1502 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1503 // _SHUF_DC00
1504 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1505 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1506 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1507 __ emit_data64(0x0b0a090803020100, relocInfo::none);
1508 }
1509
1510 return start;
1511 }
1512
1513 //Mask for byte-swapping a couple of qwords in an XMM register using (v)pshufb.
1514 address StubGenerator::generate_pshuffle_byte_flip_mask_sha512() {
1515 __ align32();
1516 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask_sha512");
1517 address start = __ pc();
1518
1519 if (VM_Version::supports_avx2()) {
1520 __ emit_data64(0x0001020304050607, relocInfo::none); // PSHUFFLE_BYTE_FLIP_MASK
1521 __ emit_data64(0x08090a0b0c0d0e0f, relocInfo::none);
1522 __ emit_data64(0x1011121314151617, relocInfo::none);
1523 __ emit_data64(0x18191a1b1c1d1e1f, relocInfo::none);
1524 __ emit_data64(0x0000000000000000, relocInfo::none); //MASK_YMM_LO
1525 __ emit_data64(0x0000000000000000, relocInfo::none);
1526 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1527 __ emit_data64(0xFFFFFFFFFFFFFFFF, relocInfo::none);
1528 }
1529
1530 return start;
1531 }
1532
1533 // ofs and limit are use for multi-block byte array.
1534 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
1535 address StubGenerator::generate_sha256_implCompress(bool multi_block, const char *name) {
1536 assert(VM_Version::supports_sha() || VM_Version::supports_avx2(), "");
1537 __ align(CodeEntryAlignment);
1538 StubCodeMark mark(this, "StubRoutines", name);
1539 address start = __ pc();
1540
1541 Register buf = c_rarg0;
1542 Register state = c_rarg1;
1543 Register ofs = c_rarg2;
1544 Register limit = c_rarg3;
1545
1546 const XMMRegister msg = xmm0;
1547 const XMMRegister state0 = xmm1;
1548 const XMMRegister state1 = xmm2;
1549 const XMMRegister msgtmp0 = xmm3;
1550
1551 const XMMRegister msgtmp1 = xmm4;
1552 const XMMRegister msgtmp2 = xmm5;
1553 const XMMRegister msgtmp3 = xmm6;
1554 const XMMRegister msgtmp4 = xmm7;
1555
1556 const XMMRegister shuf_mask = xmm8;
1557
1558 __ enter();
1559
1560 __ subptr(rsp, 4 * wordSize);
1561
1562 if (VM_Version::supports_sha()) {
1563 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1564 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1565 } else if (VM_Version::supports_avx2()) {
1566 __ sha256_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1567 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1568 }
1569 __ addptr(rsp, 4 * wordSize);
1570 __ vzeroupper();
1571 __ leave();
1572 __ ret(0);
1573
1574 return start;
1575 }
1576
1577 address StubGenerator::generate_sha512_implCompress(bool multi_block, const char *name) {
1578 assert(VM_Version::supports_avx2(), "");
1579 assert(VM_Version::supports_bmi2(), "");
1580 __ align(CodeEntryAlignment);
1581 StubCodeMark mark(this, "StubRoutines", name);
1582 address start = __ pc();
1583
1584 Register buf = c_rarg0;
1585 Register state = c_rarg1;
1586 Register ofs = c_rarg2;
1587 Register limit = c_rarg3;
1588
1589 const XMMRegister msg = xmm0;
1590 const XMMRegister state0 = xmm1;
1591 const XMMRegister state1 = xmm2;
1592 const XMMRegister msgtmp0 = xmm3;
1593 const XMMRegister msgtmp1 = xmm4;
1594 const XMMRegister msgtmp2 = xmm5;
1595 const XMMRegister msgtmp3 = xmm6;
1596 const XMMRegister msgtmp4 = xmm7;
1597
1598 const XMMRegister shuf_mask = xmm8;
1599
1600 __ enter();
1601
1602 __ sha512_AVX2(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
1603 buf, state, ofs, limit, rsp, multi_block, shuf_mask);
1604
1605 __ vzeroupper();
1606 __ leave();
1607 __ ret(0);
1608
1609 return start;
1610 }
1611
1612 address StubGenerator::base64_shuffle_addr() {
1613 __ align64();
1614 StubCodeMark mark(this, "StubRoutines", "shuffle_base64");
1615 address start = __ pc();
1616
1617 assert(((unsigned long long)start & 0x3f) == 0,
1618 "Alignment problem (0x%08llx)", (unsigned long long)start);
1619 __ emit_data64(0x0405030401020001, relocInfo::none);
1620 __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1621 __ emit_data64(0x10110f100d0e0c0d, relocInfo::none);
1622 __ emit_data64(0x1617151613141213, relocInfo::none);
1623 __ emit_data64(0x1c1d1b1c191a1819, relocInfo::none);
1624 __ emit_data64(0x222321221f201e1f, relocInfo::none);
1625 __ emit_data64(0x2829272825262425, relocInfo::none);
1626 __ emit_data64(0x2e2f2d2e2b2c2a2b, relocInfo::none);
1627
1628 return start;
1629 }
1630
1631 address StubGenerator::base64_avx2_shuffle_addr() {
1632 __ align32();
1633 StubCodeMark mark(this, "StubRoutines", "avx2_shuffle_base64");
1634 address start = __ pc();
1635
1636 __ emit_data64(0x0809070805060405, relocInfo::none);
1637 __ emit_data64(0x0e0f0d0e0b0c0a0b, relocInfo::none);
1638 __ emit_data64(0x0405030401020001, relocInfo::none);
1639 __ emit_data64(0x0a0b090a07080607, relocInfo::none);
1640
1641 return start;
1642 }
1643
1644 address StubGenerator::base64_avx2_input_mask_addr() {
1645 __ align32();
1646 StubCodeMark mark(this, "StubRoutines", "avx2_input_mask_base64");
1647 address start = __ pc();
1648
1649 __ emit_data64(0x8000000000000000, relocInfo::none);
1650 __ emit_data64(0x8000000080000000, relocInfo::none);
1651 __ emit_data64(0x8000000080000000, relocInfo::none);
1652 __ emit_data64(0x8000000080000000, relocInfo::none);
1653
1654 return start;
1655 }
1656
1657 address StubGenerator::base64_avx2_lut_addr() {
1658 __ align32();
1659 StubCodeMark mark(this, "StubRoutines", "avx2_lut_base64");
1660 address start = __ pc();
1661
1662 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1663 __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1664 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1665 __ emit_data64(0x0000f0edfcfcfcfc, relocInfo::none);
1666
1667 // URL LUT
1668 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1669 __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1670 __ emit_data64(0xfcfcfcfcfcfc4741, relocInfo::none);
1671 __ emit_data64(0x000020effcfcfcfc, relocInfo::none);
1672
1673 return start;
1674 }
1675
1676 address StubGenerator::base64_encoding_table_addr() {
1677 __ align64();
1678 StubCodeMark mark(this, "StubRoutines", "encoding_table_base64");
1679 address start = __ pc();
1680
1681 assert(((unsigned long long)start & 0x3f) == 0, "Alignment problem (0x%08llx)", (unsigned long long)start);
1682 __ emit_data64(0x4847464544434241, relocInfo::none);
1683 __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1684 __ emit_data64(0x5857565554535251, relocInfo::none);
1685 __ emit_data64(0x6665646362615a59, relocInfo::none);
1686 __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1687 __ emit_data64(0x767574737271706f, relocInfo::none);
1688 __ emit_data64(0x333231307a797877, relocInfo::none);
1689 __ emit_data64(0x2f2b393837363534, relocInfo::none);
1690
1691 // URL table
1692 __ emit_data64(0x4847464544434241, relocInfo::none);
1693 __ emit_data64(0x504f4e4d4c4b4a49, relocInfo::none);
1694 __ emit_data64(0x5857565554535251, relocInfo::none);
1695 __ emit_data64(0x6665646362615a59, relocInfo::none);
1696 __ emit_data64(0x6e6d6c6b6a696867, relocInfo::none);
1697 __ emit_data64(0x767574737271706f, relocInfo::none);
1698 __ emit_data64(0x333231307a797877, relocInfo::none);
1699 __ emit_data64(0x5f2d393837363534, relocInfo::none);
1700
1701 return start;
1702 }
1703
1704 // Code for generating Base64 encoding.
1705 // Intrinsic function prototype in Base64.java:
1706 // private void encodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp,
1707 // boolean isURL) {
1708 address StubGenerator::generate_base64_encodeBlock()
1709 {
1710 __ align(CodeEntryAlignment);
1711 StubCodeMark mark(this, "StubRoutines", "implEncode");
1712 address start = __ pc();
1713
1714 __ enter();
1715
1716 // Save callee-saved registers before using them
1717 __ push(r12);
1718 __ push(r13);
1719 __ push(r14);
1720 __ push(r15);
1721
1722 // arguments
1723 const Register source = c_rarg0; // Source Array
1724 const Register start_offset = c_rarg1; // start offset
1725 const Register end_offset = c_rarg2; // end offset
1726 const Register dest = c_rarg3; // destination array
1727
1728 #ifndef _WIN64
1729 const Register dp = c_rarg4; // Position for writing to dest array
1730 const Register isURL = c_rarg5; // Base64 or URL character set
1731 #else
1732 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
1733 const Address isURL_mem(rbp, 7 * wordSize);
1734 const Register isURL = r10; // pick the volatile windows register
1735 const Register dp = r12;
1736 __ movl(dp, dp_mem);
1737 __ movl(isURL, isURL_mem);
1738 #endif
1739
1740 const Register length = r14;
1741 const Register encode_table = r13;
1742 Label L_process3, L_exit, L_processdata, L_vbmiLoop, L_not512, L_32byteLoop;
1743
1744 // calculate length from offsets
1745 __ movl(length, end_offset);
1746 __ subl(length, start_offset);
1747 __ jcc(Assembler::lessEqual, L_exit);
1748
1749 // Code for 512-bit VBMI encoding. Encodes 48 input bytes into 64
1750 // output bytes. We read 64 input bytes and ignore the last 16, so be
1751 // sure not to read past the end of the input buffer.
1752 if (VM_Version::supports_avx512_vbmi()) {
1753 __ cmpl(length, 64); // Do not overrun input buffer.
1754 __ jcc(Assembler::below, L_not512);
1755
1756 __ shll(isURL, 6); // index into decode table based on isURL
1757 __ lea(encode_table, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
1758 __ addptr(encode_table, isURL);
1759 __ shrl(isURL, 6); // restore isURL
1760
1761 __ mov64(rax, 0x3036242a1016040aull); // Shifts
1762 __ evmovdquq(xmm3, ExternalAddress(StubRoutines::x86::base64_shuffle_addr()), Assembler::AVX_512bit, r15);
1763 __ evmovdquq(xmm2, Address(encode_table, 0), Assembler::AVX_512bit);
1764 __ evpbroadcastq(xmm1, rax, Assembler::AVX_512bit);
1765
1766 __ align32();
1767 __ BIND(L_vbmiLoop);
1768
1769 __ vpermb(xmm0, xmm3, Address(source, start_offset), Assembler::AVX_512bit);
1770 __ subl(length, 48);
1771
1772 // Put the input bytes into the proper lanes for writing, then
1773 // encode them.
1774 __ evpmultishiftqb(xmm0, xmm1, xmm0, Assembler::AVX_512bit);
1775 __ vpermb(xmm0, xmm0, xmm2, Assembler::AVX_512bit);
1776
1777 // Write to destination
1778 __ evmovdquq(Address(dest, dp), xmm0, Assembler::AVX_512bit);
1779
1780 __ addptr(dest, 64);
1781 __ addptr(source, 48);
1782 __ cmpl(length, 64);
1783 __ jcc(Assembler::aboveEqual, L_vbmiLoop);
1784
1785 __ vzeroupper();
1786 }
1787
1788 __ BIND(L_not512);
1789 if (VM_Version::supports_avx2()) {
1790 /*
1791 ** This AVX2 encoder is based off the paper at:
1792 ** https://dl.acm.org/doi/10.1145/3132709
1793 **
1794 ** We use AVX2 SIMD instructions to encode 24 bytes into 32
1795 ** output bytes.
1796 **
1797 */
1798 // Lengths under 32 bytes are done with scalar routine
1799 __ cmpl(length, 31);
1800 __ jcc(Assembler::belowEqual, L_process3);
1801
1802 // Set up supporting constant table data
1803 __ vmovdqu(xmm9, ExternalAddress(StubRoutines::x86::base64_avx2_shuffle_addr()), rax);
1804 // 6-bit mask for 2nd and 4th (and multiples) 6-bit values
1805 __ movl(rax, 0x0fc0fc00);
1806 __ movdl(xmm8, rax);
1807 __ vmovdqu(xmm1, ExternalAddress(StubRoutines::x86::base64_avx2_input_mask_addr()), rax);
1808 __ vpbroadcastd(xmm8, xmm8, Assembler::AVX_256bit);
1809
1810 // Multiplication constant for "shifting" right by 6 and 10
1811 // bits
1812 __ movl(rax, 0x04000040);
1813
1814 __ subl(length, 24);
1815 __ movdl(xmm7, rax);
1816 __ vpbroadcastd(xmm7, xmm7, Assembler::AVX_256bit);
1817
1818 // For the first load, we mask off reading of the first 4
1819 // bytes into the register. This is so we can get 4 3-byte
1820 // chunks into each lane of the register, avoiding having to
1821 // handle end conditions. We then shuffle these bytes into a
1822 // specific order so that manipulation is easier.
1823 //
1824 // The initial read loads the XMM register like this:
1825 //
1826 // Lower 128-bit lane:
1827 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1828 // | XX | XX | XX | XX | A0 | A1 | A2 | B0 | B1 | B2 | C0 | C1
1829 // | C2 | D0 | D1 | D2 |
1830 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1831 //
1832 // Upper 128-bit lane:
1833 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1834 // | E0 | E1 | E2 | F0 | F1 | F2 | G0 | G1 | G2 | H0 | H1 | H2
1835 // | XX | XX | XX | XX |
1836 // +----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+----+
1837 //
1838 // Where A0 is the first input byte, B0 is the fourth, etc.
1839 // The alphabetical significance denotes the 3 bytes to be
1840 // consumed and encoded into 4 bytes.
1841 //
1842 // We then shuffle the register so each 32-bit word contains
1843 // the sequence:
1844 // A1 A0 A2 A1, B1, B0, B2, B1, etc.
1845 // Each of these byte sequences are then manipulated into 4
1846 // 6-bit values ready for encoding.
1847 //
1848 // If we focus on one set of 3-byte chunks, changing the
1849 // nomenclature such that A0 => a, A1 => b, and A2 => c, we
1850 // shuffle such that each 24-bit chunk contains:
1851 //
1852 // b7 b6 b5 b4 b3 b2 b1 b0 | a7 a6 a5 a4 a3 a2 a1 a0 | c7 c6
1853 // c5 c4 c3 c2 c1 c0 | b7 b6 b5 b4 b3 b2 b1 b0
1854 // Explain this step.
1855 // b3 b2 b1 b0 c5 c4 c3 c2 | c1 c0 d5 d4 d3 d2 d1 d0 | a5 a4
1856 // a3 a2 a1 a0 b5 b4 | b3 b2 b1 b0 c5 c4 c3 c2
1857 //
1858 // W first and off all but bits 4-9 and 16-21 (c5..c0 and
1859 // a5..a0) and shift them using a vector multiplication
1860 // operation (vpmulhuw) which effectively shifts c right by 6
1861 // bits and a right by 10 bits. We similarly mask bits 10-15
1862 // (d5..d0) and 22-27 (b5..b0) and shift them left by 8 and 4
1863 // bits respectively. This is done using vpmullw. We end up
1864 // with 4 6-bit values, thus splitting the 3 input bytes,
1865 // ready for encoding:
1866 // 0 0 d5..d0 0 0 c5..c0 0 0 b5..b0 0 0 a5..a0
1867 //
1868 // For translation, we recognize that there are 5 distinct
1869 // ranges of legal Base64 characters as below:
1870 //
1871 // +-------------+-------------+------------+
1872 // | 6-bit value | ASCII range | offset |
1873 // +-------------+-------------+------------+
1874 // | 0..25 | A..Z | 65 |
1875 // | 26..51 | a..z | 71 |
1876 // | 52..61 | 0..9 | -4 |
1877 // | 62 | + or - | -19 or -17 |
1878 // | 63 | / or _ | -16 or 32 |
1879 // +-------------+-------------+------------+
1880 //
1881 // We note that vpshufb does a parallel lookup in a
1882 // destination register using the lower 4 bits of bytes from a
1883 // source register. If we use a saturated subtraction and
1884 // subtract 51 from each 6-bit value, bytes from [0,51]
1885 // saturate to 0, and [52,63] map to a range of [1,12]. We
1886 // distinguish the [0,25] and [26,51] ranges by assigning a
1887 // value of 13 for all 6-bit values less than 26. We end up
1888 // with:
1889 //
1890 // +-------------+-------------+------------+
1891 // | 6-bit value | Reduced | offset |
1892 // +-------------+-------------+------------+
1893 // | 0..25 | 13 | 65 |
1894 // | 26..51 | 0 | 71 |
1895 // | 52..61 | 0..9 | -4 |
1896 // | 62 | 11 | -19 or -17 |
1897 // | 63 | 12 | -16 or 32 |
1898 // +-------------+-------------+------------+
1899 //
1900 // We then use a final vpshufb to add the appropriate offset,
1901 // translating the bytes.
