1 /*
2 * Copyright (c) 1999, 2019, 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/barrierSetAssembler.hpp"
30 #include "gc/shared/barrierSetNMethod.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "memory/universe.hpp"
33 #include "nativeInst_x86.hpp"
34 #include "oops/instanceOop.hpp"
35 #include "oops/method.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/methodHandles.hpp"
39 #include "runtime/frame.inline.hpp"
40 #include "runtime/handles.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubCodeGenerator.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/thread.inline.hpp"
45 #ifdef COMPILER2
46 #include "opto/runtime.hpp"
47 #endif
48
49 // Declaration and definition of StubGenerator (no .hpp file).
50 // For a more detailed description of the stub routine structure
51 // see the comment in stubRoutines.hpp
52
53 #define __ _masm->
54 #define a__ ((Assembler*)_masm)->
55
56 #ifdef PRODUCT
57 #define BLOCK_COMMENT(str) /* nothing */
58 #else
59 #define BLOCK_COMMENT(str) __ block_comment(str)
60 #endif
61
62 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
63
64 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
65 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
66
67 // -------------------------------------------------------------------------------------------------------------------------
68 // Stub Code definitions
69
70 class StubGenerator: public StubCodeGenerator {
71 private:
72
73 #ifdef PRODUCT
74 #define inc_counter_np(counter) ((void)0)
75 #else
76 void inc_counter_np_(int& counter) {
77 __ incrementl(ExternalAddress((address)&counter));
78 }
79 #define inc_counter_np(counter) \
80 BLOCK_COMMENT("inc_counter " #counter); \
81 inc_counter_np_(counter);
82 #endif //PRODUCT
83
84 void inc_copy_counter_np(BasicType t) {
85 #ifndef PRODUCT
86 switch (t) {
87 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
88 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
89 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
90 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
91 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
92 default: ShouldNotReachHere();
93 }
94 #endif //PRODUCT
95 }
96
97 //------------------------------------------------------------------------------------------------------------------------
98 // Call stubs are used to call Java from C
99 //
100 // [ return_from_Java ] <--- rsp
101 // [ argument word n ]
102 // ...
103 // -N [ argument word 1 ]
104 // -7 [ Possible padding for stack alignment ]
105 // -6 [ Possible padding for stack alignment ]
106 // -5 [ Possible padding for stack alignment ]
107 // -4 [ mxcsr save ] <--- rsp_after_call
108 // -3 [ saved rbx, ]
109 // -2 [ saved rsi ]
110 // -1 [ saved rdi ]
111 // 0 [ saved rbp, ] <--- rbp,
112 // 1 [ return address ]
113 // 2 [ ptr. to call wrapper ]
114 // 3 [ result ]
115 // 4 [ result_type ]
116 // 5 [ method ]
117 // 6 [ entry_point ]
118 // 7 [ parameters ]
119 // 8 [ parameter_size ]
120 // 9 [ thread ]
121
122
123 address generate_call_stub(address& return_address) {
124 StubCodeMark mark(this, "StubRoutines", "call_stub");
125 address start = __ pc();
126
127 // stub code parameters / addresses
128 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
129 bool sse_save = false;
130 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
131 const int locals_count_in_bytes (4*wordSize);
132 const Address mxcsr_save (rbp, -4 * wordSize);
133 const Address saved_rbx (rbp, -3 * wordSize);
134 const Address saved_rsi (rbp, -2 * wordSize);
135 const Address saved_rdi (rbp, -1 * wordSize);
136 const Address result (rbp, 3 * wordSize);
137 const Address result_type (rbp, 4 * wordSize);
138 const Address method (rbp, 5 * wordSize);
139 const Address entry_point (rbp, 6 * wordSize);
140 const Address parameters (rbp, 7 * wordSize);
141 const Address parameter_size(rbp, 8 * wordSize);
142 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
143 sse_save = UseSSE > 0;
144
145 // stub code
146 __ enter();
147 __ movptr(rcx, parameter_size); // parameter counter
148 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
149 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
150 __ subptr(rsp, rcx);
151 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
152
153 // save rdi, rsi, & rbx, according to C calling conventions
154 __ movptr(saved_rdi, rdi);
155 __ movptr(saved_rsi, rsi);
156 __ movptr(saved_rbx, rbx);
157
158 // save and initialize %mxcsr
159 if (sse_save) {
160 Label skip_ldmx;
161 __ stmxcsr(mxcsr_save);
162 __ movl(rax, mxcsr_save);
163 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
164 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
165 __ cmp32(rax, mxcsr_std);
166 __ jcc(Assembler::equal, skip_ldmx);
167 __ ldmxcsr(mxcsr_std);
168 __ bind(skip_ldmx);
169 }
170
171 // make sure the control word is correct.
172 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
173
174 #ifdef ASSERT
175 // make sure we have no pending exceptions
176 { Label L;
177 __ movptr(rcx, thread);
178 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
179 __ jcc(Assembler::equal, L);
180 __ stop("StubRoutines::call_stub: entered with pending exception");
181 __ bind(L);
182 }
183 #endif
184
185 // pass parameters if any
186 BLOCK_COMMENT("pass parameters if any");
187 Label parameters_done;
188 __ movl(rcx, parameter_size); // parameter counter
189 __ testl(rcx, rcx);
190 __ jcc(Assembler::zero, parameters_done);
191
192 // parameter passing loop
193
194 Label loop;
195 // Copy Java parameters in reverse order (receiver last)
196 // Note that the argument order is inverted in the process
197 // source is rdx[rcx: N-1..0]
198 // dest is rsp[rbx: 0..N-1]
199
200 __ movptr(rdx, parameters); // parameter pointer
201 __ xorptr(rbx, rbx);
202
203 __ BIND(loop);
204
205 // get parameter
206 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
207 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
208 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
209 __ increment(rbx);
210 __ decrement(rcx);
211 __ jcc(Assembler::notZero, loop);
212
213 // call Java function
214 __ BIND(parameters_done);
215 __ movptr(rbx, method); // get Method*
216 __ movptr(rax, entry_point); // get entry_point
217 __ mov(rsi, rsp); // set sender sp
218 BLOCK_COMMENT("call Java function");
219 __ call(rax);
220
221 BLOCK_COMMENT("call_stub_return_address:");
222 return_address = __ pc();
223
224 #ifdef COMPILER2
225 {
226 Label L_skip;
227 if (UseSSE >= 2) {
228 __ verify_FPU(0, "call_stub_return");
229 } else {
230 for (int i = 1; i < 8; i++) {
231 __ ffree(i);
232 }
233
234 // UseSSE <= 1 so double result should be left on TOS
235 __ movl(rsi, result_type);
236 __ cmpl(rsi, T_DOUBLE);
237 __ jcc(Assembler::equal, L_skip);
238 if (UseSSE == 0) {
239 // UseSSE == 0 so float result should be left on TOS
240 __ cmpl(rsi, T_FLOAT);
241 __ jcc(Assembler::equal, L_skip);
242 }
243 __ ffree(0);
244 }
245 __ BIND(L_skip);
246 }
247 #endif // COMPILER2
248
249 // store result depending on type
250 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
251 __ movptr(rdi, result);
252 Label is_long, is_float, is_double, exit;
253 __ movl(rsi, result_type);
254 __ cmpl(rsi, T_LONG);
255 __ jcc(Assembler::equal, is_long);
256 __ cmpl(rsi, T_FLOAT);
257 __ jcc(Assembler::equal, is_float);
258 __ cmpl(rsi, T_DOUBLE);
259 __ jcc(Assembler::equal, is_double);
260
261 // handle T_INT case
262 __ movl(Address(rdi, 0), rax);
263 __ BIND(exit);
264
265 // check that FPU stack is empty
266 __ verify_FPU(0, "generate_call_stub");
267
268 // pop parameters
269 __ lea(rsp, rsp_after_call);
270
271 // restore %mxcsr
272 if (sse_save) {
273 __ ldmxcsr(mxcsr_save);
274 }
275
276 // restore rdi, rsi and rbx,
277 __ movptr(rbx, saved_rbx);
278 __ movptr(rsi, saved_rsi);
279 __ movptr(rdi, saved_rdi);
280 __ addptr(rsp, 4*wordSize);
281
282 // return
283 __ pop(rbp);
284 __ ret(0);
285
286 // handle return types different from T_INT
287 __ BIND(is_long);
288 __ movl(Address(rdi, 0 * wordSize), rax);
289 __ movl(Address(rdi, 1 * wordSize), rdx);
290 __ jmp(exit);
291
292 __ BIND(is_float);
293 // interpreter uses xmm0 for return values
294 if (UseSSE >= 1) {
295 __ movflt(Address(rdi, 0), xmm0);
296 } else {
297 __ fstp_s(Address(rdi, 0));
298 }
299 __ jmp(exit);
300
301 __ BIND(is_double);
302 // interpreter uses xmm0 for return values
303 if (UseSSE >= 2) {
304 __ movdbl(Address(rdi, 0), xmm0);
305 } else {
306 __ fstp_d(Address(rdi, 0));
307 }
308 __ jmp(exit);
309
310 return start;
311 }
312
313
314 //------------------------------------------------------------------------------------------------------------------------
315 // Return point for a Java call if there's an exception thrown in Java code.
316 // The exception is caught and transformed into a pending exception stored in
317 // JavaThread that can be tested from within the VM.
318 //
319 // Note: Usually the parameters are removed by the callee. In case of an exception
320 // crossing an activation frame boundary, that is not the case if the callee
321 // is compiled code => need to setup the rsp.
322 //
323 // rax,: exception oop
324
325 address generate_catch_exception() {
326 StubCodeMark mark(this, "StubRoutines", "catch_exception");
327 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
328 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
329 address start = __ pc();
330
331 // get thread directly
332 __ movptr(rcx, thread);
333 #ifdef ASSERT
334 // verify that threads correspond
335 { Label L;
336 __ get_thread(rbx);
337 __ cmpptr(rbx, rcx);
338 __ jcc(Assembler::equal, L);
339 __ stop("StubRoutines::catch_exception: threads must correspond");
340 __ bind(L);
341 }
342 #endif
343 // set pending exception
344 __ verify_oop(rax);
345 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
346 __ lea(Address(rcx, Thread::exception_file_offset ()),
347 ExternalAddress((address)__FILE__));
348 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
349 // complete return to VM
350 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
351 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
352
353 return start;
354 }
355
356
357 //------------------------------------------------------------------------------------------------------------------------
358 // Continuation point for runtime calls returning with a pending exception.
359 // The pending exception check happened in the runtime or native call stub.
