1 /*
2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "gc/shared/barrierSet.hpp"
29 #include "gc/shared/barrierSetCodeGen.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "nativeInst_x86.hpp"
32 #include "oops/instanceOop.hpp"
33 #include "oops/method.hpp"
34 #include "oops/objArrayKlass.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/methodHandles.hpp"
37 #include "runtime/frame.inline.hpp"
38 #include "runtime/handles.inline.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubCodeGenerator.hpp"
41 #include "runtime/stubRoutines.hpp"
42 #include "runtime/thread.inline.hpp"
43 #ifdef COMPILER2
44 #include "opto/runtime.hpp"
45 #endif
46
47 // Declaration and definition of StubGenerator (no .hpp file).
48 // For a more detailed description of the stub routine structure
49 // see the comment in stubRoutines.hpp
50
51 #define __ _masm->
52 #define a__ ((Assembler*)_masm)->
53
54 #ifdef PRODUCT
55 #define BLOCK_COMMENT(str) /* nothing */
56 #else
57 #define BLOCK_COMMENT(str) __ block_comment(str)
58 #endif
59
60 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
61
62 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
63 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
64
65 // -------------------------------------------------------------------------------------------------------------------------
66 // Stub Code definitions
67
68 class StubGenerator: public StubCodeGenerator {
69 private:
70
71 #ifdef PRODUCT
72 #define inc_counter_np(counter) ((void)0)
73 #else
74 void inc_counter_np_(int& counter) {
75 __ incrementl(ExternalAddress((address)&counter));
76 }
77 #define inc_counter_np(counter) \
78 BLOCK_COMMENT("inc_counter " #counter); \
79 inc_counter_np_(counter);
80 #endif //PRODUCT
81
82 void inc_copy_counter_np(BasicType t) {
83 #ifndef PRODUCT
84 switch (t) {
85 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
86 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
87 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
88 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
89 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
90 }
91 ShouldNotReachHere();
92 #endif //PRODUCT
93 }
94
95 //------------------------------------------------------------------------------------------------------------------------
96 // Call stubs are used to call Java from C
97 //
98 // [ return_from_Java ] <--- rsp
99 // [ argument word n ]
100 // ...
101 // -N [ argument word 1 ]
102 // -7 [ Possible padding for stack alignment ]
103 // -6 [ Possible padding for stack alignment ]
104 // -5 [ Possible padding for stack alignment ]
105 // -4 [ mxcsr save ] <--- rsp_after_call
106 // -3 [ saved rbx, ]
107 // -2 [ saved rsi ]
108 // -1 [ saved rdi ]
109 // 0 [ saved rbp, ] <--- rbp,
110 // 1 [ return address ]
111 // 2 [ ptr. to call wrapper ]
112 // 3 [ result ]
113 // 4 [ result_type ]
114 // 5 [ method ]
115 // 6 [ entry_point ]
116 // 7 [ parameters ]
117 // 8 [ parameter_size ]
118 // 9 [ thread ]
119
120
121 address generate_call_stub(address& return_address) {
122 StubCodeMark mark(this, "StubRoutines", "call_stub");
123 address start = __ pc();
124
125 // stub code parameters / addresses
126 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
127 bool sse_save = false;
128 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
129 const int locals_count_in_bytes (4*wordSize);
130 const Address mxcsr_save (rbp, -4 * wordSize);
131 const Address saved_rbx (rbp, -3 * wordSize);
132 const Address saved_rsi (rbp, -2 * wordSize);
133 const Address saved_rdi (rbp, -1 * wordSize);
134 const Address result (rbp, 3 * wordSize);
135 const Address result_type (rbp, 4 * wordSize);
136 const Address method (rbp, 5 * wordSize);
137 const Address entry_point (rbp, 6 * wordSize);
138 const Address parameters (rbp, 7 * wordSize);
139 const Address parameter_size(rbp, 8 * wordSize);
140 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
141 sse_save = UseSSE > 0;
142
143 // stub code
144 __ enter();
145 __ movptr(rcx, parameter_size); // parameter counter
146 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
147 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
148 __ subptr(rsp, rcx);
149 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
150
151 // save rdi, rsi, & rbx, according to C calling conventions
152 __ movptr(saved_rdi, rdi);
153 __ movptr(saved_rsi, rsi);
154 __ movptr(saved_rbx, rbx);
155
156 // provide initial value for required masks
157 if (UseAVX > 2) {
158 __ movl(rbx, 0xffff);
159 __ kmovwl(k1, rbx);
160 }
161
162 // save and initialize %mxcsr
163 if (sse_save) {
164 Label skip_ldmx;
165 __ stmxcsr(mxcsr_save);
166 __ movl(rax, mxcsr_save);
167 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
168 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
169 __ cmp32(rax, mxcsr_std);
170 __ jcc(Assembler::equal, skip_ldmx);
171 __ ldmxcsr(mxcsr_std);
172 __ bind(skip_ldmx);
173 }
174
175 // make sure the control word is correct.
176 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
177
178 #ifdef ASSERT
179 // make sure we have no pending exceptions
180 { Label L;
181 __ movptr(rcx, thread);
182 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
183 __ jcc(Assembler::equal, L);
184 __ stop("StubRoutines::call_stub: entered with pending exception");
185 __ bind(L);
186 }
187 #endif
188
189 // pass parameters if any
190 BLOCK_COMMENT("pass parameters if any");
191 Label parameters_done;
192 __ movl(rcx, parameter_size); // parameter counter
193 __ testl(rcx, rcx);
194 __ jcc(Assembler::zero, parameters_done);
195
196 // parameter passing loop
197
198 Label loop;
199 // Copy Java parameters in reverse order (receiver last)
200 // Note that the argument order is inverted in the process
201 // source is rdx[rcx: N-1..0]
202 // dest is rsp[rbx: 0..N-1]
203
204 __ movptr(rdx, parameters); // parameter pointer
205 __ xorptr(rbx, rbx);
206
207 __ BIND(loop);
208
209 // get parameter
210 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
211 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
212 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
213 __ increment(rbx);
214 __ decrement(rcx);
215 __ jcc(Assembler::notZero, loop);
216
217 // call Java function
218 __ BIND(parameters_done);
219 __ movptr(rbx, method); // get Method*
220 __ movptr(rax, entry_point); // get entry_point
221 __ mov(rsi, rsp); // set sender sp
222 BLOCK_COMMENT("call Java function");
223 __ call(rax);
224
225 BLOCK_COMMENT("call_stub_return_address:");
226 return_address = __ pc();
227
228 #ifdef COMPILER2
229 {
230 Label L_skip;
231 if (UseSSE >= 2) {
232 __ verify_FPU(0, "call_stub_return");
233 } else {
234 for (int i = 1; i < 8; i++) {
235 __ ffree(i);
236 }
237
238 // UseSSE <= 1 so double result should be left on TOS
239 __ movl(rsi, result_type);
240 __ cmpl(rsi, T_DOUBLE);
241 __ jcc(Assembler::equal, L_skip);
242 if (UseSSE == 0) {
243 // UseSSE == 0 so float result should be left on TOS
244 __ cmpl(rsi, T_FLOAT);
245 __ jcc(Assembler::equal, L_skip);
246 }
247 __ ffree(0);
248 }
249 __ BIND(L_skip);
250 }
251 #endif // COMPILER2
252
253 // store result depending on type
254 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
255 __ movptr(rdi, result);
256 Label is_long, is_float, is_double, exit;
257 __ movl(rsi, result_type);
258 __ cmpl(rsi, T_LONG);
259 __ jcc(Assembler::equal, is_long);
260 __ cmpl(rsi, T_FLOAT);
261 __ jcc(Assembler::equal, is_float);
262 __ cmpl(rsi, T_DOUBLE);
263 __ jcc(Assembler::equal, is_double);
264
265 // handle T_INT case
266 __ movl(Address(rdi, 0), rax);
267 __ BIND(exit);
268
269 // check that FPU stack is empty
270 __ verify_FPU(0, "generate_call_stub");
271
272 // pop parameters
273 __ lea(rsp, rsp_after_call);
274
275 // restore %mxcsr
276 if (sse_save) {
277 __ ldmxcsr(mxcsr_save);
278 }
279
280 // restore rdi, rsi and rbx,
281 __ movptr(rbx, saved_rbx);
282 __ movptr(rsi, saved_rsi);
283 __ movptr(rdi, saved_rdi);
284 __ addptr(rsp, 4*wordSize);
285
286 // return
287 __ pop(rbp);
288 __ ret(0);
289
290 // handle return types different from T_INT
291 __ BIND(is_long);
292 __ movl(Address(rdi, 0 * wordSize), rax);
293 __ movl(Address(rdi, 1 * wordSize), rdx);
294 __ jmp(exit);
295
296 __ BIND(is_float);
297 // interpreter uses xmm0 for return values
298 if (UseSSE >= 1) {
299 __ movflt(Address(rdi, 0), xmm0);
300 } else {
301 __ fstp_s(Address(rdi, 0));
302 }
303 __ jmp(exit);
304
305 __ BIND(is_double);
306 // interpreter uses xmm0 for return values
307 if (UseSSE >= 2) {
308 __ movdbl(Address(rdi, 0), xmm0);
309 } else {
310 __ fstp_d(Address(rdi, 0));
311 }
312 __ jmp(exit);
313
314 return start;
315 }
316
317
318 //------------------------------------------------------------------------------------------------------------------------
319 // Return point for a Java call if there's an exception thrown in Java code.
320 // The exception is caught and transformed into a pending exception stored in
321 // JavaThread that can be tested from within the VM.
322 //
323 // Note: Usually the parameters are removed by the callee. In case of an exception
324 // crossing an activation frame boundary, that is not the case if the callee
325 // is compiled code => need to setup the rsp.
326 //
327 // rax,: exception oop
328
329 address generate_catch_exception() {
330 StubCodeMark mark(this, "StubRoutines", "catch_exception");
331 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
332 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
333 address start = __ pc();
334
335 // get thread directly
336 __ movptr(rcx, thread);
337 #ifdef ASSERT
338 // verify that threads correspond
339 { Label L;
340 __ get_thread(rbx);
341 __ cmpptr(rbx, rcx);
342 __ jcc(Assembler::equal, L);
343 __ stop("StubRoutines::catch_exception: threads must correspond");
344 __ bind(L);
345 }
346 #endif
347 // set pending exception
348 __ verify_oop(rax);
349 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
350 __ lea(Address(rcx, Thread::exception_file_offset ()),
351 ExternalAddress((address)__FILE__));
352 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
353 // complete return to VM
354 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
355 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
356
357 return start;
358 }
359
360
361 //------------------------------------------------------------------------------------------------------------------------
362 // Continuation point for runtime calls returning with a pending exception.
363 // The pending exception check happened in the runtime or native call stub.
364 // The pending exception in Thread is converted into a Java-level exception.
365 //
366 // Contract with Java-level exception handlers:
367 // rax: exception
368 // rdx: throwing pc
369 //
370 // NOTE: At entry of this stub, exception-pc must be on stack !!
371
372 address generate_forward_exception() {
373 StubCodeMark mark(this, "StubRoutines", "forward exception");
374 address start = __ pc();
375 const Register thread = rcx;
376
377 // other registers used in this stub
378 const Register exception_oop = rax;
379 const Register handler_addr = rbx;
380 const Register exception_pc = rdx;
381
382 // Upon entry, the sp points to the return address returning into Java
383 // (interpreted or compiled) code; i.e., the return address becomes the
384 // throwing pc.
385 //
386 // Arguments pushed before the runtime call are still on the stack but
387 // the exception handler will reset the stack pointer -> ignore them.
388 // A potential result in registers can be ignored as well.
389
390 #ifdef ASSERT
391 // make sure this code is only executed if there is a pending exception
392 { Label L;
393 __ get_thread(thread);
394 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
395 __ jcc(Assembler::notEqual, L);
396 __ stop("StubRoutines::forward exception: no pending exception (1)");
397 __ bind(L);
398 }
399 #endif
400
401 // compute exception handler into rbx,
402 __ get_thread(thread);
403 __ movptr(exception_pc, Address(rsp, 0));
404 BLOCK_COMMENT("call exception_handler_for_return_address");
405 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
406 __ mov(handler_addr, rax);
407
408 // setup rax & rdx, remove return address & clear pending exception
409 __ get_thread(thread);
410 __ pop(exception_pc);
411 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
412 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
413
414 #ifdef ASSERT
415 // make sure exception is set
416 { Label L;
417 __ testptr(exception_oop, exception_oop);
418 __ jcc(Assembler::notEqual, L);
419 __ stop("StubRoutines::forward exception: no pending exception (2)");
420 __ bind(L);
421 }
422 #endif
423
424 // Verify that there is really a valid exception in RAX.
425 __ verify_oop(exception_oop);
426
427 // continue at exception handler (return address removed)
428 // rax: exception
429 // rbx: exception handler
430 // rdx: throwing pc
431 __ jmp(handler_addr);
432
433 return start;
434 }
435
436
437 //----------------------------------------------------------------------------------------------------
438 // Support for int32_t Atomic::xchg(int32_t exchange_value, volatile int32_t* dest)
439 //
440 // xchg exists as far back as 8086, lock needed for MP only
441 // Stack layout immediately after call:
442 //
443 // 0 [ret addr ] <--- rsp
444 // 1 [ ex ]
445 // 2 [ dest ]
446 //
447 // Result: *dest <- ex, return (old *dest)
448 //
449 // Note: win32 does not currently use this code
450
451 address generate_atomic_xchg() {
452 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
453 address start = __ pc();
454
455 __ push(rdx);
456 Address exchange(rsp, 2 * wordSize);
457 Address dest_addr(rsp, 3 * wordSize);
458 __ movl(rax, exchange);
459 __ movptr(rdx, dest_addr);
460 __ xchgl(rax, Address(rdx, 0));
461 __ pop(rdx);
462 __ ret(0);
463
464 return start;
465 }
466
467 //----------------------------------------------------------------------------------------------------
468 // Support for void verify_mxcsr()
469 //
470 // This routine is used with -Xcheck:jni to verify that native
471 // JNI code does not return to Java code without restoring the
472 // MXCSR register to our expected state.
473
474
475 address generate_verify_mxcsr() {
476 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
477 address start = __ pc();
478
479 const Address mxcsr_save(rsp, 0);
480
481 if (CheckJNICalls && UseSSE > 0 ) {
482 Label ok_ret;
483 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
484 __ push(rax);
485 __ subptr(rsp, wordSize); // allocate a temp location
486 __ stmxcsr(mxcsr_save);
487 __ movl(rax, mxcsr_save);
488 __ andl(rax, MXCSR_MASK);
489 __ cmp32(rax, mxcsr_std);
490 __ jcc(Assembler::equal, ok_ret);
491
492 __ warn("MXCSR changed by native JNI code.");
493
494 __ ldmxcsr(mxcsr_std);
495
496 __ bind(ok_ret);
497 __ addptr(rsp, wordSize);
498 __ pop(rax);
499 }
500
501 __ ret(0);
502
503 return start;
504 }
505
506
507 //---------------------------------------------------------------------------
508 // Support for void verify_fpu_cntrl_wrd()
509 //
510 // This routine is used with -Xcheck:jni to verify that native
511 // JNI code does not return to Java code without restoring the
512 // FP control word to our expected state.
513
514 address generate_verify_fpu_cntrl_wrd() {
515 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
516 address start = __ pc();
517
518 const Address fpu_cntrl_wrd_save(rsp, 0);
519
520 if (CheckJNICalls) {
521 Label ok_ret;
522 __ push(rax);
523 __ subptr(rsp, wordSize); // allocate a temp location
524 __ fnstcw(fpu_cntrl_wrd_save);
525 __ movl(rax, fpu_cntrl_wrd_save);
526 __ andl(rax, FPU_CNTRL_WRD_MASK);
527 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
528 __ cmp32(rax, fpu_std);
529 __ jcc(Assembler::equal, ok_ret);
530
531 __ warn("Floating point control word changed by native JNI code.");
532
533 __ fldcw(fpu_std);
534
535 __ bind(ok_ret);
536 __ addptr(rsp, wordSize);
537 __ pop(rax);
538 }
539
540 __ ret(0);
541
542 return start;
543 }
544
545 //---------------------------------------------------------------------------
546 // Wrapper for slow-case handling of double-to-integer conversion
547 // d2i or f2i fast case failed either because it is nan or because
548 // of under/overflow.