1902 //
1903 // Load input bytes - only 28 bytes. Mask the first load to
1904 // not load into the full register.
1905 __ vpmaskmovd(xmm1, xmm1, Address(source, start_offset, Address::times_1, -4), Assembler::AVX_256bit);
1906
1907 // Move 3-byte chunks of input (12 bytes) into 16 bytes,
1908 // ordering by:
1909 // 1, 0, 2, 1; 4, 3, 5, 4; etc. This groups 6-bit chunks
1910 // for easy masking
1911 __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1912
1913 __ addl(start_offset, 24);
1914
1915 // Load masking register for first and third (and multiples)
1916 // 6-bit values.
1917 __ movl(rax, 0x003f03f0);
1918 __ movdl(xmm6, rax);
1919 __ vpbroadcastd(xmm6, xmm6, Assembler::AVX_256bit);
1920 // Multiplication constant for "shifting" left by 4 and 8 bits
1921 __ movl(rax, 0x01000010);
1922 __ movdl(xmm5, rax);
1923 __ vpbroadcastd(xmm5, xmm5, Assembler::AVX_256bit);
1924
1925 // Isolate 6-bit chunks of interest
1926 __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
1927
1928 // Load constants for encoding
1929 __ movl(rax, 0x19191919);
1930 __ movdl(xmm3, rax);
1931 __ vpbroadcastd(xmm3, xmm3, Assembler::AVX_256bit);
1932 __ movl(rax, 0x33333333);
1933 __ movdl(xmm4, rax);
1934 __ vpbroadcastd(xmm4, xmm4, Assembler::AVX_256bit);
1935
1936 // Shift output bytes 0 and 2 into proper lanes
1937 __ vpmulhuw(xmm2, xmm0, xmm7, Assembler::AVX_256bit);
1938
1939 // Mask and shift output bytes 1 and 3 into proper lanes and
1940 // combine
1941 __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
1942 __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
1943 __ vpor(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
1944
1945 // Find out which are 0..25. This indicates which input
1946 // values fall in the range of 'A'-'Z', which require an
1947 // additional offset (see comments above)
1948 __ vpcmpgtb(xmm2, xmm0, xmm3, Assembler::AVX_256bit);
1949 __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
1950 __ vpsubb(xmm1, xmm1, xmm2, Assembler::AVX_256bit);
1951
1952 // Load the proper lookup table
1953 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_avx2_lut_addr()));
1954 __ movl(r15, isURL);
1955 __ shll(r15, 5);
1956 __ vmovdqu(xmm2, Address(r11, r15));
1957
1958 // Shuffle the offsets based on the range calculation done
1959 // above. This allows us to add the correct offset to the
1960 // 6-bit value corresponding to the range documented above.
1961 __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
1962 __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
1963
1964 // Store the encoded bytes
1965 __ vmovdqu(Address(dest, dp), xmm0);
1966 __ addl(dp, 32);
1967
1968 __ cmpl(length, 31);
1969 __ jcc(Assembler::belowEqual, L_process3);
1970
1971 __ align32();
1972 __ BIND(L_32byteLoop);
1973
1974 // Get next 32 bytes
1975 __ vmovdqu(xmm1, Address(source, start_offset, Address::times_1, -4));
1976
1977 __ subl(length, 24);
1978 __ addl(start_offset, 24);
1979
1980 // This logic is identical to the above, with only constant
1981 // register loads removed. Shuffle the input, mask off 6-bit
1982 // chunks, shift them into place, then add the offset to
1983 // encode.
1984 __ vpshufb(xmm1, xmm1, xmm9, Assembler::AVX_256bit);
1985
1986 __ vpand(xmm0, xmm8, xmm1, Assembler::AVX_256bit);
1987 __ vpmulhuw(xmm10, xmm0, xmm7, Assembler::AVX_256bit);
1988 __ vpand(xmm0, xmm6, xmm1, Assembler::AVX_256bit);
1989 __ vpmullw(xmm0, xmm5, xmm0, Assembler::AVX_256bit);
1990 __ vpor(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
1991 __ vpcmpgtb(xmm10, xmm0, xmm3, Assembler::AVX_256bit);
1992 __ vpsubusb(xmm1, xmm0, xmm4, Assembler::AVX_256bit);
1993 __ vpsubb(xmm1, xmm1, xmm10, Assembler::AVX_256bit);
1994 __ vpshufb(xmm1, xmm2, xmm1, Assembler::AVX_256bit);
1995 __ vpaddb(xmm0, xmm1, xmm0, Assembler::AVX_256bit);
1996
1997 // Store the encoded bytes
1998 __ vmovdqu(Address(dest, dp), xmm0);
1999 __ addl(dp, 32);
2000
2001 __ cmpl(length, 31);
2002 __ jcc(Assembler::above, L_32byteLoop);
2003
2004 __ BIND(L_process3);
2005 __ vzeroupper();
2006 } else {
2007 __ BIND(L_process3);
2008 }
2009
2010 __ cmpl(length, 3);
2011 __ jcc(Assembler::below, L_exit);
2012
2013 // Load the encoding table based on isURL
2014 __ lea(r11, ExternalAddress(StubRoutines::x86::base64_encoding_table_addr()));
2015 __ movl(r15, isURL);
2016 __ shll(r15, 6);
2017 __ addptr(r11, r15);
2018
2019 __ BIND(L_processdata);
2020
2021 // Load 3 bytes
2022 __ load_unsigned_byte(r15, Address(source, start_offset));
2023 __ load_unsigned_byte(r10, Address(source, start_offset, Address::times_1, 1));
2024 __ load_unsigned_byte(r13, Address(source, start_offset, Address::times_1, 2));
2025
2026 // Build a 32-bit word with bytes 1, 2, 0, 1
2027 __ movl(rax, r10);
2028 __ shll(r10, 24);
2029 __ orl(rax, r10);
2030
2031 __ subl(length, 3);
2032
2033 __ shll(r15, 8);
2034 __ shll(r13, 16);
2035 __ orl(rax, r15);
2036
2037 __ addl(start_offset, 3);
2038
2039 __ orl(rax, r13);
2040 // At this point, rax contains | byte1 | byte2 | byte0 | byte1
2041 // r13 has byte2 << 16 - need low-order 6 bits to translate.
2042 // This translated byte is the fourth output byte.
2043 __ shrl(r13, 16);
2044 __ andl(r13, 0x3f);
2045
2046 // The high-order 6 bits of r15 (byte0) is translated.
2047 // The translated byte is the first output byte.
2048 __ shrl(r15, 10);
2049
2050 __ load_unsigned_byte(r13, Address(r11, r13));
2051 __ load_unsigned_byte(r15, Address(r11, r15));
2052
2053 __ movb(Address(dest, dp, Address::times_1, 3), r13);
2054
2055 // Extract high-order 4 bits of byte1 and low-order 2 bits of byte0.
2056 // This translated byte is the second output byte.
2057 __ shrl(rax, 4);
2058 __ movl(r10, rax);
2059 __ andl(rax, 0x3f);
2060
2061 __ movb(Address(dest, dp, Address::times_1, 0), r15);
2062
2063 __ load_unsigned_byte(rax, Address(r11, rax));
2064
2065 // Extract low-order 2 bits of byte1 and high-order 4 bits of byte2.
2066 // This translated byte is the third output byte.
2067 __ shrl(r10, 18);
2068 __ andl(r10, 0x3f);
2069
2070 __ load_unsigned_byte(r10, Address(r11, r10));
2071
2072 __ movb(Address(dest, dp, Address::times_1, 1), rax);
2073 __ movb(Address(dest, dp, Address::times_1, 2), r10);
2074
2075 __ addl(dp, 4);
2076 __ cmpl(length, 3);
2077 __ jcc(Assembler::aboveEqual, L_processdata);
2078
2079 __ BIND(L_exit);
2080 __ pop(r15);
2081 __ pop(r14);
2082 __ pop(r13);
2083 __ pop(r12);
2084 __ leave();
2085 __ ret(0);
2086
2087 return start;
2088 }
2089
2090 // base64 AVX512vbmi tables
2091 address StubGenerator::base64_vbmi_lookup_lo_addr() {
2092 __ align64();
2093 StubCodeMark mark(this, "StubRoutines", "lookup_lo_base64");
2094 address start = __ pc();
2095
2096 assert(((unsigned long long)start & 0x3f) == 0,
2097 "Alignment problem (0x%08llx)", (unsigned long long)start);
2098 __ emit_data64(0x8080808080808080, relocInfo::none);
2099 __ emit_data64(0x8080808080808080, relocInfo::none);
2100 __ emit_data64(0x8080808080808080, relocInfo::none);
2101 __ emit_data64(0x8080808080808080, relocInfo::none);
2102 __ emit_data64(0x8080808080808080, relocInfo::none);
2103 __ emit_data64(0x3f8080803e808080, relocInfo::none);
2104 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2105 __ emit_data64(0x8080808080803d3c, relocInfo::none);
2106
2107 return start;
2108 }
2109
2110 address StubGenerator::base64_vbmi_lookup_hi_addr() {
2111 __ align64();
2112 StubCodeMark mark(this, "StubRoutines", "lookup_hi_base64");
2113 address start = __ pc();
2114
2115 assert(((unsigned long long)start & 0x3f) == 0,
2116 "Alignment problem (0x%08llx)", (unsigned long long)start);
2117 __ emit_data64(0x0605040302010080, relocInfo::none);
2118 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2119 __ emit_data64(0x161514131211100f, relocInfo::none);
2120 __ emit_data64(0x8080808080191817, relocInfo::none);
2121 __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2122 __ emit_data64(0x2827262524232221, relocInfo::none);
2123 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2124 __ emit_data64(0x8080808080333231, relocInfo::none);
2125
2126 return start;
2127 }
2128 address StubGenerator::base64_vbmi_lookup_lo_url_addr() {
2129 __ align64();
2130 StubCodeMark mark(this, "StubRoutines", "lookup_lo_base64url");
2131 address start = __ pc();
2132
2133 assert(((unsigned long long)start & 0x3f) == 0,
2134 "Alignment problem (0x%08llx)", (unsigned long long)start);
2135 __ emit_data64(0x8080808080808080, relocInfo::none);
2136 __ emit_data64(0x8080808080808080, relocInfo::none);
2137 __ emit_data64(0x8080808080808080, relocInfo::none);
2138 __ emit_data64(0x8080808080808080, relocInfo::none);
2139 __ emit_data64(0x8080808080808080, relocInfo::none);
2140 __ emit_data64(0x80803e8080808080, relocInfo::none);
2141 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2142 __ emit_data64(0x8080808080803d3c, relocInfo::none);
2143
2144 return start;
2145 }
2146
2147 address StubGenerator::base64_vbmi_lookup_hi_url_addr() {
2148 __ align64();
2149 StubCodeMark mark(this, "StubRoutines", "lookup_hi_base64url");
2150 address start = __ pc();
2151
2152 assert(((unsigned long long)start & 0x3f) == 0,
2153 "Alignment problem (0x%08llx)", (unsigned long long)start);
2154 __ emit_data64(0x0605040302010080, relocInfo::none);
2155 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2156 __ emit_data64(0x161514131211100f, relocInfo::none);
2157 __ emit_data64(0x3f80808080191817, relocInfo::none);
2158 __ emit_data64(0x201f1e1d1c1b1a80, relocInfo::none);
2159 __ emit_data64(0x2827262524232221, relocInfo::none);
2160 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2161 __ emit_data64(0x8080808080333231, relocInfo::none);
2162
2163 return start;
2164 }
2165
2166 address StubGenerator::base64_vbmi_pack_vec_addr() {
2167 __ align64();
2168 StubCodeMark mark(this, "StubRoutines", "pack_vec_base64");
2169 address start = __ pc();
2170
2171 assert(((unsigned long long)start & 0x3f) == 0,
2172 "Alignment problem (0x%08llx)", (unsigned long long)start);
2173 __ emit_data64(0x090a040506000102, relocInfo::none);
2174 __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2175 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2176 __ emit_data64(0x292a242526202122, relocInfo::none);
2177 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2178 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2179 __ emit_data64(0x0000000000000000, relocInfo::none);
2180 __ emit_data64(0x0000000000000000, relocInfo::none);
2181
2182 return start;
2183 }
2184
2185 address StubGenerator::base64_vbmi_join_0_1_addr() {
2186 __ align64();
2187 StubCodeMark mark(this, "StubRoutines", "join_0_1_base64");
2188 address start = __ pc();
2189
2190 assert(((unsigned long long)start & 0x3f) == 0,
2191 "Alignment problem (0x%08llx)", (unsigned long long)start);
2192 __ emit_data64(0x090a040506000102, relocInfo::none);
2193 __ emit_data64(0x161011120c0d0e08, relocInfo::none);
2194 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2195 __ emit_data64(0x292a242526202122, relocInfo::none);
2196 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2197 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2198 __ emit_data64(0x494a444546404142, relocInfo::none);
2199 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2200
2201 return start;
2202 }
2203
2204 address StubGenerator::base64_vbmi_join_1_2_addr() {
2205 __ align64();
2206 StubCodeMark mark(this, "StubRoutines", "join_1_2_base64");
2207 address start = __ pc();
2208
2209 assert(((unsigned long long)start & 0x3f) == 0,
2210 "Alignment problem (0x%08llx)", (unsigned long long)start);
2211 __ emit_data64(0x1c1d1e18191a1415, relocInfo::none);
2212 __ emit_data64(0x292a242526202122, relocInfo::none);
2213 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2214 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2215 __ emit_data64(0x494a444546404142, relocInfo::none);
2216 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2217 __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2218 __ emit_data64(0x696a646566606162, relocInfo::none);
2219
2220 return start;
2221 }
2222
2223 address StubGenerator::base64_vbmi_join_2_3_addr() {
2224 __ align64();
2225 StubCodeMark mark(this, "StubRoutines", "join_2_3_base64");
2226 address start = __ pc();
2227
2228 assert(((unsigned long long)start & 0x3f) == 0,
2229 "Alignment problem (0x%08llx)", (unsigned long long)start);
2230 __ emit_data64(0x363031322c2d2e28, relocInfo::none);
2231 __ emit_data64(0x3c3d3e38393a3435, relocInfo::none);
2232 __ emit_data64(0x494a444546404142, relocInfo::none);
2233 __ emit_data64(0x565051524c4d4e48, relocInfo::none);
2234 __ emit_data64(0x5c5d5e58595a5455, relocInfo::none);
2235 __ emit_data64(0x696a646566606162, relocInfo::none);
2236 __ emit_data64(0x767071726c6d6e68, relocInfo::none);
2237 __ emit_data64(0x7c7d7e78797a7475, relocInfo::none);
2238
2239 return start;
2240 }
2241
2242 address StubGenerator::base64_AVX2_decode_tables_addr() {
2243 __ align64();
2244 StubCodeMark mark(this, "StubRoutines", "AVX2_tables_base64");
2245 address start = __ pc();
2246
2247 assert(((unsigned long long)start & 0x3f) == 0,
2248 "Alignment problem (0x%08llx)", (unsigned long long)start);
2249 __ emit_data(0x2f2f2f2f, relocInfo::none, 0);
2250 __ emit_data(0x5f5f5f5f, relocInfo::none, 0); // for URL
2251
2252 __ emit_data(0xffffffff, relocInfo::none, 0);
2253 __ emit_data(0xfcfcfcfc, relocInfo::none, 0); // for URL
2254
2255 // Permute table
2256 __ emit_data64(0x0000000100000000, relocInfo::none);
2257 __ emit_data64(0x0000000400000002, relocInfo::none);
2258 __ emit_data64(0x0000000600000005, relocInfo::none);
2259 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2260
2261 // Shuffle table
2262 __ emit_data64(0x090a040506000102, relocInfo::none);
2263 __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2264 __ emit_data64(0x090a040506000102, relocInfo::none);
2265 __ emit_data64(0xffffffff0c0d0e08, relocInfo::none);
2266
2267 // merge table
2268 __ emit_data(0x01400140, relocInfo::none, 0);
2269
2270 // merge multiplier
2271 __ emit_data(0x00011000, relocInfo::none, 0);
2272
2273 return start;
2274 }
2275
2276 address StubGenerator::base64_AVX2_decode_LUT_tables_addr() {
2277 __ align64();
2278 StubCodeMark mark(this, "StubRoutines", "AVX2_tables_URL_base64");
2279 address start = __ pc();
2280
2281 assert(((unsigned long long)start & 0x3f) == 0,
2282 "Alignment problem (0x%08llx)", (unsigned long long)start);
2283 // lut_lo
2284 __ emit_data64(0x1111111111111115, relocInfo::none);
2285 __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2286 __ emit_data64(0x1111111111111115, relocInfo::none);
2287 __ emit_data64(0x1a1b1b1b1a131111, relocInfo::none);
2288
2289 // lut_roll
2290 __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2291 __ emit_data64(0x0000000000000000, relocInfo::none);
2292 __ emit_data64(0xb9b9bfbf04131000, relocInfo::none);
2293 __ emit_data64(0x0000000000000000, relocInfo::none);
2294
2295 // lut_lo URL
2296 __ emit_data64(0x1111111111111115, relocInfo::none);
2297 __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2298 __ emit_data64(0x1111111111111115, relocInfo::none);
2299 __ emit_data64(0x1b1b1a1b1b131111, relocInfo::none);
2300
2301 // lut_roll URL
2302 __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2303 __ emit_data64(0x0000000000000000, relocInfo::none);
2304 __ emit_data64(0xb9b9bfbf0411e000, relocInfo::none);
2305 __ emit_data64(0x0000000000000000, relocInfo::none);
2306
2307 // lut_hi
2308 __ emit_data64(0x0804080402011010, relocInfo::none);
2309 __ emit_data64(0x1010101010101010, relocInfo::none);
2310 __ emit_data64(0x0804080402011010, relocInfo::none);
2311 __ emit_data64(0x1010101010101010, relocInfo::none);
2312
2313 return start;
2314 }
2315
2316 address StubGenerator::base64_decoding_table_addr() {
2317 StubCodeMark mark(this, "StubRoutines", "decoding_table_base64");
2318 address start = __ pc();
2319
2320 __ emit_data64(0xffffffffffffffff, 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(0x3fffffff3effffff, relocInfo::none);
2326 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2327 __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2328 __ emit_data64(0x06050403020100ff, relocInfo::none);
2329 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2330 __ emit_data64(0x161514131211100f, relocInfo::none);
2331 __ emit_data64(0xffffffffff191817, relocInfo::none);
2332 __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2333 __ emit_data64(0x2827262524232221, relocInfo::none);
2334 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2335 __ emit_data64(0xffffffffff333231, relocInfo::none);
2336 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2337 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2338 __ emit_data64(0xffffffffffffffff, relocInfo::none);
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(0xffffffffffffffff, relocInfo::none);
2345 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2346 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2347 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2348 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2349 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2350 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2351 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2352
2353 // URL table
2354 __ emit_data64(0xffffffffffffffff, 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(0xffff3effffffffff, relocInfo::none);
2360 __ emit_data64(0x3b3a393837363534, relocInfo::none);
2361 __ emit_data64(0xffffffffffff3d3c, relocInfo::none);
2362 __ emit_data64(0x06050403020100ff, relocInfo::none);
2363 __ emit_data64(0x0e0d0c0b0a090807, relocInfo::none);
2364 __ emit_data64(0x161514131211100f, relocInfo::none);
2365 __ emit_data64(0x3fffffffff191817, relocInfo::none);
2366 __ emit_data64(0x201f1e1d1c1b1aff, relocInfo::none);
2367 __ emit_data64(0x2827262524232221, relocInfo::none);
2368 __ emit_data64(0x302f2e2d2c2b2a29, relocInfo::none);
2369 __ emit_data64(0xffffffffff333231, relocInfo::none);
2370 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2371 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2372 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2373 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2374 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2375 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2376 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2377 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2378 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2379 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2380 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2381 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2382 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2383 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2384 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2385 __ emit_data64(0xffffffffffffffff, relocInfo::none);
2386
2387 return start;
2388 }
2389
2390
2391 // Code for generating Base64 decoding.