360 // The pending exception in Thread is converted into a Java-level exception.
361 //
362 // Contract with Java-level exception handlers:
363 // rax: exception
364 // rdx: throwing pc
365 //
366 // NOTE: At entry of this stub, exception-pc must be on stack !!
367
368 address generate_forward_exception() {
369 StubCodeMark mark(this, "StubRoutines", "forward exception");
370 address start = __ pc();
371 const Register thread = rcx;
372
373 // other registers used in this stub
374 const Register exception_oop = rax;
375 const Register handler_addr = rbx;
376 const Register exception_pc = rdx;
377
378 // Upon entry, the sp points to the return address returning into Java
379 // (interpreted or compiled) code; i.e., the return address becomes the
380 // throwing pc.
381 //
382 // Arguments pushed before the runtime call are still on the stack but
383 // the exception handler will reset the stack pointer -> ignore them.
384 // A potential result in registers can be ignored as well.
385
386 #ifdef ASSERT
387 // make sure this code is only executed if there is a pending exception
388 { Label L;
389 __ get_thread(thread);
390 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
391 __ jcc(Assembler::notEqual, L);
392 __ stop("StubRoutines::forward exception: no pending exception (1)");
393 __ bind(L);
394 }
395 #endif
396
397 // compute exception handler into rbx,
398 __ get_thread(thread);
399 __ movptr(exception_pc, Address(rsp, 0));
400 BLOCK_COMMENT("call exception_handler_for_return_address");
401 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
402 __ mov(handler_addr, rax);
403
404 // setup rax & rdx, remove return address & clear pending exception
405 __ get_thread(thread);
406 __ pop(exception_pc);
407 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
408 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
409
410 #ifdef ASSERT
411 // make sure exception is set
412 { Label L;
413 __ testptr(exception_oop, exception_oop);
414 __ jcc(Assembler::notEqual, L);
415 __ stop("StubRoutines::forward exception: no pending exception (2)");
416 __ bind(L);
417 }
418 #endif
419
420 // Verify that there is really a valid exception in RAX.
421 __ verify_oop(exception_oop);
422
423 // continue at exception handler (return address removed)
424 // rax: exception
425 // rbx: exception handler
426 // rdx: throwing pc
427 __ jmp(handler_addr);
428
429 return start;
430 }
431
432
433 //----------------------------------------------------------------------------------------------------
434 // Implementation of int32_t atomic_xchg(int32_t exchange_value, volatile int32_t* dest)
435 // used by Atomic::xchg(volatile int32_t* dest, int32_t exchange_value)
436 //
437 // xchg exists as far back as 8086, lock needed for MP only
438 // Stack layout immediately after call:
439 //
440 // 0 [ret addr ] <--- rsp
441 // 1 [ ex ]
442 // 2 [ dest ]
443 //
444 // Result: *dest <- ex, return (old *dest)
445 //
446 // Note: win32 does not currently use this code
447
448 address generate_atomic_xchg() {
449 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
450 address start = __ pc();
451
452 __ push(rdx);
453 Address exchange(rsp, 2 * wordSize);
454 Address dest_addr(rsp, 3 * wordSize);
455 __ movl(rax, exchange);
456 __ movptr(rdx, dest_addr);
457 __ xchgl(rax, Address(rdx, 0));
458 __ pop(rdx);
459 __ ret(0);
460
461 return start;
462 }
463
464 //----------------------------------------------------------------------------------------------------
465 // Support for void verify_mxcsr()
466 //
467 // This routine is used with -Xcheck:jni to verify that native
468 // JNI code does not return to Java code without restoring the
469 // MXCSR register to our expected state.
470
471
472 address generate_verify_mxcsr() {
473 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
474 address start = __ pc();
475
476 const Address mxcsr_save(rsp, 0);
477
478 if (CheckJNICalls && UseSSE > 0 ) {
479 Label ok_ret;
480 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
481 __ push(rax);
482 __ subptr(rsp, wordSize); // allocate a temp location
483 __ stmxcsr(mxcsr_save);
484 __ movl(rax, mxcsr_save);
485 __ andl(rax, MXCSR_MASK);
486 __ cmp32(rax, mxcsr_std);
487 __ jcc(Assembler::equal, ok_ret);
488
489 __ warn("MXCSR changed by native JNI code.");
490
491 __ ldmxcsr(mxcsr_std);
492
493 __ bind(ok_ret);
494 __ addptr(rsp, wordSize);
495 __ pop(rax);
496 }
497
498 __ ret(0);
499
500 return start;
501 }
502
503
504 //---------------------------------------------------------------------------
505 // Support for void verify_fpu_cntrl_wrd()
506 //
507 // This routine is used with -Xcheck:jni to verify that native
508 // JNI code does not return to Java code without restoring the
509 // FP control word to our expected state.
510
511 address generate_verify_fpu_cntrl_wrd() {
512 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
513 address start = __ pc();
514
515 const Address fpu_cntrl_wrd_save(rsp, 0);
516
517 if (CheckJNICalls) {
518 Label ok_ret;
519 __ push(rax);
520 __ subptr(rsp, wordSize); // allocate a temp location
521 __ fnstcw(fpu_cntrl_wrd_save);
522 __ movl(rax, fpu_cntrl_wrd_save);
523 __ andl(rax, FPU_CNTRL_WRD_MASK);
524 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
525 __ cmp32(rax, fpu_std);
526 __ jcc(Assembler::equal, ok_ret);
527
528 __ warn("Floating point control word changed by native JNI code.");
529
530 __ fldcw(fpu_std);
531
532 __ bind(ok_ret);
533 __ addptr(rsp, wordSize);
534 __ pop(rax);
535 }
536
537 __ ret(0);
538
539 return start;
540 }
541
542 //---------------------------------------------------------------------------
543 // Wrapper for slow-case handling of double-to-integer conversion
544 // d2i or f2i fast case failed either because it is nan or because
545 // of under/overflow.
546 // Input: FPU TOS: float value
547 // Output: rax, (rdx): integer (long) result
548
549 address generate_d2i_wrapper(BasicType t, address fcn) {
550 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
551 address start = __ pc();
552
553 // Capture info about frame layout
554 enum layout { FPUState_off = 0,
555 rbp_off = FPUStateSizeInWords,
556 rdi_off,
557 rsi_off,
558 rcx_off,
559 rbx_off,
560 saved_argument_off,
561 saved_argument_off2, // 2nd half of double
562 framesize
563 };
564
565 assert(FPUStateSizeInWords == 27, "update stack layout");
566
567 // Save outgoing argument to stack across push_FPU_state()
568 __ subptr(rsp, wordSize * 2);
569 __ fstp_d(Address(rsp, 0));
570
571 // Save CPU & FPU state
572 __ push(rbx);
573 __ push(rcx);
574 __ push(rsi);
575 __ push(rdi);
576 __ push(rbp);
577 __ push_FPU_state();
578
579 // push_FPU_state() resets the FP top of stack
580 // Load original double into FP top of stack
581 __ fld_d(Address(rsp, saved_argument_off * wordSize));
582 // Store double into stack as outgoing argument
583 __ subptr(rsp, wordSize*2);
584 __ fst_d(Address(rsp, 0));
585
586 // Prepare FPU for doing math in C-land
587 __ empty_FPU_stack();
588 // Call the C code to massage the double. Result in EAX
589 if (t == T_INT)
590 { BLOCK_COMMENT("SharedRuntime::d2i"); }
591 else if (t == T_LONG)
592 { BLOCK_COMMENT("SharedRuntime::d2l"); }
593 __ call_VM_leaf( fcn, 2 );
594
595 // Restore CPU & FPU state
596 __ pop_FPU_state();
597 __ pop(rbp);
598 __ pop(rdi);
599 __ pop(rsi);
600 __ pop(rcx);
601 __ pop(rbx);
602 __ addptr(rsp, wordSize * 2);
603
604 __ ret(0);
605
606 return start;
607 }
608 //---------------------------------------------------------------------------------------------------
609
610 address generate_vector_mask(const char *stub_name, int32_t mask) {
611 __ align(CodeEntryAlignment);
612 StubCodeMark mark(this, "StubRoutines", stub_name);
613 address start = __ pc();
614
615 for (int i = 0; i < 16; i++) {
616 __ emit_data(mask, relocInfo::none, 0);
617 }
618
619 return start;
620 }
621
622 address generate_vector_mask_long_double(const char *stub_name, int32_t maskhi, int32_t masklo) {
623 __ align(CodeEntryAlignment);
624 StubCodeMark mark(this, "StubRoutines", stub_name);
625 address start = __ pc();
626
627 for (int i = 0; i < 8; i++) {
628 __ emit_data(masklo, relocInfo::none, 0);
629 __ emit_data(maskhi, relocInfo::none, 0);
630 }
631
632 return start;
633 }
634
635 //----------------------------------------------------------------------------------------------------
636
637 address generate_vector_byte_perm_mask(const char *stub_name) {
638 __ align(CodeEntryAlignment);
639 StubCodeMark mark(this, "StubRoutines", stub_name);
640 address start = __ pc();
641
642 __ emit_data(0x00000001, relocInfo::none, 0);
643 __ emit_data(0x00000000, relocInfo::none, 0);
644 __ emit_data(0x00000003, relocInfo::none, 0);
645 __ emit_data(0x00000000, relocInfo::none, 0);
646 __ emit_data(0x00000005, relocInfo::none, 0);
647 __ emit_data(0x00000000, relocInfo::none, 0);
648 __ emit_data(0x00000007, relocInfo::none, 0);
649 __ emit_data(0x00000000, relocInfo::none, 0);
650 __ emit_data(0x00000000, relocInfo::none, 0);
651 __ emit_data(0x00000000, relocInfo::none, 0);
652 __ emit_data(0x00000002, relocInfo::none, 0);
653 __ emit_data(0x00000000, relocInfo::none, 0);
654 __ emit_data(0x00000004, relocInfo::none, 0);
655 __ emit_data(0x00000000, relocInfo::none, 0);
656 __ emit_data(0x00000006, relocInfo::none, 0);
657 __ emit_data(0x00000000, relocInfo::none, 0);
658
659 return start;
660 }
661
662 //----------------------------------------------------------------------------------------------------
663 // Non-destructive plausibility checks for oops
664
665 address generate_verify_oop() {
666 StubCodeMark mark(this, "StubRoutines", "verify_oop");
667 address start = __ pc();
668
669 // Incoming arguments on stack after saving rax,:
670 //
671 // [tos ]: saved rdx
672 // [tos + 1]: saved EFLAGS
673 // [tos + 2]: return address
674 // [tos + 3]: char* error message
675 // [tos + 4]: oop object to verify
676 // [tos + 5]: saved rax, - saved by caller and bashed
677
678 Label exit, error;
679 __ pushf();
680 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
681 __ push(rdx); // save rdx
682 // make sure object is 'reasonable'
683 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
684 __ testptr(rax, rax);
685 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
686
687 // Check if the oop is in the right area of memory
688 const int oop_mask = Universe::verify_oop_mask();
689 const int oop_bits = Universe::verify_oop_bits();
690 __ mov(rdx, rax);
691 __ andptr(rdx, oop_mask);
692 __ cmpptr(rdx, oop_bits);
693 __ jcc(Assembler::notZero, error);
694
695 // make sure klass is 'reasonable', which is not zero.