549 // Input: FPU TOS: float value
550 // Output: rax, (rdx): integer (long) result
551
552 address generate_d2i_wrapper(BasicType t, address fcn) {
553 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
554 address start = __ pc();
555
556 // Capture info about frame layout
557 enum layout { FPUState_off = 0,
558 rbp_off = FPUStateSizeInWords,
559 rdi_off,
560 rsi_off,
561 rcx_off,
562 rbx_off,
563 saved_argument_off,
564 saved_argument_off2, // 2nd half of double
565 framesize
566 };
567
568 assert(FPUStateSizeInWords == 27, "update stack layout");
569
570 // Save outgoing argument to stack across push_FPU_state()
571 __ subptr(rsp, wordSize * 2);
572 __ fstp_d(Address(rsp, 0));
573
574 // Save CPU & FPU state
575 __ push(rbx);
576 __ push(rcx);
577 __ push(rsi);
578 __ push(rdi);
579 __ push(rbp);
580 __ push_FPU_state();
581
582 // push_FPU_state() resets the FP top of stack
583 // Load original double into FP top of stack
584 __ fld_d(Address(rsp, saved_argument_off * wordSize));
585 // Store double into stack as outgoing argument
586 __ subptr(rsp, wordSize*2);
587 __ fst_d(Address(rsp, 0));
588
589 // Prepare FPU for doing math in C-land
590 __ empty_FPU_stack();
591 // Call the C code to massage the double. Result in EAX
592 if (t == T_INT)
593 { BLOCK_COMMENT("SharedRuntime::d2i"); }
594 else if (t == T_LONG)
595 { BLOCK_COMMENT("SharedRuntime::d2l"); }
596 __ call_VM_leaf( fcn, 2 );
597
598 // Restore CPU & FPU state
599 __ pop_FPU_state();
600 __ pop(rbp);
601 __ pop(rdi);
602 __ pop(rsi);
603 __ pop(rcx);
604 __ pop(rbx);
605 __ addptr(rsp, wordSize * 2);
606
607 __ ret(0);
608
609 return start;
610 }
611
612
613 //----------------------------------------------------------------------------------------------------
614 // Non-destructive plausibility checks for oops
615
616 address generate_verify_oop() {
617 StubCodeMark mark(this, "StubRoutines", "verify_oop");
618 address start = __ pc();
619
620 // Incoming arguments on stack after saving rax,:
621 //
622 // [tos ]: saved rdx
623 // [tos + 1]: saved EFLAGS
624 // [tos + 2]: return address
625 // [tos + 3]: char* error message
626 // [tos + 4]: oop object to verify
627 // [tos + 5]: saved rax, - saved by caller and bashed
628
629 Label exit, error;
630 __ pushf();
631 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
632 __ push(rdx); // save rdx
633 // make sure object is 'reasonable'
634 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
635 __ testptr(rax, rax);
636 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
637
638 // Check if the oop is in the right area of memory
639 const int oop_mask = Universe::verify_oop_mask();
640 const int oop_bits = Universe::verify_oop_bits();
641 __ mov(rdx, rax);
642 __ andptr(rdx, oop_mask);
643 __ cmpptr(rdx, oop_bits);
644 __ jcc(Assembler::notZero, error);
645
646 // make sure klass is 'reasonable', which is not zero.
647 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
648 __ testptr(rax, rax);
649 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
650
651 // return if everything seems ok
652 __ bind(exit);
653 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
654 __ pop(rdx); // restore rdx
655 __ popf(); // restore EFLAGS
656 __ ret(3 * wordSize); // pop arguments
657
658 // handle errors
659 __ bind(error);
660 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
661 __ pop(rdx); // get saved rdx back
662 __ popf(); // get saved EFLAGS off stack -- will be ignored
663 __ pusha(); // push registers (eip = return address & msg are already pushed)
664 BLOCK_COMMENT("call MacroAssembler::debug");
665 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
666 __ popa();
667 __ ret(3 * wordSize); // pop arguments
668 return start;
669 }
670
671
672 // Copy 64 bytes chunks
673 //
674 // Inputs:
675 // from - source array address
676 // to_from - destination array address - from
677 // qword_count - 8-bytes element count, negative
678 //
679 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
680 assert( UseSSE >= 2, "supported cpu only" );
681 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
682 if (UseAVX > 2) {
683 __ push(rbx);
684 __ movl(rbx, 0xffff);
685 __ kmovwl(k1, rbx);
686 __ pop(rbx);
687 }
688 // Copy 64-byte chunks
689 __ jmpb(L_copy_64_bytes);
690 __ align(OptoLoopAlignment);
691 __ BIND(L_copy_64_bytes_loop);
692
693 if (UseUnalignedLoadStores) {
694 if (UseAVX > 2) {
695 __ evmovdqul(xmm0, Address(from, 0), Assembler::AVX_512bit);
696 __ evmovdqul(Address(from, to_from, Address::times_1, 0), xmm0, Assembler::AVX_512bit);
697 } else if (UseAVX == 2) {
698 __ vmovdqu(xmm0, Address(from, 0));
699 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
700 __ vmovdqu(xmm1, Address(from, 32));
701 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
702 } else {
703 __ movdqu(xmm0, Address(from, 0));
704 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
705 __ movdqu(xmm1, Address(from, 16));
706 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
707 __ movdqu(xmm2, Address(from, 32));
708 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
709 __ movdqu(xmm3, Address(from, 48));
710 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
711 }
712 } else {
713 __ movq(xmm0, Address(from, 0));
714 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
715 __ movq(xmm1, Address(from, 8));
716 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
717 __ movq(xmm2, Address(from, 16));
718 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
719 __ movq(xmm3, Address(from, 24));
720 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
721 __ movq(xmm4, Address(from, 32));
722 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
723 __ movq(xmm5, Address(from, 40));
724 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
725 __ movq(xmm6, Address(from, 48));
726 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
727 __ movq(xmm7, Address(from, 56));
728 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
729 }
730
731 __ addl(from, 64);
732 __ BIND(L_copy_64_bytes);
733 __ subl(qword_count, 8);
734 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
735
736 if (UseUnalignedLoadStores && (UseAVX == 2)) {
737 // clean upper bits of YMM registers
738 __ vpxor(xmm0, xmm0);
739 __ vpxor(xmm1, xmm1);
740 }
741 __ addl(qword_count, 8);
742 __ jccb(Assembler::zero, L_exit);
743 //
744 // length is too short, just copy qwords
745 //
746 __ BIND(L_copy_8_bytes);
747 __ movq(xmm0, Address(from, 0));
748 __ movq(Address(from, to_from, Address::times_1), xmm0);
749 __ addl(from, 8);
750 __ decrement(qword_count);
751 __ jcc(Assembler::greater, L_copy_8_bytes);
752 __ BIND(L_exit);
753 }
754
755 // Copy 64 bytes chunks
756 //
757 // Inputs:
758 // from - source array address
759 // to_from - destination array address - from
760 // qword_count - 8-bytes element count, negative
761 //
762 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
763 assert( VM_Version::supports_mmx(), "supported cpu only" );
764 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
765 // Copy 64-byte chunks
766 __ jmpb(L_copy_64_bytes);
767 __ align(OptoLoopAlignment);
768 __ BIND(L_copy_64_bytes_loop);
769 __ movq(mmx0, Address(from, 0));
770 __ movq(mmx1, Address(from, 8));
771 __ movq(mmx2, Address(from, 16));
772 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
773 __ movq(mmx3, Address(from, 24));
774 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
775 __ movq(mmx4, Address(from, 32));
776 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
777 __ movq(mmx5, Address(from, 40));
778 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
779 __ movq(mmx6, Address(from, 48));
780 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
781 __ movq(mmx7, Address(from, 56));
782 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
783 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
784 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
785 __ addptr(from, 64);
786 __ BIND(L_copy_64_bytes);
787 __ subl(qword_count, 8);
788 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
789 __ addl(qword_count, 8);
790 __ jccb(Assembler::zero, L_exit);
791 //
792 // length is too short, just copy qwords
793 //
794 __ BIND(L_copy_8_bytes);
795 __ movq(mmx0, Address(from, 0));
796 __ movq(Address(from, to_from, Address::times_1), mmx0);
797 __ addptr(from, 8);
798 __ decrement(qword_count);
799 __ jcc(Assembler::greater, L_copy_8_bytes);
800 __ BIND(L_exit);
801 __ emms();
802 }
803
804 address generate_disjoint_copy(BasicType t, bool aligned,
805 Address::ScaleFactor sf,
806 address* entry, const char *name,
807 bool dest_uninitialized = false) {
808 __ align(CodeEntryAlignment);
809 StubCodeMark mark(this, "StubRoutines", name);
810 address start = __ pc();
811
812 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
813 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
814
815 int shift = Address::times_ptr - sf;
816
817 const Register from = rsi; // source array address
818 const Register to = rdi; // destination array address
819 const Register count = rcx; // elements count
820 const Register to_from = to; // (to - from)
821 const Register saved_to = rdx; // saved destination array address
822
823 __ enter(); // required for proper stackwalking of RuntimeStub frame
824 __ push(rsi);
825 __ push(rdi);
826 __ movptr(from , Address(rsp, 12+ 4));
827 __ movptr(to , Address(rsp, 12+ 8));
828 __ movl(count, Address(rsp, 12+ 12));
829
830 if (entry != NULL) {
831 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
832 BLOCK_COMMENT("Entry:");
833 }
834
835 if (t == T_OBJECT) {
836 __ testl(count, count);
837 __ jcc(Assembler::zero, L_0_count);
838 }
839
840 BarrierSetCodeGen *bs = Universe::heap()->barrier_set()->code_gen();
841 DecoratorSet decorators = ARRAYCOPY_DISJOINT;
842 if (dest_uninitialized) {
843 decorators |= AS_DEST_NOT_INITIALIZED;
844 }
845 if (aligned) {
846 decorators |= ARRAYCOPY_ALIGNED;
847 }
848
849 bs->arraycopy_prologue(_masm, decorators, t, from, to, count);
850
851 __ subptr(to, from); // to --> to_from
852 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
853 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
854 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
855 // align source address at 4 bytes address boundary
856 if (t == T_BYTE) {
857 // One byte misalignment happens only for byte arrays
858 __ testl(from, 1);
859 __ jccb(Assembler::zero, L_skip_align1);
860 __ movb(rax, Address(from, 0));
861 __ movb(Address(from, to_from, Address::times_1, 0), rax);
862 __ increment(from);
863 __ decrement(count);
864 __ BIND(L_skip_align1);
865 }
866 // Two bytes misalignment happens only for byte and short (char) arrays
867 __ testl(from, 2);
868 __ jccb(Assembler::zero, L_skip_align2);
869 __ movw(rax, Address(from, 0));
870 __ movw(Address(from, to_from, Address::times_1, 0), rax);
871 __ addptr(from, 2);
872 __ subl(count, 1<<(shift-1));
873 __ BIND(L_skip_align2);
874 }
875 if (!VM_Version::supports_mmx()) {
876 __ mov(rax, count); // save 'count'
877 __ shrl(count, shift); // bytes count
878 __ addptr(to_from, from);// restore 'to'
879 __ rep_mov();
880 __ subptr(to_from, from);// restore 'to_from'
881 __ mov(count, rax); // restore 'count'
882 __ jmpb(L_copy_2_bytes); // all dwords were copied
883 } else {
884 if (!UseUnalignedLoadStores) {
885 // align to 8 bytes, we know we are 4 byte aligned to start
886 __ testptr(from, 4);
887 __ jccb(Assembler::zero, L_copy_64_bytes);
888 __ movl(rax, Address(from, 0));
889 __ movl(Address(from, to_from, Address::times_1, 0), rax);
890 __ addptr(from, 4);
891 __ subl(count, 1<<shift);
892 }
893 __ BIND(L_copy_64_bytes);
894 __ mov(rax, count);
895 __ shrl(rax, shift+1); // 8 bytes chunk count
896 //
897 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
898 //
899 if (UseXMMForArrayCopy) {
900 xmm_copy_forward(from, to_from, rax);
901 } else {
902 mmx_copy_forward(from, to_from, rax);
903 }
904 }
905 // copy tailing dword
906 __ BIND(L_copy_4_bytes);
907 __ testl(count, 1<<shift);
908 __ jccb(Assembler::zero, L_copy_2_bytes);
909 __ movl(rax, Address(from, 0));
910 __ movl(Address(from, to_from, Address::times_1, 0), rax);
911 if (t == T_BYTE || t == T_SHORT) {
912 __ addptr(from, 4);
913 __ BIND(L_copy_2_bytes);
914 // copy tailing word
915 __ testl(count, 1<<(shift-1));
916 __ jccb(Assembler::zero, L_copy_byte);
917 __ movw(rax, Address(from, 0));
918 __ movw(Address(from, to_from, Address::times_1, 0), rax);
919 if (t == T_BYTE) {
920 __ addptr(from, 2);
921 __ BIND(L_copy_byte);
922 // copy tailing byte
923 __ testl(count, 1);
924 __ jccb(Assembler::zero, L_exit);
925 __ movb(rax, Address(from, 0));
926 __ movb(Address(from, to_from, Address::times_1, 0), rax);
927 __ BIND(L_exit);
928 } else {
929 __ BIND(L_copy_byte);
930 }
931 } else {
932 __ BIND(L_copy_2_bytes);
933 }
934
935 __ movl(count, Address(rsp, 12+12)); // reread 'count'
936 bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
937
938 if (t == T_OBJECT) {
939 __ BIND(L_0_count);
940 }
941 inc_copy_counter_np(t);
942 __ pop(rdi);
943 __ pop(rsi);
944 __ leave(); // required for proper stackwalking of RuntimeStub frame
945 __ vzeroupper();
946 __ xorptr(rax, rax); // return 0
947 __ ret(0);
948 return start;
949 }
950
951
952 address generate_fill(BasicType t, bool aligned, const char *name) {
953 __ align(CodeEntryAlignment);
954 StubCodeMark mark(this, "StubRoutines", name);
955 address start = __ pc();
956
957 BLOCK_COMMENT("Entry:");
958
959 const Register to = rdi; // source array address
960 const Register value = rdx; // value
961 const Register count = rsi; // elements count
962
963 __ enter(); // required for proper stackwalking of RuntimeStub frame
964 __ push(rsi);
965 __ push(rdi);
966 __ movptr(to , Address(rsp, 12+ 4));
967 __ movl(value, Address(rsp, 12+ 8));
968 __ movl(count, Address(rsp, 12+ 12));
969
970 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
971
972 __ pop(rdi);
973 __ pop(rsi);
974 __ leave(); // required for proper stackwalking of RuntimeStub frame
975 __ ret(0);
976 return start;
977 }
978
979 address generate_conjoint_copy(BasicType t, bool aligned,
980 Address::ScaleFactor sf,
981 address nooverlap_target,
982 address* entry, const char *name,
983 bool dest_uninitialized = false) {
984 __ align(CodeEntryAlignment);
985 StubCodeMark mark(this, "StubRoutines", name);
986 address start = __ pc();
987
988 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
989 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
990
991 int shift = Address::times_ptr - sf;
992
993 const Register src = rax; // source array address
994 const Register dst = rdx; // destination array address
995 const Register from = rsi; // source array address
996 const Register to = rdi; // destination array address
997 const Register count = rcx; // elements count
998 const Register end = rax; // array end address
999
1000 __ enter(); // required for proper stackwalking of RuntimeStub frame
1001 __ push(rsi);
1002 __ push(rdi);
1003 __ movptr(src , Address(rsp, 12+ 4)); // from
1004 __ movptr(dst , Address(rsp, 12+ 8)); // to
1005 __ movl2ptr(count, Address(rsp, 12+12)); // count
1006
1007 if (entry != NULL) {
1008 *entry = __ pc(); // Entry point from generic arraycopy stub.
1009 BLOCK_COMMENT("Entry:");
1010 }
1011
1012 // nooverlap_target expects arguments in rsi and rdi.
1013 __ mov(from, src);
1014 __ mov(to , dst);
1015
1016 // arrays overlap test: dispatch to disjoint stub if necessary.