2392 //
2393 // Based on the article (and associated code) from https://arxiv.org/abs/1910.05109.
2394 //
2395 // Intrinsic function prototype in Base64.java:
2396 // private void decodeBlock(byte[] src, int sp, int sl, byte[] dst, int dp, boolean isURL, isMIME) {
2397 address StubGenerator::generate_base64_decodeBlock() {
2398 __ align(CodeEntryAlignment);
2399 StubCodeMark mark(this, "StubRoutines", "implDecode");
2400 address start = __ pc();
2401
2402 __ enter();
2403
2404 // Save callee-saved registers before using them
2405 __ push(r12);
2406 __ push(r13);
2407 __ push(r14);
2408 __ push(r15);
2409 __ push(rbx);
2410
2411 // arguments
2412 const Register source = c_rarg0; // Source Array
2413 const Register start_offset = c_rarg1; // start offset
2414 const Register end_offset = c_rarg2; // end offset
2415 const Register dest = c_rarg3; // destination array
2416 const Register isMIME = rbx;
2417
2418 #ifndef _WIN64
2419 const Register dp = c_rarg4; // Position for writing to dest array
2420 const Register isURL = c_rarg5;// Base64 or URL character set
2421 __ movl(isMIME, Address(rbp, 2 * wordSize));
2422 #else
2423 const Address dp_mem(rbp, 6 * wordSize); // length is on stack on Win64
2424 const Address isURL_mem(rbp, 7 * wordSize);
2425 const Register isURL = r10; // pick the volatile windows register
2426 const Register dp = r12;
2427 __ movl(dp, dp_mem);
2428 __ movl(isURL, isURL_mem);
2429 __ movl(isMIME, Address(rbp, 8 * wordSize));
2430 #endif
2431
2432 const XMMRegister lookup_lo = xmm5;
2433 const XMMRegister lookup_hi = xmm6;
2434 const XMMRegister errorvec = xmm7;
2435 const XMMRegister pack16_op = xmm9;
2436 const XMMRegister pack32_op = xmm8;
2437 const XMMRegister input0 = xmm3;
2438 const XMMRegister input1 = xmm20;
2439 const XMMRegister input2 = xmm21;
2440 const XMMRegister input3 = xmm19;
2441 const XMMRegister join01 = xmm12;
2442 const XMMRegister join12 = xmm11;
2443 const XMMRegister join23 = xmm10;
2444 const XMMRegister translated0 = xmm2;
2445 const XMMRegister translated1 = xmm1;
2446 const XMMRegister translated2 = xmm0;
2447 const XMMRegister translated3 = xmm4;
2448
2449 const XMMRegister merged0 = xmm2;
2450 const XMMRegister merged1 = xmm1;
2451 const XMMRegister merged2 = xmm0;
2452 const XMMRegister merged3 = xmm4;
2453 const XMMRegister merge_ab_bc0 = xmm2;
2454 const XMMRegister merge_ab_bc1 = xmm1;
2455 const XMMRegister merge_ab_bc2 = xmm0;
2456 const XMMRegister merge_ab_bc3 = xmm4;
2457
2458 const XMMRegister pack24bits = xmm4;
2459
2460 const Register length = r14;
2461 const Register output_size = r13;
2462 const Register output_mask = r15;
2463 const KRegister input_mask = k1;
2464
2465 const XMMRegister input_initial_valid_b64 = xmm0;
2466 const XMMRegister tmp = xmm10;
2467 const XMMRegister mask = xmm0;
2468 const XMMRegister invalid_b64 = xmm1;
2469
2470 Label L_process256, L_process64, L_process64Loop, L_exit, L_processdata, L_loadURL;
2471 Label L_continue, L_finalBit, L_padding, L_donePadding, L_bruteForce;
2472 Label L_forceLoop, L_bottomLoop, L_checkMIME, L_exit_no_vzero, L_lastChunk;
2473
2474 // calculate length from offsets
2475 __ movl(length, end_offset);
2476 __ subl(length, start_offset);
2477 __ push(dest); // Save for return value calc
2478
2479 // If AVX512 VBMI not supported, just compile non-AVX code
2480 if(VM_Version::supports_avx512_vbmi() &&
2481 VM_Version::supports_avx512bw()) {
2482 __ cmpl(length, 31); // 32-bytes is break-even for AVX-512
2483 __ jcc(Assembler::lessEqual, L_lastChunk);
2484
2485 __ cmpl(isMIME, 0);
2486 __ jcc(Assembler::notEqual, L_lastChunk);
2487
2488 // Load lookup tables based on isURL
2489 __ cmpl(isURL, 0);
2490 __ jcc(Assembler::notZero, L_loadURL);
2491
2492 __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_addr()), Assembler::AVX_512bit, r13);
2493 __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_addr()), Assembler::AVX_512bit, r13);
2494
2495 __ BIND(L_continue);
2496
2497 __ movl(r15, 0x01400140);
2498 __ evpbroadcastd(pack16_op, r15, Assembler::AVX_512bit);
2499
2500 __ movl(r15, 0x00011000);
2501 __ evpbroadcastd(pack32_op, r15, Assembler::AVX_512bit);
2502
2503 __ cmpl(length, 0xff);
2504 __ jcc(Assembler::lessEqual, L_process64);
2505
2506 // load masks required for decoding data
2507 __ BIND(L_processdata);
2508 __ evmovdquq(join01, ExternalAddress(StubRoutines::x86::base64_vbmi_join_0_1_addr()), Assembler::AVX_512bit,r13);
2509 __ evmovdquq(join12, ExternalAddress(StubRoutines::x86::base64_vbmi_join_1_2_addr()), Assembler::AVX_512bit, r13);
2510 __ evmovdquq(join23, ExternalAddress(StubRoutines::x86::base64_vbmi_join_2_3_addr()), Assembler::AVX_512bit, r13);
2511
2512 __ align32();
2513 __ BIND(L_process256);
2514 // Grab input data
2515 __ evmovdquq(input0, Address(source, start_offset, Address::times_1, 0x00), Assembler::AVX_512bit);
2516 __ evmovdquq(input1, Address(source, start_offset, Address::times_1, 0x40), Assembler::AVX_512bit);
2517 __ evmovdquq(input2, Address(source, start_offset, Address::times_1, 0x80), Assembler::AVX_512bit);
2518 __ evmovdquq(input3, Address(source, start_offset, Address::times_1, 0xc0), Assembler::AVX_512bit);
2519
2520 // Copy the low part of the lookup table into the destination of the permutation
2521 __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2522 __ evmovdquq(translated1, lookup_lo, Assembler::AVX_512bit);
2523 __ evmovdquq(translated2, lookup_lo, Assembler::AVX_512bit);
2524 __ evmovdquq(translated3, lookup_lo, Assembler::AVX_512bit);
2525
2526 // Translate the base64 input into "decoded" bytes
2527 __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2528 __ evpermt2b(translated1, input1, lookup_hi, Assembler::AVX_512bit);
2529 __ evpermt2b(translated2, input2, lookup_hi, Assembler::AVX_512bit);
2530 __ evpermt2b(translated3, input3, lookup_hi, Assembler::AVX_512bit);
2531
2532 // OR all of the translations together to check for errors (high-order bit of byte set)
2533 __ vpternlogd(input0, 0xfe, input1, input2, Assembler::AVX_512bit);
2534
2535 __ vpternlogd(input3, 0xfe, translated0, translated1, Assembler::AVX_512bit);
2536 __ vpternlogd(input0, 0xfe, translated2, translated3, Assembler::AVX_512bit);
2537 __ vpor(errorvec, input3, input0, Assembler::AVX_512bit);
2538
2539 // Check if there was an error - if so, try 64-byte chunks
2540 __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2541 __ kortestql(k3, k3);
2542 __ jcc(Assembler::notZero, L_process64);
2543
2544 // The merging and shuffling happens here
2545 // We multiply each byte pair [00dddddd | 00cccccc | 00bbbbbb | 00aaaaaa]
2546 // Multiply [00cccccc] by 2^6 added to [00dddddd] to get [0000cccc | ccdddddd]
2547 // The pack16_op is a vector of 0x01400140, so multiply D by 1 and C by 0x40
2548 __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2549 __ vpmaddubsw(merge_ab_bc1, translated1, pack16_op, Assembler::AVX_512bit);
2550 __ vpmaddubsw(merge_ab_bc2, translated2, pack16_op, Assembler::AVX_512bit);
2551 __ vpmaddubsw(merge_ab_bc3, translated3, pack16_op, Assembler::AVX_512bit);
2552
2553 // Now do the same with packed 16-bit values.
2554 // We start with [0000cccc | ccdddddd | 0000aaaa | aabbbbbb]
2555 // pack32_op is 0x00011000 (2^12, 1), so this multiplies [0000aaaa | aabbbbbb] by 2^12
2556 // and adds [0000cccc | ccdddddd] to yield [00000000 | aaaaaabb | bbbbcccc | ccdddddd]
2557 __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2558 __ vpmaddwd(merged1, merge_ab_bc1, pack32_op, Assembler::AVX_512bit);
2559 __ vpmaddwd(merged2, merge_ab_bc2, pack32_op, Assembler::AVX_512bit);
2560 __ vpmaddwd(merged3, merge_ab_bc3, pack32_op, Assembler::AVX_512bit);
2561
2562 // The join vectors specify which byte from which vector goes into the outputs
2563 // One of every 4 bytes in the extended vector is zero, so we pack them into their
2564 // final positions in the register for storing (256 bytes in, 192 bytes out)
2565 __ evpermt2b(merged0, join01, merged1, Assembler::AVX_512bit);
2566 __ evpermt2b(merged1, join12, merged2, Assembler::AVX_512bit);
2567 __ evpermt2b(merged2, join23, merged3, Assembler::AVX_512bit);
2568
2569 // Store result
2570 __ evmovdquq(Address(dest, dp, Address::times_1, 0x00), merged0, Assembler::AVX_512bit);
2571 __ evmovdquq(Address(dest, dp, Address::times_1, 0x40), merged1, Assembler::AVX_512bit);
2572 __ evmovdquq(Address(dest, dp, Address::times_1, 0x80), merged2, Assembler::AVX_512bit);
2573
2574 __ addptr(source, 0x100);
2575 __ addptr(dest, 0xc0);
2576 __ subl(length, 0x100);
2577 __ cmpl(length, 64 * 4);
2578 __ jcc(Assembler::greaterEqual, L_process256);
2579
2580 // At this point, we've decoded 64 * 4 * n bytes.
2581 // The remaining length will be <= 64 * 4 - 1.
2582 // UNLESS there was an error decoding the first 256-byte chunk. In this
2583 // case, the length will be arbitrarily long.
2584 //
2585 // Note that this will be the path for MIME-encoded strings.
2586
2587 __ BIND(L_process64);
2588
2589 __ evmovdquq(pack24bits, ExternalAddress(StubRoutines::x86::base64_vbmi_pack_vec_addr()), Assembler::AVX_512bit, r13);
2590
2591 __ cmpl(length, 63);
2592 __ jcc(Assembler::lessEqual, L_finalBit);
2593
2594 __ mov64(rax, 0x0000ffffffffffff);
2595 __ kmovql(k2, rax);
2596
2597 __ align32();
2598 __ BIND(L_process64Loop);
2599
2600 // Handle first 64-byte block
2601
2602 __ evmovdquq(input0, Address(source, start_offset), Assembler::AVX_512bit);
2603 __ evmovdquq(translated0, lookup_lo, Assembler::AVX_512bit);
2604 __ evpermt2b(translated0, input0, lookup_hi, Assembler::AVX_512bit);
2605
2606 __ vpor(errorvec, translated0, input0, Assembler::AVX_512bit);
2607
2608 // Check for error and bomb out before updating dest
2609 __ evpmovb2m(k3, errorvec, Assembler::AVX_512bit);
2610 __ kortestql(k3, k3);
2611 __ jcc(Assembler::notZero, L_exit);
2612
2613 // Pack output register, selecting correct byte ordering
2614 __ vpmaddubsw(merge_ab_bc0, translated0, pack16_op, Assembler::AVX_512bit);
2615 __ vpmaddwd(merged0, merge_ab_bc0, pack32_op, Assembler::AVX_512bit);
2616 __ vpermb(merged0, pack24bits, merged0, Assembler::AVX_512bit);
2617
2618 __ evmovdqub(Address(dest, dp), k2, merged0, true, Assembler::AVX_512bit);
2619
2620 __ subl(length, 64);
2621 __ addptr(source, 64);
2622 __ addptr(dest, 48);
2623
2624 __ cmpl(length, 64);
2625 __ jcc(Assembler::greaterEqual, L_process64Loop);
2626
2627 __ cmpl(length, 0);
2628 __ jcc(Assembler::lessEqual, L_exit);
2629
2630 __ BIND(L_finalBit);
2631 // Now have 1 to 63 bytes left to decode
2632
2633 // I was going to let Java take care of the final fragment
2634 // however it will repeatedly call this routine for every 4 bytes
2635 // of input data, so handle the rest here.