696 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
697 __ testptr(rax, rax);
698 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
699
700 // return if everything seems ok
701 __ bind(exit);
702 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
703 __ pop(rdx); // restore rdx
704 __ popf(); // restore EFLAGS
705 __ ret(3 * wordSize); // pop arguments
706
707 // handle errors
708 __ bind(error);
709 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
710 __ pop(rdx); // get saved rdx back
711 __ popf(); // get saved EFLAGS off stack -- will be ignored
712 __ pusha(); // push registers (eip = return address & msg are already pushed)
713 BLOCK_COMMENT("call MacroAssembler::debug");
714 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
715 __ hlt();
716 return start;
717 }
718
719
720 // Copy 64 bytes chunks
721 //
722 // Inputs:
723 // from - source array address
724 // to_from - destination array address - from
725 // qword_count - 8-bytes element count, negative
726 //
727 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
728 assert( UseSSE >= 2, "supported cpu only" );
729 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
730
731 // Copy 64-byte chunks
732 __ jmpb(L_copy_64_bytes);
733 __ align(OptoLoopAlignment);
734 __ BIND(L_copy_64_bytes_loop);
735
736 if (UseUnalignedLoadStores) {
737 if (UseAVX > 2) {
738 __ evmovdqul(xmm0, Address(from, 0), Assembler::AVX_512bit);
739 __ evmovdqul(Address(from, to_from, Address::times_1, 0), xmm0, Assembler::AVX_512bit);
740 } else if (UseAVX == 2) {
741 __ vmovdqu(xmm0, Address(from, 0));
742 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
743 __ vmovdqu(xmm1, Address(from, 32));
744 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
745 } else {
746 __ movdqu(xmm0, Address(from, 0));
747 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
748 __ movdqu(xmm1, Address(from, 16));
749 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
750 __ movdqu(xmm2, Address(from, 32));
751 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
752 __ movdqu(xmm3, Address(from, 48));
753 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
754 }
755 } else {
756 __ movq(xmm0, Address(from, 0));
757 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
758 __ movq(xmm1, Address(from, 8));
759 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
760 __ movq(xmm2, Address(from, 16));
761 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
762 __ movq(xmm3, Address(from, 24));
763 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
764 __ movq(xmm4, Address(from, 32));
765 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
766 __ movq(xmm5, Address(from, 40));
767 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
768 __ movq(xmm6, Address(from, 48));
769 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
770 __ movq(xmm7, Address(from, 56));
771 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
772 }
773
774 __ addl(from, 64);
775 __ BIND(L_copy_64_bytes);
776 __ subl(qword_count, 8);
777 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
778
779 if (UseUnalignedLoadStores && (UseAVX == 2)) {
780 // clean upper bits of YMM registers
781 __ vpxor(xmm0, xmm0);
782 __ vpxor(xmm1, xmm1);
783 }
784 __ addl(qword_count, 8);
785 __ jccb(Assembler::zero, L_exit);
786 //
787 // length is too short, just copy qwords
788 //
789 __ BIND(L_copy_8_bytes);
790 __ movq(xmm0, Address(from, 0));
791 __ movq(Address(from, to_from, Address::times_1), xmm0);
792 __ addl(from, 8);
793 __ decrement(qword_count);
794 __ jcc(Assembler::greater, L_copy_8_bytes);
795 __ BIND(L_exit);
796 }
797
798 // Copy 64 bytes chunks
799 //
800 // Inputs:
801 // from - source array address
802 // to_from - destination array address - from
803 // qword_count - 8-bytes element count, negative
804 //
805 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
806 assert( VM_Version::supports_mmx(), "supported cpu only" );
807 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
808 // Copy 64-byte chunks
809 __ jmpb(L_copy_64_bytes);
810 __ align(OptoLoopAlignment);
811 __ BIND(L_copy_64_bytes_loop);
812 __ movq(mmx0, Address(from, 0));
813 __ movq(mmx1, Address(from, 8));
814 __ movq(mmx2, Address(from, 16));
815 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
816 __ movq(mmx3, Address(from, 24));
817 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
818 __ movq(mmx4, Address(from, 32));
819 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
820 __ movq(mmx5, Address(from, 40));
821 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
822 __ movq(mmx6, Address(from, 48));
823 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
824 __ movq(mmx7, Address(from, 56));
825 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
826 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
827 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
828 __ addptr(from, 64);
829 __ BIND(L_copy_64_bytes);
830 __ subl(qword_count, 8);
831 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
832 __ addl(qword_count, 8);
833 __ jccb(Assembler::zero, L_exit);
834 //
835 // length is too short, just copy qwords
836 //
837 __ BIND(L_copy_8_bytes);
838 __ movq(mmx0, Address(from, 0));
839 __ movq(Address(from, to_from, Address::times_1), mmx0);
840 __ addptr(from, 8);
841 __ decrement(qword_count);
842 __ jcc(Assembler::greater, L_copy_8_bytes);
843 __ BIND(L_exit);
844 __ emms();
845 }
846
847 address generate_disjoint_copy(BasicType t, bool aligned,
848 Address::ScaleFactor sf,
849 address* entry, const char *name,
850 bool dest_uninitialized = false) {
851 __ align(CodeEntryAlignment);
852 StubCodeMark mark(this, "StubRoutines", name);
853 address start = __ pc();
854
855 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
856 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
857
858 int shift = Address::times_ptr - sf;
859
860 const Register from = rsi; // source array address
861 const Register to = rdi; // destination array address
862 const Register count = rcx; // elements count
863 const Register to_from = to; // (to - from)
864 const Register saved_to = rdx; // saved destination array address
865
866 __ enter(); // required for proper stackwalking of RuntimeStub frame
867 __ push(rsi);
868 __ push(rdi);
869 __ movptr(from , Address(rsp, 12+ 4));
870 __ movptr(to , Address(rsp, 12+ 8));
871 __ movl(count, Address(rsp, 12+ 12));
872
873 if (entry != NULL) {
874 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
875 BLOCK_COMMENT("Entry:");
876 }
877
878 if (t == T_OBJECT) {
879 __ testl(count, count);
880 __ jcc(Assembler::zero, L_0_count);
881 }
882
883 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
884 if (dest_uninitialized) {
885 decorators |= IS_DEST_UNINITIALIZED;
886 }
887 if (aligned) {
888 decorators |= ARRAYCOPY_ALIGNED;
889 }
890
891 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
892 bs->arraycopy_prologue(_masm, decorators, t, from, to, count);
893 {
894 bool add_entry = (t != T_OBJECT && (!aligned || t == T_INT));
895 // UnsafeCopyMemory page error: continue after ucm
896 UnsafeCopyMemoryMark ucmm(this, add_entry, true);
897 __ subptr(to, from); // to --> to_from
898 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
899 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
900 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
901 // align source address at 4 bytes address boundary
902 if (t == T_BYTE) {
903 // One byte misalignment happens only for byte arrays
904 __ testl(from, 1);
905 __ jccb(Assembler::zero, L_skip_align1);
906 __ movb(rax, Address(from, 0));
907 __ movb(Address(from, to_from, Address::times_1, 0), rax);
908 __ increment(from);
909 __ decrement(count);
910 __ BIND(L_skip_align1);
911 }
912 // Two bytes misalignment happens only for byte and short (char) arrays
913 __ testl(from, 2);
914 __ jccb(Assembler::zero, L_skip_align2);
915 __ movw(rax, Address(from, 0));
916 __ movw(Address(from, to_from, Address::times_1, 0), rax);
917 __ addptr(from, 2);
918 __ subl(count, 1<<(shift-1));
919 __ BIND(L_skip_align2);
920 }
921 if (!VM_Version::supports_mmx()) {
922 __ mov(rax, count); // save 'count'
923 __ shrl(count, shift); // bytes count
924 __ addptr(to_from, from);// restore 'to'
925 __ rep_mov();
926 __ subptr(to_from, from);// restore 'to_from'
927 __ mov(count, rax); // restore 'count'
928 __ jmpb(L_copy_2_bytes); // all dwords were copied
929 } else {
930 if (!UseUnalignedLoadStores) {
931 // align to 8 bytes, we know we are 4 byte aligned to start
932 __ testptr(from, 4);
933 __ jccb(Assembler::zero, L_copy_64_bytes);
934 __ movl(rax, Address(from, 0));
935 __ movl(Address(from, to_from, Address::times_1, 0), rax);
936 __ addptr(from, 4);
937 __ subl(count, 1<<shift);
938 }
939 __ BIND(L_copy_64_bytes);
940 __ mov(rax, count);
941 __ shrl(rax, shift+1); // 8 bytes chunk count
942 //
943 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
944 //
945 if (UseXMMForArrayCopy) {
946 xmm_copy_forward(from, to_from, rax);
947 } else {
948 mmx_copy_forward(from, to_from, rax);
949 }
950 }
951 // copy tailing dword
952 __ BIND(L_copy_4_bytes);
953 __ testl(count, 1<<shift);
954 __ jccb(Assembler::zero, L_copy_2_bytes);
955 __ movl(rax, Address(from, 0));
956 __ movl(Address(from, to_from, Address::times_1, 0), rax);
957 if (t == T_BYTE || t == T_SHORT) {
958 __ addptr(from, 4);
959 __ BIND(L_copy_2_bytes);
960 // copy tailing word
961 __ testl(count, 1<<(shift-1));
962 __ jccb(Assembler::zero, L_copy_byte);
963 __ movw(rax, Address(from, 0));
964 __ movw(Address(from, to_from, Address::times_1, 0), rax);
965 if (t == T_BYTE) {
966 __ addptr(from, 2);
967 __ BIND(L_copy_byte);
968 // copy tailing byte
969 __ testl(count, 1);
970 __ jccb(Assembler::zero, L_exit);
971 __ movb(rax, Address(from, 0));
972 __ movb(Address(from, to_from, Address::times_1, 0), rax);
973 __ BIND(L_exit);
974 } else {
975 __ BIND(L_copy_byte);
976 }
977 } else {
978 __ BIND(L_copy_2_bytes);
979 }
980 }
981
982 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
983 __ emms();
984 }
985 __ movl(count, Address(rsp, 12+12)); // reread 'count'
986 bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
987
988 if (t == T_OBJECT) {
989 __ BIND(L_0_count);
990 }
991 inc_copy_counter_np(t);
992 __ pop(rdi);
993 __ pop(rsi);
994 __ leave(); // required for proper stackwalking of RuntimeStub frame
995 __ vzeroupper();
996 __ xorptr(rax, rax); // return 0
997 __ ret(0);
998 return start;
999 }
1000
1001
1002 address generate_fill(BasicType t, bool aligned, const char *name) {
1003 __ align(CodeEntryAlignment);
1004 StubCodeMark mark(this, "StubRoutines", name);
1005 address start = __ pc();
1006
1007 BLOCK_COMMENT("Entry:");
1008
1009 const Register to = rdi; // source array address
1010 const Register value = rdx; // value
1011 const Register count = rsi; // elements count
1012
1013 __ enter(); // required for proper stackwalking of RuntimeStub frame
1014 __ push(rsi);
1015 __ push(rdi);
1016 __ movptr(to , Address(rsp, 12+ 4));
1017 __ movl(value, Address(rsp, 12+ 8));
1018 __ movl(count, Address(rsp, 12+ 12));
1019
1020 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1021
1022 __ pop(rdi);
1023 __ pop(rsi);
1024 __ leave(); // required for proper stackwalking of RuntimeStub frame
1025 __ ret(0);
1026 return start;
1027 }
1028
1029 address generate_conjoint_copy(BasicType t, bool aligned,
1030 Address::ScaleFactor sf,
1031 address nooverlap_target,
1032 address* entry, const char *name,
1033 bool dest_uninitialized = false) {
1034 __ align(CodeEntryAlignment);
1035 StubCodeMark mark(this, "StubRoutines", name);
1036 address start = __ pc();
1037
1038 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1039 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1040
1041 int shift = Address::times_ptr - sf;
1042
1043 const Register src = rax; // source array address
1044 const Register dst = rdx; // destination array address
1045 const Register from = rsi; // source array address
1046 const Register to = rdi; // destination array address
1047 const Register count = rcx; // elements count
1048 const Register end = rax; // array end address
1049
1050 __ enter(); // required for proper stackwalking of RuntimeStub frame
1051 __ push(rsi);
1052 __ push(rdi);
1053 __ movptr(src , Address(rsp, 12+ 4)); // from
1054 __ movptr(dst , Address(rsp, 12+ 8)); // to
1055 __ movl2ptr(count, Address(rsp, 12+12)); // count
1056
1057 if (entry != NULL) {
1058 *entry = __ pc(); // Entry point from generic arraycopy stub.