1017 RuntimeAddress nooverlap(nooverlap_target);
1018 __ cmpptr(dst, src);
1019 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1020 __ jump_cc(Assembler::belowEqual, nooverlap);
1021 __ cmpptr(dst, end);
1022 __ jump_cc(Assembler::aboveEqual, nooverlap);
1023
1024 if (t == T_OBJECT) {
1025 __ testl(count, count);
1026 __ jcc(Assembler::zero, L_0_count);
1027 }
1028
1029 BarrierSetCodeGen *bs = Universe::heap()->barrier_set()->code_gen();
1030 DecoratorSet decorators = 0;
1031 if (dest_uninitialized) {
1032 decorators |= AS_DEST_NOT_INITIALIZED;
1033 }
1034 if (aligned) {
1035 decorators |= ARRAYCOPY_ALIGNED;
1036 }
1037 bs->arraycopy_prologue(_masm, decorators, t, from, to, count);
1038
1039 // copy from high to low
1040 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1041 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1042 if (t == T_BYTE || t == T_SHORT) {
1043 // Align the end of destination array at 4 bytes address boundary
1044 __ lea(end, Address(dst, count, sf, 0));
1045 if (t == T_BYTE) {
1046 // One byte misalignment happens only for byte arrays
1047 __ testl(end, 1);
1048 __ jccb(Assembler::zero, L_skip_align1);
1049 __ decrement(count);
1050 __ movb(rdx, Address(from, count, sf, 0));
1051 __ movb(Address(to, count, sf, 0), rdx);
1052 __ BIND(L_skip_align1);
1053 }
1054 // Two bytes misalignment happens only for byte and short (char) arrays
1055 __ testl(end, 2);
1056 __ jccb(Assembler::zero, L_skip_align2);
1057 __ subptr(count, 1<<(shift-1));
1058 __ movw(rdx, Address(from, count, sf, 0));
1059 __ movw(Address(to, count, sf, 0), rdx);
1060 __ BIND(L_skip_align2);
1061 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1062 __ jcc(Assembler::below, L_copy_4_bytes);
1063 }
1064
1065 if (!VM_Version::supports_mmx()) {
1066 __ std();
1067 __ mov(rax, count); // Save 'count'
1068 __ mov(rdx, to); // Save 'to'
1069 __ lea(rsi, Address(from, count, sf, -4));
1070 __ lea(rdi, Address(to , count, sf, -4));
1071 __ shrptr(count, shift); // bytes count
1072 __ rep_mov();
1073 __ cld();
1074 __ mov(count, rax); // restore 'count'
1075 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1076 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1077 __ mov(to, rdx); // restore 'to'
1078 __ jmpb(L_copy_2_bytes); // all dword were copied
1079 } else {
1080 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1081 __ testptr(end, 4);
1082 __ jccb(Assembler::zero, L_copy_8_bytes);
1083 __ subl(count, 1<<shift);
1084 __ movl(rdx, Address(from, count, sf, 0));
1085 __ movl(Address(to, count, sf, 0), rdx);
1086 __ jmpb(L_copy_8_bytes);
1087
1088 __ align(OptoLoopAlignment);
1089 // Move 8 bytes
1090 __ BIND(L_copy_8_bytes_loop);
1091 if (UseXMMForArrayCopy) {
1092 __ movq(xmm0, Address(from, count, sf, 0));
1093 __ movq(Address(to, count, sf, 0), xmm0);
1094 } else {
1095 __ movq(mmx0, Address(from, count, sf, 0));
1096 __ movq(Address(to, count, sf, 0), mmx0);
1097 }
1098 __ BIND(L_copy_8_bytes);
1099 __ subl(count, 2<<shift);
1100 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1101 __ addl(count, 2<<shift);
1102 if (!UseXMMForArrayCopy) {
1103 __ emms();
1104 }
1105 }
1106 __ BIND(L_copy_4_bytes);
1107 // copy prefix qword
1108 __ testl(count, 1<<shift);
1109 __ jccb(Assembler::zero, L_copy_2_bytes);
1110 __ movl(rdx, Address(from, count, sf, -4));
1111 __ movl(Address(to, count, sf, -4), rdx);
1112
1113 if (t == T_BYTE || t == T_SHORT) {
1114 __ subl(count, (1<<shift));
1115 __ BIND(L_copy_2_bytes);
1116 // copy prefix dword
1117 __ testl(count, 1<<(shift-1));
1118 __ jccb(Assembler::zero, L_copy_byte);
1119 __ movw(rdx, Address(from, count, sf, -2));
1120 __ movw(Address(to, count, sf, -2), rdx);
1121 if (t == T_BYTE) {
1122 __ subl(count, 1<<(shift-1));
1123 __ BIND(L_copy_byte);
1124 // copy prefix byte
1125 __ testl(count, 1);
1126 __ jccb(Assembler::zero, L_exit);
1127 __ movb(rdx, Address(from, 0));
1128 __ movb(Address(to, 0), rdx);
1129 __ BIND(L_exit);
1130 } else {
1131 __ BIND(L_copy_byte);
1132 }
1133 } else {
1134 __ BIND(L_copy_2_bytes);
1135 }
1136
1137 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1138 bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
1139
1140 if (t == T_OBJECT) {
1141 __ BIND(L_0_count);
1142 }
1143 inc_copy_counter_np(t);
1144 __ pop(rdi);
1145 __ pop(rsi);
1146 __ leave(); // required for proper stackwalking of RuntimeStub frame
1147 __ xorptr(rax, rax); // return 0
1148 __ ret(0);
1149 return start;
1150 }
1151
1152
1153 address generate_disjoint_long_copy(address* entry, const char *name) {
1154 __ align(CodeEntryAlignment);
1155 StubCodeMark mark(this, "StubRoutines", name);
1156 address start = __ pc();
1157
1158 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1159 const Register from = rax; // source array address
1160 const Register to = rdx; // destination array address
1161 const Register count = rcx; // elements count
1162 const Register to_from = rdx; // (to - from)
1163
1164 __ enter(); // required for proper stackwalking of RuntimeStub frame
1165 __ movptr(from , Address(rsp, 8+0)); // from
1166 __ movptr(to , Address(rsp, 8+4)); // to
1167 __ movl2ptr(count, Address(rsp, 8+8)); // count
1168
1169 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1170 BLOCK_COMMENT("Entry:");
1171
1172 __ subptr(to, from); // to --> to_from
1173 if (VM_Version::supports_mmx()) {
1174 if (UseXMMForArrayCopy) {
1175 xmm_copy_forward(from, to_from, count);
1176 } else {
1177 mmx_copy_forward(from, to_from, count);
1178 }
1179 } else {
1180 __ jmpb(L_copy_8_bytes);
1181 __ align(OptoLoopAlignment);
1182 __ BIND(L_copy_8_bytes_loop);
1183 __ fild_d(Address(from, 0));
1184 __ fistp_d(Address(from, to_from, Address::times_1));
1185 __ addptr(from, 8);
1186 __ BIND(L_copy_8_bytes);
1187 __ decrement(count);
1188 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1189 }
1190 inc_copy_counter_np(T_LONG);
1191 __ leave(); // required for proper stackwalking of RuntimeStub frame
1192 __ vzeroupper();
1193 __ xorptr(rax, rax); // return 0
1194 __ ret(0);
1195 return start;
1196 }
1197
1198 address generate_conjoint_long_copy(address nooverlap_target,
1199 address* entry, const char *name) {
1200 __ align(CodeEntryAlignment);
1201 StubCodeMark mark(this, "StubRoutines", name);
1202 address start = __ pc();
1203
1204 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1205 const Register from = rax; // source array address
1206 const Register to = rdx; // destination array address
1207 const Register count = rcx; // elements count
1208 const Register end_from = rax; // source array end address
1209
1210 __ enter(); // required for proper stackwalking of RuntimeStub frame
1211 __ movptr(from , Address(rsp, 8+0)); // from
1212 __ movptr(to , Address(rsp, 8+4)); // to
1213 __ movl2ptr(count, Address(rsp, 8+8)); // count
1214
1215 *entry = __ pc(); // Entry point from generic arraycopy stub.
1216 BLOCK_COMMENT("Entry:");
1217
1218 // arrays overlap test
1219 __ cmpptr(to, from);
1220 RuntimeAddress nooverlap(nooverlap_target);
1221 __ jump_cc(Assembler::belowEqual, nooverlap);
1222 __ lea(end_from, Address(from, count, Address::times_8, 0));
1223 __ cmpptr(to, end_from);
1224 __ movptr(from, Address(rsp, 8)); // from
1225 __ jump_cc(Assembler::aboveEqual, nooverlap);
1226
1227 __ jmpb(L_copy_8_bytes);
1228
1229 __ align(OptoLoopAlignment);
1230 __ BIND(L_copy_8_bytes_loop);
1231 if (VM_Version::supports_mmx()) {
1232 if (UseXMMForArrayCopy) {
1233 __ movq(xmm0, Address(from, count, Address::times_8));
1234 __ movq(Address(to, count, Address::times_8), xmm0);
1235 } else {
1236 __ movq(mmx0, Address(from, count, Address::times_8));
1237 __ movq(Address(to, count, Address::times_8), mmx0);
1238 }
1239 } else {
1240 __ fild_d(Address(from, count, Address::times_8));
1241 __ fistp_d(Address(to, count, Address::times_8));
1242 }
1243 __ BIND(L_copy_8_bytes);
1244 __ decrement(count);
1245 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1246
1247 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1248 __ emms();
1249 }
1250 inc_copy_counter_np(T_LONG);
1251 __ leave(); // required for proper stackwalking of RuntimeStub frame
1252 __ xorptr(rax, rax); // return 0
1253 __ ret(0);
1254 return start;
1255 }
1256
1257
1258 // Helper for generating a dynamic type check.
1259 // The sub_klass must be one of {rbx, rdx, rsi}.
1260 // The temp is killed.
1261 void generate_type_check(Register sub_klass,
1262 Address& super_check_offset_addr,
1263 Address& super_klass_addr,
1264 Register temp,
1265 Label* L_success, Label* L_failure) {
1266 BLOCK_COMMENT("type_check:");
1267
1268 Label L_fallthrough;
1269 #define LOCAL_JCC(assembler_con, label_ptr) \
1270 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1271 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1272
1273 // The following is a strange variation of the fast path which requires
1274 // one less register, because needed values are on the argument stack.
1275 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1276 // L_success, L_failure, NULL);
1277 assert_different_registers(sub_klass, temp);
1278
1279 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1280
1281 // if the pointers are equal, we are done (e.g., String[] elements)
1282 __ cmpptr(sub_klass, super_klass_addr);
1283 LOCAL_JCC(Assembler::equal, L_success);
1284
1285 // check the supertype display:
1286 __ movl2ptr(temp, super_check_offset_addr);
1287 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1288 __ movptr(temp, super_check_addr); // load displayed supertype
1289 __ cmpptr(temp, super_klass_addr); // test the super type
1290 LOCAL_JCC(Assembler::equal, L_success);
1291
1292 // if it was a primary super, we can just fail immediately
1293 __ cmpl(super_check_offset_addr, sc_offset);
1294 LOCAL_JCC(Assembler::notEqual, L_failure);
1295
1296 // The repne_scan instruction uses fixed registers, which will get spilled.
1297 // We happen to know this works best when super_klass is in rax.
1298 Register super_klass = temp;
1299 __ movptr(super_klass, super_klass_addr);
1300 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1301 L_success, L_failure);
1302
1303 __ bind(L_fallthrough);
1304
1305 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1306 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1307
1308 #undef LOCAL_JCC
1309 }
1310
1311 //
1312 // Generate checkcasting array copy stub
1313 //
1314 // Input:
1315 // 4(rsp) - source array address
1316 // 8(rsp) - destination array address
1317 // 12(rsp) - element count, can be zero
1318 // 16(rsp) - size_t ckoff (super_check_offset)
1319 // 20(rsp) - oop ckval (super_klass)
1320 //
1321 // Output:
1322 // rax, == 0 - success
1323 // rax, == -1^K - failure, where K is partial transfer count
1324 //
1325 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1326 __ align(CodeEntryAlignment);
1327 StubCodeMark mark(this, "StubRoutines", name);
1328 address start = __ pc();
1329
1330 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1331
1332 // register use:
1333 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1334 // rdi, rsi -- element access (oop, klass)
1335 // rbx, -- temp
1336 const Register from = rax; // source array address
1337 const Register to = rdx; // destination array address
1338 const Register length = rcx; // elements count
1339 const Register elem = rdi; // each oop copied
1340 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1341 const Register temp = rbx; // lone remaining temp
1342
1343 __ enter(); // required for proper stackwalking of RuntimeStub frame
1344
1345 __ push(rsi);
1346 __ push(rdi);
1347 __ push(rbx);
1348
1349 Address from_arg(rsp, 16+ 4); // from
1350 Address to_arg(rsp, 16+ 8); // to
1351 Address length_arg(rsp, 16+12); // elements count
1352 Address ckoff_arg(rsp, 16+16); // super_check_offset
1353 Address ckval_arg(rsp, 16+20); // super_klass
1354
1355 // Load up:
1356 __ movptr(from, from_arg);
1357 __ movptr(to, to_arg);
1358 __ movl2ptr(length, length_arg);
1359
1360 if (entry != NULL) {
1361 *entry = __ pc(); // Entry point from generic arraycopy stub.
1362 BLOCK_COMMENT("Entry:");
1363 }
1364
1365 //---------------------------------------------------------------
1366 // Assembler stub will be used for this call to arraycopy
1367 // if the two arrays are subtypes of Object[] but the
1368 // destination array type is not equal to or a supertype
1369 // of the source type. Each element must be separately
1370 // checked.
1371
1372 // Loop-invariant addresses. They are exclusive end pointers.
1373 Address end_from_addr(from, length, Address::times_ptr, 0);
1374 Address end_to_addr(to, length, Address::times_ptr, 0);
1375
1376 Register end_from = from; // re-use
1377 Register end_to = to; // re-use
1378 Register count = length; // re-use
1379
1380 // Loop-variant addresses. They assume post-incremented count < 0.
1381 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1382 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1383 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1384
1385 BarrierSetCodeGen *bs = Universe::heap()->barrier_set()->code_gen();
1386 DecoratorSet decorators = ARRAYCOPY_CHECKCAST;
1387 BasicType type = T_OBJECT;
1388 if (dest_uninitialized) {
1389 decorators |= AS_DEST_NOT_INITIALIZED;
1390 }
1391
1392 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1393
1394 // Copy from low to high addresses, indexed from the end of each array.
1395 __ lea(end_from, end_from_addr);
1396 __ lea(end_to, end_to_addr);
1397 assert(length == count, ""); // else fix next line:
1398 __ negptr(count); // negate and test the length
1399 __ jccb(Assembler::notZero, L_load_element);
1400
1401 // Empty array: Nothing to do.
1402 __ xorptr(rax, rax); // return 0 on (trivial) success
1403 __ jmp(L_done);
1404
1405 // ======== begin loop ========
1406 // (Loop is rotated; its entry is L_load_element.)
1407 // Loop control:
1408 // for (count = -count; count != 0; count++)
1409 // Base pointers src, dst are biased by 8*count,to last element.
1410 __ align(OptoLoopAlignment);
1411
1412 __ BIND(L_store_element);
1413 __ movptr(to_element_addr, elem); // store the oop
1414 __ increment(count); // increment the count toward zero
1415 __ jccb(Assembler::zero, L_do_card_marks);
1416
1417 // ======== loop entry is here ========
1418 __ BIND(L_load_element);
1419 __ movptr(elem, from_element_addr); // load the oop
1420 __ testptr(elem, elem);
1421 __ jccb(Assembler::zero, L_store_element);
1422
1423 // (Could do a trick here: Remember last successful non-null
1424 // element stored and make a quick oop equality check on it.)
1425
1426 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1427 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1428 &L_store_element, NULL);
1429 // (On fall-through, we have failed the element type check.)
1430 // ======== end loop ========
1431
1432 // It was a real error; we must depend on the caller to finish the job.
1433 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1434 // Emit GC store barriers for the oops we have copied (length_arg + count),
1435 // and report their number to the caller.
1436 assert_different_registers(to, count, rax);
1437 Label L_post_barrier;
1438 __ addl(count, length_arg); // transfers = (length - remaining)
1439 __ movl2ptr(rax, count); // save the value
1440 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1441 __ jccb(Assembler::notZero, L_post_barrier);
1442 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1443
1444 // Come here on success only.
1445 __ BIND(L_do_card_marks);
1446 __ xorptr(rax, rax); // return 0 on success
1447 __ movl2ptr(count, length_arg);
1448
1449 __ BIND(L_post_barrier);
1450 __ movptr(to, to_arg); // reload
1451 bs->arraycopy_epilogue(_masm, decorators, type, from, to, count);
1452
1453 // Common exit point (success or failure).
1454 __ BIND(L_done);
1455 __ pop(rbx);
1456 __ pop(rdi);
1457 __ pop(rsi);
1458 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1459 __ leave(); // required for proper stackwalking of RuntimeStub frame
1460 __ ret(0);
1461
1462 return start;
1463 }
1464
1465 //
1466 // Generate 'unsafe' array copy stub
1467 // Though just as safe as the other stubs, it takes an unscaled
1468 // size_t argument instead of an element count.
1469 //
1470 // Input:
1471 // 4(rsp) - source array address
1472 // 8(rsp) - destination array address
1473 // 12(rsp) - byte count, can be zero
1474 //
1475 // Output:
1476 // rax, == 0 - success
1477 // rax, == -1 - need to call System.arraycopy
1478 //
1479 // Examines the alignment of the operands and dispatches
1480 // to a long, int, short, or byte copy loop.
1481 //
1482 address generate_unsafe_copy(const char *name,
1483 address byte_copy_entry,
1484 address short_copy_entry,
1485 address int_copy_entry,
1486 address long_copy_entry) {
1487
1488 Label L_long_aligned, L_int_aligned, L_short_aligned;
1489
1490 __ align(CodeEntryAlignment);
1491 StubCodeMark mark(this, "StubRoutines", name);
1492 address start = __ pc();
1493
1494 const Register from = rax; // source array address
1495 const Register to = rdx; // destination array address
1496 const Register count = rcx; // elements count
1497
1498 __ enter(); // required for proper stackwalking of RuntimeStub frame
1499 __ push(rsi);
1500 __ push(rdi);
1501 Address from_arg(rsp, 12+ 4); // from
1502 Address to_arg(rsp, 12+ 8); // to
1503 Address count_arg(rsp, 12+12); // byte count
1504
1505 // Load up:
1506 __ movptr(from , from_arg);
1507 __ movptr(to , to_arg);
1508 __ movl2ptr(count, count_arg);
1509
1510 // bump this on entry, not on exit:
1511 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1512
1513 const Register bits = rsi;
1514 __ mov(bits, from);
1515 __ orptr(bits, to);
1516 __ orptr(bits, count);
1517
1518 __ testl(bits, BytesPerLong-1);
1519 __ jccb(Assembler::zero, L_long_aligned);
1520
1521 __ testl(bits, BytesPerInt-1);
1522 __ jccb(Assembler::zero, L_int_aligned);
1523
1524 __ testl(bits, BytesPerShort-1);
1525 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1526
1527 __ BIND(L_short_aligned);
1528 __ shrptr(count, LogBytesPerShort); // size => short_count
1529 __ movl(count_arg, count); // update 'count'
1530 __ jump(RuntimeAddress(short_copy_entry));
1531
1532 __ BIND(L_int_aligned);
1533 __ shrptr(count, LogBytesPerInt); // size => int_count
1534 __ movl(count_arg, count); // update 'count'
1535 __ jump(RuntimeAddress(int_copy_entry));
1536
1537 __ BIND(L_long_aligned);
1538 __ shrptr(count, LogBytesPerLong); // size => qword_count
1539 __ movl(count_arg, count); // update 'count'
1540 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1541 __ pop(rsi);
1542 __ jump(RuntimeAddress(long_copy_entry));
1543
1544 return start;
1545 }
1546
1547
1548 // Perform range checks on the proposed arraycopy.