2636 __ movq(rax, -1);
2637 __ bzhiq(rax, rax, length); // Input mask in rax
2638
2639 __ movl(output_size, length);
2640 __ shrl(output_size, 2); // Find (len / 4) * 3 (output length)
2641 __ lea(output_size, Address(output_size, output_size, Address::times_2, 0));
2642 // output_size in r13
2643
2644 // Strip pad characters, if any, and adjust length and mask
2645 __ addq(length, start_offset);
2646 __ cmpb(Address(source, length, Address::times_1, -1), '=');
2647 __ jcc(Assembler::equal, L_padding);
2648
2649 __ BIND(L_donePadding);
2650 __ subq(length, start_offset);
2651
2652 // Output size is (64 - output_size), output mask is (all 1s >> output_size).
2653 __ kmovql(input_mask, rax);
2654 __ movq(output_mask, -1);
2655 __ bzhiq(output_mask, output_mask, output_size);
2656
2657 // Load initial input with all valid base64 characters. Will be used
2658 // in merging source bytes to avoid masking when determining if an error occurred.
2659 __ movl(rax, 0x61616161);
2660 __ evpbroadcastd(input_initial_valid_b64, rax, Assembler::AVX_512bit);
2661
2662 // A register containing all invalid base64 decoded values
2663 __ movl(rax, 0x80808080);
2664 __ evpbroadcastd(invalid_b64, rax, Assembler::AVX_512bit);
2665
2666 // input_mask is in k1
2667 // output_size is in r13
2668 // output_mask is in r15
2669 // zmm0 - free
2670 // zmm1 - 0x00011000
2671 // zmm2 - 0x01400140
2672 // zmm3 - errorvec
2673 // zmm4 - pack vector
2674 // zmm5 - lookup_lo
2675 // zmm6 - lookup_hi
2676 // zmm7 - errorvec
2677 // zmm8 - 0x61616161
2678 // zmm9 - 0x80808080
2679
2680 // Load only the bytes from source, merging into our "fully-valid" register
2681 __ evmovdqub(input_initial_valid_b64, input_mask, Address(source, start_offset, Address::times_1, 0x0), true, Assembler::AVX_512bit);
2682
2683 // Decode all bytes within our merged input
2684 __ evmovdquq(tmp, lookup_lo, Assembler::AVX_512bit);
2685 __ evpermt2b(tmp, input_initial_valid_b64, lookup_hi, Assembler::AVX_512bit);
2686 __ evporq(mask, tmp, input_initial_valid_b64, Assembler::AVX_512bit);
2687
2688 // Check for error. Compare (decoded | initial) to all invalid.
2689 // If any bytes have their high-order bit set, then we have an error.
2690 __ evptestmb(k2, mask, invalid_b64, Assembler::AVX_512bit);
2691 __ kortestql(k2, k2);
2692
2693 // If we have an error, use the brute force loop to decode what we can (4-byte chunks).
2694 __ jcc(Assembler::notZero, L_bruteForce);
2695
2696 // Shuffle output bytes
2697 __ vpmaddubsw(tmp, tmp, pack16_op, Assembler::AVX_512bit);
2698 __ vpmaddwd(tmp, tmp, pack32_op, Assembler::AVX_512bit);
2699
2700 __ vpermb(tmp, pack24bits, tmp, Assembler::AVX_512bit);
2701 __ kmovql(k1, output_mask);
2702 __ evmovdqub(Address(dest, dp), k1, tmp, true, Assembler::AVX_512bit);
2703
2704 __ addptr(dest, output_size);
2705
2706 __ BIND(L_exit);
2707 __ vzeroupper();
2708 __ pop(rax); // Get original dest value
2709 __ subptr(dest, rax); // Number of bytes converted
2710 __ movptr(rax, dest);
2711 __ pop(rbx);
2712 __ pop(r15);
2713 __ pop(r14);
2714 __ pop(r13);
2715 __ pop(r12);
2716 __ leave();
2717 __ ret(0);
2718
2719 __ BIND(L_loadURL);
2720 __ evmovdquq(lookup_lo, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_lo_url_addr()), Assembler::AVX_512bit, r13);
2721 __ evmovdquq(lookup_hi, ExternalAddress(StubRoutines::x86::base64_vbmi_lookup_hi_url_addr()), Assembler::AVX_512bit, r13);
2722 __ jmp(L_continue);
2723
2724 __ BIND(L_padding);
2725 __ decrementq(output_size, 1);
2726 __ shrq(rax, 1);
2727
2728 __ cmpb(Address(source, length, Address::times_1, -2), '=');
2729 __ jcc(Assembler::notEqual, L_donePadding);
2730
2731 __ decrementq(output_size, 1);
2732 __ shrq(rax, 1);
2733 __ jmp(L_donePadding);
2734
2735 __ align32();
2736 __ BIND(L_bruteForce);
2737 } // End of if(avx512_vbmi)
2738
2739 if (VM_Version::supports_avx2()) {
2740 Label L_tailProc, L_topLoop, L_enterLoop;
2741
2742 __ cmpl(isMIME, 0);
2743 __ jcc(Assembler::notEqual, L_lastChunk);
2744
2745 // Check for buffer too small (for algorithm)
2746 __ subl(length, 0x2c);
2747 __ jcc(Assembler::less, L_tailProc);
2748
2749 __ shll(isURL, 2);
2750
2751 // Algorithm adapted from https://arxiv.org/abs/1704.00605, "Faster Base64
2752 // Encoding and Decoding using AVX2 Instructions". URL modifications added.
2753
2754 // Set up constants
2755 __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_tables_addr()));
2756 __ vpbroadcastd(xmm4, Address(r13, isURL, Address::times_1), Assembler::AVX_256bit); // 2F or 5F
2757 __ vpbroadcastd(xmm10, Address(r13, isURL, Address::times_1, 0x08), Assembler::AVX_256bit); // -1 or -4
2758 __ vmovdqu(xmm12, Address(r13, 0x10)); // permute
2759 __ vmovdqu(xmm13, Address(r13, 0x30)); // shuffle
2760 __ vpbroadcastd(xmm7, Address(r13, 0x50), Assembler::AVX_256bit); // merge
2761 __ vpbroadcastd(xmm6, Address(r13, 0x54), Assembler::AVX_256bit); // merge mult
2762
2763 __ lea(r13, ExternalAddress(StubRoutines::x86::base64_AVX2_decode_LUT_tables_addr()));
2764 __ shll(isURL, 4);
2765 __ vmovdqu(xmm11, Address(r13, isURL, Address::times_1, 0x00)); // lut_lo
2766 __ vmovdqu(xmm8, Address(r13, isURL, Address::times_1, 0x20)); // lut_roll
2767 __ shrl(isURL, 6); // restore isURL
2768 __ vmovdqu(xmm9, Address(r13, 0x80)); // lut_hi
2769 __ jmp(L_enterLoop);
2770
2771 __ align32();
2772 __ bind(L_topLoop);
2773 // Add in the offset value (roll) to get 6-bit out values
2774 __ vpaddb(xmm0, xmm0, xmm2, Assembler::AVX_256bit);
2775 // Merge and permute the output bits into appropriate output byte lanes
2776 __ vpmaddubsw(xmm0, xmm0, xmm7, Assembler::AVX_256bit);
2777 __ vpmaddwd(xmm0, xmm0, xmm6, Assembler::AVX_256bit);
2778 __ vpshufb(xmm0, xmm0, xmm13, Assembler::AVX_256bit);
2779 __ vpermd(xmm0, xmm12, xmm0, Assembler::AVX_256bit);
2780 // Store the output bytes
2781 __ vmovdqu(Address(dest, dp, Address::times_1, 0), xmm0);
2782 __ addptr(source, 0x20);
2783 __ addptr(dest, 0x18);
2784 __ subl(length, 0x20);
2785 __ jcc(Assembler::less, L_tailProc);
2786
2787 __ bind(L_enterLoop);
2788
2789 // Load in encoded string (32 bytes)
2790 __ vmovdqu(xmm2, Address(source, start_offset, Address::times_1, 0x0));
2791 // Extract the high nibble for indexing into the lut tables. High 4 bits are don't care.
2792 __ vpsrld(xmm1, xmm2, 0x4, Assembler::AVX_256bit);
2793 __ vpand(xmm1, xmm4, xmm1, Assembler::AVX_256bit);
2794 // Extract the low nibble. 5F/2F will isolate the low-order 4 bits. High 4 bits are don't care.
2795 __ vpand(xmm3, xmm2, xmm4, Assembler::AVX_256bit);
2796 // Check for special-case (0x2F or 0x5F (URL))
2797 __ vpcmpeqb(xmm0, xmm4, xmm2, Assembler::AVX_256bit);
2798 // Get the bitset based on the low nibble. vpshufb uses low-order 4 bits only.
2799 __ vpshufb(xmm3, xmm11, xmm3, Assembler::AVX_256bit);
2800 // Get the bit value of the high nibble
2801 __ vpshufb(xmm5, xmm9, xmm1, Assembler::AVX_256bit);
2802 // Make sure 2F / 5F shows as valid
2803 __ vpandn(xmm3, xmm0, xmm3, Assembler::AVX_256bit);
2804 // Make adjustment for roll index. For non-URL, this is a no-op,
2805 // for URL, this adjusts by -4. This is to properly index the
2806 // roll value for 2F / 5F.
2807 __ vpand(xmm0, xmm0, xmm10, Assembler::AVX_256bit);
2808 // If the and of the two is non-zero, we have an invalid input character
2809 __ vptest(xmm3, xmm5);
2810 // Extract the "roll" value - value to add to the input to get 6-bit out value
2811 __ vpaddb(xmm0, xmm0, xmm1, Assembler::AVX_256bit); // Handle 2F / 5F
2812 __ vpshufb(xmm0, xmm8, xmm0, Assembler::AVX_256bit);
2813 __ jcc(Assembler::equal, L_topLoop); // Fall through on error
2814
2815 __ bind(L_tailProc);
2816
2817 __ addl(length, 0x2c);
2818
2819 __ vzeroupper();
2820 }
2821
2822 // Use non-AVX code to decode 4-byte chunks into 3 bytes of output
2823
2824 // Register state (Linux):
2825 // r12-15 - saved on stack
2826 // rdi - src
2827 // rsi - sp
2828 // rdx - sl
2829 // rcx - dst
2830 // r8 - dp
2831 // r9 - isURL
2832
2833 // Register state (Windows):
2834 // r12-15 - saved on stack
2835 // rcx - src
2836 // rdx - sp
2837 // r8 - sl
2838 // r9 - dst
2839 // r12 - dp
2840 // r10 - isURL
2841
2842 // Registers (common):
2843 // length (r14) - bytes in src
2844
2845 const Register decode_table = r11;
2846 const Register out_byte_count = rbx;
2847 const Register byte1 = r13;
2848 const Register byte2 = r15;
2849 const Register byte3 = WIN64_ONLY(r8) NOT_WIN64(rdx);
2850 const Register byte4 = WIN64_ONLY(r10) NOT_WIN64(r9);
2851
2852 __ bind(L_lastChunk);
2853
2854 __ shrl(length, 2); // Multiple of 4 bytes only - length is # 4-byte chunks
2855 __ cmpl(length, 0);
2856 __ jcc(Assembler::lessEqual, L_exit_no_vzero);
2857
2858 __ shll(isURL, 8); // index into decode table based on isURL
2859 __ lea(decode_table, ExternalAddress(StubRoutines::x86::base64_decoding_table_addr()));
2860 __ addptr(decode_table, isURL);
2861
2862 __ jmp(L_bottomLoop);
2863
2864 __ align32();
2865 __ BIND(L_forceLoop);
2866 __ shll(byte1, 18);
2867 __ shll(byte2, 12);
2868 __ shll(byte3, 6);
2869 __ orl(byte1, byte2);
2870 __ orl(byte1, byte3);
2871 __ orl(byte1, byte4);
2872
2873 __ addptr(source, 4);
2874
2875 __ movb(Address(dest, dp, Address::times_1, 2), byte1);
2876 __ shrl(byte1, 8);
2877 __ movb(Address(dest, dp, Address::times_1, 1), byte1);
2878 __ shrl(byte1, 8);
2879 __ movb(Address(dest, dp, Address::times_1, 0), byte1);
2880
2881 __ addptr(dest, 3);
2882 __ decrementl(length, 1);
2883 __ jcc(Assembler::zero, L_exit_no_vzero);
2884
2885 __ BIND(L_bottomLoop);
2886 __ load_unsigned_byte(byte1, Address(source, start_offset, Address::times_1, 0x00));
2887 __ load_unsigned_byte(byte2, Address(source, start_offset, Address::times_1, 0x01));
2888 __ load_signed_byte(byte1, Address(decode_table, byte1));
2889 __ load_signed_byte(byte2, Address(decode_table, byte2));
2890 __ load_unsigned_byte(byte3, Address(source, start_offset, Address::times_1, 0x02));
2891 __ load_unsigned_byte(byte4, Address(source, start_offset, Address::times_1, 0x03));
2892 __ load_signed_byte(byte3, Address(decode_table, byte3));
2893 __ load_signed_byte(byte4, Address(decode_table, byte4));
2894
2895 __ mov(rax, byte1);
2896 __ orl(rax, byte2);
2897 __ orl(rax, byte3);
2898 __ orl(rax, byte4);
2899 __ jcc(Assembler::positive, L_forceLoop);
2900
2901 __ BIND(L_exit_no_vzero);
2902 __ pop(rax); // Get original dest value
2903 __ subptr(dest, rax); // Number of bytes converted
2904 __ movptr(rax, dest);
2905 __ pop(rbx);
2906 __ pop(r15);
2907 __ pop(r14);
2908 __ pop(r13);
2909 __ pop(r12);
2910 __ leave();
2911 __ ret(0);
2912
2913 return start;
2914 }
2915
2916
2917 /**
2918 * Arguments:
2919 *
2920 * Inputs:
2921 * c_rarg0 - int crc
2922 * c_rarg1 - byte* buf
2923 * c_rarg2 - int length
2924 *
2925 * Output:
2926 * rax - int crc result
2927 */
2928 address StubGenerator::generate_updateBytesCRC32() {
2929 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2930
2931 __ align(CodeEntryAlignment);
2932 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2933
2934 address start = __ pc();
2935
2936 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
2937 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
2938 // rscratch1: r10
2939 const Register crc = c_rarg0; // crc
2940 const Register buf = c_rarg1; // source java byte array address
2941 const Register len = c_rarg2; // length
2942 const Register table = c_rarg3; // crc_table address (reuse register)
2943 const Register tmp1 = r11;
2944 const Register tmp2 = r10;
2945 assert_different_registers(crc, buf, len, table, tmp1, tmp2, rax);
2946
2947 BLOCK_COMMENT("Entry:");
2948 __ enter(); // required for proper stackwalking of RuntimeStub frame
2949
2950 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
2951 VM_Version::supports_avx512bw() &&
2952 VM_Version::supports_avx512vl()) {
2953 // The constants used in the CRC32 algorithm requires the 1's compliment of the initial crc value.
2954 // However, the constant table for CRC32-C assumes the original crc value. Account for this
2955 // difference before calling and after returning.