1059 BLOCK_COMMENT("Entry:");
1060 }
1061
1062 // nooverlap_target expects arguments in rsi and rdi.
1063 __ mov(from, src);
1064 __ mov(to , dst);
1065
1066 // arrays overlap test: dispatch to disjoint stub if necessary.
1067 RuntimeAddress nooverlap(nooverlap_target);
1068 __ cmpptr(dst, src);
1069 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1070 __ jump_cc(Assembler::belowEqual, nooverlap);
1071 __ cmpptr(dst, end);
1072 __ jump_cc(Assembler::aboveEqual, nooverlap);
1073
1074 if (t == T_OBJECT) {
1075 __ testl(count, count);
1076 __ jcc(Assembler::zero, L_0_count);
1077 }
1078
1079 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
1080 if (dest_uninitialized) {
1081 decorators |= IS_DEST_UNINITIALIZED;
1082 }
1083 if (aligned) {
1084 decorators |= ARRAYCOPY_ALIGNED;
1085 }
1086
1087 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1088 bs->arraycopy_prologue(_masm, decorators, t, from, to, count);
1089
1090 {
1091 bool add_entry = (t != T_OBJECT && (!aligned || t == T_INT));
1092 // UnsafeCopyMemory page error: continue after ucm
1093 UnsafeCopyMemoryMark ucmm(this, add_entry, true);
1094 // copy from high to low
1095 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1096 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1097 if (t == T_BYTE || t == T_SHORT) {
1098 // Align the end of destination array at 4 bytes address boundary
1099 __ lea(end, Address(dst, count, sf, 0));
1100 if (t == T_BYTE) {
1101 // One byte misalignment happens only for byte arrays
1102 __ testl(end, 1);
1103 __ jccb(Assembler::zero, L_skip_align1);
1104 __ decrement(count);
1105 __ movb(rdx, Address(from, count, sf, 0));
1106 __ movb(Address(to, count, sf, 0), rdx);
1107 __ BIND(L_skip_align1);
1108 }
1109 // Two bytes misalignment happens only for byte and short (char) arrays
1110 __ testl(end, 2);
1111 __ jccb(Assembler::zero, L_skip_align2);
1112 __ subptr(count, 1<<(shift-1));
1113 __ movw(rdx, Address(from, count, sf, 0));
1114 __ movw(Address(to, count, sf, 0), rdx);
1115 __ BIND(L_skip_align2);
1116 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1117 __ jcc(Assembler::below, L_copy_4_bytes);
1118 }
1119
1120 if (!VM_Version::supports_mmx()) {
1121 __ std();
1122 __ mov(rax, count); // Save 'count'
1123 __ mov(rdx, to); // Save 'to'
1124 __ lea(rsi, Address(from, count, sf, -4));
1125 __ lea(rdi, Address(to , count, sf, -4));
1126 __ shrptr(count, shift); // bytes count
1127 __ rep_mov();
1128 __ cld();
1129 __ mov(count, rax); // restore 'count'
1130 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1131 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1132 __ mov(to, rdx); // restore 'to'
1133 __ jmpb(L_copy_2_bytes); // all dword were copied
1134 } else {
1135 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1136 __ testptr(end, 4);
1137 __ jccb(Assembler::zero, L_copy_8_bytes);
1138 __ subl(count, 1<<shift);
1139 __ movl(rdx, Address(from, count, sf, 0));
1140 __ movl(Address(to, count, sf, 0), rdx);
1141 __ jmpb(L_copy_8_bytes);
1142
1143 __ align(OptoLoopAlignment);
1144 // Move 8 bytes
1145 __ BIND(L_copy_8_bytes_loop);
1146 if (UseXMMForArrayCopy) {
1147 __ movq(xmm0, Address(from, count, sf, 0));
1148 __ movq(Address(to, count, sf, 0), xmm0);
1149 } else {
1150 __ movq(mmx0, Address(from, count, sf, 0));
1151 __ movq(Address(to, count, sf, 0), mmx0);
1152 }
1153 __ BIND(L_copy_8_bytes);
1154 __ subl(count, 2<<shift);
1155 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1156 __ addl(count, 2<<shift);
1157 if (!UseXMMForArrayCopy) {
1158 __ emms();
1159 }
1160 }
1161 __ BIND(L_copy_4_bytes);
1162 // copy prefix qword
1163 __ testl(count, 1<<shift);
1164 __ jccb(Assembler::zero, L_copy_2_bytes);
1165 __ movl(rdx, Address(from, count, sf, -4));
1166 __ movl(Address(to, count, sf, -4), rdx);
1167
1168 if (t == T_BYTE || t == T_SHORT) {
1169 __ subl(count, (1<<shift));
1170 __ BIND(L_copy_2_bytes);
1171 // copy prefix dword
1172 __ testl(count, 1<<(shift-1));
1173 __ jccb(Assembler::zero, L_copy_byte);
1174 __ movw(rdx, Address(from, count, sf, -2));
1175 __ movw(Address(to, count, sf, -2), rdx);
1176 if (t == T_BYTE) {
1177 __ subl(count, 1<<(shift-1));
1178 __ BIND(L_copy_byte);
1179 // copy prefix byte
1180 __ testl(count, 1);
1181 __ jccb(Assembler::zero, L_exit);
1182 __ movb(rdx, Address(from, 0));
1183 __ movb(Address(to, 0), rdx);
1184 __ BIND(L_exit);
1185 } else {
1186 __ BIND(L_copy_byte);
1187 }
1188 } else {
1189 __ BIND(L_copy_2_bytes);
1190 }
1191 }
1192
1193 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1194 __ emms();
1195 }
1196 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1197 bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
1198
1199 if (t == T_OBJECT) {
1200 __ BIND(L_0_count);
1201 }
1202 inc_copy_counter_np(t);
1203 __ pop(rdi);
1204 __ pop(rsi);
1205 __ leave(); // required for proper stackwalking of RuntimeStub frame
1206 __ xorptr(rax, rax); // return 0
1207 __ ret(0);
1208 return start;
1209 }
1210
1211
1212 address generate_disjoint_long_copy(address* entry, const char *name) {
1213 __ align(CodeEntryAlignment);
1214 StubCodeMark mark(this, "StubRoutines", name);
1215 address start = __ pc();
1216
1217 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1218 const Register from = rax; // source array address
1219 const Register to = rdx; // destination array address
1220 const Register count = rcx; // elements count
1221 const Register to_from = rdx; // (to - from)
1222
1223 __ enter(); // required for proper stackwalking of RuntimeStub frame
1224 __ movptr(from , Address(rsp, 8+0)); // from
1225 __ movptr(to , Address(rsp, 8+4)); // to
1226 __ movl2ptr(count, Address(rsp, 8+8)); // count
1227
1228 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1229 BLOCK_COMMENT("Entry:");
1230
1231 {
1232 // UnsafeCopyMemory page error: continue after ucm
1233 UnsafeCopyMemoryMark ucmm(this, true, true);
1234 __ subptr(to, from); // to --> to_from
1235 if (VM_Version::supports_mmx()) {
1236 if (UseXMMForArrayCopy) {
1237 xmm_copy_forward(from, to_from, count);
1238 } else {
1239 mmx_copy_forward(from, to_from, count);
1240 }
1241 } else {
1242 __ jmpb(L_copy_8_bytes);
1243 __ align(OptoLoopAlignment);
1244 __ BIND(L_copy_8_bytes_loop);
1245 __ fild_d(Address(from, 0));
1246 __ fistp_d(Address(from, to_from, Address::times_1));
1247 __ addptr(from, 8);
1248 __ BIND(L_copy_8_bytes);
1249 __ decrement(count);
1250 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1251 }
1252 }
1253 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1254 __ emms();
1255 }
1256 inc_copy_counter_np(T_LONG);
1257 __ leave(); // required for proper stackwalking of RuntimeStub frame
1258 __ vzeroupper();
1259 __ xorptr(rax, rax); // return 0
1260 __ ret(0);
1261 return start;
1262 }
1263
1264 address generate_conjoint_long_copy(address nooverlap_target,
1265 address* entry, const char *name) {
1266 __ align(CodeEntryAlignment);
1267 StubCodeMark mark(this, "StubRoutines", name);
1268 address start = __ pc();
1269
1270 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1271 const Register from = rax; // source array address
1272 const Register to = rdx; // destination array address
1273 const Register count = rcx; // elements count
1274 const Register end_from = rax; // source array end address
1275
1276 __ enter(); // required for proper stackwalking of RuntimeStub frame
1277 __ movptr(from , Address(rsp, 8+0)); // from
1278 __ movptr(to , Address(rsp, 8+4)); // to
1279 __ movl2ptr(count, Address(rsp, 8+8)); // count
1280
1281 *entry = __ pc(); // Entry point from generic arraycopy stub.