1549 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1550 void arraycopy_range_checks(Register src,
1551 Register src_pos,
1552 Register dst,
1553 Register dst_pos,
1554 Address& length,
1555 Label& L_failed) {
1556 BLOCK_COMMENT("arraycopy_range_checks:");
1557 const Register src_end = src_pos; // source array end position
1558 const Register dst_end = dst_pos; // destination array end position
1559 __ addl(src_end, length); // src_pos + length
1560 __ addl(dst_end, length); // dst_pos + length
1561
1562 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1563 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1564 __ jcc(Assembler::above, L_failed);
1565
1566 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1567 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1568 __ jcc(Assembler::above, L_failed);
1569
1570 BLOCK_COMMENT("arraycopy_range_checks done");
1571 }
1572
1573
1574 //
1575 // Generate generic array copy stubs
1576 //
1577 // Input:
1578 // 4(rsp) - src oop
1579 // 8(rsp) - src_pos
1580 // 12(rsp) - dst oop
1581 // 16(rsp) - dst_pos
1582 // 20(rsp) - element count
1583 //
1584 // Output:
1585 // rax, == 0 - success
1586 // rax, == -1^K - failure, where K is partial transfer count
1587 //
1588 address generate_generic_copy(const char *name,
1589 address entry_jbyte_arraycopy,
1590 address entry_jshort_arraycopy,
1591 address entry_jint_arraycopy,
1592 address entry_oop_arraycopy,
1593 address entry_jlong_arraycopy,
1594 address entry_checkcast_arraycopy) {
1595 Label L_failed, L_failed_0, L_objArray;
1596
1597 { int modulus = CodeEntryAlignment;
1598 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1599 int advance = target - (__ offset() % modulus);
1600 if (advance < 0) advance += modulus;
1601 if (advance > 0) __ nop(advance);
1602 }
1603 StubCodeMark mark(this, "StubRoutines", name);
1604
1605 // Short-hop target to L_failed. Makes for denser prologue code.
1606 __ BIND(L_failed_0);
1607 __ jmp(L_failed);
1608 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1609
1610 __ align(CodeEntryAlignment);
1611 address start = __ pc();
1612
1613 __ enter(); // required for proper stackwalking of RuntimeStub frame
1614 __ push(rsi);
1615 __ push(rdi);
1616
1617 // bump this on entry, not on exit:
1618 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1619
1620 // Input values
1621 Address SRC (rsp, 12+ 4);
1622 Address SRC_POS (rsp, 12+ 8);
1623 Address DST (rsp, 12+12);
1624 Address DST_POS (rsp, 12+16);
1625 Address LENGTH (rsp, 12+20);
1626
1627 //-----------------------------------------------------------------------
1628 // Assembler stub will be used for this call to arraycopy
1629 // if the following conditions are met:
1630 //
1631 // (1) src and dst must not be null.
1632 // (2) src_pos must not be negative.
1633 // (3) dst_pos must not be negative.
1634 // (4) length must not be negative.
1635 // (5) src klass and dst klass should be the same and not NULL.
1636 // (6) src and dst should be arrays.
1637 // (7) src_pos + length must not exceed length of src.
1638 // (8) dst_pos + length must not exceed length of dst.
1639 //
1640
1641 const Register src = rax; // source array oop
1642 const Register src_pos = rsi;
1643 const Register dst = rdx; // destination array oop
1644 const Register dst_pos = rdi;
1645 const Register length = rcx; // transfer count
1646
1647 // if (src == NULL) return -1;
1648 __ movptr(src, SRC); // src oop
1649 __ testptr(src, src);
1650 __ jccb(Assembler::zero, L_failed_0);
1651
1652 // if (src_pos < 0) return -1;
1653 __ movl2ptr(src_pos, SRC_POS); // src_pos
1654 __ testl(src_pos, src_pos);
1655 __ jccb(Assembler::negative, L_failed_0);
1656
1657 // if (dst == NULL) return -1;
1658 __ movptr(dst, DST); // dst oop
1659 __ testptr(dst, dst);
1660 __ jccb(Assembler::zero, L_failed_0);
1661
1662 // if (dst_pos < 0) return -1;
1663 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1664 __ testl(dst_pos, dst_pos);
1665 __ jccb(Assembler::negative, L_failed_0);
1666
1667 // if (length < 0) return -1;
1668 __ movl2ptr(length, LENGTH); // length
1669 __ testl(length, length);
1670 __ jccb(Assembler::negative, L_failed_0);
1671
1672 // if (src->klass() == NULL) return -1;
1673 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1674 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1675 const Register rcx_src_klass = rcx; // array klass
1676 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1677
1678 #ifdef ASSERT
1679 // assert(src->klass() != NULL);
1680 BLOCK_COMMENT("assert klasses not null");
1681 { Label L1, L2;
1682 __ testptr(rcx_src_klass, rcx_src_klass);
1683 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1684 __ bind(L1);
1685 __ stop("broken null klass");
1686 __ bind(L2);
1687 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1688 __ jccb(Assembler::equal, L1); // this would be broken also
1689 BLOCK_COMMENT("assert done");
1690 }
1691 #endif //ASSERT
1692
1693 // Load layout helper (32-bits)
1694 //
1695 // |array_tag| | header_size | element_type | |log2_element_size|
1696 // 32 30 24 16 8 2 0
1697 //
1698 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1699 //
1700
1701 int lh_offset = in_bytes(Klass::layout_helper_offset());
1702 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1703
1704 // Handle objArrays completely differently...
1705 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1706 __ cmpl(src_klass_lh_addr, objArray_lh);
1707 __ jcc(Assembler::equal, L_objArray);
1708
1709 // if (src->klass() != dst->klass()) return -1;
1710 __ cmpptr(rcx_src_klass, dst_klass_addr);
1711 __ jccb(Assembler::notEqual, L_failed_0);
1712
1713 const Register rcx_lh = rcx; // layout helper
1714 assert(rcx_lh == rcx_src_klass, "known alias");
1715 __ movl(rcx_lh, src_klass_lh_addr);
1716
1717 // if (!src->is_Array()) return -1;
1718 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1719 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1720
1721 // At this point, it is known to be a typeArray (array_tag 0x3).
1722 #ifdef ASSERT
1723 { Label L;
1724 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1725 __ jcc(Assembler::greaterEqual, L); // signed cmp
1726 __ stop("must be a primitive array");
1727 __ bind(L);
1728 }
1729 #endif
1730
1731 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1732 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1733
1734 // TypeArrayKlass
1735 //
1736 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1737 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1738 //
1739 const Register rsi_offset = rsi; // array offset
1740 const Register src_array = src; // src array offset
1741 const Register dst_array = dst; // dst array offset
1742 const Register rdi_elsize = rdi; // log2 element size
1743
1744 __ mov(rsi_offset, rcx_lh);
1745 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1746 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1747 __ addptr(src_array, rsi_offset); // src array offset
1748 __ addptr(dst_array, rsi_offset); // dst array offset
1749 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1750
1751 // next registers should be set before the jump to corresponding stub
1752 const Register from = src; // source array address
1753 const Register to = dst; // destination array address
1754 const Register count = rcx; // elements count
1755 // some of them should be duplicated on stack
1756 #define FROM Address(rsp, 12+ 4)
1757 #define TO Address(rsp, 12+ 8) // Not used now
1758 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1759
1760 BLOCK_COMMENT("scale indexes to element size");
1761 __ movl2ptr(rsi, SRC_POS); // src_pos
1762 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1763 assert(src_array == from, "");
1764 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1765 __ movl2ptr(rdi, DST_POS); // dst_pos
1766 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1767 assert(dst_array == to, "");
1768 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1769 __ movptr(FROM, from); // src_addr
1770 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1771 __ movl2ptr(count, LENGTH); // elements count
1772
1773 BLOCK_COMMENT("choose copy loop based on element size");
1774 __ cmpl(rdi_elsize, 0);
1775
1776 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1777 __ cmpl(rdi_elsize, LogBytesPerShort);
1778 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1779 __ cmpl(rdi_elsize, LogBytesPerInt);
1780 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1781 #ifdef ASSERT
1782 __ cmpl(rdi_elsize, LogBytesPerLong);
1783 __ jccb(Assembler::notEqual, L_failed);
1784 #endif
1785 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1786 __ pop(rsi);
1787 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1788
1789 __ BIND(L_failed);
1790 __ xorptr(rax, rax);
1791 __ notptr(rax); // return -1
1792 __ pop(rdi);
1793 __ pop(rsi);
1794 __ leave(); // required for proper stackwalking of RuntimeStub frame
1795 __ ret(0);
1796
1797 // ObjArrayKlass
1798 __ BIND(L_objArray);
1799 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1800
1801 Label L_plain_copy, L_checkcast_copy;
1802 // test array classes for subtyping
1803 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1804 __ jccb(Assembler::notEqual, L_checkcast_copy);
1805
1806 // Identically typed arrays can be copied without element-wise checks.
1807 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1808 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1809
1810 __ BIND(L_plain_copy);
1811 __ movl2ptr(count, LENGTH); // elements count
1812 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1813 __ lea(from, Address(src, src_pos, Address::times_ptr,
1814 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1815 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1816 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1817 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1818 __ movptr(FROM, from); // src_addr
1819 __ movptr(TO, to); // dst_addr
1820 __ movl(COUNT, count); // count
1821 __ jump(RuntimeAddress(entry_oop_arraycopy));
1822
1823 __ BIND(L_checkcast_copy);
1824 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1825 {
1826 // Handy offsets:
1827 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1828 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1829
1830 Register rsi_dst_klass = rsi;
1831 Register rdi_temp = rdi;
1832 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1833 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1834 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1835
1836 // Before looking at dst.length, make sure dst is also an objArray.
1837 __ movptr(rsi_dst_klass, dst_klass_addr);
1838 __ cmpl(dst_klass_lh_addr, objArray_lh);
1839 __ jccb(Assembler::notEqual, L_failed);
1840
1841 // It is safe to examine both src.length and dst.length.
1842 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1843 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1844 // (Now src_pos and dst_pos are killed, but not src and dst.)
1845
1846 // We'll need this temp (don't forget to pop it after the type check).
1847 __ push(rbx);
1848 Register rbx_src_klass = rbx;
1849
1850 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1851 __ movptr(rsi_dst_klass, dst_klass_addr);
1852 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1853 Label L_fail_array_check;
1854 generate_type_check(rbx_src_klass,
1855 super_check_offset_addr, dst_klass_addr,
1856 rdi_temp, NULL, &L_fail_array_check);
1857 // (On fall-through, we have passed the array type check.)
1858 __ pop(rbx);
1859 __ jmp(L_plain_copy);
1860
1861 __ BIND(L_fail_array_check);
1862 // Reshuffle arguments so we can call checkcast_arraycopy:
1863
1864 // match initial saves for checkcast_arraycopy
1865 // push(rsi); // already done; see above
1866 // push(rdi); // already done; see above
1867 // push(rbx); // already done; see above
1868
1869 // Marshal outgoing arguments now, freeing registers.
1870 Address from_arg(rsp, 16+ 4); // from
1871 Address to_arg(rsp, 16+ 8); // to
1872 Address length_arg(rsp, 16+12); // elements count
1873 Address ckoff_arg(rsp, 16+16); // super_check_offset
1874 Address ckval_arg(rsp, 16+20); // super_klass
1875
1876 Address SRC_POS_arg(rsp, 16+ 8);
1877 Address DST_POS_arg(rsp, 16+16);
1878 Address LENGTH_arg(rsp, 16+20);
1879 // push rbx, changed the incoming offsets (why not just use rbp,??)
1880 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1881
1882 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1883 __ movl2ptr(length, LENGTH_arg); // reload elements count
1884 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1885 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1886
1887 __ movptr(ckval_arg, rbx); // destination element type
1888 __ movl(rbx, Address(rbx, sco_offset));
1889 __ movl(ckoff_arg, rbx); // corresponding class check offset
1890
1891 __ movl(length_arg, length); // outgoing length argument
1892
1893 __ lea(from, Address(src, src_pos, Address::times_ptr,
1894 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1895 __ movptr(from_arg, from);
1896
1897 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1898 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1899 __ movptr(to_arg, to);
1900 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1901 }
1902
1903 return start;
1904 }
1905
1906 void generate_arraycopy_stubs() {
1907 address entry;
1908 address entry_jbyte_arraycopy;
1909 address entry_jshort_arraycopy;
1910 address entry_jint_arraycopy;
1911 address entry_oop_arraycopy;
1912 address entry_jlong_arraycopy;
1913 address entry_checkcast_arraycopy;
1914
1915 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1916 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1917 "arrayof_jbyte_disjoint_arraycopy");
1918 StubRoutines::_arrayof_jbyte_arraycopy =
1919 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1920 NULL, "arrayof_jbyte_arraycopy");
1921 StubRoutines::_jbyte_disjoint_arraycopy =
1922 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1923 "jbyte_disjoint_arraycopy");
1924 StubRoutines::_jbyte_arraycopy =
1925 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1926 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1927
1928 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1929 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
1930 "arrayof_jshort_disjoint_arraycopy");
1931 StubRoutines::_arrayof_jshort_arraycopy =
1932 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
1933 NULL, "arrayof_jshort_arraycopy");
1934 StubRoutines::_jshort_disjoint_arraycopy =
1935 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
1936 "jshort_disjoint_arraycopy");
1937 StubRoutines::_jshort_arraycopy =
1938 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
1939 &entry_jshort_arraycopy, "jshort_arraycopy");
1940
1941 // Next arrays are always aligned on 4 bytes at least.