2956 __ lea(table, ExternalAddress(StubRoutines::x86::crc_table_avx512_addr()));
2957 __ notl(crc);
2958 __ kernel_crc32_avx512(crc, buf, len, table, tmp1, tmp2);
2959 __ notl(crc);
2960 } else {
2961 __ kernel_crc32(crc, buf, len, table, tmp1);
2962 }
2963
2964 __ movl(rax, crc);
2965 __ vzeroupper();
2966 __ leave(); // required for proper stackwalking of RuntimeStub frame
2967 __ ret(0);
2968
2969 return start;
2970 }
2971
2972 /**
2973 * Arguments:
2974 *
2975 * Inputs:
2976 * c_rarg0 - int crc
2977 * c_rarg1 - byte* buf
2978 * c_rarg2 - long length
2979 * c_rarg3 - table_start - optional (present only when doing a library_call,
2980 * not used by x86 algorithm)
2981 *
2982 * Output:
2983 * rax - int crc result
2984 */
2985 address StubGenerator::generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
2986 assert(UseCRC32CIntrinsics, "need SSE4_2");
2987 __ align(CodeEntryAlignment);
2988 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
2989 address start = __ pc();
2990
2991 //reg.arg int#0 int#1 int#2 int#3 int#4 int#5 float regs
2992 //Windows RCX RDX R8 R9 none none XMM0..XMM3
2993 //Lin / Sol RDI RSI RDX RCX R8 R9 XMM0..XMM7
2994 const Register crc = c_rarg0; // crc
2995 const Register buf = c_rarg1; // source java byte array address
2996 const Register len = c_rarg2; // length
2997 const Register a = rax;
2998 const Register j = r9;
2999 const Register k = r10;
3000 const Register l = r11;
3001 #ifdef _WIN64
3002 const Register y = rdi;
3003 const Register z = rsi;
3004 #else
3005 const Register y = rcx;
3006 const Register z = r8;
3007 #endif
3008 assert_different_registers(crc, buf, len, a, j, k, l, y, z);
3009
3010 BLOCK_COMMENT("Entry:");
3011 __ enter(); // required for proper stackwalking of RuntimeStub frame
3012 Label L_continue;
3013
3014 if (VM_Version::supports_sse4_1() && VM_Version::supports_avx512_vpclmulqdq() &&
3015 VM_Version::supports_avx512bw() &&
3016 VM_Version::supports_avx512vl()) {
3017 Label L_doSmall;
3018
3019 __ cmpl(len, 384);
3020 __ jcc(Assembler::lessEqual, L_doSmall);
3021
3022 __ lea(j, ExternalAddress(StubRoutines::x86::crc32c_table_avx512_addr()));
3023 __ kernel_crc32_avx512(crc, buf, len, j, l, k);
3024
3025 __ jmp(L_continue);
3026
3027 __ bind(L_doSmall);
3028 }
3029 #ifdef _WIN64
3030 __ push(y);
3031 __ push(z);
3032 #endif
3033 __ crc32c_ipl_alg2_alt2(crc, buf, len,
3034 a, j, k,
3035 l, y, z,
3036 c_farg0, c_farg1, c_farg2,
3037 is_pclmulqdq_supported);
3038 #ifdef _WIN64
3039 __ pop(z);
3040 __ pop(y);
3041 #endif
3042
3043 __ bind(L_continue);
3044 __ movl(rax, crc);
3045 __ vzeroupper();
3046 __ leave(); // required for proper stackwalking of RuntimeStub frame
3047 __ ret(0);
3048
3049 return start;
3050 }
3051
3052
3053 /**
3054 * Arguments:
3055 *
3056 * Input:
3057 * c_rarg0 - x address
3058 * c_rarg1 - x length
3059 * c_rarg2 - y address
3060 * c_rarg3 - y length
3061 * not Win64
3062 * c_rarg4 - z address
3063 * c_rarg5 - z length
3064 * Win64
3065 * rsp+40 - z address
3066 * rsp+48 - z length
3067 */
3068 address StubGenerator::generate_multiplyToLen() {
3069 __ align(CodeEntryAlignment);
3070 StubCodeMark mark(this, "StubRoutines", "multiplyToLen");
3071 address start = __ pc();
3072
3073 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3074 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3075 const Register x = rdi;
3076 const Register xlen = rax;
3077 const Register y = rsi;
3078 const Register ylen = rcx;
3079 const Register z = r8;
3080 const Register zlen = r11;
3081
3082 // Next registers will be saved on stack in multiply_to_len().
3083 const Register tmp1 = r12;
3084 const Register tmp2 = r13;
3085 const Register tmp3 = r14;
3086 const Register tmp4 = r15;
3087 const Register tmp5 = rbx;
3088
3089 BLOCK_COMMENT("Entry:");
3090 __ enter(); // required for proper stackwalking of RuntimeStub frame
3091
3092 #ifndef _WIN64
3093 __ movptr(zlen, r9); // Save r9 in r11 - zlen
3094 #endif
3095 setup_arg_regs(4); // x => rdi, xlen => rsi, y => rdx
3096 // ylen => rcx, z => r8, zlen => r11
3097 // r9 and r10 may be used to save non-volatile registers
3098 #ifdef _WIN64
3099 // last 2 arguments (#4, #5) are on stack on Win64
3100 __ movptr(z, Address(rsp, 6 * wordSize));
3101 __ movptr(zlen, Address(rsp, 7 * wordSize));
3102 #endif
3103
3104 __ movptr(xlen, rsi);
3105 __ movptr(y, rdx);
3106 __ multiply_to_len(x, xlen, y, ylen, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5);
3107
3108 restore_arg_regs();
3109
3110 __ leave(); // required for proper stackwalking of RuntimeStub frame
3111 __ ret(0);
3112
3113 return start;
3114 }
3115
3116 /**
3117 * Arguments:
3118 *
3119 * Input:
3120 * c_rarg0 - obja address
3121 * c_rarg1 - objb address
3122 * c_rarg3 - length length
3123 * c_rarg4 - scale log2_array_indxscale
3124 *
3125 * Output:
3126 * rax - int >= mismatched index, < 0 bitwise complement of tail
3127 */
3128 address StubGenerator::generate_vectorizedMismatch() {
3129 __ align(CodeEntryAlignment);
3130 StubCodeMark mark(this, "StubRoutines", "vectorizedMismatch");
3131 address start = __ pc();
3132
3133 BLOCK_COMMENT("Entry:");
3134 __ enter();
3135
3136 #ifdef _WIN64 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3137 const Register scale = c_rarg0; //rcx, will exchange with r9
3138 const Register objb = c_rarg1; //rdx
3139 const Register length = c_rarg2; //r8
3140 const Register obja = c_rarg3; //r9
3141 __ xchgq(obja, scale); //now obja and scale contains the correct contents
3142
3143 const Register tmp1 = r10;
3144 const Register tmp2 = r11;
3145 #endif
3146 #ifndef _WIN64 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3147 const Register obja = c_rarg0; //U:rdi
3148 const Register objb = c_rarg1; //U:rsi
3149 const Register length = c_rarg2; //U:rdx
3150 const Register scale = c_rarg3; //U:rcx
3151 const Register tmp1 = r8;
3152 const Register tmp2 = r9;
3153 #endif
3154 const Register result = rax; //return value
3155 const XMMRegister vec0 = xmm0;
3156 const XMMRegister vec1 = xmm1;
3157 const XMMRegister vec2 = xmm2;
3158
3159 __ vectorized_mismatch(obja, objb, length, scale, result, tmp1, tmp2, vec0, vec1, vec2);
3160
3161 __ vzeroupper();
3162 __ leave();
3163 __ ret(0);
3164
3165 return start;
3166 }
3167
3168 /**
3169 * Arguments:
3170 *
3171 // Input:
3172 // c_rarg0 - x address
3173 // c_rarg1 - x length
3174 // c_rarg2 - z address
3175 // c_rarg3 - z length
3176 *
3177 */
3178 address StubGenerator::generate_squareToLen() {
3179
3180 __ align(CodeEntryAlignment);
3181 StubCodeMark mark(this, "StubRoutines", "squareToLen");
3182 address start = __ pc();
3183
3184 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3185 // Unix: rdi, rsi, rdx, rcx (c_rarg0, c_rarg1, ...)
3186 const Register x = rdi;
3187 const Register len = rsi;
3188 const Register z = r8;
3189 const Register zlen = rcx;
3190
3191 const Register tmp1 = r12;
3192 const Register tmp2 = r13;
3193 const Register tmp3 = r14;
3194 const Register tmp4 = r15;
3195 const Register tmp5 = rbx;
3196
3197 BLOCK_COMMENT("Entry:");
3198 __ enter(); // required for proper stackwalking of RuntimeStub frame
3199
3200 setup_arg_regs(4); // x => rdi, len => rsi, z => rdx
3201 // zlen => rcx
3202 // r9 and r10 may be used to save non-volatile registers
3203 __ movptr(r8, rdx);
3204 __ square_to_len(x, len, z, zlen, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3205
3206 restore_arg_regs();
3207
3208 __ leave(); // required for proper stackwalking of RuntimeStub frame
3209 __ ret(0);
3210
3211 return start;
3212 }
3213
3214 address StubGenerator::generate_method_entry_barrier() {
3215 __ align(CodeEntryAlignment);
3216 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
3217 address start = __ pc();
3218
3219 Label deoptimize_label;
3220
3221 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
3222
3223 BLOCK_COMMENT("Entry:");
3224 __ enter(); // save rbp
3225
3226 // save c_rarg0, because we want to use that value.
3227 // We could do without it but then we depend on the number of slots used by pusha
3228 __ push(c_rarg0);
3229
3230 __ lea(c_rarg0, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for c_rarg0 - this should be the return address
3231
3232 __ pusha();
3233
3234 // The method may have floats as arguments, and we must spill them before calling
3235 // the VM runtime.
3236 assert(Argument::n_float_register_parameters_j == 8, "Assumption");
3237 const int xmm_size = wordSize * 2;
3238 const int xmm_spill_size = xmm_size * Argument::n_float_register_parameters_j;
3239 __ subptr(rsp, xmm_spill_size);
3240 __ movdqu(Address(rsp, xmm_size * 7), xmm7);
3241 __ movdqu(Address(rsp, xmm_size * 6), xmm6);
3242 __ movdqu(Address(rsp, xmm_size * 5), xmm5);
3243 __ movdqu(Address(rsp, xmm_size * 4), xmm4);
3244 __ movdqu(Address(rsp, xmm_size * 3), xmm3);
3245 __ movdqu(Address(rsp, xmm_size * 2), xmm2);
3246 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
3247 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
3248
3249 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), 1);
3250
3251 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
3252 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
3253 __ movdqu(xmm2, Address(rsp, xmm_size * 2));
3254 __ movdqu(xmm3, Address(rsp, xmm_size * 3));
3255 __ movdqu(xmm4, Address(rsp, xmm_size * 4));
3256 __ movdqu(xmm5, Address(rsp, xmm_size * 5));
3257 __ movdqu(xmm6, Address(rsp, xmm_size * 6));
3258 __ movdqu(xmm7, Address(rsp, xmm_size * 7));
3259 __ addptr(rsp, xmm_spill_size);
3260
3261 __ cmpl(rax, 1); // 1 means deoptimize
3262 __ jcc(Assembler::equal, deoptimize_label);
3263
3264 __ popa();
3265 __ pop(c_rarg0);
3266
3267 __ leave();
3268
3269 __ addptr(rsp, 1 * wordSize); // cookie
3270 __ ret(0);
3271
3272
3273 __ BIND(deoptimize_label);
3274
3275 __ popa();
3276 __ pop(c_rarg0);
3277
3278 __ leave();
3279
3280 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
3281 // here while still having a correct stack is valuable
3282 __ testptr(rsp, Address(rsp, 0));
3283
3284 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
3285 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
3286
3287 return start;
3288 }
3289
3290 /**
3291 * Arguments:
3292 *
3293 * Input:
3294 * c_rarg0 - out address
3295 * c_rarg1 - in address
3296 * c_rarg2 - offset
3297 * c_rarg3 - len
3298 * not Win64
3299 * c_rarg4 - k
3300 * Win64
3301 * rsp+40 - k
3302 */
3303 address StubGenerator::generate_mulAdd() {
3304 __ align(CodeEntryAlignment);
3305 StubCodeMark mark(this, "StubRoutines", "mulAdd");
3306 address start = __ pc();
3307
3308 // Win64: rcx, rdx, r8, r9 (c_rarg0, c_rarg1, ...)
3309 // Unix: rdi, rsi, rdx, rcx, r8, r9 (c_rarg0, c_rarg1, ...)
3310 const Register out = rdi;
3311 const Register in = rsi;
3312 const Register offset = r11;
3313 const Register len = rcx;
3314 const Register k = r8;
3315
3316 // Next registers will be saved on stack in mul_add().
3317 const Register tmp1 = r12;
3318 const Register tmp2 = r13;
3319 const Register tmp3 = r14;
3320 const Register tmp4 = r15;
3321 const Register tmp5 = rbx;
3322
3323 BLOCK_COMMENT("Entry:");
3324 __ enter(); // required for proper stackwalking of RuntimeStub frame
3325
3326 setup_arg_regs(4); // out => rdi, in => rsi, offset => rdx
3327 // len => rcx, k => r8
3328 // r9 and r10 may be used to save non-volatile registers
3329 #ifdef _WIN64
3330 // last argument is on stack on Win64
3331 __ movl(k, Address(rsp, 6 * wordSize));
3332 #endif
3333 __ movptr(r11, rdx); // move offset in rdx to offset(r11)
3334 __ mul_add(out, in, offset, len, k, tmp1, tmp2, tmp3, tmp4, tmp5, rdx, rax);
3335
3336 restore_arg_regs();
3337
3338 __ leave(); // required for proper stackwalking of RuntimeStub frame
3339 __ ret(0);
3340
3341 return start;
3342 }
3343
3344 address StubGenerator::generate_bigIntegerRightShift() {
3345 __ align(CodeEntryAlignment);
3346 StubCodeMark mark(this, "StubRoutines", "bigIntegerRightShiftWorker");
3347 address start = __ pc();
3348
3349 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3350 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3351 const Register newArr = rdi;
3352 const Register oldArr = rsi;
3353 const Register newIdx = rdx;
3354 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3355 const Register totalNumIter = r8;
3356
3357 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3358 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3359 const Register tmp1 = r11; // Caller save.
3360 const Register tmp2 = rax; // Caller save.
3361 const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9); // Windows: Callee save. Linux: Caller save.
3362 const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10); // Windows: Callee save. Linux: Caller save.
3363 const Register tmp5 = r14; // Callee save.
3364 const Register tmp6 = r15;
3365
3366 const XMMRegister x0 = xmm0;
3367 const XMMRegister x1 = xmm1;
3368 const XMMRegister x2 = xmm2;
3369
3370 BLOCK_COMMENT("Entry:");
3371 __ enter(); // required for proper stackwalking of RuntimeStub frame
3372
3373 #ifdef _WIN64
3374 setup_arg_regs(4);
3375 // For windows, since last argument is on stack, we need to move it to the appropriate register.
3376 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3377 // Save callee save registers.
3378 __ push(tmp3);
3379 __ push(tmp4);
3380 #endif
3381 __ push(tmp5);
3382
3383 // Rename temps used throughout the code.
3384 const Register idx = tmp1;
3385 const Register nIdx = tmp2;
3386
3387 __ xorl(idx, idx);
3388
3389 // Start right shift from end of the array.
3390 // For example, if #iteration = 4 and newIdx = 1
3391 // then dest[4] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
3392 // if #iteration = 4 and newIdx = 0
3393 // then dest[3] = src[4] >> shiftCount | src[3] <<< (shiftCount - 32)
3394 __ movl(idx, totalNumIter);
3395 __ movl(nIdx, idx);
3396 __ addl(nIdx, newIdx);
3397
3398 // If vectorization is enabled, check if the number of iterations is at least 64
3399 // If not, then go to ShifTwo processing 2 iterations
3400 if (VM_Version::supports_avx512_vbmi2()) {
3401 __ cmpptr(totalNumIter, (AVX3Threshold/64));
3402 __ jcc(Assembler::less, ShiftTwo);
3403
3404 if (AVX3Threshold < 16 * 64) {
3405 __ cmpl(totalNumIter, 16);
3406 __ jcc(Assembler::less, ShiftTwo);
3407 }
3408 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3409 __ subl(idx, 16);
3410 __ subl(nIdx, 16);
3411 __ BIND(Shift512Loop);
3412 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 4), Assembler::AVX_512bit);
3413 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3414 __ vpshrdvd(x2, x1, x0, Assembler::AVX_512bit);
3415 __ evmovdqul(Address(newArr, nIdx, Address::times_4), x2, Assembler::AVX_512bit);
3416 __ subl(nIdx, 16);
3417 __ subl(idx, 16);
3418 __ jcc(Assembler::greaterEqual, Shift512Loop);
3419 __ addl(idx, 16);
3420 __ addl(nIdx, 16);
3421 }
3422 __ BIND(ShiftTwo);
3423 __ cmpl(idx, 2);
3424 __ jcc(Assembler::less, ShiftOne);
3425 __ subl(idx, 2);
3426 __ subl(nIdx, 2);
3427 __ BIND(ShiftTwoLoop);
3428 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 8));
3429 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3430 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3431 __ shrdl(tmp5, tmp4);
3432 __ shrdl(tmp4, tmp3);
3433 __ movl(Address(newArr, nIdx, Address::times_4, 4), tmp5);
3434 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3435 __ subl(nIdx, 2);
3436 __ subl(idx, 2);
3437 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3438 __ addl(idx, 2);
3439 __ addl(nIdx, 2);
3440
3441 // Do the last iteration
3442 __ BIND(ShiftOne);
3443 __ cmpl(idx, 1);
3444 __ jcc(Assembler::less, Exit);
3445 __ subl(idx, 1);
3446 __ subl(nIdx, 1);
3447 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 4));
3448 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3449 __ shrdl(tmp4, tmp3);
3450 __ movl(Address(newArr, nIdx, Address::times_4), tmp4);
3451 __ BIND(Exit);
3452 __ vzeroupper();
3453 // Restore callee save registers.