1282 BLOCK_COMMENT("Entry:");
1283
1284 // arrays overlap test
1285 __ cmpptr(to, from);
1286 RuntimeAddress nooverlap(nooverlap_target);
1287 __ jump_cc(Assembler::belowEqual, nooverlap);
1288 __ lea(end_from, Address(from, count, Address::times_8, 0));
1289 __ cmpptr(to, end_from);
1290 __ movptr(from, Address(rsp, 8)); // from
1291 __ jump_cc(Assembler::aboveEqual, nooverlap);
1292
1293 {
1294 // UnsafeCopyMemory page error: continue after ucm
1295 UnsafeCopyMemoryMark ucmm(this, true, true);
1296
1297 __ jmpb(L_copy_8_bytes);
1298
1299 __ align(OptoLoopAlignment);
1300 __ BIND(L_copy_8_bytes_loop);
1301 if (VM_Version::supports_mmx()) {
1302 if (UseXMMForArrayCopy) {
1303 __ movq(xmm0, Address(from, count, Address::times_8));
1304 __ movq(Address(to, count, Address::times_8), xmm0);
1305 } else {
1306 __ movq(mmx0, Address(from, count, Address::times_8));
1307 __ movq(Address(to, count, Address::times_8), mmx0);
1308 }
1309 } else {
1310 __ fild_d(Address(from, count, Address::times_8));
1311 __ fistp_d(Address(to, count, Address::times_8));
1312 }
1313 __ BIND(L_copy_8_bytes);
1314 __ decrement(count);
1315 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1316
1317 }
1318 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1319 __ emms();
1320 }
1321 inc_copy_counter_np(T_LONG);
1322 __ leave(); // required for proper stackwalking of RuntimeStub frame
1323 __ xorptr(rax, rax); // return 0
1324 __ ret(0);
1325 return start;
1326 }
1327
1328
1329 // Helper for generating a dynamic type check.
1330 // The sub_klass must be one of {rbx, rdx, rsi}.
1331 // The temp is killed.
1332 void generate_type_check(Register sub_klass,
1333 Address& super_check_offset_addr,
1334 Address& super_klass_addr,
1335 Register temp,
1336 Label* L_success, Label* L_failure) {
1337 BLOCK_COMMENT("type_check:");
1338
1339 Label L_fallthrough;
1340 #define LOCAL_JCC(assembler_con, label_ptr) \
1341 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1342 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1343
1344 // The following is a strange variation of the fast path which requires
1345 // one less register, because needed values are on the argument stack.
1346 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1347 // L_success, L_failure, NULL);
1348 assert_different_registers(sub_klass, temp);
1349
1350 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1351
1352 // if the pointers are equal, we are done (e.g., String[] elements)
1353 __ cmpptr(sub_klass, super_klass_addr);
1354 LOCAL_JCC(Assembler::equal, L_success);
1355
1356 // check the supertype display:
1357 __ movl2ptr(temp, super_check_offset_addr);
1358 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1359 __ movptr(temp, super_check_addr); // load displayed supertype
1360 __ cmpptr(temp, super_klass_addr); // test the super type
1361 LOCAL_JCC(Assembler::equal, L_success);
1362
1363 // if it was a primary super, we can just fail immediately
1364 __ cmpl(super_check_offset_addr, sc_offset);
1365 LOCAL_JCC(Assembler::notEqual, L_failure);
1366
1367 // The repne_scan instruction uses fixed registers, which will get spilled.
1368 // We happen to know this works best when super_klass is in rax.
1369 Register super_klass = temp;
1370 __ movptr(super_klass, super_klass_addr);
1371 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1372 L_success, L_failure);
1373
1374 __ bind(L_fallthrough);
1375
1376 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1377 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1378
1379 #undef LOCAL_JCC
1380 }
1381
1382 //
1383 // Generate checkcasting array copy stub
1384 //
1385 // Input:
1386 // 4(rsp) - source array address
1387 // 8(rsp) - destination array address
1388 // 12(rsp) - element count, can be zero
1389 // 16(rsp) - size_t ckoff (super_check_offset)
1390 // 20(rsp) - oop ckval (super_klass)
1391 //
1392 // Output:
1393 // rax, == 0 - success
1394 // rax, == -1^K - failure, where K is partial transfer count
1395 //
1396 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1397 __ align(CodeEntryAlignment);
1398 StubCodeMark mark(this, "StubRoutines", name);
1399 address start = __ pc();
1400
1401 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1402
1403 // register use:
1404 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1405 // rdi, rsi -- element access (oop, klass)
1406 // rbx, -- temp
1407 const Register from = rax; // source array address
1408 const Register to = rdx; // destination array address
1409 const Register length = rcx; // elements count
1410 const Register elem = rdi; // each oop copied
1411 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1412 const Register temp = rbx; // lone remaining temp
1413
1414 __ enter(); // required for proper stackwalking of RuntimeStub frame
1415
1416 __ push(rsi);
1417 __ push(rdi);
1418 __ push(rbx);
1419
1420 Address from_arg(rsp, 16+ 4); // from
1421 Address to_arg(rsp, 16+ 8); // to
1422 Address length_arg(rsp, 16+12); // elements count
1423 Address ckoff_arg(rsp, 16+16); // super_check_offset
1424 Address ckval_arg(rsp, 16+20); // super_klass
1425
1426 // Load up:
1427 __ movptr(from, from_arg);
1428 __ movptr(to, to_arg);
1429 __ movl2ptr(length, length_arg);
1430
1431 if (entry != NULL) {
1432 *entry = __ pc(); // Entry point from generic arraycopy stub.
1433 BLOCK_COMMENT("Entry:");
1434 }
1435
1436 //---------------------------------------------------------------
1437 // Assembler stub will be used for this call to arraycopy
1438 // if the two arrays are subtypes of Object[] but the
1439 // destination array type is not equal to or a supertype
1440 // of the source type. Each element must be separately
1441 // checked.
1442
1443 // Loop-invariant addresses. They are exclusive end pointers.
1444 Address end_from_addr(from, length, Address::times_ptr, 0);
1445 Address end_to_addr(to, length, Address::times_ptr, 0);
1446
1447 Register end_from = from; // re-use
1448 Register end_to = to; // re-use
1449 Register count = length; // re-use
1450
1451 // Loop-variant addresses. They assume post-incremented count < 0.
1452 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1453 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1454 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1455
1456 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST;
1457 if (dest_uninitialized) {
1458 decorators |= IS_DEST_UNINITIALIZED;
1459 }
1460
1461 BasicType type = T_OBJECT;
1462 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1463 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1464
1465 // Copy from low to high addresses, indexed from the end of each array.
1466 __ lea(end_from, end_from_addr);
1467 __ lea(end_to, end_to_addr);
1468 assert(length == count, ""); // else fix next line:
1469 __ negptr(count); // negate and test the length
1470 __ jccb(Assembler::notZero, L_load_element);
1471
1472 // Empty array: Nothing to do.
1473 __ xorptr(rax, rax); // return 0 on (trivial) success
1474 __ jmp(L_done);
1475
1476 // ======== begin loop ========
1477 // (Loop is rotated; its entry is L_load_element.)
1478 // Loop control:
1479 // for (count = -count; count != 0; count++)
1480 // Base pointers src, dst are biased by 8*count,to last element.
1481 __ align(OptoLoopAlignment);
1482
1483 __ BIND(L_store_element);
1484 __ movptr(to_element_addr, elem); // store the oop
1485 __ increment(count); // increment the count toward zero
1486 __ jccb(Assembler::zero, L_do_card_marks);
1487
1488 // ======== loop entry is here ========
1489 __ BIND(L_load_element);
1490 __ movptr(elem, from_element_addr); // load the oop
1491 __ testptr(elem, elem);
1492 __ jccb(Assembler::zero, L_store_element);
1493
1494 // (Could do a trick here: Remember last successful non-null
1495 // element stored and make a quick oop equality check on it.)
1496
1497 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1498 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1499 &L_store_element, NULL);
1500 // (On fall-through, we have failed the element type check.)
1501 // ======== end loop ========
1502
1503 // It was a real error; we must depend on the caller to finish the job.
1504 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1505 // Emit GC store barriers for the oops we have copied (length_arg + count),
1506 // and report their number to the caller.
1507 assert_different_registers(to, count, rax);
1508 Label L_post_barrier;
1509 __ addl(count, length_arg); // transfers = (length - remaining)
1510 __ movl2ptr(rax, count); // save the value
1511 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1512 __ jccb(Assembler::notZero, L_post_barrier);
1513 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1514
1515 // Come here on success only.
1516 __ BIND(L_do_card_marks);
1517 __ xorptr(rax, rax); // return 0 on success
1518 __ movl2ptr(count, length_arg);
1519
1520 __ BIND(L_post_barrier);
1521 __ movptr(to, to_arg); // reload
1522 bs->arraycopy_epilogue(_masm, decorators, type, from, to, count);
1523
1524 // Common exit point (success or failure).
1525 __ BIND(L_done);
1526 __ pop(rbx);
1527 __ pop(rdi);
1528 __ pop(rsi);
1529 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1530 __ leave(); // required for proper stackwalking of RuntimeStub frame
1531 __ ret(0);
1532
1533 return start;
1534 }
1535
1536 //
1537 // Generate 'unsafe' array copy stub
1538 // Though just as safe as the other stubs, it takes an unscaled
1539 // size_t argument instead of an element count.