1942 StubRoutines::_jint_disjoint_arraycopy =
1943 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
1944 "jint_disjoint_arraycopy");
1945 StubRoutines::_jint_arraycopy =
1946 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
1947 &entry_jint_arraycopy, "jint_arraycopy");
1948
1949 StubRoutines::_oop_disjoint_arraycopy =
1950 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
1951 "oop_disjoint_arraycopy");
1952 StubRoutines::_oop_arraycopy =
1953 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
1954 &entry_oop_arraycopy, "oop_arraycopy");
1955
1956 StubRoutines::_oop_disjoint_arraycopy_uninit =
1957 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
1958 "oop_disjoint_arraycopy_uninit",
1959 /*dest_uninitialized*/true);
1960 StubRoutines::_oop_arraycopy_uninit =
1961 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
1962 NULL, "oop_arraycopy_uninit",
1963 /*dest_uninitialized*/true);
1964
1965 StubRoutines::_jlong_disjoint_arraycopy =
1966 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
1967 StubRoutines::_jlong_arraycopy =
1968 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
1969 "jlong_arraycopy");
1970
1971 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
1972 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
1973 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
1974 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
1975 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
1976 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
1977
1978 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
1979 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
1980 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
1981 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
1982
1983 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
1984 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
1985 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
1986 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
1987
1988 StubRoutines::_checkcast_arraycopy =
1989 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
1990 StubRoutines::_checkcast_arraycopy_uninit =
1991 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
1992
1993 StubRoutines::_unsafe_arraycopy =
1994 generate_unsafe_copy("unsafe_arraycopy",
1995 entry_jbyte_arraycopy,
1996 entry_jshort_arraycopy,
1997 entry_jint_arraycopy,
1998 entry_jlong_arraycopy);
1999
2000 StubRoutines::_generic_arraycopy =
2001 generate_generic_copy("generic_arraycopy",
2002 entry_jbyte_arraycopy,
2003 entry_jshort_arraycopy,
2004 entry_jint_arraycopy,
2005 entry_oop_arraycopy,
2006 entry_jlong_arraycopy,
2007 entry_checkcast_arraycopy);
2008 }
2009
2010 // AES intrinsic stubs
2011 enum {AESBlockSize = 16};
2012
2013 address generate_key_shuffle_mask() {
2014 __ align(16);
2015 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2016 address start = __ pc();
2017 __ emit_data(0x00010203, relocInfo::none, 0 );
2018 __ emit_data(0x04050607, relocInfo::none, 0 );
2019 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2020 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2021 return start;
2022 }
2023
2024 address generate_counter_shuffle_mask() {
2025 __ align(16);
2026 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2027 address start = __ pc();
2028 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
2029 __ emit_data(0x08090a0b, relocInfo::none, 0);
2030 __ emit_data(0x04050607, relocInfo::none, 0);
2031 __ emit_data(0x00010203, relocInfo::none, 0);
2032 return start;
2033 }
2034
2035 // Utility routine for loading a 128-bit key word in little endian format
2036 // can optionally specify that the shuffle mask is already in an xmmregister
2037 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2038 __ movdqu(xmmdst, Address(key, offset));
2039 if (xmm_shuf_mask != NULL) {
2040 __ pshufb(xmmdst, xmm_shuf_mask);
2041 } else {
2042 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2043 }
2044 }
2045
2046 // aesenc using specified key+offset
2047 // can optionally specify that the shuffle mask is already in an xmmregister
2048 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2049 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2050 __ aesenc(xmmdst, xmmtmp);
2051 }
2052
2053 // aesdec using specified key+offset
2054 // can optionally specify that the shuffle mask is already in an xmmregister
2055 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2056 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2057 __ aesdec(xmmdst, xmmtmp);
2058 }
2059
2060 // Utility routine for increase 128bit counter (iv in CTR mode)
2061 // XMM_128bit, D3, D2, D1, D0
2062 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
2063 __ pextrd(reg, xmmdst, 0x0);
2064 __ addl(reg, inc_delta);
2065 __ pinsrd(xmmdst, reg, 0x0);
2066 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2067
2068 __ pextrd(reg, xmmdst, 0x01); // Carry-> D1
2069 __ addl(reg, 0x01);
2070 __ pinsrd(xmmdst, reg, 0x01);
2071 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2072
2073 __ pextrd(reg, xmmdst, 0x02); // Carry-> D2
2074 __ addl(reg, 0x01);
2075 __ pinsrd(xmmdst, reg, 0x02);
2076 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2077
2078 __ pextrd(reg, xmmdst, 0x03); // Carry -> D3
2079 __ addl(reg, 0x01);
2080 __ pinsrd(xmmdst, reg, 0x03);
2081
2082 __ BIND(next_block); // next instruction
2083 }
2084
2085
2086 // Arguments:
2087 //
2088 // Inputs:
2089 // c_rarg0 - source byte array address
2090 // c_rarg1 - destination byte array address
2091 // c_rarg2 - K (key) in little endian int array
2092 //
2093 address generate_aescrypt_encryptBlock() {
2094 assert(UseAES, "need AES instructions and misaligned SSE support");
2095 __ align(CodeEntryAlignment);
2096 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2097 Label L_doLast;
2098 address start = __ pc();
2099
2100 const Register from = rdx; // source array address
2101 const Register to = rdx; // destination array address
2102 const Register key = rcx; // key array address
2103 const Register keylen = rax;
2104 const Address from_param(rbp, 8+0);
2105 const Address to_param (rbp, 8+4);
2106 const Address key_param (rbp, 8+8);
2107
2108 const XMMRegister xmm_result = xmm0;
2109 const XMMRegister xmm_key_shuf_mask = xmm1;
2110 const XMMRegister xmm_temp1 = xmm2;
2111 const XMMRegister xmm_temp2 = xmm3;
2112 const XMMRegister xmm_temp3 = xmm4;
2113 const XMMRegister xmm_temp4 = xmm5;
2114
2115 __ enter(); // required for proper stackwalking of RuntimeStub frame
2116
2117 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2118 // context for the registers used, where all instructions below are using 128-bit mode
2119 // On EVEX without VL and BW, these instructions will all be AVX.
2120 if (VM_Version::supports_avx512vlbw()) {
2121 __ movl(rdx, 0xffff);
2122 __ kmovdl(k1, rdx);
2123 }
2124
2125 __ movptr(from, from_param);
2126 __ movptr(key, key_param);
2127
2128 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2129 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2130
2131 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2132 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2133 __ movptr(to, to_param);
2134
2135 // For encryption, the java expanded key ordering is just what we need
2136
2137 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2138 __ pxor(xmm_result, xmm_temp1);
2139
2140 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2141 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2142 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2143 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2144
2145 __ aesenc(xmm_result, xmm_temp1);
2146 __ aesenc(xmm_result, xmm_temp2);
2147 __ aesenc(xmm_result, xmm_temp3);
2148 __ aesenc(xmm_result, xmm_temp4);
2149
2150 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2151 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2152 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2153 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2154
2155 __ aesenc(xmm_result, xmm_temp1);
2156 __ aesenc(xmm_result, xmm_temp2);
2157 __ aesenc(xmm_result, xmm_temp3);
2158 __ aesenc(xmm_result, xmm_temp4);
2159
2160 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2161 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2162
2163 __ cmpl(keylen, 44);
2164 __ jccb(Assembler::equal, L_doLast);
2165
2166 __ aesenc(xmm_result, xmm_temp1);
2167 __ aesenc(xmm_result, xmm_temp2);
2168
2169 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2170 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2171
2172 __ cmpl(keylen, 52);
2173 __ jccb(Assembler::equal, L_doLast);
2174
2175 __ aesenc(xmm_result, xmm_temp1);
2176 __ aesenc(xmm_result, xmm_temp2);
2177
2178 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2179 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2180
2181 __ BIND(L_doLast);
2182 __ aesenc(xmm_result, xmm_temp1);
2183 __ aesenclast(xmm_result, xmm_temp2);
2184 __ movdqu(Address(to, 0), xmm_result); // store the result
2185 __ xorptr(rax, rax); // return 0
2186 __ leave(); // required for proper stackwalking of RuntimeStub frame
2187 __ ret(0);
2188
2189 return start;
2190 }
2191
2192
2193 // Arguments:
2194 //
2195 // Inputs:
2196 // c_rarg0 - source byte array address
2197 // c_rarg1 - destination byte array address
2198 // c_rarg2 - K (key) in little endian int array
2199 //
2200 address generate_aescrypt_decryptBlock() {
2201 assert(UseAES, "need AES instructions and misaligned SSE support");
2202 __ align(CodeEntryAlignment);
2203 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2204 Label L_doLast;
2205 address start = __ pc();
2206
2207 const Register from = rdx; // source array address
2208 const Register to = rdx; // destination array address
2209 const Register key = rcx; // key array address
2210 const Register keylen = rax;
2211 const Address from_param(rbp, 8+0);
2212 const Address to_param (rbp, 8+4);
2213 const Address key_param (rbp, 8+8);
2214
2215 const XMMRegister xmm_result = xmm0;
2216 const XMMRegister xmm_key_shuf_mask = xmm1;
2217 const XMMRegister xmm_temp1 = xmm2;
2218 const XMMRegister xmm_temp2 = xmm3;
2219 const XMMRegister xmm_temp3 = xmm4;
2220 const XMMRegister xmm_temp4 = xmm5;
2221
2222 __ enter(); // required for proper stackwalking of RuntimeStub frame
2223
2224 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2225 // context for the registers used, where all instructions below are using 128-bit mode
2226 // On EVEX without VL and BW, these instructions will all be AVX.
2227 if (VM_Version::supports_avx512vlbw()) {
2228 __ movl(rdx, 0xffff);
2229 __ kmovdl(k1, rdx);
2230 }
2231
2232 __ movptr(from, from_param);
2233 __ movptr(key, key_param);
2234
2235 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2236 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2237
2238 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2239 __ movdqu(xmm_result, Address(from, 0));
2240 __ movptr(to, to_param);
2241
2242 // for decryption java expanded key ordering is rotated one position from what we want
2243 // so we start from 0x10 here and hit 0x00 last
2244 // we don't know if the key is aligned, hence not using load-execute form
2245 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2246 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2247 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2248 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2249
2250 __ pxor (xmm_result, xmm_temp1);
2251 __ aesdec(xmm_result, xmm_temp2);
2252 __ aesdec(xmm_result, xmm_temp3);
2253 __ aesdec(xmm_result, xmm_temp4);
2254
2255 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2256 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2257 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2258 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2259
2260 __ aesdec(xmm_result, xmm_temp1);
2261 __ aesdec(xmm_result, xmm_temp2);
2262 __ aesdec(xmm_result, xmm_temp3);
2263 __ aesdec(xmm_result, xmm_temp4);
2264
2265 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2266 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2267 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2268
2269 __ cmpl(keylen, 44);
2270 __ jccb(Assembler::equal, L_doLast);
2271
2272 __ aesdec(xmm_result, xmm_temp1);
2273 __ aesdec(xmm_result, xmm_temp2);
2274
2275 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2276 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2277
2278 __ cmpl(keylen, 52);
2279 __ jccb(Assembler::equal, L_doLast);
2280
2281 __ aesdec(xmm_result, xmm_temp1);
2282 __ aesdec(xmm_result, xmm_temp2);
2283
2284 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2285 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2286
2287 __ BIND(L_doLast);
2288 __ aesdec(xmm_result, xmm_temp1);
2289 __ aesdec(xmm_result, xmm_temp2);
2290
2291 // for decryption the aesdeclast operation is always on key+0x00
2292 __ aesdeclast(xmm_result, xmm_temp3);
2293 __ movdqu(Address(to, 0), xmm_result); // store the result
2294 __ xorptr(rax, rax); // return 0
2295 __ leave(); // required for proper stackwalking of RuntimeStub frame
2296 __ ret(0);
2297
2298 return start;
2299 }
2300
2301 void handleSOERegisters(bool saving) {
2302 const int saveFrameSizeInBytes = 4 * wordSize;
2303 const Address saved_rbx (rbp, -3 * wordSize);
2304 const Address saved_rsi (rbp, -2 * wordSize);
2305 const Address saved_rdi (rbp, -1 * wordSize);
2306
2307 if (saving) {
2308 __ subptr(rsp, saveFrameSizeInBytes);
2309 __ movptr(saved_rsi, rsi);
2310 __ movptr(saved_rdi, rdi);
2311 __ movptr(saved_rbx, rbx);
2312 } else {
2313 // restoring
2314 __ movptr(rsi, saved_rsi);
2315 __ movptr(rdi, saved_rdi);
2316 __ movptr(rbx, saved_rbx);
2317 }
2318 }
2319
2320 // Arguments:
2321 //
2322 // Inputs:
2323 // c_rarg0 - source byte array address
2324 // c_rarg1 - destination byte array address
2325 // c_rarg2 - K (key) in little endian int array
2326 // c_rarg3 - r vector byte array address
2327 // c_rarg4 - input length
2328 //
2329 // Output:
2330 // rax - input length
2331 //
2332 address generate_cipherBlockChaining_encryptAESCrypt() {
2333 assert(UseAES, "need AES instructions and misaligned SSE support");
2334 __ align(CodeEntryAlignment);
2335 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2336 address start = __ pc();
2337
2338 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2339 const Register from = rsi; // source array address
2340 const Register to = rdx; // destination array address
2341 const Register key = rcx; // key array address
2342 const Register rvec = rdi; // r byte array initialized from initvector array address
2343 // and left with the results of the last encryption block
2344 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2345 const Register pos = rax;
2346
2347 // xmm register assignments for the loops below
2348 const XMMRegister xmm_result = xmm0;
2349 const XMMRegister xmm_temp = xmm1;
2350 // first 6 keys preloaded into xmm2-xmm7
2351 const int XMM_REG_NUM_KEY_FIRST = 2;
2352 const int XMM_REG_NUM_KEY_LAST = 7;
2353 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2354
2355 __ enter(); // required for proper stackwalking of RuntimeStub frame
2356 handleSOERegisters(true /*saving*/);
2357
2358 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2359 // context for the registers used, where all instructions below are using 128-bit mode
2360 // On EVEX without VL and BW, these instructions will all be AVX.
2361 if (VM_Version::supports_avx512vlbw()) {
2362 __ movl(rdx, 0xffff);
2363 __ kmovdl(k1, rdx);
2364 }
2365
2366 // load registers from incoming parameters
2367 const Address from_param(rbp, 8+0);
2368 const Address to_param (rbp, 8+4);
2369 const Address key_param (rbp, 8+8);
2370 const Address rvec_param (rbp, 8+12);
2371 const Address len_param (rbp, 8+16);
2372 __ movptr(from , from_param);
2373 __ movptr(to , to_param);
2374 __ movptr(key , key_param);
2375 __ movptr(rvec , rvec_param);
2376 __ movptr(len_reg , len_param);
2377
2378 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2379 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2380 // load up xmm regs 2 thru 7 with keys 0-5
2381 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2382 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2383 offset += 0x10;
2384 }
2385
2386 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2387
2388 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2389 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2390 __ cmpl(rax, 44);
2391 __ jcc(Assembler::notEqual, L_key_192_256);
2392
2393 // 128 bit code follows here
2394 __ movl(pos, 0);
2395 __ align(OptoLoopAlignment);
2396 __ BIND(L_loopTop_128);
2397 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2398 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2399
2400 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2401 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2402 __ aesenc(xmm_result, as_XMMRegister(rnum));
2403 }
2404 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2405 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2406 }
2407 load_key(xmm_temp, key, 0xa0);
2408 __ aesenclast(xmm_result, xmm_temp);
2409
2410 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2411 // no need to store r to memory until we exit
2412 __ addptr(pos, AESBlockSize);
2413 __ subptr(len_reg, AESBlockSize);
2414 __ jcc(Assembler::notEqual, L_loopTop_128);
2415
2416 __ BIND(L_exit);
2417 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2418
2419 handleSOERegisters(false /*restoring*/);
2420 __ movptr(rax, len_param); // return length
2421 __ leave(); // required for proper stackwalking of RuntimeStub frame
2422 __ ret(0);
2423
2424 __ BIND(L_key_192_256);
2425 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2426 __ cmpl(rax, 52);
2427 __ jcc(Assembler::notEqual, L_key_256);
2428
2429 // 192-bit code follows here (could be changed to use more xmm registers)
2430 __ movl(pos, 0);
2431 __ align(OptoLoopAlignment);
2432 __ BIND(L_loopTop_192);
2433 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2434 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2435
2436 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2437 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2438 __ aesenc(xmm_result, as_XMMRegister(rnum));
2439 }
2440 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2441 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2442 }
2443 load_key(xmm_temp, key, 0xc0);
2444 __ aesenclast(xmm_result, xmm_temp);
2445
2446 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2447 // no need to store r to memory until we exit
2448 __ addptr(pos, AESBlockSize);
2449 __ subptr(len_reg, AESBlockSize);
2450 __ jcc(Assembler::notEqual, L_loopTop_192);
2451 __ jmp(L_exit);
2452
2453 __ BIND(L_key_256);
2454 // 256-bit code follows here (could be changed to use more xmm registers)
2455 __ movl(pos, 0);
2456 __ align(OptoLoopAlignment);
2457 __ BIND(L_loopTop_256);
2458 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2459 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2460
2461 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2462 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2463 __ aesenc(xmm_result, as_XMMRegister(rnum));
2464 }
2465 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2466 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2467 }
2468 load_key(xmm_temp, key, 0xe0);
2469 __ aesenclast(xmm_result, xmm_temp);
2470
2471 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2472 // no need to store r to memory until we exit
2473 __ addptr(pos, AESBlockSize);
2474 __ subptr(len_reg, AESBlockSize);
2475 __ jcc(Assembler::notEqual, L_loopTop_256);
2476 __ jmp(L_exit);
2477
2478 return start;
2479 }
2480
2481
2482 // CBC AES Decryption.
2483 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2484 //
2485 // Arguments:
2486 //
2487 // Inputs:
2488 // c_rarg0 - source byte array address
2489 // c_rarg1 - destination byte array address
2490 // c_rarg2 - K (key) in little endian int array
2491 // c_rarg3 - r vector byte array address
2492 // c_rarg4 - input length
2493 //
2494 // Output:
2495 // rax - input length
2496 //
2497
2498 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
2499 assert(UseAES, "need AES instructions and misaligned SSE support");
2500 __ align(CodeEntryAlignment);
2501 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2502 address start = __ pc();
2503
2504 const Register from = rsi; // source array address
2505 const Register to = rdx; // destination array address
2506 const Register key = rcx; // key array address
2507 const Register rvec = rdi; // r byte array initialized from initvector array address
2508 // and left with the results of the last encryption block
2509 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2510 const Register pos = rax;
2511
2512 const int PARALLEL_FACTOR = 4;
2513 const int ROUNDS[3] = { 10, 12, 14 }; //aes rounds for key128, key192, key256
2514
2515 Label L_exit;
2516 Label L_singleBlock_loopTop[3]; //128, 192, 256
2517 Label L_multiBlock_loopTop[3]; //128, 192, 256
2518
2519 const XMMRegister xmm_prev_block_cipher = xmm0; // holds cipher of previous block
2520 const XMMRegister xmm_key_shuf_mask = xmm1;
2521
2522 const XMMRegister xmm_key_tmp0 = xmm2;
2523 const XMMRegister xmm_key_tmp1 = xmm3;
2524
2525 // registers holding the six results in the parallelized loop
2526 const XMMRegister xmm_result0 = xmm4;
2527 const XMMRegister xmm_result1 = xmm5;
2528 const XMMRegister xmm_result2 = xmm6;
2529 const XMMRegister xmm_result3 = xmm7;
2530
2531 __ enter(); // required for proper stackwalking of RuntimeStub frame
2532 handleSOERegisters(true /*saving*/);
2533
2534 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2535 // context for the registers used, where all instructions below are using 128-bit mode
2536 // On EVEX without VL and BW, these instructions will all be AVX.