3454 __ pop(tmp5);
3455 #ifdef _WIN64
3456 __ pop(tmp4);
3457 __ pop(tmp3);
3458 restore_arg_regs();
3459 #endif
3460 __ leave(); // required for proper stackwalking of RuntimeStub frame
3461 __ ret(0);
3462
3463 return start;
3464 }
3465
3466 /**
3467 * Arguments:
3468 *
3469 * Input:
3470 * c_rarg0 - newArr address
3471 * c_rarg1 - oldArr address
3472 * c_rarg2 - newIdx
3473 * c_rarg3 - shiftCount
3474 * not Win64
3475 * c_rarg4 - numIter
3476 * Win64
3477 * rsp40 - numIter
3478 */
3479 address StubGenerator::generate_bigIntegerLeftShift() {
3480 __ align(CodeEntryAlignment);
3481 StubCodeMark mark(this, "StubRoutines", "bigIntegerLeftShiftWorker");
3482 address start = __ pc();
3483
3484 Label Shift512Loop, ShiftTwo, ShiftTwoLoop, ShiftOne, Exit;
3485 // For Unix, the arguments are as follows: rdi, rsi, rdx, rcx, r8.
3486 const Register newArr = rdi;
3487 const Register oldArr = rsi;
3488 const Register newIdx = rdx;
3489 const Register shiftCount = rcx; // It was intentional to have shiftCount in rcx since it is used implicitly for shift.
3490 const Register totalNumIter = r8;
3491 // For windows, we use r9 and r10 as temps to save rdi and rsi. Thus we cannot allocate them for our temps.
3492 // For everything else, we prefer using r9 and r10 since we do not have to save them before use.
3493 const Register tmp1 = r11; // Caller save.
3494 const Register tmp2 = rax; // Caller save.
3495 const Register tmp3 = WIN64_ONLY(r12) NOT_WIN64(r9); // Windows: Callee save. Linux: Caller save.
3496 const Register tmp4 = WIN64_ONLY(r13) NOT_WIN64(r10); // Windows: Callee save. Linux: Caller save.
3497 const Register tmp5 = r14; // Callee save.
3498
3499 const XMMRegister x0 = xmm0;
3500 const XMMRegister x1 = xmm1;
3501 const XMMRegister x2 = xmm2;
3502 BLOCK_COMMENT("Entry:");
3503 __ enter(); // required for proper stackwalking of RuntimeStub frame
3504
3505 #ifdef _WIN64
3506 setup_arg_regs(4);
3507 // For windows, since last argument is on stack, we need to move it to the appropriate register.
3508 __ movl(totalNumIter, Address(rsp, 6 * wordSize));
3509 // Save callee save registers.
3510 __ push(tmp3);
3511 __ push(tmp4);
3512 #endif
3513 __ push(tmp5);
3514
3515 // Rename temps used throughout the code
3516 const Register idx = tmp1;
3517 const Register numIterTmp = tmp2;
3518
3519 // Start idx from zero.
3520 __ xorl(idx, idx);
3521 // Compute interior pointer for new array. We do this so that we can use same index for both old and new arrays.
3522 __ lea(newArr, Address(newArr, newIdx, Address::times_4));
3523 __ movl(numIterTmp, totalNumIter);
3524
3525 // If vectorization is enabled, check if the number of iterations is at least 64
3526 // If not, then go to ShiftTwo shifting two numbers at a time
3527 if (VM_Version::supports_avx512_vbmi2()) {
3528 __ cmpl(totalNumIter, (AVX3Threshold/64));
3529 __ jcc(Assembler::less, ShiftTwo);
3530
3531 if (AVX3Threshold < 16 * 64) {
3532 __ cmpl(totalNumIter, 16);
3533 __ jcc(Assembler::less, ShiftTwo);
3534 }
3535 __ evpbroadcastd(x0, shiftCount, Assembler::AVX_512bit);
3536 __ subl(numIterTmp, 16);
3537 __ BIND(Shift512Loop);
3538 __ evmovdqul(x1, Address(oldArr, idx, Address::times_4), Assembler::AVX_512bit);
3539 __ evmovdqul(x2, Address(oldArr, idx, Address::times_4, 0x4), Assembler::AVX_512bit);
3540 __ vpshldvd(x1, x2, x0, Assembler::AVX_512bit);
3541 __ evmovdqul(Address(newArr, idx, Address::times_4), x1, Assembler::AVX_512bit);
3542 __ addl(idx, 16);
3543 __ subl(numIterTmp, 16);
3544 __ jcc(Assembler::greaterEqual, Shift512Loop);
3545 __ addl(numIterTmp, 16);
3546 }
3547 __ BIND(ShiftTwo);
3548 __ cmpl(totalNumIter, 1);
3549 __ jcc(Assembler::less, Exit);
3550 __ movl(tmp3, Address(oldArr, idx, Address::times_4));
3551 __ subl(numIterTmp, 2);
3552 __ jcc(Assembler::less, ShiftOne);
3553
3554 __ BIND(ShiftTwoLoop);
3555 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3556 __ movl(tmp5, Address(oldArr, idx, Address::times_4, 0x8));
3557 __ shldl(tmp3, tmp4);
3558 __ shldl(tmp4, tmp5);
3559 __ movl(Address(newArr, idx, Address::times_4), tmp3);
3560 __ movl(Address(newArr, idx, Address::times_4, 0x4), tmp4);
3561 __ movl(tmp3, tmp5);
3562 __ addl(idx, 2);
3563 __ subl(numIterTmp, 2);
3564 __ jcc(Assembler::greaterEqual, ShiftTwoLoop);
3565
3566 // Do the last iteration
3567 __ BIND(ShiftOne);
3568 __ addl(numIterTmp, 2);
3569 __ cmpl(numIterTmp, 1);
3570 __ jcc(Assembler::less, Exit);
3571 __ movl(tmp4, Address(oldArr, idx, Address::times_4, 0x4));
3572 __ shldl(tmp3, tmp4);
3573 __ movl(Address(newArr, idx, Address::times_4), tmp3);
3574
3575 __ BIND(Exit);
3576 __ vzeroupper();
3577 // Restore callee save registers.
3578 __ pop(tmp5);
3579 #ifdef _WIN64
3580 __ pop(tmp4);
3581 __ pop(tmp3);
3582 restore_arg_regs();
3583 #endif
3584 __ leave(); // required for proper stackwalking of RuntimeStub frame
3585 __ ret(0);
3586
3587 return start;
3588 }
3589
3590 void StubGenerator::generate_libm_stubs() {
3591 if (UseLibmIntrinsic && InlineIntrinsics) {
3592 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3593 StubRoutines::_dsin = generate_libmSin(); // from stubGenerator_x86_64_sin.cpp
3594 }
3595 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3596 StubRoutines::_dcos = generate_libmCos(); // from stubGenerator_x86_64_cos.cpp
3597 }
3598 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3599 StubRoutines::_dtan = generate_libmTan(); // from stubGenerator_x86_64_tan.cpp
3600 }
3601 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3602 StubRoutines::_dexp = generate_libmExp(); // from stubGenerator_x86_64_exp.cpp
3603 }
3604 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3605 StubRoutines::_dpow = generate_libmPow(); // from stubGenerator_x86_64_pow.cpp
3606 }
3607 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3608 StubRoutines::_dlog = generate_libmLog(); // from stubGenerator_x86_64_log.cpp
3609 }
3610 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3611 StubRoutines::_dlog10 = generate_libmLog10(); // from stubGenerator_x86_64_log.cpp
3612 }
3613 }
3614 }
3615
3616 /**
3617 * Arguments:
3618 *
3619 * Input:
3620 * c_rarg0 - float16 jshort
3621 *
3622 * Output:
3623 * xmm0 - float
3624 */
3625 address StubGenerator::generate_float16ToFloat() {
3626 StubCodeMark mark(this, "StubRoutines", "float16ToFloat");
3627
3628 address start = __ pc();
3629
3630 BLOCK_COMMENT("Entry:");
3631 // No need for RuntimeStub frame since it is called only during JIT compilation
3632
3633 // Load value into xmm0 and convert
3634 __ flt16_to_flt(xmm0, c_rarg0);
3635
3636 __ ret(0);
3637
3638 return start;
3639 }
3640
3641 /**
3642 * Arguments:
3643 *
3644 * Input:
3645 * xmm0 - float
3646 *
3647 * Output:
3648 * rax - float16 jshort
3649 */
3650 address StubGenerator::generate_floatToFloat16() {
3651 StubCodeMark mark(this, "StubRoutines", "floatToFloat16");
3652
3653 address start = __ pc();
3654
3655 BLOCK_COMMENT("Entry:");
3656 // No need for RuntimeStub frame since it is called only during JIT compilation
3657
3658 // Convert and put result into rax
3659 __ flt_to_flt16(rax, xmm0, xmm1);
3660
3661 __ ret(0);
3662
3663 return start;
3664 }
3665
3666 address StubGenerator::generate_cont_thaw(const char* label, Continuation::thaw_kind kind) {
3667 if (!Continuations::enabled()) return nullptr;
3668
3669 bool return_barrier = Continuation::is_thaw_return_barrier(kind);
3670 bool return_barrier_exception = Continuation::is_thaw_return_barrier_exception(kind);
3671
3672 StubCodeMark mark(this, "StubRoutines", label);
3673 address start = __ pc();
3674
3675 // TODO: Handle Valhalla return types. May require generating different return barriers.
3676
3677 if (!return_barrier) {
3678 // Pop return address. If we don't do this, we get a drift,
3679 // where the bottom-most frozen frame continuously grows.
3680 __ pop(c_rarg3);
3681 } else {
3682 __ movptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3683 }
3684
3685 #ifdef ASSERT
3686 {
3687 Label L_good_sp;
3688 __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3689 __ jcc(Assembler::equal, L_good_sp);
3690 __ stop("Incorrect rsp at thaw entry");
3691 __ BIND(L_good_sp);
3692 }
3693 #endif // ASSERT
3694
3695 if (return_barrier) {
3696 // Preserve possible return value from a method returning to the return barrier.
3697 __ push(rax);
3698 __ push_d(xmm0);
3699 }
3700
3701 __ movptr(c_rarg0, r15_thread);
3702 __ movptr(c_rarg1, (return_barrier ? 1 : 0));
3703 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Continuation::prepare_thaw), 2);
3704 __ movptr(rbx, rax);
3705
3706 if (return_barrier) {
3707 // Restore return value from a method returning to the return barrier.
3708 // No safepoint in the call to thaw, so even an oop return value should be OK.
3709 __ pop_d(xmm0);
3710 __ pop(rax);
3711 }
3712
3713 #ifdef ASSERT
3714 {
3715 Label L_good_sp;
3716 __ cmpptr(rsp, Address(r15_thread, JavaThread::cont_entry_offset()));
3717 __ jcc(Assembler::equal, L_good_sp);
3718 __ stop("Incorrect rsp after prepare thaw");
3719 __ BIND(L_good_sp);
3720 }
3721 #endif // ASSERT
3722
3723 // rbx contains the size of the frames to thaw, 0 if overflow or no more frames
3724 Label L_thaw_success;
3725 __ testptr(rbx, rbx);
3726 __ jccb(Assembler::notZero, L_thaw_success);
3727 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
3728 __ bind(L_thaw_success);
3729
3730 // Make room for the thawed frames and align the stack.
3731 __ subptr(rsp, rbx);
3732 __ andptr(rsp, -StackAlignmentInBytes);
3733
3734 if (return_barrier) {
3735 // Preserve possible return value from a method returning to the return barrier. (Again.)
3736 __ push(rax);
3737 __ push_d(xmm0);
3738 }
3739
3740 // If we want, we can templatize thaw by kind, and have three different entries.
3741 __ movptr(c_rarg0, r15_thread);
3742 __ movptr(c_rarg1, kind);
3743 __ call_VM_leaf(Continuation::thaw_entry(), 2);
3744 __ movptr(rbx, rax);
3745
3746 if (return_barrier) {
3747 // Restore return value from a method returning to the return barrier. (Again.)
3748 // No safepoint in the call to thaw, so even an oop return value should be OK.
3749 __ pop_d(xmm0);
3750 __ pop(rax);
3751 } else {
3752 // Return 0 (success) from doYield.
3753 __ xorptr(rax, rax);
3754 }
3755
3756 // After thawing, rbx is the SP of the yielding frame.
3757 // Move there, and then to saved RBP slot.
3758 __ movptr(rsp, rbx);
3759 __ subptr(rsp, 2*wordSize);
3760
3761 if (return_barrier_exception) {
3762 __ movptr(c_rarg0, r15_thread);
3763 __ movptr(c_rarg1, Address(rsp, wordSize)); // return address
3764
3765 // rax still holds the original exception oop, save it before the call
3766 __ push(rax);
3767
3768 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), 2);
3769 __ movptr(rbx, rax);
3770
3771 // Continue at exception handler:
3772 // rax: exception oop
3773 // rbx: exception handler
3774 // rdx: exception pc
3775 __ pop(rax);
3776 __ verify_oop(rax);
3777 __ pop(rbp); // pop out RBP here too
3778 __ pop(rdx);
3779 __ jmp(rbx);
3780 } else {
3781 // We are "returning" into the topmost thawed frame; see Thaw::push_return_frame
3782 __ pop(rbp);
3783 __ ret(0);
3784 }
3785
3786 return start;
3787 }
3788
3789 address StubGenerator::generate_cont_thaw() {
3790 return generate_cont_thaw("Cont thaw", Continuation::thaw_top);
3791 }
3792
3793 // TODO: will probably need multiple return barriers depending on return type
3794
3795 address StubGenerator::generate_cont_returnBarrier() {
3796 return generate_cont_thaw("Cont thaw return barrier", Continuation::thaw_return_barrier);
3797 }
3798
3799 address StubGenerator::generate_cont_returnBarrier_exception() {
3800 return generate_cont_thaw("Cont thaw return barrier exception", Continuation::thaw_return_barrier_exception);
3801 }
3802
3803 #if INCLUDE_JFR
3804
3805 // For c2: c_rarg0 is junk, call to runtime to write a checkpoint.
3806 // It returns a jobject handle to the event writer.
3807 // The handle is dereferenced and the return value is the event writer oop.
3808 RuntimeStub* StubGenerator::generate_jfr_write_checkpoint() {
3809 enum layout {
3810 rbp_off,
3811 rbpH_off,
3812 return_off,
3813 return_off2,
3814 framesize // inclusive of return address
3815 };
3816
3817 CodeBuffer code("jfr_write_checkpoint", 1024, 64);
3818 MacroAssembler* _masm = new MacroAssembler(&code);
3819 address start = __ pc();
3820
3821 __ enter();
3822 address the_pc = __ pc();
3823
3824 int frame_complete = the_pc - start;
3825
3826 __ set_last_Java_frame(rsp, rbp, the_pc, rscratch1);
3827 __ movptr(c_rarg0, r15_thread);
3828 __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::write_checkpoint), 1);
3829 __ reset_last_Java_frame(true);
3830
3831 // rax is jobject handle result, unpack and process it through a barrier.
3832 __ resolve_global_jobject(rax, r15_thread, c_rarg0);
3833
3834 __ leave();
3835 __ ret(0);
3836
3837 OopMapSet* oop_maps = new OopMapSet();
3838 OopMap* map = new OopMap(framesize, 1);
3839 oop_maps->add_gc_map(frame_complete, map);
3840
3841 RuntimeStub* stub =
3842 RuntimeStub::new_runtime_stub(code.name(),
3843 &code,
3844 frame_complete,
3845 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3846 oop_maps,
3847 false);
3848 return stub;
3849 }
3850
3851 // For c2: call to return a leased buffer.
3852 RuntimeStub* StubGenerator::generate_jfr_return_lease() {
3853 enum layout {
3854 rbp_off,
3855 rbpH_off,
3856 return_off,
3857 return_off2,
3858 framesize // inclusive of return address
3859 };
3860
3861 CodeBuffer code("jfr_return_lease", 1024, 64);
3862 MacroAssembler* _masm = new MacroAssembler(&code);
3863 address start = __ pc();
3864
3865 __ enter();
3866 address the_pc = __ pc();
3867
3868 int frame_complete = the_pc - start;
3869
3870 __ set_last_Java_frame(rsp, rbp, the_pc, rscratch2);
3871 __ movptr(c_rarg0, r15_thread);
3872 __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::return_lease), 1);
3873 __ reset_last_Java_frame(true);
3874
3875 __ leave();
3876 __ ret(0);
3877
3878 OopMapSet* oop_maps = new OopMapSet();
3879 OopMap* map = new OopMap(framesize, 1);
3880 oop_maps->add_gc_map(frame_complete, map);
3881
3882 RuntimeStub* stub =
3883 RuntimeStub::new_runtime_stub(code.name(),
3884 &code,
3885 frame_complete,
3886 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3887 oop_maps,
3888 false);
3889 return stub;
3890 }
3891
3892 #endif // INCLUDE_JFR
3893
3894 // Continuation point for throwing of implicit exceptions that are
3895 // not handled in the current activation. Fabricates an exception
3896 // oop and initiates normal exception dispatching in this
3897 // frame. Since we need to preserve callee-saved values (currently
3898 // only for C2, but done for C1 as well) we need a callee-saved oop
3899 // map and therefore have to make these stubs into RuntimeStubs
3900 // rather than BufferBlobs. If the compiler needs all registers to
3901 // be preserved between the fault point and the exception handler
3902 // then it must assume responsibility for that in
3903 // AbstractCompiler::continuation_for_implicit_null_exception or
3904 // continuation_for_implicit_division_by_zero_exception. All other
3905 // implicit exceptions (e.g., NullPointerException or
3906 // AbstractMethodError on entry) are either at call sites or
3907 // otherwise assume that stack unwinding will be initiated, so
3908 // caller saved registers were assumed volatile in the compiler.