1540 //
1541 // Input:
1542 // 4(rsp) - source array address
1543 // 8(rsp) - destination array address
1544 // 12(rsp) - byte count, can be zero
1545 //
1546 // Output:
1547 // rax, == 0 - success
1548 // rax, == -1 - need to call System.arraycopy
1549 //
1550 // Examines the alignment of the operands and dispatches
1551 // to a long, int, short, or byte copy loop.
1552 //
1553 address generate_unsafe_copy(const char *name,
1554 address byte_copy_entry,
1555 address short_copy_entry,
1556 address int_copy_entry,
1557 address long_copy_entry) {
1558
1559 Label L_long_aligned, L_int_aligned, L_short_aligned;
1560
1561 __ align(CodeEntryAlignment);
1562 StubCodeMark mark(this, "StubRoutines", name);
1563 address start = __ pc();
1564
1565 const Register from = rax; // source array address
1566 const Register to = rdx; // destination array address
1567 const Register count = rcx; // elements count
1568
1569 __ enter(); // required for proper stackwalking of RuntimeStub frame
1570 __ push(rsi);
1571 __ push(rdi);
1572 Address from_arg(rsp, 12+ 4); // from
1573 Address to_arg(rsp, 12+ 8); // to
1574 Address count_arg(rsp, 12+12); // byte count
1575
1576 // Load up:
1577 __ movptr(from , from_arg);
1578 __ movptr(to , to_arg);
1579 __ movl2ptr(count, count_arg);
1580
1581 // bump this on entry, not on exit:
1582 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1583
1584 const Register bits = rsi;
1585 __ mov(bits, from);
1586 __ orptr(bits, to);
1587 __ orptr(bits, count);
1588
1589 __ testl(bits, BytesPerLong-1);
1590 __ jccb(Assembler::zero, L_long_aligned);
1591
1592 __ testl(bits, BytesPerInt-1);
1593 __ jccb(Assembler::zero, L_int_aligned);
1594
1595 __ testl(bits, BytesPerShort-1);
1596 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1597
1598 __ BIND(L_short_aligned);
1599 __ shrptr(count, LogBytesPerShort); // size => short_count
1600 __ movl(count_arg, count); // update 'count'
1601 __ jump(RuntimeAddress(short_copy_entry));
1602
1603 __ BIND(L_int_aligned);
1604 __ shrptr(count, LogBytesPerInt); // size => int_count
1605 __ movl(count_arg, count); // update 'count'
1606 __ jump(RuntimeAddress(int_copy_entry));
1607
1608 __ BIND(L_long_aligned);
1609 __ shrptr(count, LogBytesPerLong); // size => qword_count
1610 __ movl(count_arg, count); // update 'count'
1611 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1612 __ pop(rsi);
1613 __ jump(RuntimeAddress(long_copy_entry));
1614
1615 return start;
1616 }
1617
1618
1619 // Perform range checks on the proposed arraycopy.
1620 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1621 void arraycopy_range_checks(Register src,
1622 Register src_pos,
1623 Register dst,
1624 Register dst_pos,
1625 Address& length,
1626 Label& L_failed) {
1627 BLOCK_COMMENT("arraycopy_range_checks:");
1628 const Register src_end = src_pos; // source array end position
1629 const Register dst_end = dst_pos; // destination array end position
1630 __ addl(src_end, length); // src_pos + length
1631 __ addl(dst_end, length); // dst_pos + length
1632
1633 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1634 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1635 __ jcc(Assembler::above, L_failed);
1636
1637 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1638 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1639 __ jcc(Assembler::above, L_failed);
1640
1641 BLOCK_COMMENT("arraycopy_range_checks done");
1642 }
1643
1644
1645 //
1646 // Generate generic array copy stubs
1647 //
1648 // Input:
1649 // 4(rsp) - src oop
1650 // 8(rsp) - src_pos
1651 // 12(rsp) - dst oop
1652 // 16(rsp) - dst_pos
1653 // 20(rsp) - element count
1654 //
1655 // Output:
1656 // rax, == 0 - success
1657 // rax, == -1^K - failure, where K is partial transfer count
1658 //
1659 address generate_generic_copy(const char *name,
1660 address entry_jbyte_arraycopy,
1661 address entry_jshort_arraycopy,
1662 address entry_jint_arraycopy,
1663 address entry_oop_arraycopy,
1664 address entry_jlong_arraycopy,
1665 address entry_checkcast_arraycopy) {
1666 Label L_failed, L_failed_0, L_objArray;
1667
1668 { int modulus = CodeEntryAlignment;
1669 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1670 int advance = target - (__ offset() % modulus);
1671 if (advance < 0) advance += modulus;
1672 if (advance > 0) __ nop(advance);
1673 }
1674 StubCodeMark mark(this, "StubRoutines", name);
1675
1676 // Short-hop target to L_failed. Makes for denser prologue code.
1677 __ BIND(L_failed_0);
1678 __ jmp(L_failed);
1679 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1680
1681 __ align(CodeEntryAlignment);
1682 address start = __ pc();
1683
1684 __ enter(); // required for proper stackwalking of RuntimeStub frame
1685 __ push(rsi);
1686 __ push(rdi);
1687
1688 // bump this on entry, not on exit:
1689 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1690
1691 // Input values
1692 Address SRC (rsp, 12+ 4);
1693 Address SRC_POS (rsp, 12+ 8);
1694 Address DST (rsp, 12+12);
1695 Address DST_POS (rsp, 12+16);
1696 Address LENGTH (rsp, 12+20);
1697
1698 //-----------------------------------------------------------------------
1699 // Assembler stub will be used for this call to arraycopy
1700 // if the following conditions are met:
1701 //
1702 // (1) src and dst must not be null.
1703 // (2) src_pos must not be negative.
1704 // (3) dst_pos must not be negative.
1705 // (4) length must not be negative.
1706 // (5) src klass and dst klass should be the same and not NULL.
1707 // (6) src and dst should be arrays.
1708 // (7) src_pos + length must not exceed length of src.
1709 // (8) dst_pos + length must not exceed length of dst.
1710 //
1711
1712 const Register src = rax; // source array oop
1713 const Register src_pos = rsi;
1714 const Register dst = rdx; // destination array oop
1715 const Register dst_pos = rdi;
1716 const Register length = rcx; // transfer count
1717
1718 // if (src == NULL) return -1;
1719 __ movptr(src, SRC); // src oop
1720 __ testptr(src, src);
1721 __ jccb(Assembler::zero, L_failed_0);
1722
1723 // if (src_pos < 0) return -1;
1724 __ movl2ptr(src_pos, SRC_POS); // src_pos
1725 __ testl(src_pos, src_pos);
1726 __ jccb(Assembler::negative, L_failed_0);
1727
1728 // if (dst == NULL) return -1;
1729 __ movptr(dst, DST); // dst oop
1730 __ testptr(dst, dst);
1731 __ jccb(Assembler::zero, L_failed_0);
1732
1733 // if (dst_pos < 0) return -1;
1734 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1735 __ testl(dst_pos, dst_pos);
1736 __ jccb(Assembler::negative, L_failed_0);
1737
1738 // if (length < 0) return -1;
1739 __ movl2ptr(length, LENGTH); // length
1740 __ testl(length, length);
1741 __ jccb(Assembler::negative, L_failed_0);
1742
1743 // if (src->klass() == NULL) return -1;
1744 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1745 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1746 const Register rcx_src_klass = rcx; // array klass
1747 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1748
1749 #ifdef ASSERT
1750 // assert(src->klass() != NULL);
1751 BLOCK_COMMENT("assert klasses not null");
1752 { Label L1, L2;
1753 __ testptr(rcx_src_klass, rcx_src_klass);
1754 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1755 __ bind(L1);
1756 __ stop("broken null klass");
1757 __ bind(L2);
1758 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1759 __ jccb(Assembler::equal, L1); // this would be broken also
1760 BLOCK_COMMENT("assert done");
1761 }
1762 #endif //ASSERT
1763
1764 // Load layout helper (32-bits)
1765 //
1766 // |array_tag| | header_size | element_type | |log2_element_size|
1767 // 32 30 24 16 8 2 0
1768 //
1769 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1770 //
1771
1772 int lh_offset = in_bytes(Klass::layout_helper_offset());
1773 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1774
1775 // Handle objArrays completely differently...
1776 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1777 __ cmpl(src_klass_lh_addr, objArray_lh);
1778 __ jcc(Assembler::equal, L_objArray);
1779
1780 // if (src->klass() != dst->klass()) return -1;
1781 __ cmpptr(rcx_src_klass, dst_klass_addr);
1782 __ jccb(Assembler::notEqual, L_failed_0);
1783
1784 const Register rcx_lh = rcx; // layout helper
1785 assert(rcx_lh == rcx_src_klass, "known alias");
1786 __ movl(rcx_lh, src_klass_lh_addr);
1787
1788 // if (!src->is_Array()) return -1;
1789 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1790 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1791
1792 // At this point, it is known to be a typeArray (array_tag 0x3).
1793 #ifdef ASSERT
1794 { Label L;
1795 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1796 __ jcc(Assembler::greaterEqual, L); // signed cmp
1797 __ stop("must be a primitive array");
1798 __ bind(L);
1799 }
1800 #endif
1801
1802 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1803 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1804
1805 // TypeArrayKlass
1806 //
1807 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1808 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1809 //
1810 const Register rsi_offset = rsi; // array offset
1811 const Register src_array = src; // src array offset
1812 const Register dst_array = dst; // dst array offset
1813 const Register rdi_elsize = rdi; // log2 element size
1814
1815 __ mov(rsi_offset, rcx_lh);
1816 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1817 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1818 __ addptr(src_array, rsi_offset); // src array offset
1819 __ addptr(dst_array, rsi_offset); // dst array offset
1820 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1821
1822 // next registers should be set before the jump to corresponding stub
1823 const Register from = src; // source array address
1824 const Register to = dst; // destination array address
1825 const Register count = rcx; // elements count
1826 // some of them should be duplicated on stack
1827 #define FROM Address(rsp, 12+ 4)
1828 #define TO Address(rsp, 12+ 8) // Not used now
1829 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1830
1831 BLOCK_COMMENT("scale indexes to element size");
1832 __ movl2ptr(rsi, SRC_POS); // src_pos
1833 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1834 assert(src_array == from, "");
1835 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1836 __ movl2ptr(rdi, DST_POS); // dst_pos
1837 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1838 assert(dst_array == to, "");
1839 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1840 __ movptr(FROM, from); // src_addr
1841 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1842 __ movl2ptr(count, LENGTH); // elements count
1843
1844 BLOCK_COMMENT("choose copy loop based on element size");
1845 __ cmpl(rdi_elsize, 0);
1846
1847 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1848 __ cmpl(rdi_elsize, LogBytesPerShort);
1849 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1850 __ cmpl(rdi_elsize, LogBytesPerInt);
1851 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1852 #ifdef ASSERT
1853 __ cmpl(rdi_elsize, LogBytesPerLong);
1854 __ jccb(Assembler::notEqual, L_failed);
1855 #endif
1856 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1857 __ pop(rsi);
1858 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1859
1860 __ BIND(L_failed);
1861 __ xorptr(rax, rax);
1862 __ notptr(rax); // return -1
1863 __ pop(rdi);
1864 __ pop(rsi);
1865 __ leave(); // required for proper stackwalking of RuntimeStub frame
1866 __ ret(0);
1867
1868 // ObjArrayKlass
1869 __ BIND(L_objArray);
1870 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1871
1872 Label L_plain_copy, L_checkcast_copy;
1873 // test array classes for subtyping
1874 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1875 __ jccb(Assembler::notEqual, L_checkcast_copy);
1876
1877 // Identically typed arrays can be copied without element-wise checks.