2537 if (VM_Version::supports_avx512vlbw()) {
2538 __ movl(rdx, 0xffff);
2539 __ kmovdl(k1, rdx);
2540 }
2541
2542 // load registers from incoming parameters
2543 const Address from_param(rbp, 8+0);
2544 const Address to_param (rbp, 8+4);
2545 const Address key_param (rbp, 8+8);
2546 const Address rvec_param (rbp, 8+12);
2547 const Address len_param (rbp, 8+16);
2548
2549 __ movptr(from , from_param);
2550 __ movptr(to , to_param);
2551 __ movptr(key , key_param);
2552 __ movptr(rvec , rvec_param);
2553 __ movptr(len_reg , len_param);
2554
2555 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2556 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
2557
2558 __ xorptr(pos, pos);
2559
2560 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2561 // rvec is reused
2562 __ movl(rvec, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2563 __ cmpl(rvec, 52);
2564 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
2565 __ cmpl(rvec, 60);
2566 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
2567
2568 #define DoFour(opc, src_reg) \
2569 __ opc(xmm_result0, src_reg); \
2570 __ opc(xmm_result1, src_reg); \
2571 __ opc(xmm_result2, src_reg); \
2572 __ opc(xmm_result3, src_reg); \
2573
2574 for (int k = 0; k < 3; ++k) {
2575 __ align(OptoLoopAlignment);
2576 __ BIND(L_multiBlock_loopTop[k]);
2577 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
2578 __ jcc(Assembler::less, L_singleBlock_loopTop[k]);
2579
2580 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
2581 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2582 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2583 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2584
2585 // the java expanded key ordering is rotated one position from what we want
2586 // so we start from 0x10 here and hit 0x00 last
2587 load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2588 DoFour(pxor, xmm_key_tmp0); //xor with first key
2589 // do the aes dec rounds
2590 for (int rnum = 1; rnum <= ROUNDS[k];) {
2591 //load two keys at a time
2592 //k1->0x20, ..., k9->0xa0, k10->0x00
2593 load_key(xmm_key_tmp1, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2594 load_key(xmm_key_tmp0, key, ((rnum + 2) % (ROUNDS[k] + 1)) * 0x10, xmm_key_shuf_mask); // hit 0x00 last!
2595 DoFour(aesdec, xmm_key_tmp1);
2596 rnum++;
2597 if (rnum != ROUNDS[k]) {
2598 DoFour(aesdec, xmm_key_tmp0);
2599 }
2600 else {
2601 DoFour(aesdeclast, xmm_key_tmp0);
2602 }
2603 rnum++;
2604 }
2605
2606 // for each result, xor with the r vector of previous cipher block
2607 __ pxor(xmm_result0, xmm_prev_block_cipher);
2608 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2609 __ pxor(xmm_result1, xmm_prev_block_cipher);
2610 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2611 __ pxor(xmm_result2, xmm_prev_block_cipher);
2612 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2613 __ pxor(xmm_result3, xmm_prev_block_cipher);
2614 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
2615
2616 // store 4 results into the next 64 bytes of output
2617 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2618 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2619 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2620 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2621
2622 __ addptr(pos, 4 * AESBlockSize);
2623 __ subptr(len_reg, 4 * AESBlockSize);
2624 __ jmp(L_multiBlock_loopTop[k]);
2625
2626 //singleBlock starts here
2627 __ align(OptoLoopAlignment);
2628 __ BIND(L_singleBlock_loopTop[k]);
2629 __ cmpptr(len_reg, 0); // any blocks left?
2630 __ jcc(Assembler::equal, L_exit);
2631 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2632 __ movdqa(xmm_result1, xmm_result0);
2633
2634 load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2635 __ pxor(xmm_result0, xmm_key_tmp0);
2636 // do the aes dec rounds
2637 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
2638 // the java expanded key ordering is rotated one position from what we want
2639 load_key(xmm_key_tmp0, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2640 __ aesdec(xmm_result0, xmm_key_tmp0);
2641 }
2642 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
2643 __ aesdeclast(xmm_result0, xmm_key_tmp0);
2644 __ pxor(xmm_result0, xmm_prev_block_cipher); // xor with the current r vector
2645 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result0); // store into the next 16 bytes of output
2646 // no need to store r to memory until we exit
2647 __ movdqa(xmm_prev_block_cipher, xmm_result1); // set up next r vector with cipher input from this block
2648
2649 __ addptr(pos, AESBlockSize);
2650 __ subptr(len_reg, AESBlockSize);
2651 __ jmp(L_singleBlock_loopTop[k]);
2652 }//for 128/192/256
2653
2654 __ BIND(L_exit);
2655 __ movptr(rvec, rvec_param); // restore this since reused earlier
2656 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
2657 handleSOERegisters(false /*restoring*/);
2658 __ movptr(rax, len_param); // return length
2659 __ leave(); // required for proper stackwalking of RuntimeStub frame
2660 __ ret(0);
2661
2662 return start;
2663 }
2664
2665 // CTR AES crypt.
2666 // In 32-bit stub, parallelize 4 blocks at a time
2667 // Arguments:
2668 //
2669 // Inputs:
2670 // c_rarg0 - source byte array address
2671 // c_rarg1 - destination byte array address
2672 // c_rarg2 - K (key) in little endian int array
2673 // c_rarg3 - counter vector byte array address
2674 // c_rarg4 - input length
2675 //
2676 // Output:
2677 // rax - input length
2678 //
2679 address generate_counterMode_AESCrypt_Parallel() {
2680 assert(UseAES, "need AES instructions and misaligned SSE support");
2681 __ align(CodeEntryAlignment);
2682 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
2683 address start = __ pc();
2684 const Register from = rsi; // source array address
2685 const Register to = rdx; // destination array address
2686 const Register key = rcx; // key array address
2687 const Register counter = rdi; // counter byte array initialized from initvector array address
2688 // and updated with the incremented counter in the end
2689 const Register len_reg = rbx;
2690 const Register pos = rax;
2691
2692 __ enter(); // required for proper stackwalking of RuntimeStub frame
2693 handleSOERegisters(true /*saving*/); // save rbx, rsi, rdi
2694
2695 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
2696 // context for the registers used, where all instructions below are using 128-bit mode
2697 // On EVEX without VL and BW, these instructions will all be AVX.
2698 if (VM_Version::supports_avx512vlbw()) {
2699 __ movl(rdx, 0xffff);
2700 __ kmovdl(k1, rdx);
2701 }
2702
2703 // load registers from incoming parameters
2704 const Address from_param(rbp, 8+0);
2705 const Address to_param (rbp, 8+4);
2706 const Address key_param (rbp, 8+8);
2707 const Address rvec_param (rbp, 8+12);
2708 const Address len_param (rbp, 8+16);
2709 const Address saved_counter_param(rbp, 8 + 20);
2710 const Address used_addr_param(rbp, 8 + 24);
2711
2712 __ movptr(from , from_param);
2713 __ movptr(to , to_param);
2714 __ movptr(len_reg , len_param);
2715
2716 // Use the partially used encrpyted counter from last invocation
2717 Label L_exit_preLoop, L_preLoop_start;
2718
2719 // Use the registers 'counter' and 'key' here in this preloop
2720 // to hold of last 2 params 'used' and 'saved_encCounter_start'
2721 Register used = counter;
2722 Register saved_encCounter_start = key;
2723 Register used_addr = saved_encCounter_start;
2724
2725 __ movptr(used_addr, used_addr_param);
2726 __ movptr(used, Address(used_addr, 0));
2727 __ movptr(saved_encCounter_start, saved_counter_param);
2728
2729 __ BIND(L_preLoop_start);
2730 __ cmpptr(used, 16);
2731 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
2732 __ cmpptr(len_reg, 0);
2733 __ jcc(Assembler::lessEqual, L_exit_preLoop);
2734 __ movb(rax, Address(saved_encCounter_start, used));
2735 __ xorb(rax, Address(from, 0));
2736 __ movb(Address(to, 0), rax);
2737 __ addptr(from, 1);
2738 __ addptr(to, 1);
2739 __ addptr(used, 1);
2740 __ subptr(len_reg, 1);
2741
2742 __ jmp(L_preLoop_start);
2743
2744 __ BIND(L_exit_preLoop);
2745 __ movptr(used_addr, used_addr_param);
2746 __ movptr(used_addr, used_addr_param);
2747 __ movl(Address(used_addr, 0), used);
2748
2749 // load the parameters 'key' and 'counter'
2750 __ movptr(key, key_param);
2751 __ movptr(counter, rvec_param);
2752
2753 // xmm register assignments for the loops below
2754 const XMMRegister xmm_curr_counter = xmm0;
2755 const XMMRegister xmm_counter_shuf_mask = xmm1; // need to be reloaded
2756 const XMMRegister xmm_key_shuf_mask = xmm2; // need to be reloaded
2757 const XMMRegister xmm_key = xmm3;
2758 const XMMRegister xmm_result0 = xmm4;
2759 const XMMRegister xmm_result1 = xmm5;
2760 const XMMRegister xmm_result2 = xmm6;
2761 const XMMRegister xmm_result3 = xmm7;
2762 const XMMRegister xmm_from0 = xmm1; //reuse XMM register
2763 const XMMRegister xmm_from1 = xmm2;
2764 const XMMRegister xmm_from2 = xmm3;
2765 const XMMRegister xmm_from3 = xmm4;
2766
2767 //for key_128, key_192, key_256
2768 const int rounds[3] = {10, 12, 14};
2769 Label L_singleBlockLoopTop[3];
2770 Label L_multiBlock_loopTop[3];
2771 Label L_key192_top, L_key256_top;
2772 Label L_incCounter[3][4]; // 3: different key length, 4: 4 blocks at a time
2773 Label L_incCounter_single[3]; //for single block, key128, key192, key256
2774 Label L_processTail_insr[3], L_processTail_4_insr[3], L_processTail_2_insr[3], L_processTail_1_insr[3], L_processTail_exit_insr[3];
2775 Label L_processTail_extr[3], L_processTail_4_extr[3], L_processTail_2_extr[3], L_processTail_1_extr[3], L_processTail_exit_extr[3];
2776
2777 Label L_exit;
2778 const int PARALLEL_FACTOR = 4; //because of the limited register number
2779
2780 // initialize counter with initial counter
2781 __ movdqu(xmm_curr_counter, Address(counter, 0x00));
2782 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2783 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled for increase
2784
2785 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
2786 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2787 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2788 __ cmpl(rax, 52);
2789 __ jcc(Assembler::equal, L_key192_top);
2790 __ cmpl(rax, 60);
2791 __ jcc(Assembler::equal, L_key256_top);
2792
2793 //key128 begins here
2794 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2795
2796 #define CTR_DoFour(opc, src_reg) \
2797 __ opc(xmm_result0, src_reg); \
2798 __ opc(xmm_result1, src_reg); \
2799 __ opc(xmm_result2, src_reg); \
2800 __ opc(xmm_result3, src_reg);
2801
2802 // k == 0 : generate code for key_128
2803 // k == 1 : generate code for key_192
2804 // k == 2 : generate code for key_256
2805 for (int k = 0; k < 3; ++k) {
2806 //multi blocks starts here
2807 __ align(OptoLoopAlignment);
2808 __ BIND(L_multiBlock_loopTop[k]);
2809 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
2810 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
2811
2812 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2813 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2814
2815 //load, then increase counters
2816 CTR_DoFour(movdqa, xmm_curr_counter);
2817 __ push(rbx);
2818 inc_counter(rbx, xmm_result1, 0x01, L_incCounter[k][0]);
2819 inc_counter(rbx, xmm_result2, 0x02, L_incCounter[k][1]);
2820 inc_counter(rbx, xmm_result3, 0x03, L_incCounter[k][2]);
2821 inc_counter(rbx, xmm_curr_counter, 0x04, L_incCounter[k][3]);
2822 __ pop (rbx);
2823
2824 load_key(xmm_key, key, 0x00, xmm_key_shuf_mask); // load Round 0 key. interleaving for better performance
2825
2826 CTR_DoFour(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
2827 CTR_DoFour(pxor, xmm_key); //PXOR with Round 0 key
2828
2829 for (int i = 1; i < rounds[k]; ++i) {
2830 load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2831 CTR_DoFour(aesenc, xmm_key);
2832 }
2833 load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2834 CTR_DoFour(aesenclast, xmm_key);
2835
2836 // get next PARALLEL_FACTOR blocks into xmm_from registers
2837 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2838 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2839 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2840
2841 // PXOR with input text
2842 __ pxor(xmm_result0, xmm_from0); //result0 is xmm4
2843 __ pxor(xmm_result1, xmm_from1);
2844 __ pxor(xmm_result2, xmm_from2);
2845
2846 // store PARALLEL_FACTOR results into the next 64 bytes of output
2847 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2848 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2849 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2850
2851 // do it here after xmm_result0 is saved, because xmm_from3 reuse the same register of xmm_result0.
2852 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2853 __ pxor(xmm_result3, xmm_from3);
2854 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2855
2856 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
2857 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
2858 __ jmp(L_multiBlock_loopTop[k]);
2859
2860 // singleBlock starts here
2861 __ align(OptoLoopAlignment);
2862 __ BIND(L_singleBlockLoopTop[k]);
2863 __ cmpptr(len_reg, 0);
2864 __ jcc(Assembler::equal, L_exit);
2865 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2866 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2867 __ movdqa(xmm_result0, xmm_curr_counter);
2868 load_key(xmm_key, key, 0x00, xmm_key_shuf_mask);
2869 __ push(rbx);//rbx is used for increasing counter
2870 inc_counter(rbx, xmm_curr_counter, 0x01, L_incCounter_single[k]);
2871 __ pop (rbx);
2872 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
2873 __ pxor(xmm_result0, xmm_key);
2874 for (int i = 1; i < rounds[k]; i++) {
2875 load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2876 __ aesenc(xmm_result0, xmm_key);
2877 }
2878 load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2879 __ aesenclast(xmm_result0, xmm_key);
2880 __ cmpptr(len_reg, AESBlockSize);
2881 __ jcc(Assembler::less, L_processTail_insr[k]);
2882 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2883 __ pxor(xmm_result0, xmm_from0);
2884 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2885 __ addptr(pos, AESBlockSize);
2886 __ subptr(len_reg, AESBlockSize);
2887 __ jmp(L_singleBlockLoopTop[k]);
2888
2889 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
2890 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
2891 __ testptr(len_reg, 8);
2892 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
2893 __ subptr(pos,8);
2894 __ pinsrd(xmm_from0, Address(from, pos), 0);
2895 __ pinsrd(xmm_from0, Address(from, pos, Address::times_1, 4), 1);
2896 __ BIND(L_processTail_4_insr[k]);
2897 __ testptr(len_reg, 4);
2898 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
2899 __ subptr(pos,4);
2900 __ pslldq(xmm_from0, 4);
2901 __ pinsrd(xmm_from0, Address(from, pos), 0);
2902 __ BIND(L_processTail_2_insr[k]);
2903 __ testptr(len_reg, 2);
2904 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
2905 __ subptr(pos, 2);
2906 __ pslldq(xmm_from0, 2);
2907 __ pinsrw(xmm_from0, Address(from, pos), 0);
2908 __ BIND(L_processTail_1_insr[k]);
2909 __ testptr(len_reg, 1);
2910 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
2911 __ subptr(pos, 1);
2912 __ pslldq(xmm_from0, 1);
2913 __ pinsrb(xmm_from0, Address(from, pos), 0);
2914 __ BIND(L_processTail_exit_insr[k]);
2915
2916 __ movptr(saved_encCounter_start, saved_counter_param);
2917 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
2918 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
2919
2920 __ testptr(len_reg, 8);
2921 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
2922 __ pextrd(Address(to, pos), xmm_result0, 0);
2923 __ pextrd(Address(to, pos, Address::times_1, 4), xmm_result0, 1);
2924 __ psrldq(xmm_result0, 8);
2925 __ addptr(pos, 8);
2926 __ BIND(L_processTail_4_extr[k]);
2927 __ testptr(len_reg, 4);
2928 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
2929 __ pextrd(Address(to, pos), xmm_result0, 0);
2930 __ psrldq(xmm_result0, 4);
2931 __ addptr(pos, 4);
2932 __ BIND(L_processTail_2_extr[k]);
2933 __ testptr(len_reg, 2);
2934 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
2935 __ pextrb(Address(to, pos), xmm_result0, 0);
2936 __ pextrb(Address(to, pos, Address::times_1, 1), xmm_result0, 1);
2937 __ psrldq(xmm_result0, 2);
2938 __ addptr(pos, 2);
2939 __ BIND(L_processTail_1_extr[k]);
2940 __ testptr(len_reg, 1);
2941 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
2942 __ pextrb(Address(to, pos), xmm_result0, 0);
2943
2944 __ BIND(L_processTail_exit_extr[k]);
2945 __ movptr(used_addr, used_addr_param);
2946 __ movl(Address(used_addr, 0), len_reg);
2947 __ jmp(L_exit);
2948 }
2949
2950 __ BIND(L_exit);
2951 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2952 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
2953 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
2954 handleSOERegisters(false /*restoring*/);
2955 __ movptr(rax, len_param); // return length
2956 __ leave(); // required for proper stackwalking of RuntimeStub frame
2957 __ ret(0);
2958
2959 __ BIND (L_key192_top);
2960 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2961 __ jmp(L_multiBlock_loopTop[1]); //key192
2962
2963 __ BIND (L_key256_top);
2964 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2965 __ jmp(L_multiBlock_loopTop[2]); //key192
2966
2967 return start;
2968 }
2969
2970 address generate_upper_word_mask() {
2971 __ align(64);
2972 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
2973 address start = __ pc();
2974 __ emit_data(0x00000000, relocInfo::none, 0);
2975 __ emit_data(0x00000000, relocInfo::none, 0);
2976 __ emit_data(0x00000000, relocInfo::none, 0);
2977 __ emit_data(0xFFFFFFFF, relocInfo::none, 0);
2978 return start;
2979 }
2980
2981 address generate_shuffle_byte_flip_mask() {
2982 __ align(64);
2983 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
2984 address start = __ pc();
2985 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
2986 __ emit_data(0x08090a0b, relocInfo::none, 0);
2987 __ emit_data(0x04050607, relocInfo::none, 0);
2988 __ emit_data(0x00010203, relocInfo::none, 0);
2989 return start;
2990 }
2991
2992 // ofs and limit are use for multi-block byte array.