3909 address StubGenerator::generate_throw_exception(const char* name,
3910 address runtime_entry,
3911 Register arg1,
3912 Register arg2) {
3913 // Information about frame layout at time of blocking runtime call.
3914 // Note that we only have to preserve callee-saved registers since
3915 // the compilers are responsible for supplying a continuation point
3916 // if they expect all registers to be preserved.
3917 enum layout {
3918 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
3919 rbp_off2,
3920 return_off,
3921 return_off2,
3922 framesize // inclusive of return address
3923 };
3924
3925 int insts_size = 512;
3926 int locs_size = 64;
3927
3928 CodeBuffer code(name, insts_size, locs_size);
3929 OopMapSet* oop_maps = new OopMapSet();
3930 MacroAssembler* _masm = new MacroAssembler(&code);
3931
3932 address start = __ pc();
3933
3934 // This is an inlined and slightly modified version of call_VM
3935 // which has the ability to fetch the return PC out of
3936 // thread-local storage and also sets up last_Java_sp slightly
3937 // differently than the real call_VM
3938
3939 __ enter(); // required for proper stackwalking of RuntimeStub frame
3940
3941 assert(is_even(framesize/2), "sp not 16-byte aligned");
3942
3943 // return address and rbp are already in place
3944 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
3945
3946 int frame_complete = __ pc() - start;
3947
3948 // Set up last_Java_sp and last_Java_fp
3949 address the_pc = __ pc();
3950 __ set_last_Java_frame(rsp, rbp, the_pc, rscratch1);
3951 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
3952
3953 // Call runtime
3954 if (arg1 != noreg) {
3955 assert(arg2 != c_rarg1, "clobbered");
3956 __ movptr(c_rarg1, arg1);
3957 }
3958 if (arg2 != noreg) {
3959 __ movptr(c_rarg2, arg2);
3960 }
3961 __ movptr(c_rarg0, r15_thread);
3962 BLOCK_COMMENT("call runtime_entry");
3963 __ call(RuntimeAddress(runtime_entry));
3964
3965 // Generate oop map
3966 OopMap* map = new OopMap(framesize, 0);
3967
3968 oop_maps->add_gc_map(the_pc - start, map);
3969
3970 __ reset_last_Java_frame(true);
3971
3972 __ leave(); // required for proper stackwalking of RuntimeStub frame
3973
3974 // check for pending exceptions
3975 #ifdef ASSERT
3976 Label L;
3977 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
3978 __ jcc(Assembler::notEqual, L);
3979 __ should_not_reach_here();
3980 __ bind(L);
3981 #endif // ASSERT
3982 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3983
3984
3985 // codeBlob framesize is in words (not VMRegImpl::slot_size)
3986 RuntimeStub* stub =
3987 RuntimeStub::new_runtime_stub(name,
3988 &code,
3989 frame_complete,
3990 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3991 oop_maps, false);
3992 return stub->entry_point();
3993 }
3994
3995 // exception handler for upcall stubs
3996 address StubGenerator::generate_upcall_stub_exception_handler() {
3997 StubCodeMark mark(this, "StubRoutines", "upcall stub exception handler");
3998 address start = __ pc();
3999
4000 // native caller has no idea how to handle exceptions
4001 // we just crash here. Up to callee to catch exceptions.
4002 __ verify_oop(rax);
4003 __ vzeroupper();
4004 __ mov(c_rarg0, rax);
4005 __ andptr(rsp, -StackAlignmentInBytes); // align stack as required by ABI
4006 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
4007 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, UpcallLinker::handle_uncaught_exception)));
4008 __ should_not_reach_here();
4009
4010 return start;
4011 }
4012
4013 void StubGenerator::create_control_words() {
4014 // Round to nearest, 64-bit mode, exceptions masked, flags specialized
4015 StubRoutines::x86::_mxcsr_std = EnableX86ECoreOpts ? 0x1FBF : 0x1F80;
4016 // Round to zero, 64-bit mode, exceptions masked, flags specialized
4017 StubRoutines::x86::_mxcsr_rz = EnableX86ECoreOpts ? 0x7FBF : 0x7F80;
4018 }
4019
4020 // Initialization
4021 void StubGenerator::generate_initial_stubs() {
4022 // Generates all stubs and initializes the entry points
4023
4024 // This platform-specific settings are needed by generate_call_stub()
4025 create_control_words();
4026
4027 // Initialize table for unsafe copy memeory check.
4028 if (UnsafeCopyMemory::_table == nullptr) {
4029 UnsafeCopyMemory::create_table(16);
4030 }
4031
4032 // entry points that exist in all platforms Note: This is code
4033 // that could be shared among different platforms - however the
4034 // benefit seems to be smaller than the disadvantage of having a
4035 // much more complicated generator structure. See also comment in
4036 // stubRoutines.hpp.
4037
4038 StubRoutines::_forward_exception_entry = generate_forward_exception();
4039
4040 // Generate these first because they are called from other stubs
4041 if (InlineTypeReturnedAsFields) {
4042 StubRoutines::_load_inline_type_fields_in_regs =
4043 generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::load_inline_type_fields_in_regs),
4044 "load_inline_type_fields_in_regs", false);
4045 StubRoutines::_store_inline_type_fields_to_buf =
4046 generate_return_value_stub(CAST_FROM_FN_PTR(address, SharedRuntime::store_inline_type_fields_to_buf),
4047 "store_inline_type_fields_to_buf", true);
4048 }
4049
4050 StubRoutines::_call_stub_entry =
4051 generate_call_stub(StubRoutines::_call_stub_return_address);
4052
4053 // is referenced by megamorphic call
4054 StubRoutines::_catch_exception_entry = generate_catch_exception();
4055
4056 // atomic calls
4057 StubRoutines::_fence_entry = generate_orderaccess_fence();
4058
4059 // platform dependent
4060 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
4061
4062 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
4063
4064 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
4065 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
4066 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
4067 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
4068
4069 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
4070 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
4071 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4072 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
4073
4074 // Build this early so it's available for the interpreter.
4075 StubRoutines::_throw_StackOverflowError_entry =
4076 generate_throw_exception("StackOverflowError throw_exception",
4077 CAST_FROM_FN_PTR(address,
4078 SharedRuntime::
4079 throw_StackOverflowError));
4080 StubRoutines::_throw_delayed_StackOverflowError_entry =
4081 generate_throw_exception("delayed StackOverflowError throw_exception",
4082 CAST_FROM_FN_PTR(address,
4083 SharedRuntime::
4084 throw_delayed_StackOverflowError));
4085 if (UseCRC32Intrinsics) {
4086 // set table address before stub generation which use it
4087 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
4088 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
4089 }
4090
4091 if (UseCRC32CIntrinsics) {
4092 bool supports_clmul = VM_Version::supports_clmul();
4093 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
4094 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
4095 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
4096 }
4097
4098 if (VM_Version::supports_float16()) {
4099 // For results consistency both intrinsics should be enabled.
4100 // vmIntrinsics checks InlineIntrinsics flag, no need to check it here.
4101 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_float16ToFloat) &&
4102 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_floatToFloat16)) {
4103 StubRoutines::_hf2f = generate_float16ToFloat();
4104 StubRoutines::_f2hf = generate_floatToFloat16();
4105 }
4106 }
4107
4108 generate_libm_stubs();
4109
4110 StubRoutines::_fmod = generate_libmFmod(); // from stubGenerator_x86_64_fmod.cpp
4111 }
4112
4113 // Call here from the interpreter or compiled code to either load
4114 // multiple returned values from the inline type instance being
4115 // returned to registers or to store returned values to a newly
4116 // allocated inline type instance.
4117 // Register is a class, but it would be assigned numerical value.
4118 // "0" is assigned for xmm0. Thus we need to ignore -Wnonnull.
4119 PRAGMA_DIAG_PUSH
4120 PRAGMA_NONNULL_IGNORED
4121 address StubGenerator::generate_return_value_stub(address destination, const char* name, bool has_res) {
4122 // We need to save all registers the calling convention may use so
4123 // the runtime calls read or update those registers. This needs to
4124 // be in sync with SharedRuntime::java_return_convention().
4125 enum layout {
4126 pad_off = frame::arg_reg_save_area_bytes/BytesPerInt, pad_off_2,
4127 rax_off, rax_off_2,
4128 j_rarg5_off, j_rarg5_2,
4129 j_rarg4_off, j_rarg4_2,
4130 j_rarg3_off, j_rarg3_2,
4131 j_rarg2_off, j_rarg2_2,
4132 j_rarg1_off, j_rarg1_2,
4133 j_rarg0_off, j_rarg0_2,
4134 j_farg0_off, j_farg0_2,
4135 j_farg1_off, j_farg1_2,
4136 j_farg2_off, j_farg2_2,
4137 j_farg3_off, j_farg3_2,
4138 j_farg4_off, j_farg4_2,
4139 j_farg5_off, j_farg5_2,
4140 j_farg6_off, j_farg6_2,
4141 j_farg7_off, j_farg7_2,
4142 rbp_off, rbp_off_2,
4143 return_off, return_off_2,
4144
4145 framesize
4146 };
4147
4148 CodeBuffer buffer(name, 1000, 512);
4149 MacroAssembler* _masm = new MacroAssembler(&buffer);
4150
4151 int frame_size_in_bytes = align_up(framesize*BytesPerInt, 16);
4152 assert(frame_size_in_bytes == framesize*BytesPerInt, "misaligned");
4153 int frame_size_in_slots = frame_size_in_bytes / BytesPerInt;
4154 int frame_size_in_words = frame_size_in_bytes / wordSize;
4155
4156 OopMapSet *oop_maps = new OopMapSet();
4157 OopMap* map = new OopMap(frame_size_in_slots, 0);
4158
4159 map->set_callee_saved(VMRegImpl::stack2reg(rax_off), rax->as_VMReg());
4160 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg5_off), j_rarg5->as_VMReg());
4161 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg4_off), j_rarg4->as_VMReg());
4162 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg3_off), j_rarg3->as_VMReg());
4163 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg2_off), j_rarg2->as_VMReg());
4164 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg1_off), j_rarg1->as_VMReg());
4165 map->set_callee_saved(VMRegImpl::stack2reg(j_rarg0_off), j_rarg0->as_VMReg());
4166 map->set_callee_saved(VMRegImpl::stack2reg(j_farg0_off), j_farg0->as_VMReg());
4167 map->set_callee_saved(VMRegImpl::stack2reg(j_farg1_off), j_farg1->as_VMReg());
4168 map->set_callee_saved(VMRegImpl::stack2reg(j_farg2_off), j_farg2->as_VMReg());
4169 map->set_callee_saved(VMRegImpl::stack2reg(j_farg3_off), j_farg3->as_VMReg());
4170 map->set_callee_saved(VMRegImpl::stack2reg(j_farg4_off), j_farg4->as_VMReg());
4171 map->set_callee_saved(VMRegImpl::stack2reg(j_farg5_off), j_farg5->as_VMReg());
4172 map->set_callee_saved(VMRegImpl::stack2reg(j_farg6_off), j_farg6->as_VMReg());
4173 map->set_callee_saved(VMRegImpl::stack2reg(j_farg7_off), j_farg7->as_VMReg());
4174
4175 int start = __ offset();
4176
4177 __ subptr(rsp, frame_size_in_bytes - 8 /* return address*/);
4178
4179 __ movptr(Address(rsp, rbp_off * BytesPerInt), rbp);
4180 __ movdbl(Address(rsp, j_farg7_off * BytesPerInt), j_farg7);
4181 __ movdbl(Address(rsp, j_farg6_off * BytesPerInt), j_farg6);
4182 __ movdbl(Address(rsp, j_farg5_off * BytesPerInt), j_farg5);
4183 __ movdbl(Address(rsp, j_farg4_off * BytesPerInt), j_farg4);
4184 __ movdbl(Address(rsp, j_farg3_off * BytesPerInt), j_farg3);
4185 __ movdbl(Address(rsp, j_farg2_off * BytesPerInt), j_farg2);
4186 __ movdbl(Address(rsp, j_farg1_off * BytesPerInt), j_farg1);
4187 __ movdbl(Address(rsp, j_farg0_off * BytesPerInt), j_farg0);
4188
4189 __ movptr(Address(rsp, j_rarg0_off * BytesPerInt), j_rarg0);
4190 __ movptr(Address(rsp, j_rarg1_off * BytesPerInt), j_rarg1);
4191 __ movptr(Address(rsp, j_rarg2_off * BytesPerInt), j_rarg2);
4192 __ movptr(Address(rsp, j_rarg3_off * BytesPerInt), j_rarg3);
4193 __ movptr(Address(rsp, j_rarg4_off * BytesPerInt), j_rarg4);
4194 __ movptr(Address(rsp, j_rarg5_off * BytesPerInt), j_rarg5);
4195 __ movptr(Address(rsp, rax_off * BytesPerInt), rax);
4196
4197 int frame_complete = __ offset();
4198
4199 __ set_last_Java_frame(noreg, noreg, nullptr, rscratch1);
4200
4201 __ mov(c_rarg0, r15_thread);
4202 __ mov(c_rarg1, rax);
4203
4204 __ call(RuntimeAddress(destination));
4205
4206 // Set an oopmap for the call site.
4207
4208 oop_maps->add_gc_map( __ offset() - start, map);
4209
4210 // clear last_Java_sp
4211 __ reset_last_Java_frame(false);
4212
4213 __ movptr(rbp, Address(rsp, rbp_off * BytesPerInt));
4214 __ movdbl(j_farg7, Address(rsp, j_farg7_off * BytesPerInt));
4215 __ movdbl(j_farg6, Address(rsp, j_farg6_off * BytesPerInt));
4216 __ movdbl(j_farg5, Address(rsp, j_farg5_off * BytesPerInt));
4217 __ movdbl(j_farg4, Address(rsp, j_farg4_off * BytesPerInt));
4218 __ movdbl(j_farg3, Address(rsp, j_farg3_off * BytesPerInt));
4219 __ movdbl(j_farg2, Address(rsp, j_farg2_off * BytesPerInt));
4220 __ movdbl(j_farg1, Address(rsp, j_farg1_off * BytesPerInt));
4221 __ movdbl(j_farg0, Address(rsp, j_farg0_off * BytesPerInt));
4222
4223 __ movptr(j_rarg0, Address(rsp, j_rarg0_off * BytesPerInt));
4224 __ movptr(j_rarg1, Address(rsp, j_rarg1_off * BytesPerInt));
4225 __ movptr(j_rarg2, Address(rsp, j_rarg2_off * BytesPerInt));
4226 __ movptr(j_rarg3, Address(rsp, j_rarg3_off * BytesPerInt));
4227 __ movptr(j_rarg4, Address(rsp, j_rarg4_off * BytesPerInt));
4228 __ movptr(j_rarg5, Address(rsp, j_rarg5_off * BytesPerInt));
4229 __ movptr(rax, Address(rsp, rax_off * BytesPerInt));
4230
4231 __ addptr(rsp, frame_size_in_bytes-8);
4232
4233 // check for pending exceptions
4234 Label pending;
4235 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
4236 __ jcc(Assembler::notEqual, pending);
4237
4238 if (has_res) {
4239 __ get_vm_result(rax, r15_thread);
4240 }
4241
4242 __ ret(0);
4243
4244 __ bind(pending);
4245
4246 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
4247 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
4248
4249 // -------------
4250 // make sure all code is generated
4251 _masm->flush();
4252
4253 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, false);
4254 return stub->entry_point();
4255 }
4256
4257 void StubGenerator::generate_continuation_stubs() {
4258 // Continuation stubs:
4259 StubRoutines::_cont_thaw = generate_cont_thaw();
4260 StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
4261 StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
4262
4263 JFR_ONLY(generate_jfr_stubs();)
4264 }
4265
4266 #if INCLUDE_JFR
4267 void StubGenerator::generate_jfr_stubs() {
4268 StubRoutines::_jfr_write_checkpoint_stub = generate_jfr_write_checkpoint();
4269 StubRoutines::_jfr_write_checkpoint = StubRoutines::_jfr_write_checkpoint_stub->entry_point();
4270 StubRoutines::_jfr_return_lease_stub = generate_jfr_return_lease();
4271 StubRoutines::_jfr_return_lease = StubRoutines::_jfr_return_lease_stub->entry_point();
4272 }
4273 #endif
4274
4275 void StubGenerator::generate_final_stubs() {
4276 // Generates the rest of stubs and initializes the entry points
4277
4278 // These entry points require SharedInfo::stack0 to be set up in
4279 // non-core builds and need to be relocatable, so they each
4280 // fabricate a RuntimeStub internally.