1878 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1879 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1880
1881 __ BIND(L_plain_copy);
1882 __ movl2ptr(count, LENGTH); // elements count
1883 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1884 __ lea(from, Address(src, src_pos, Address::times_ptr,
1885 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1886 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1887 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1888 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1889 __ movptr(FROM, from); // src_addr
1890 __ movptr(TO, to); // dst_addr
1891 __ movl(COUNT, count); // count
1892 __ jump(RuntimeAddress(entry_oop_arraycopy));
1893
1894 __ BIND(L_checkcast_copy);
1895 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1896 {
1897 // Handy offsets:
1898 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1899 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1900
1901 Register rsi_dst_klass = rsi;
1902 Register rdi_temp = rdi;
1903 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1904 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1905 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1906
1907 // Before looking at dst.length, make sure dst is also an objArray.
1908 __ movptr(rsi_dst_klass, dst_klass_addr);
1909 __ cmpl(dst_klass_lh_addr, objArray_lh);
1910 __ jccb(Assembler::notEqual, L_failed);
1911
1912 // It is safe to examine both src.length and dst.length.
1913 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1914 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1915 // (Now src_pos and dst_pos are killed, but not src and dst.)
1916
1917 // We'll need this temp (don't forget to pop it after the type check).
1918 __ push(rbx);
1919 Register rbx_src_klass = rbx;
1920
1921 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1922 __ movptr(rsi_dst_klass, dst_klass_addr);
1923 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1924 Label L_fail_array_check;
1925 generate_type_check(rbx_src_klass,
1926 super_check_offset_addr, dst_klass_addr,
1927 rdi_temp, NULL, &L_fail_array_check);
1928 // (On fall-through, we have passed the array type check.)
1929 __ pop(rbx);
1930 __ jmp(L_plain_copy);
1931
1932 __ BIND(L_fail_array_check);
1933 // Reshuffle arguments so we can call checkcast_arraycopy:
1934
1935 // match initial saves for checkcast_arraycopy
1936 // push(rsi); // already done; see above
1937 // push(rdi); // already done; see above
1938 // push(rbx); // already done; see above
1939
1940 // Marshal outgoing arguments now, freeing registers.
1941 Address from_arg(rsp, 16+ 4); // from
1942 Address to_arg(rsp, 16+ 8); // to
1943 Address length_arg(rsp, 16+12); // elements count
1944 Address ckoff_arg(rsp, 16+16); // super_check_offset
1945 Address ckval_arg(rsp, 16+20); // super_klass
1946
1947 Address SRC_POS_arg(rsp, 16+ 8);
1948 Address DST_POS_arg(rsp, 16+16);
1949 Address LENGTH_arg(rsp, 16+20);
1950 // push rbx, changed the incoming offsets (why not just use rbp,??)
1951 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1952
1953 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1954 __ movl2ptr(length, LENGTH_arg); // reload elements count
1955 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1956 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1957
1958 __ movptr(ckval_arg, rbx); // destination element type
1959 __ movl(rbx, Address(rbx, sco_offset));
1960 __ movl(ckoff_arg, rbx); // corresponding class check offset
1961
1962 __ movl(length_arg, length); // outgoing length argument
1963
1964 __ lea(from, Address(src, src_pos, Address::times_ptr,
1965 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1966 __ movptr(from_arg, from);
1967
1968 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1969 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1970 __ movptr(to_arg, to);
1971 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1972 }
1973
1974 return start;
1975 }
1976
1977 void generate_arraycopy_stubs() {
1978 address entry;
1979 address entry_jbyte_arraycopy;
1980 address entry_jshort_arraycopy;
1981 address entry_jint_arraycopy;
1982 address entry_oop_arraycopy;
1983 address entry_jlong_arraycopy;
1984 address entry_checkcast_arraycopy;
1985
1986 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1987 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1988 "arrayof_jbyte_disjoint_arraycopy");
1989 StubRoutines::_arrayof_jbyte_arraycopy =
1990 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1991 NULL, "arrayof_jbyte_arraycopy");
1992 StubRoutines::_jbyte_disjoint_arraycopy =
1993 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1994 "jbyte_disjoint_arraycopy");
1995 StubRoutines::_jbyte_arraycopy =
1996 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1997 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1998
1999 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2000 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
2001 "arrayof_jshort_disjoint_arraycopy");
2002 StubRoutines::_arrayof_jshort_arraycopy =
2003 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2004 NULL, "arrayof_jshort_arraycopy");
2005 StubRoutines::_jshort_disjoint_arraycopy =
2006 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2007 "jshort_disjoint_arraycopy");
2008 StubRoutines::_jshort_arraycopy =
2009 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2010 &entry_jshort_arraycopy, "jshort_arraycopy");
2011
2012 // Next arrays are always aligned on 4 bytes at least.
2013 StubRoutines::_jint_disjoint_arraycopy =
2014 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2015 "jint_disjoint_arraycopy");
2016 StubRoutines::_jint_arraycopy =
2017 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2018 &entry_jint_arraycopy, "jint_arraycopy");
2019
2020 StubRoutines::_oop_disjoint_arraycopy =
2021 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2022 "oop_disjoint_arraycopy");
2023 StubRoutines::_oop_arraycopy =
2024 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2025 &entry_oop_arraycopy, "oop_arraycopy");
2026
2027 StubRoutines::_oop_disjoint_arraycopy_uninit =
2028 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2029 "oop_disjoint_arraycopy_uninit",
2030 /*dest_uninitialized*/true);
2031 StubRoutines::_oop_arraycopy_uninit =
2032 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2033 NULL, "oop_arraycopy_uninit",
2034 /*dest_uninitialized*/true);
2035
2036 StubRoutines::_jlong_disjoint_arraycopy =
2037 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2038 StubRoutines::_jlong_arraycopy =
2039 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2040 "jlong_arraycopy");
2041
2042 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2043 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2044 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2045 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2046 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2047 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2048
2049 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2050 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2051 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2052 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2053
2054 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2055 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2056 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2057 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2058
2059 StubRoutines::_checkcast_arraycopy =
2060 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2061 StubRoutines::_checkcast_arraycopy_uninit =
2062 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2063
2064 StubRoutines::_unsafe_arraycopy =
2065 generate_unsafe_copy("unsafe_arraycopy",
2066 entry_jbyte_arraycopy,
2067 entry_jshort_arraycopy,
2068 entry_jint_arraycopy,
2069 entry_jlong_arraycopy);
2070
2071 StubRoutines::_generic_arraycopy =
2072 generate_generic_copy("generic_arraycopy",
2073 entry_jbyte_arraycopy,
2074 entry_jshort_arraycopy,
2075 entry_jint_arraycopy,
2076 entry_oop_arraycopy,
2077 entry_jlong_arraycopy,
2078 entry_checkcast_arraycopy);
2079 }
2080
2081 // AES intrinsic stubs
2082 enum {AESBlockSize = 16};
2083
2084 address generate_key_shuffle_mask() {
2085 __ align(16);
2086 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2087 address start = __ pc();
2088 __ emit_data(0x00010203, relocInfo::none, 0 );
2089 __ emit_data(0x04050607, relocInfo::none, 0 );
2090 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2091 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2092 return start;
2093 }
2094
2095 address generate_counter_shuffle_mask() {
2096 __ align(16);
2097 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2098 address start = __ pc();
2099 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
2100 __ emit_data(0x08090a0b, relocInfo::none, 0);
2101 __ emit_data(0x04050607, relocInfo::none, 0);
2102 __ emit_data(0x00010203, relocInfo::none, 0);
2103 return start;
2104 }
2105
2106 // Utility routine for loading a 128-bit key word in little endian format
2107 // can optionally specify that the shuffle mask is already in an xmmregister
2108 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2109 __ movdqu(xmmdst, Address(key, offset));
2110 if (xmm_shuf_mask != NULL) {
2111 __ pshufb(xmmdst, xmm_shuf_mask);
2112 } else {
2113 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2114 }
2115 }
2116
2117 // aesenc using specified key+offset
2118 // can optionally specify that the shuffle mask is already in an xmmregister
2119 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2120 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2121 __ aesenc(xmmdst, xmmtmp);
2122 }
2123
2124 // aesdec using specified key+offset
2125 // can optionally specify that the shuffle mask is already in an xmmregister
2126 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2127 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2128 __ aesdec(xmmdst, xmmtmp);
2129 }
2130
2131 // Utility routine for increase 128bit counter (iv in CTR mode)
2132 // XMM_128bit, D3, D2, D1, D0
2133 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
2134 __ pextrd(reg, xmmdst, 0x0);
2135 __ addl(reg, inc_delta);
2136 __ pinsrd(xmmdst, reg, 0x0);
2137 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2138
2139 __ pextrd(reg, xmmdst, 0x01); // Carry-> D1
2140 __ addl(reg, 0x01);
2141 __ pinsrd(xmmdst, reg, 0x01);
2142 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2143
2144 __ pextrd(reg, xmmdst, 0x02); // Carry-> D2
2145 __ addl(reg, 0x01);
2146 __ pinsrd(xmmdst, reg, 0x02);
2147 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2148
2149 __ pextrd(reg, xmmdst, 0x03); // Carry -> D3
2150 __ addl(reg, 0x01);
2151 __ pinsrd(xmmdst, reg, 0x03);
2152
2153 __ BIND(next_block); // next instruction
2154 }
2155
2156
2157 // Arguments:
2158 //
2159 // Inputs:
2160 // c_rarg0 - source byte array address
2161 // c_rarg1 - destination byte array address