2993 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
2994 address generate_sha1_implCompress(bool multi_block, const char *name) {
2995 __ align(CodeEntryAlignment);
2996 StubCodeMark mark(this, "StubRoutines", name);
2997 address start = __ pc();
2998
2999 Register buf = rax;
3000 Register state = rdx;
3001 Register ofs = rcx;
3002 Register limit = rdi;
3003
3004 const Address buf_param(rbp, 8 + 0);
3005 const Address state_param(rbp, 8 + 4);
3006 const Address ofs_param(rbp, 8 + 8);
3007 const Address limit_param(rbp, 8 + 12);
3008
3009 const XMMRegister abcd = xmm0;
3010 const XMMRegister e0 = xmm1;
3011 const XMMRegister e1 = xmm2;
3012 const XMMRegister msg0 = xmm3;
3013
3014 const XMMRegister msg1 = xmm4;
3015 const XMMRegister msg2 = xmm5;
3016 const XMMRegister msg3 = xmm6;
3017 const XMMRegister shuf_mask = xmm7;
3018
3019 __ enter();
3020 __ subptr(rsp, 8 * wordSize);
3021 if (multi_block) {
3022 __ push(limit);
3023 }
3024 __ movptr(buf, buf_param);
3025 __ movptr(state, state_param);
3026 if (multi_block) {
3027 __ movptr(ofs, ofs_param);
3028 __ movptr(limit, limit_param);
3029 }
3030
3031 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3032 buf, state, ofs, limit, rsp, multi_block);
3033
3034 if (multi_block) {
3035 __ pop(limit);
3036 }
3037 __ addptr(rsp, 8 * wordSize);
3038 __ leave();
3039 __ ret(0);
3040 return start;
3041 }
3042
3043 address generate_pshuffle_byte_flip_mask() {
3044 __ align(64);
3045 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3046 address start = __ pc();
3047 __ emit_data(0x00010203, relocInfo::none, 0);
3048 __ emit_data(0x04050607, relocInfo::none, 0);
3049 __ emit_data(0x08090a0b, relocInfo::none, 0);
3050 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3051 return start;
3052 }
3053
3054 // ofs and limit are use for multi-block byte array.
3055 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3056 address generate_sha256_implCompress(bool multi_block, const char *name) {
3057 __ align(CodeEntryAlignment);
3058 StubCodeMark mark(this, "StubRoutines", name);
3059 address start = __ pc();
3060
3061 Register buf = rbx;
3062 Register state = rsi;
3063 Register ofs = rdx;
3064 Register limit = rcx;
3065
3066 const Address buf_param(rbp, 8 + 0);
3067 const Address state_param(rbp, 8 + 4);
3068 const Address ofs_param(rbp, 8 + 8);
3069 const Address limit_param(rbp, 8 + 12);
3070
3071 const XMMRegister msg = xmm0;
3072 const XMMRegister state0 = xmm1;
3073 const XMMRegister state1 = xmm2;
3074 const XMMRegister msgtmp0 = xmm3;
3075
3076 const XMMRegister msgtmp1 = xmm4;
3077 const XMMRegister msgtmp2 = xmm5;
3078 const XMMRegister msgtmp3 = xmm6;
3079 const XMMRegister msgtmp4 = xmm7;
3080
3081 __ enter();
3082 __ subptr(rsp, 8 * wordSize);
3083 handleSOERegisters(true /*saving*/);
3084 __ movptr(buf, buf_param);
3085 __ movptr(state, state_param);
3086 if (multi_block) {
3087 __ movptr(ofs, ofs_param);
3088 __ movptr(limit, limit_param);
3089 }
3090
3091 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3092 buf, state, ofs, limit, rsp, multi_block);
3093
3094 handleSOERegisters(false);
3095 __ addptr(rsp, 8 * wordSize);
3096 __ leave();
3097 __ ret(0);
3098 return start;
3099 }
3100
3101 // byte swap x86 long
3102 address generate_ghash_long_swap_mask() {
3103 __ align(CodeEntryAlignment);
3104 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
3105 address start = __ pc();
3106 __ emit_data(0x0b0a0908, relocInfo::none, 0);
3107 __ emit_data(0x0f0e0d0c, relocInfo::none, 0);
3108 __ emit_data(0x03020100, relocInfo::none, 0);
3109 __ emit_data(0x07060504, relocInfo::none, 0);
3110
3111 return start;
3112 }
3113
3114 // byte swap x86 byte array
3115 address generate_ghash_byte_swap_mask() {
3116 __ align(CodeEntryAlignment);
3117 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
3118 address start = __ pc();
3119 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3120 __ emit_data(0x08090a0b, relocInfo::none, 0);
3121 __ emit_data(0x04050607, relocInfo::none, 0);
3122 __ emit_data(0x00010203, relocInfo::none, 0);
3123 return start;
3124 }
3125
3126 /* Single and multi-block ghash operations */
3127 address generate_ghash_processBlocks() {
3128 assert(UseGHASHIntrinsics, "need GHASH intrinsics and CLMUL support");
3129 __ align(CodeEntryAlignment);
3130 Label L_ghash_loop, L_exit;
3131 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
3132 address start = __ pc();
3133
3134 const Register state = rdi;
3135 const Register subkeyH = rsi;
3136 const Register data = rdx;
3137 const Register blocks = rcx;
3138
3139 const Address state_param(rbp, 8+0);
3140 const Address subkeyH_param(rbp, 8+4);
3141 const Address data_param(rbp, 8+8);
3142 const Address blocks_param(rbp, 8+12);
3143
3144 const XMMRegister xmm_temp0 = xmm0;
3145 const XMMRegister xmm_temp1 = xmm1;
3146 const XMMRegister xmm_temp2 = xmm2;
3147 const XMMRegister xmm_temp3 = xmm3;
3148 const XMMRegister xmm_temp4 = xmm4;
3149 const XMMRegister xmm_temp5 = xmm5;
3150 const XMMRegister xmm_temp6 = xmm6;
3151 const XMMRegister xmm_temp7 = xmm7;
3152
3153 __ enter();
3154 handleSOERegisters(true); // Save registers
3155
3156 // For EVEX with VL and BW, provide a standard mask, VL = 128 will guide the merge
3157 // context for the registers used, where all instructions below are using 128-bit mode
3158 // On EVEX without VL and BW, these instructions will all be AVX.
3159 if (VM_Version::supports_avx512vlbw()) {
3160 __ movl(rdx, 0xffff);
3161 __ kmovdl(k1, rdx);
3162 }
3163
3164 __ movptr(state, state_param);
3165 __ movptr(subkeyH, subkeyH_param);
3166 __ movptr(data, data_param);
3167 __ movptr(blocks, blocks_param);
3168
3169 __ movdqu(xmm_temp0, Address(state, 0));
3170 __ pshufb(xmm_temp0, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3171
3172 __ movdqu(xmm_temp1, Address(subkeyH, 0));
3173 __ pshufb(xmm_temp1, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3174
3175 __ BIND(L_ghash_loop);
3176 __ movdqu(xmm_temp2, Address(data, 0));
3177 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
3178
3179 __ pxor(xmm_temp0, xmm_temp2);
3180
3181 //
3182 // Multiply with the hash key
3183 //
3184 __ movdqu(xmm_temp3, xmm_temp0);
3185 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
3186 __ movdqu(xmm_temp4, xmm_temp0);
3187 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
3188
3189 __ movdqu(xmm_temp5, xmm_temp0);
3190 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
3191 __ movdqu(xmm_temp6, xmm_temp0);
3192 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
3193
3194 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
3195
3196 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
3197 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
3198 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
3199 __ pxor(xmm_temp3, xmm_temp5);
3200 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
3201 // of the carry-less multiplication of
3202 // xmm0 by xmm1.
3203
3204 // We shift the result of the multiplication by one bit position
3205 // to the left to cope for the fact that the bits are reversed.
3206 __ movdqu(xmm_temp7, xmm_temp3);
3207 __ movdqu(xmm_temp4, xmm_temp6);
3208 __ pslld (xmm_temp3, 1);
3209 __ pslld(xmm_temp6, 1);
3210 __ psrld(xmm_temp7, 31);
3211 __ psrld(xmm_temp4, 31);
3212 __ movdqu(xmm_temp5, xmm_temp7);
3213 __ pslldq(xmm_temp4, 4);
3214 __ pslldq(xmm_temp7, 4);
3215 __ psrldq(xmm_temp5, 12);
3216 __ por(xmm_temp3, xmm_temp7);
3217 __ por(xmm_temp6, xmm_temp4);
3218 __ por(xmm_temp6, xmm_temp5);
3219
3220 //
3221 // First phase of the reduction
3222 //
3223 // Move xmm3 into xmm4, xmm5, xmm7 in order to perform the shifts
3224 // independently.
3225 __ movdqu(xmm_temp7, xmm_temp3);
3226 __ movdqu(xmm_temp4, xmm_temp3);
3227 __ movdqu(xmm_temp5, xmm_temp3);
3228 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
3229 __ pslld(xmm_temp4, 30); // packed right shift shifting << 30
3230 __ pslld(xmm_temp5, 25); // packed right shift shifting << 25
3231 __ pxor(xmm_temp7, xmm_temp4); // xor the shifted versions
3232 __ pxor(xmm_temp7, xmm_temp5);
3233 __ movdqu(xmm_temp4, xmm_temp7);
3234 __ pslldq(xmm_temp7, 12);
3235 __ psrldq(xmm_temp4, 4);
3236 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
3237
3238 //
3239 // Second phase of the reduction
3240 //
3241 // Make 3 copies of xmm3 in xmm2, xmm5, xmm7 for doing these
3242 // shift operations.
3243 __ movdqu(xmm_temp2, xmm_temp3);
3244 __ movdqu(xmm_temp7, xmm_temp3);
3245 __ movdqu(xmm_temp5, xmm_temp3);
3246 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
3247 __ psrld(xmm_temp7, 2); // packed left shifting >> 2
3248 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
3249 __ pxor(xmm_temp2, xmm_temp7); // xor the shifted versions
3250 __ pxor(xmm_temp2, xmm_temp5);
3251 __ pxor(xmm_temp2, xmm_temp4);
3252 __ pxor(xmm_temp3, xmm_temp2);
3253 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
3254
3255 __ decrement(blocks);
3256 __ jcc(Assembler::zero, L_exit);
3257 __ movdqu(xmm_temp0, xmm_temp6);
3258 __ addptr(data, 16);
3259 __ jmp(L_ghash_loop);
3260
3261 __ BIND(L_exit);
3262 // Byte swap 16-byte result
3263 __ pshufb(xmm_temp6, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3264 __ movdqu(Address(state, 0), xmm_temp6); // store the result
3265
3266 handleSOERegisters(false); // restore registers
3267 __ leave();
3268 __ ret(0);
3269 return start;
3270 }
3271
3272 /**
3273 * Arguments:
3274 *
3275 * Inputs:
3276 * rsp(4) - int crc
3277 * rsp(8) - byte* buf
3278 * rsp(12) - int length
3279 *
3280 * Ouput:
3281 * rax - int crc result
3282 */
3283 address generate_updateBytesCRC32() {
3284 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3285
3286 __ align(CodeEntryAlignment);
3287 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
3288
3289 address start = __ pc();
3290
3291 const Register crc = rdx; // crc
3292 const Register buf = rsi; // source java byte array address
3293 const Register len = rcx; // length
3294 const Register table = rdi; // crc_table address (reuse register)
3295 const Register tmp = rbx;
3296 assert_different_registers(crc, buf, len, table, tmp, rax);
3297
3298 BLOCK_COMMENT("Entry:");
3299 __ enter(); // required for proper stackwalking of RuntimeStub frame
3300 __ push(rsi);
3301 __ push(rdi);
3302 __ push(rbx);
3303
3304 Address crc_arg(rbp, 8 + 0);
3305 Address buf_arg(rbp, 8 + 4);
3306 Address len_arg(rbp, 8 + 8);
3307
3308 // Load up:
3309 __ movl(crc, crc_arg);
3310 __ movptr(buf, buf_arg);
3311 __ movl(len, len_arg);
3312
3313 __ kernel_crc32(crc, buf, len, table, tmp);
3314
3315 __ movl(rax, crc);
3316 __ pop(rbx);
3317 __ pop(rdi);
3318 __ pop(rsi);
3319 __ vzeroupper();
3320 __ leave(); // required for proper stackwalking of RuntimeStub frame
3321 __ ret(0);
3322
3323 return start;
3324 }
3325
3326 /**
3327 * Arguments:
3328 *
3329 * Inputs:
3330 * rsp(4) - int crc
3331 * rsp(8) - byte* buf
3332 * rsp(12) - int length
3333 * rsp(16) - table_start - optional (present only when doing a library_calll,
3334 * not used by x86 algorithm)
3335 *
3336 * Ouput:
3337 * rax - int crc result
3338 */
3339 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
3340 assert(UseCRC32CIntrinsics, "need SSE4_2");
3341 __ align(CodeEntryAlignment);
3342 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
3343 address start = __ pc();
3344 const Register crc = rax; // crc
3345 const Register buf = rcx; // source java byte array address
3346 const Register len = rdx; // length
3347 const Register d = rbx;
3348 const Register g = rsi;
3349 const Register h = rdi;
3350 const Register empty = 0; // will never be used, in order not
3351 // to change a signature for crc32c_IPL_Alg2_Alt2
3352 // between 64/32 I'm just keeping it here
3353 assert_different_registers(crc, buf, len, d, g, h);
3354
3355 BLOCK_COMMENT("Entry:");
3356 __ enter(); // required for proper stackwalking of RuntimeStub frame
3357 Address crc_arg(rsp, 4 + 4 + 0); // ESP+4 +
3358 // we need to add additional 4 because __ enter
3359 // have just pushed ebp on a stack
3360 Address buf_arg(rsp, 4 + 4 + 4);
3361 Address len_arg(rsp, 4 + 4 + 8);
3362 // Load up:
3363 __ movl(crc, crc_arg);
3364 __ movl(buf, buf_arg);
3365 __ movl(len, len_arg);
3366 __ push(d);
3367 __ push(g);
3368 __ push(h);
3369 __ crc32c_ipl_alg2_alt2(crc, buf, len,
3370 d, g, h,
3371 empty, empty, empty,
3372 xmm0, xmm1, xmm2,
3373 is_pclmulqdq_supported);
3374 __ pop(h);
3375 __ pop(g);
3376 __ pop(d);
3377 __ vzeroupper();
3378 __ leave(); // required for proper stackwalking of RuntimeStub frame
3379 __ ret(0);
3380
3381 return start;
3382 }
3383
3384 address generate_libmExp() {
3385 StubCodeMark mark(this, "StubRoutines", "libmExp");
3386
3387 address start = __ pc();
3388
3389 const XMMRegister x0 = xmm0;
3390 const XMMRegister x1 = xmm1;
3391 const XMMRegister x2 = xmm2;
3392 const XMMRegister x3 = xmm3;
3393
3394 const XMMRegister x4 = xmm4;
3395 const XMMRegister x5 = xmm5;
3396 const XMMRegister x6 = xmm6;
3397 const XMMRegister x7 = xmm7;
3398
3399 const Register tmp = rbx;
3400
3401 BLOCK_COMMENT("Entry:");
3402 __ enter(); // required for proper stackwalking of RuntimeStub frame
3403 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3404 __ leave(); // required for proper stackwalking of RuntimeStub frame
3405 __ ret(0);
3406
3407 return start;
3408
3409 }
3410
3411 address generate_libmLog() {
3412 StubCodeMark mark(this, "StubRoutines", "libmLog");
3413
3414 address start = __ pc();
3415
3416 const XMMRegister x0 = xmm0;
3417 const XMMRegister x1 = xmm1;
3418 const XMMRegister x2 = xmm2;
3419 const XMMRegister x3 = xmm3;
3420
3421 const XMMRegister x4 = xmm4;
3422 const XMMRegister x5 = xmm5;
3423 const XMMRegister x6 = xmm6;
3424 const XMMRegister x7 = xmm7;
3425
3426 const Register tmp = rbx;
3427
3428 BLOCK_COMMENT("Entry:");
3429 __ enter(); // required for proper stackwalking of RuntimeStub frame
3430 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3431 __ leave(); // required for proper stackwalking of RuntimeStub frame
3432 __ ret(0);
3433
3434 return start;
3435
3436 }
3437
3438 address generate_libmLog10() {
3439 StubCodeMark mark(this, "StubRoutines", "libmLog10");
3440
3441 address start = __ pc();
3442
3443 const XMMRegister x0 = xmm0;
3444 const XMMRegister x1 = xmm1;
3445 const XMMRegister x2 = xmm2;
3446 const XMMRegister x3 = xmm3;
3447
3448 const XMMRegister x4 = xmm4;
3449 const XMMRegister x5 = xmm5;
3450 const XMMRegister x6 = xmm6;
3451 const XMMRegister x7 = xmm7;
3452
3453 const Register tmp = rbx;
3454
3455 BLOCK_COMMENT("Entry:");
3456 __ enter(); // required for proper stackwalking of RuntimeStub frame
3457 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3458 __ leave(); // required for proper stackwalking of RuntimeStub frame
3459 __ ret(0);
3460
3461 return start;
3462
3463 }
3464
3465 address generate_libmPow() {
3466 StubCodeMark mark(this, "StubRoutines", "libmPow");
3467
3468 address start = __ pc();
3469
3470 const XMMRegister x0 = xmm0;
3471 const XMMRegister x1 = xmm1;
3472 const XMMRegister x2 = xmm2;
3473 const XMMRegister x3 = xmm3;
3474
3475 const XMMRegister x4 = xmm4;
3476 const XMMRegister x5 = xmm5;
3477 const XMMRegister x6 = xmm6;
3478 const XMMRegister x7 = xmm7;
3479
3480 const Register tmp = rbx;
3481
3482 BLOCK_COMMENT("Entry:");
3483 __ enter(); // required for proper stackwalking of RuntimeStub frame
3484 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3485 __ leave(); // required for proper stackwalking of RuntimeStub frame
3486 __ ret(0);
3487
3488 return start;
3489
3490 }
3491
3492 address generate_libm_reduce_pi04l() {
3493 StubCodeMark mark(this, "StubRoutines", "libm_reduce_pi04l");
3494
3495 address start = __ pc();
3496
3497 BLOCK_COMMENT("Entry:");
3498 __ libm_reduce_pi04l(rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3499
3500 return start;
3501
3502 }
3503
3504 address generate_libm_sin_cos_huge() {
3505 StubCodeMark mark(this, "StubRoutines", "libm_sin_cos_huge");
3506
3507 address start = __ pc();
3508
3509 const XMMRegister x0 = xmm0;
3510 const XMMRegister x1 = xmm1;
3511
3512 BLOCK_COMMENT("Entry:");
3513 __ libm_sincos_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3514
3515 return start;
3516
3517 }
3518
3519 address generate_libmSin() {