4281 StubRoutines::_throw_AbstractMethodError_entry =
4282 generate_throw_exception("AbstractMethodError throw_exception",
4283 CAST_FROM_FN_PTR(address,
4284 SharedRuntime::
4285 throw_AbstractMethodError));
4286
4287 StubRoutines::_throw_IncompatibleClassChangeError_entry =
4288 generate_throw_exception("IncompatibleClassChangeError throw_exception",
4289 CAST_FROM_FN_PTR(address,
4290 SharedRuntime::
4291 throw_IncompatibleClassChangeError));
4292
4293 StubRoutines::_throw_NullPointerException_at_call_entry =
4294 generate_throw_exception("NullPointerException at call throw_exception",
4295 CAST_FROM_FN_PTR(address,
4296 SharedRuntime::
4297 throw_NullPointerException_at_call));
4298
4299 // support for verify_oop (must happen after universe_init)
4300 if (VerifyOops) {
4301 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
4302 }
4303
4304 // data cache line writeback
4305 StubRoutines::_data_cache_writeback = generate_data_cache_writeback();
4306 StubRoutines::_data_cache_writeback_sync = generate_data_cache_writeback_sync();
4307
4308 // arraycopy stubs used by compilers
4309 generate_arraycopy_stubs();
4310
4311 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
4312 if (bs_nm != nullptr) {
4313 StubRoutines::_method_entry_barrier = generate_method_entry_barrier();
4314 }
4315
4316 if (UseVectorizedMismatchIntrinsic) {
4317 StubRoutines::_vectorizedMismatch = generate_vectorizedMismatch();
4318 }
4319
4320 StubRoutines::_upcall_stub_exception_handler = generate_upcall_stub_exception_handler();
4321 }
4322
4323 void StubGenerator::generate_compiler_stubs() {
4324 #if COMPILER2_OR_JVMCI
4325
4326 // Entry points that are C2 compiler specific.
4327
4328 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF7FFFFFFF);
4329 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x8000000080000000);
4330 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask("vector_double_sign_mask", 0x7FFFFFFFFFFFFFFF);
4331 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask("vector_double_sign_flip", 0x8000000000000000);
4332 StubRoutines::x86::_vector_all_bits_set = generate_vector_mask("vector_all_bits_set", 0xFFFFFFFFFFFFFFFF);
4333 StubRoutines::x86::_vector_int_mask_cmp_bits = generate_vector_mask("vector_int_mask_cmp_bits", 0x0000000100000001);
4334 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff00ff00ff);
4335 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
4336 StubRoutines::x86::_vector_int_to_byte_mask = generate_vector_mask("vector_int_to_byte_mask", 0x000000ff000000ff);
4337 StubRoutines::x86::_vector_int_to_short_mask = generate_vector_mask("vector_int_to_short_mask", 0x0000ffff0000ffff);
4338 StubRoutines::x86::_vector_32_bit_mask = generate_vector_custom_i32("vector_32_bit_mask", Assembler::AVX_512bit,
4339 0xFFFFFFFF, 0, 0, 0);
4340 StubRoutines::x86::_vector_64_bit_mask = generate_vector_custom_i32("vector_64_bit_mask", Assembler::AVX_512bit,
4341 0xFFFFFFFF, 0xFFFFFFFF, 0, 0);
4342 StubRoutines::x86::_vector_int_shuffle_mask = generate_vector_mask("vector_int_shuffle_mask", 0x0302010003020100);
4343 StubRoutines::x86::_vector_byte_shuffle_mask = generate_vector_byte_shuffle_mask("vector_byte_shuffle_mask");
4344 StubRoutines::x86::_vector_short_shuffle_mask = generate_vector_mask("vector_short_shuffle_mask", 0x0100010001000100);
4345 StubRoutines::x86::_vector_long_shuffle_mask = generate_vector_mask("vector_long_shuffle_mask", 0x0000000100000000);
4346 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask("vector_long_sign_mask", 0x8000000000000000);
4347 StubRoutines::x86::_vector_iota_indices = generate_iota_indices("iota_indices");
4348 StubRoutines::x86::_vector_count_leading_zeros_lut = generate_count_leading_zeros_lut("count_leading_zeros_lut");
4349 StubRoutines::x86::_vector_reverse_bit_lut = generate_vector_reverse_bit_lut("reverse_bit_lut");
4350 StubRoutines::x86::_vector_reverse_byte_perm_mask_long = generate_vector_reverse_byte_perm_mask_long("perm_mask_long");
4351 StubRoutines::x86::_vector_reverse_byte_perm_mask_int = generate_vector_reverse_byte_perm_mask_int("perm_mask_int");
4352 StubRoutines::x86::_vector_reverse_byte_perm_mask_short = generate_vector_reverse_byte_perm_mask_short("perm_mask_short");
4353
4354 if (VM_Version::supports_avx2() && !VM_Version::supports_avx512vl()) {
4355 StubRoutines::x86::_compress_perm_table32 = generate_compress_perm_table("compress_perm_table32", 32);
4356 StubRoutines::x86::_compress_perm_table64 = generate_compress_perm_table("compress_perm_table64", 64);
4357 StubRoutines::x86::_expand_perm_table32 = generate_expand_perm_table("expand_perm_table32", 32);
4358 StubRoutines::x86::_expand_perm_table64 = generate_expand_perm_table("expand_perm_table64", 64);
4359 }
4360
4361 if (VM_Version::supports_avx2() && !VM_Version::supports_avx512_vpopcntdq()) {
4362 // lut implementation influenced by counting 1s algorithm from section 5-1 of Hackers' Delight.
4363 StubRoutines::x86::_vector_popcount_lut = generate_popcount_avx_lut("popcount_lut");
4364 }
4365
4366 generate_aes_stubs();
4367
4368 generate_ghash_stubs();
4369
4370 generate_chacha_stubs();
4371
4372 if (UseAdler32Intrinsics) {
4373 StubRoutines::_updateBytesAdler32 = generate_updateBytesAdler32();
4374 }
4375
4376 if (UsePoly1305Intrinsics) {
4377 StubRoutines::_poly1305_processBlocks = generate_poly1305_processBlocks();
4378 }
4379
4380 if (UseMD5Intrinsics) {
4381 StubRoutines::_md5_implCompress = generate_md5_implCompress(false, "md5_implCompress");
4382 StubRoutines::_md5_implCompressMB = generate_md5_implCompress(true, "md5_implCompressMB");
4383 }
4384
4385 if (UseSHA1Intrinsics) {
4386 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
4387 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
4388 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
4389 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
4390 }
4391
4392 if (UseSHA256Intrinsics) {
4393 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
4394 char* dst = (char*)StubRoutines::x86::_k256_W;
4395 char* src = (char*)StubRoutines::x86::_k256;
4396 for (int ii = 0; ii < 16; ++ii) {
4397 memcpy(dst + 32 * ii, src + 16 * ii, 16);
4398 memcpy(dst + 32 * ii + 16, src + 16 * ii, 16);
4399 }
4400 StubRoutines::x86::_k256_W_adr = (address)StubRoutines::x86::_k256_W;
4401 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
4402 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
4403 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
4404 }
4405
4406 if (UseSHA512Intrinsics) {
4407 StubRoutines::x86::_k512_W_addr = (address)StubRoutines::x86::_k512_W;
4408 StubRoutines::x86::_pshuffle_byte_flip_mask_addr_sha512 = generate_pshuffle_byte_flip_mask_sha512();
4409 StubRoutines::_sha512_implCompress = generate_sha512_implCompress(false, "sha512_implCompress");
4410 StubRoutines::_sha512_implCompressMB = generate_sha512_implCompress(true, "sha512_implCompressMB");
4411 }
4412
4413 if (UseBASE64Intrinsics) {
4414 if(VM_Version::supports_avx2()) {
4415 StubRoutines::x86::_avx2_shuffle_base64 = base64_avx2_shuffle_addr();
4416 StubRoutines::x86::_avx2_input_mask_base64 = base64_avx2_input_mask_addr();
4417 StubRoutines::x86::_avx2_lut_base64 = base64_avx2_lut_addr();
4418 StubRoutines::x86::_avx2_decode_tables_base64 = base64_AVX2_decode_tables_addr();
4419 StubRoutines::x86::_avx2_decode_lut_tables_base64 = base64_AVX2_decode_LUT_tables_addr();
4420 }
4421 StubRoutines::x86::_encoding_table_base64 = base64_encoding_table_addr();
4422 if (VM_Version::supports_avx512_vbmi()) {
4423 StubRoutines::x86::_shuffle_base64 = base64_shuffle_addr();
4424 StubRoutines::x86::_lookup_lo_base64 = base64_vbmi_lookup_lo_addr();
4425 StubRoutines::x86::_lookup_hi_base64 = base64_vbmi_lookup_hi_addr();
4426 StubRoutines::x86::_lookup_lo_base64url = base64_vbmi_lookup_lo_url_addr();
4427 StubRoutines::x86::_lookup_hi_base64url = base64_vbmi_lookup_hi_url_addr();
4428 StubRoutines::x86::_pack_vec_base64 = base64_vbmi_pack_vec_addr();
4429 StubRoutines::x86::_join_0_1_base64 = base64_vbmi_join_0_1_addr();
4430 StubRoutines::x86::_join_1_2_base64 = base64_vbmi_join_1_2_addr();
4431 StubRoutines::x86::_join_2_3_base64 = base64_vbmi_join_2_3_addr();
4432 }
4433 StubRoutines::x86::_decoding_table_base64 = base64_decoding_table_addr();
4434 StubRoutines::_base64_encodeBlock = generate_base64_encodeBlock();
4435 StubRoutines::_base64_decodeBlock = generate_base64_decodeBlock();
4436 }
4437
4438 #ifdef COMPILER2
4439 if (UseMultiplyToLenIntrinsic) {
4440 StubRoutines::_multiplyToLen = generate_multiplyToLen();
4441 }
4442 if (UseSquareToLenIntrinsic) {
4443 StubRoutines::_squareToLen = generate_squareToLen();
4444 }
4445 if (UseMulAddIntrinsic) {
4446 StubRoutines::_mulAdd = generate_mulAdd();
4447 }
4448 if (VM_Version::supports_avx512_vbmi2()) {
4449 StubRoutines::_bigIntegerRightShiftWorker = generate_bigIntegerRightShift();
4450 StubRoutines::_bigIntegerLeftShiftWorker = generate_bigIntegerLeftShift();
4451 }
4452 if (UseMontgomeryMultiplyIntrinsic) {
4453 StubRoutines::_montgomeryMultiply
4454 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_multiply);
4455 }
4456 if (UseMontgomerySquareIntrinsic) {
4457 StubRoutines::_montgomerySquare
4458 = CAST_FROM_FN_PTR(address, SharedRuntime::montgomery_square);
4459 }
4460
4461 // Load x86_64_sort library on supported hardware to enable SIMD sort and partition intrinsics
4462
4463 if (VM_Version::is_intel() && (VM_Version::supports_avx512dq() || VM_Version::supports_avx2())) {
4464 void *libsimdsort = nullptr;
4465 char ebuf_[1024];
4466 char dll_name_simd_sort[JVM_MAXPATHLEN];
4467 if (os::dll_locate_lib(dll_name_simd_sort, sizeof(dll_name_simd_sort), Arguments::get_dll_dir(), "simdsort")) {
4468 libsimdsort = os::dll_load(dll_name_simd_sort, ebuf_, sizeof ebuf_);
4469 }
4470 // Get addresses for SIMD sort and partition routines
4471 if (libsimdsort != nullptr) {
4472 log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "simdsort" JNI_LIB_SUFFIX, p2i(libsimdsort));
4473
4474 snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512dq() ? "avx512_sort" : "avx2_sort");
4475 StubRoutines::_array_sort = (address)os::dll_lookup(libsimdsort, ebuf_);
4476
4477 snprintf(ebuf_, sizeof(ebuf_), VM_Version::supports_avx512dq() ? "avx512_partition" : "avx2_partition");
4478 StubRoutines::_array_partition = (address)os::dll_lookup(libsimdsort, ebuf_);
4479 }
4480 }
4481
4482 // Get svml stub routine addresses
4483 void *libjsvml = nullptr;
4484 char ebuf[1024];
4485 char dll_name[JVM_MAXPATHLEN];
4486 if (os::dll_locate_lib(dll_name, sizeof(dll_name), Arguments::get_dll_dir(), "jsvml")) {
4487 libjsvml = os::dll_load(dll_name, ebuf, sizeof ebuf);
4488 }
4489 if (libjsvml != nullptr) {
4490 // SVML method naming convention
4491 // All the methods are named as __jsvml_op<T><N>_ha_<VV>
4492 // Where:
4493 // ha stands for high accuracy
4494 // <T> is optional to indicate float/double
4495 // Set to f for vector float operation
4496 // Omitted for vector double operation
4497 // <N> is the number of elements in the vector
4498 // 1, 2, 4, 8, 16
4499 // e.g. 128 bit float vector has 4 float elements
4500 // <VV> indicates the avx/sse level:
4501 // z0 is AVX512, l9 is AVX2, e9 is AVX1 and ex is for SSE2
4502 // e.g. __jsvml_expf16_ha_z0 is the method for computing 16 element vector float exp using AVX 512 insns
4503 // __jsvml_exp8_ha_z0 is the method for computing 8 element vector double exp using AVX 512 insns
4504
4505 log_info(library)("Loaded library %s, handle " INTPTR_FORMAT, JNI_LIB_PREFIX "jsvml" JNI_LIB_SUFFIX, p2i(libjsvml));
4506 if (UseAVX > 2) {
4507 for (int op = 0; op < VectorSupport::NUM_SVML_OP; op++) {
4508 int vop = VectorSupport::VECTOR_OP_SVML_START + op;
4509 if ((!VM_Version::supports_avx512dq()) &&
4510 (vop == VectorSupport::VECTOR_OP_LOG || vop == VectorSupport::VECTOR_OP_LOG10 || vop == VectorSupport::VECTOR_OP_POW)) {
4511 continue;
4512 }
4513 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf16_ha_z0", VectorSupport::svmlname[op]);
4514 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4515
4516 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s8_ha_z0", VectorSupport::svmlname[op]);
4517 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_512][op] = (address)os::dll_lookup(libjsvml, ebuf);
4518 }
4519 }
4520 const char* avx_sse_str = (UseAVX >= 2) ? "l9" : ((UseAVX == 1) ? "e9" : "ex");
4521 for (int op = 0; op < VectorSupport::NUM_SVML_OP; op++) {
4522 int vop = VectorSupport::VECTOR_OP_SVML_START + op;
4523 if (vop == VectorSupport::VECTOR_OP_POW) {
4524 continue;
4525 }
4526 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4527 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4528
4529 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4530 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4531
4532 snprintf(ebuf, sizeof(ebuf), "__jsvml_%sf8_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4533 StubRoutines::_vector_f_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4534
4535 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s1_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4536 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_64][op] = (address)os::dll_lookup(libjsvml, ebuf);
4537
4538 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s2_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4539 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_128][op] = (address)os::dll_lookup(libjsvml, ebuf);
4540
4541 snprintf(ebuf, sizeof(ebuf), "__jsvml_%s4_ha_%s", VectorSupport::svmlname[op], avx_sse_str);
4542 StubRoutines::_vector_d_math[VectorSupport::VEC_SIZE_256][op] = (address)os::dll_lookup(libjsvml, ebuf);
4543 }
4544 }
4545 #endif // COMPILER2
4546 #endif // COMPILER2_OR_JVMCI
4547 }
4548
4549 StubGenerator::StubGenerator(CodeBuffer* code, StubsKind kind) : StubCodeGenerator(code) {
4550 DEBUG_ONLY( _regs_in_thread = false; )
4551 switch(kind) {
4552 case Initial_stubs:
4553 generate_initial_stubs();
4554 break;
4555 case Continuation_stubs:
4556 generate_continuation_stubs();
4557 break;
4558 case Compiler_stubs:
4559 generate_compiler_stubs();
4560 break;
4561 case Final_stubs:
4562 generate_final_stubs();
4563 break;
4564 default:
4565 fatal("unexpected stubs kind: %d", kind);
4566 break;
4567 };
4568 }
4569
4570 void StubGenerator_generate(CodeBuffer* code, StubCodeGenerator::StubsKind kind) {
4571 StubGenerator g(code, kind);
4572 }
4573
4574 #undef __