3520 StubCodeMark mark(this, "StubRoutines", "libmSin");
3521
3522 address start = __ pc();
3523
3524 const XMMRegister x0 = xmm0;
3525 const XMMRegister x1 = xmm1;
3526 const XMMRegister x2 = xmm2;
3527 const XMMRegister x3 = xmm3;
3528
3529 const XMMRegister x4 = xmm4;
3530 const XMMRegister x5 = xmm5;
3531 const XMMRegister x6 = xmm6;
3532 const XMMRegister x7 = xmm7;
3533
3534 BLOCK_COMMENT("Entry:");
3535 __ enter(); // required for proper stackwalking of RuntimeStub frame
3536 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rdx);
3537 __ leave(); // required for proper stackwalking of RuntimeStub frame
3538 __ ret(0);
3539
3540 return start;
3541
3542 }
3543
3544 address generate_libmCos() {
3545 StubCodeMark mark(this, "StubRoutines", "libmCos");
3546
3547 address start = __ pc();
3548
3549 const XMMRegister x0 = xmm0;
3550 const XMMRegister x1 = xmm1;
3551 const XMMRegister x2 = xmm2;
3552 const XMMRegister x3 = xmm3;
3553
3554 const XMMRegister x4 = xmm4;
3555 const XMMRegister x5 = xmm5;
3556 const XMMRegister x6 = xmm6;
3557 const XMMRegister x7 = xmm7;
3558
3559 const Register tmp = rbx;
3560
3561 BLOCK_COMMENT("Entry:");
3562 __ enter(); // required for proper stackwalking of RuntimeStub frame
3563 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3564 __ leave(); // required for proper stackwalking of RuntimeStub frame
3565 __ ret(0);
3566
3567 return start;
3568
3569 }
3570
3571 address generate_libm_tan_cot_huge() {
3572 StubCodeMark mark(this, "StubRoutines", "libm_tan_cot_huge");
3573
3574 address start = __ pc();
3575
3576 const XMMRegister x0 = xmm0;
3577 const XMMRegister x1 = xmm1;
3578
3579 BLOCK_COMMENT("Entry:");
3580 __ libm_tancot_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3581
3582 return start;
3583
3584 }
3585
3586 address generate_libmTan() {
3587 StubCodeMark mark(this, "StubRoutines", "libmTan");
3588
3589 address start = __ pc();
3590
3591 const XMMRegister x0 = xmm0;
3592 const XMMRegister x1 = xmm1;
3593 const XMMRegister x2 = xmm2;
3594 const XMMRegister x3 = xmm3;
3595
3596 const XMMRegister x4 = xmm4;
3597 const XMMRegister x5 = xmm5;
3598 const XMMRegister x6 = xmm6;
3599 const XMMRegister x7 = xmm7;
3600
3601 const Register tmp = rbx;
3602
3603 BLOCK_COMMENT("Entry:");
3604 __ enter(); // required for proper stackwalking of RuntimeStub frame
3605 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3606 __ leave(); // required for proper stackwalking of RuntimeStub frame
3607 __ ret(0);
3608
3609 return start;
3610
3611 }
3612
3613 // Safefetch stubs.
3614 void generate_safefetch(const char* name, int size, address* entry,
3615 address* fault_pc, address* continuation_pc) {
3616 // safefetch signatures:
3617 // int SafeFetch32(int* adr, int errValue);
3618 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3619
3620 StubCodeMark mark(this, "StubRoutines", name);
3621
3622 // Entry point, pc or function descriptor.
3623 *entry = __ pc();
3624
3625 __ movl(rax, Address(rsp, 0x8));
3626 __ movl(rcx, Address(rsp, 0x4));
3627 // Load *adr into eax, may fault.
3628 *fault_pc = __ pc();
3629 switch (size) {
3630 case 4:
3631 // int32_t
3632 __ movl(rax, Address(rcx, 0));
3633 break;
3634 case 8:
3635 // int64_t
3636 Unimplemented();
3637 break;
3638 default:
3639 ShouldNotReachHere();
3640 }
3641
3642 // Return errValue or *adr.
3643 *continuation_pc = __ pc();
3644 __ ret(0);
3645 }
3646
3647 public:
3648 // Information about frame layout at time of blocking runtime call.
3649 // Note that we only have to preserve callee-saved registers since
3650 // the compilers are responsible for supplying a continuation point
3651 // if they expect all registers to be preserved.
3652 enum layout {
3653 thread_off, // last_java_sp
3654 arg1_off,
3655 arg2_off,
3656 rbp_off, // callee saved register
3657 ret_pc,
3658 framesize
3659 };
3660
3661 private:
3662
3663 #undef __
3664 #define __ masm->
3665
3666 //------------------------------------------------------------------------------------------------------------------------
3667 // Continuation point for throwing of implicit exceptions that are not handled in
3668 // the current activation. Fabricates an exception oop and initiates normal
3669 // exception dispatching in this frame.
3670 //
3671 // Previously the compiler (c2) allowed for callee save registers on Java calls.
3672 // This is no longer true after adapter frames were removed but could possibly
3673 // be brought back in the future if the interpreter code was reworked and it
3674 // was deemed worthwhile. The comment below was left to describe what must
3675 // happen here if callee saves were resurrected. As it stands now this stub
3676 // could actually be a vanilla BufferBlob and have now oopMap at all.
3677 // Since it doesn't make much difference we've chosen to leave it the
3678 // way it was in the callee save days and keep the comment.
3679
3680 // If we need to preserve callee-saved values we need a callee-saved oop map and
3681 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
3682 // If the compiler needs all registers to be preserved between the fault
3683 // point and the exception handler then it must assume responsibility for that in
3684 // AbstractCompiler::continuation_for_implicit_null_exception or
3685 // continuation_for_implicit_division_by_zero_exception. All other implicit
3686 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
3687 // either at call sites or otherwise assume that stack unwinding will be initiated,
3688 // so caller saved registers were assumed volatile in the compiler.
3689 address generate_throw_exception(const char* name, address runtime_entry,
3690 Register arg1 = noreg, Register arg2 = noreg) {
3691
3692 int insts_size = 256;
3693 int locs_size = 32;
3694
3695 CodeBuffer code(name, insts_size, locs_size);
3696 OopMapSet* oop_maps = new OopMapSet();
3697 MacroAssembler* masm = new MacroAssembler(&code);
3698
3699 address start = __ pc();
3700
3701 // This is an inlined and slightly modified version of call_VM
3702 // which has the ability to fetch the return PC out of
3703 // thread-local storage and also sets up last_Java_sp slightly
3704 // differently than the real call_VM
3705 Register java_thread = rbx;
3706 __ get_thread(java_thread);
3707
3708 __ enter(); // required for proper stackwalking of RuntimeStub frame
3709
3710 // pc and rbp, already pushed
3711 __ subptr(rsp, (framesize-2) * wordSize); // prolog
3712
3713 // Frame is now completed as far as size and linkage.
3714
3715 int frame_complete = __ pc() - start;
3716
3717 // push java thread (becomes first argument of C function)
3718 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
3719 if (arg1 != noreg) {
3720 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
3721 }
3722 if (arg2 != noreg) {
3723 assert(arg1 != noreg, "missing reg arg");
3724 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
3725 }
3726
3727 // Set up last_Java_sp and last_Java_fp
3728 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
3729
3730 // Call runtime
3731 BLOCK_COMMENT("call runtime_entry");
3732 __ call(RuntimeAddress(runtime_entry));
3733 // Generate oop map
3734 OopMap* map = new OopMap(framesize, 0);
3735 oop_maps->add_gc_map(__ pc() - start, map);
3736
3737 // restore the thread (cannot use the pushed argument since arguments
3738 // may be overwritten by C code generated by an optimizing compiler);
3739 // however can use the register value directly if it is callee saved.
3740 __ get_thread(java_thread);
3741
3742 __ reset_last_Java_frame(java_thread, true);
3743
3744 __ leave(); // required for proper stackwalking of RuntimeStub frame
3745
3746 // check for pending exceptions
3747 #ifdef ASSERT
3748 Label L;
3749 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3750 __ jcc(Assembler::notEqual, L);
3751 __ should_not_reach_here();
3752 __ bind(L);
3753 #endif /* ASSERT */
3754 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3755
3756
3757 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
3758 return stub->entry_point();
3759 }
3760
3761
3762 void create_control_words() {
3763 // Round to nearest, 53-bit mode, exceptions masked
3764 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
3765 // Round to zero, 53-bit mode, exception mased
3766 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
3767 // Round to nearest, 24-bit mode, exceptions masked
3768 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
3769 // Round to nearest, 64-bit mode, exceptions masked
3770 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
3771 // Round to nearest, 64-bit mode, exceptions masked
3772 StubRoutines::_mxcsr_std = 0x1F80;
3773 // Note: the following two constants are 80-bit values
3774 // layout is critical for correct loading by FPU.
3775 // Bias for strict fp multiply/divide
3776 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
3777 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
3778 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
3779 // Un-Bias for strict fp multiply/divide
3780 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
3781 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
3782 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
3783 }
3784
3785 //---------------------------------------------------------------------------
3786 // Initialization
3787
3788 void generate_initial() {
3789 // Generates all stubs and initializes the entry points
3790
3791 //------------------------------------------------------------------------------------------------------------------------
3792 // entry points that exist in all platforms
3793 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
3794 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
3795 StubRoutines::_forward_exception_entry = generate_forward_exception();
3796
3797 StubRoutines::_call_stub_entry =
3798 generate_call_stub(StubRoutines::_call_stub_return_address);
3799 // is referenced by megamorphic call
3800 StubRoutines::_catch_exception_entry = generate_catch_exception();
3801
3802 // These are currently used by Solaris/Intel
3803 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3804
3805 // platform dependent
3806 create_control_words();
3807
3808 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3809 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
3810 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
3811 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
3812 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
3813 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
3814
3815 // Build this early so it's available for the interpreter
3816 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception",
3817 CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
3818 StubRoutines::_throw_delayed_StackOverflowError_entry = generate_throw_exception("delayed StackOverflowError throw_exception",
3819 CAST_FROM_FN_PTR(address, SharedRuntime::throw_delayed_StackOverflowError));
3820
3821 if (UseCRC32Intrinsics) {
3822 // set table address before stub generation which use it
3823 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3824 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3825 }
3826
3827 if (UseCRC32CIntrinsics) {
3828 bool supports_clmul = VM_Version::supports_clmul();
3829 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
3830 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
3831 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
3832 }
3833 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
3834 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3835 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3836 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3837 StubRoutines::x86::_L_2il0floatpacket_0_adr = (address)StubRoutines::x86::_L_2il0floatpacket_0;
3838 StubRoutines::x86::_Pi4Inv_adr = (address)StubRoutines::x86::_Pi4Inv;
3839 StubRoutines::x86::_Pi4x3_adr = (address)StubRoutines::x86::_Pi4x3;
3840 StubRoutines::x86::_Pi4x4_adr = (address)StubRoutines::x86::_Pi4x4;
3841 StubRoutines::x86::_ones_adr = (address)StubRoutines::x86::_ones;
3842 }
3843 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3844 StubRoutines::_dexp = generate_libmExp();
3845 }
3846 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3847 StubRoutines::_dlog = generate_libmLog();
3848 }
3849 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3850 StubRoutines::_dlog10 = generate_libmLog10();
3851 }
3852 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3853 StubRoutines::_dpow = generate_libmPow();
3854 }
3855 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3856 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3857 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3858 StubRoutines::_dlibm_reduce_pi04l = generate_libm_reduce_pi04l();
3859 }
3860 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3861 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3862 StubRoutines::_dlibm_sin_cos_huge = generate_libm_sin_cos_huge();
3863 }
3864 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3865 StubRoutines::_dsin = generate_libmSin();
3866 }
3867 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3868 StubRoutines::_dcos = generate_libmCos();
3869 }
3870 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3871 StubRoutines::_dlibm_tan_cot_huge = generate_libm_tan_cot_huge();
3872 StubRoutines::_dtan = generate_libmTan();
3873 }
3874 }
3875 }
3876
3877 void generate_all() {
3878 // Generates all stubs and initializes the entry points
3879
3880 // These entry points require SharedInfo::stack0 to be set up in non-core builds
3881 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
3882 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
3883 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
3884 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
3885
3886 //------------------------------------------------------------------------------------------------------------------------
3887 // entry points that are platform specific
3888
3889 // support for verify_oop (must happen after universe_init)
3890 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3891
3892 // arraycopy stubs used by compilers
3893 generate_arraycopy_stubs();
3894
3895 // don't bother generating these AES intrinsic stubs unless global flag is set
3896 if (UseAESIntrinsics) {
3897 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
3898
3899 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3900 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3901 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3902 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
3903 }
3904
3905 if (UseAESCTRIntrinsics) {
3906 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
3907 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
3908 }
3909
3910 if (UseSHA1Intrinsics) {
3911 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
3912 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
3913 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
3914 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
3915 }
3916 if (UseSHA256Intrinsics) {
3917 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
3918 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
3919 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
3920 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
3921 }
3922
3923 // Generate GHASH intrinsics code
3924 if (UseGHASHIntrinsics) {
3925 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
3926 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
3927 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
3928 }
3929
3930 // Safefetch stubs.
3931 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3932 &StubRoutines::_safefetch32_fault_pc,
3933 &StubRoutines::_safefetch32_continuation_pc);
3934 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
3935 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
3936 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3937 }
3938
3939
3940 public:
3941 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3942 if (all) {
3943 generate_all();
3944 } else {
3945 generate_initial();
3946 }
3947 }
3948 }; // end class declaration
3949
3950
3951 void StubGenerator_generate(CodeBuffer* code, bool all) {
3952 StubGenerator g(code, all);
3953 }