2162 // c_rarg2 - K (key) in little endian int array
2163 //
2164 address generate_aescrypt_encryptBlock() {
2165 assert(UseAES, "need AES instructions and misaligned SSE support");
2166 __ align(CodeEntryAlignment);
2167 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2168 Label L_doLast;
2169 address start = __ pc();
2170
2171 const Register from = rdx; // source array address
2172 const Register to = rdx; // destination array address
2173 const Register key = rcx; // key array address
2174 const Register keylen = rax;
2175 const Address from_param(rbp, 8+0);
2176 const Address to_param (rbp, 8+4);
2177 const Address key_param (rbp, 8+8);
2178
2179 const XMMRegister xmm_result = xmm0;
2180 const XMMRegister xmm_key_shuf_mask = xmm1;
2181 const XMMRegister xmm_temp1 = xmm2;
2182 const XMMRegister xmm_temp2 = xmm3;
2183 const XMMRegister xmm_temp3 = xmm4;
2184 const XMMRegister xmm_temp4 = xmm5;
2185
2186 __ enter(); // required for proper stackwalking of RuntimeStub frame
2187
2188 __ movptr(from, from_param);
2189 __ movptr(key, key_param);
2190
2191 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2192 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2193
2194 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2195 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2196 __ movptr(to, to_param);
2197
2198 // For encryption, the java expanded key ordering is just what we need
2199
2200 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2201 __ pxor(xmm_result, xmm_temp1);
2202
2203 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2204 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2205 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2206 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2207
2208 __ aesenc(xmm_result, xmm_temp1);
2209 __ aesenc(xmm_result, xmm_temp2);
2210 __ aesenc(xmm_result, xmm_temp3);
2211 __ aesenc(xmm_result, xmm_temp4);
2212
2213 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2214 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2215 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2216 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2217
2218 __ aesenc(xmm_result, xmm_temp1);
2219 __ aesenc(xmm_result, xmm_temp2);
2220 __ aesenc(xmm_result, xmm_temp3);
2221 __ aesenc(xmm_result, xmm_temp4);
2222
2223 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2224 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2225
2226 __ cmpl(keylen, 44);
2227 __ jccb(Assembler::equal, L_doLast);
2228
2229 __ aesenc(xmm_result, xmm_temp1);
2230 __ aesenc(xmm_result, xmm_temp2);
2231
2232 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2233 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2234
2235 __ cmpl(keylen, 52);
2236 __ jccb(Assembler::equal, L_doLast);
2237
2238 __ aesenc(xmm_result, xmm_temp1);
2239 __ aesenc(xmm_result, xmm_temp2);
2240
2241 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2242 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2243
2244 __ BIND(L_doLast);
2245 __ aesenc(xmm_result, xmm_temp1);
2246 __ aesenclast(xmm_result, xmm_temp2);
2247 __ movdqu(Address(to, 0), xmm_result); // store the result
2248 __ xorptr(rax, rax); // return 0
2249 __ leave(); // required for proper stackwalking of RuntimeStub frame
2250 __ ret(0);
2251
2252 return start;
2253 }
2254
2255
2256 // Arguments:
2257 //
2258 // Inputs:
2259 // c_rarg0 - source byte array address
2260 // c_rarg1 - destination byte array address
2261 // c_rarg2 - K (key) in little endian int array
2262 //
2263 address generate_aescrypt_decryptBlock() {
2264 assert(UseAES, "need AES instructions and misaligned SSE support");
2265 __ align(CodeEntryAlignment);
2266 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2267 Label L_doLast;
2268 address start = __ pc();
2269
2270 const Register from = rdx; // source array address
2271 const Register to = rdx; // destination array address
2272 const Register key = rcx; // key array address
2273 const Register keylen = rax;
2274 const Address from_param(rbp, 8+0);
2275 const Address to_param (rbp, 8+4);
2276 const Address key_param (rbp, 8+8);
2277
2278 const XMMRegister xmm_result = xmm0;
2279 const XMMRegister xmm_key_shuf_mask = xmm1;
2280 const XMMRegister xmm_temp1 = xmm2;
2281 const XMMRegister xmm_temp2 = xmm3;
2282 const XMMRegister xmm_temp3 = xmm4;
2283 const XMMRegister xmm_temp4 = xmm5;
2284
2285 __ enter(); // required for proper stackwalking of RuntimeStub frame
2286
2287 __ movptr(from, from_param);
2288 __ movptr(key, key_param);
2289
2290 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2291 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2292
2293 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2294 __ movdqu(xmm_result, Address(from, 0));
2295 __ movptr(to, to_param);
2296
2297 // for decryption java expanded key ordering is rotated one position from what we want
2298 // so we start from 0x10 here and hit 0x00 last
2299 // we don't know if the key is aligned, hence not using load-execute form
2300 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2301 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2302 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2303 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2304
2305 __ pxor (xmm_result, xmm_temp1);
2306 __ aesdec(xmm_result, xmm_temp2);
2307 __ aesdec(xmm_result, xmm_temp3);
2308 __ aesdec(xmm_result, xmm_temp4);
2309
2310 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2311 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2312 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2313 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2314
2315 __ aesdec(xmm_result, xmm_temp1);
2316 __ aesdec(xmm_result, xmm_temp2);
2317 __ aesdec(xmm_result, xmm_temp3);
2318 __ aesdec(xmm_result, xmm_temp4);
2319
2320 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2321 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2322 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2323
2324 __ cmpl(keylen, 44);
2325 __ jccb(Assembler::equal, L_doLast);
2326
2327 __ aesdec(xmm_result, xmm_temp1);
2328 __ aesdec(xmm_result, xmm_temp2);
2329
2330 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2331 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2332
2333 __ cmpl(keylen, 52);
2334 __ jccb(Assembler::equal, L_doLast);
2335
2336 __ aesdec(xmm_result, xmm_temp1);
2337 __ aesdec(xmm_result, xmm_temp2);
2338
2339 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2340 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2341
2342 __ BIND(L_doLast);
2343 __ aesdec(xmm_result, xmm_temp1);
2344 __ aesdec(xmm_result, xmm_temp2);
2345
2346 // for decryption the aesdeclast operation is always on key+0x00
2347 __ aesdeclast(xmm_result, xmm_temp3);
2348 __ movdqu(Address(to, 0), xmm_result); // store the result
2349 __ xorptr(rax, rax); // return 0
2350 __ leave(); // required for proper stackwalking of RuntimeStub frame
2351 __ ret(0);
2352
2353 return start;
2354 }
2355
2356 void handleSOERegisters(bool saving) {
2357 const int saveFrameSizeInBytes = 4 * wordSize;
2358 const Address saved_rbx (rbp, -3 * wordSize);
2359 const Address saved_rsi (rbp, -2 * wordSize);
2360 const Address saved_rdi (rbp, -1 * wordSize);
2361
2362 if (saving) {
2363 __ subptr(rsp, saveFrameSizeInBytes);
2364 __ movptr(saved_rsi, rsi);
2365 __ movptr(saved_rdi, rdi);
2366 __ movptr(saved_rbx, rbx);
2367 } else {
2368 // restoring
2369 __ movptr(rsi, saved_rsi);
2370 __ movptr(rdi, saved_rdi);
2371 __ movptr(rbx, saved_rbx);
2372 }
2373 }
2374
2375 // Arguments:
2376 //
2377 // Inputs:
2378 // c_rarg0 - source byte array address
2379 // c_rarg1 - destination byte array address
2380 // c_rarg2 - K (key) in little endian int array
2381 // c_rarg3 - r vector byte array address
2382 // c_rarg4 - input length
2383 //
2384 // Output:
2385 // rax - input length
2386 //
2387 address generate_cipherBlockChaining_encryptAESCrypt() {
2388 assert(UseAES, "need AES instructions and misaligned SSE support");
2389 __ align(CodeEntryAlignment);
2390 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2391 address start = __ pc();
2392
2393 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2394 const Register from = rsi; // source array address
2395 const Register to = rdx; // destination array address
2396 const Register key = rcx; // key array address
2397 const Register rvec = rdi; // r byte array initialized from initvector array address
2398 // and left with the results of the last encryption block
2399 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2400 const Register pos = rax;
2401
2402 // xmm register assignments for the loops below
2403 const XMMRegister xmm_result = xmm0;
2404 const XMMRegister xmm_temp = xmm1;
2405 // first 6 keys preloaded into xmm2-xmm7
2406 const int XMM_REG_NUM_KEY_FIRST = 2;
2407 const int XMM_REG_NUM_KEY_LAST = 7;
2408 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2409
2410 __ enter(); // required for proper stackwalking of RuntimeStub frame
2411 handleSOERegisters(true /*saving*/);
2412
2413 // load registers from incoming parameters
2414 const Address from_param(rbp, 8+0);
2415 const Address to_param (rbp, 8+4);
2416 const Address key_param (rbp, 8+8);
2417 const Address rvec_param (rbp, 8+12);
2418 const Address len_param (rbp, 8+16);
2419 __ movptr(from , from_param);
2420 __ movptr(to , to_param);
2421 __ movptr(key , key_param);
2422 __ movptr(rvec , rvec_param);
2423 __ movptr(len_reg , len_param);
2424
2425 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2426 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2427 // load up xmm regs 2 thru 7 with keys 0-5
2428 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2429 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2430 offset += 0x10;
2431 }
2432
2433 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2434
2435 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2436 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2437 __ cmpl(rax, 44);
2438 __ jcc(Assembler::notEqual, L_key_192_256);
2439
2440 // 128 bit code follows here
2441 __ movl(pos, 0);
2442 __ align(OptoLoopAlignment);
2443 __ BIND(L_loopTop_128);
2444 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2445 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2446
2447 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2448 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2449 __ aesenc(xmm_result, as_XMMRegister(rnum));
2450 }
2451 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2452 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2453 }
2454 load_key(xmm_temp, key, 0xa0);
2455 __ aesenclast(xmm_result, xmm_temp);
2456
2457 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2458 // no need to store r to memory until we exit
2459 __ addptr(pos, AESBlockSize);
2460 __ subptr(len_reg, AESBlockSize);
2461 __ jcc(Assembler::notEqual, L_loopTop_128);
2462
2463 __ BIND(L_exit);
2464 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2465
2466 handleSOERegisters(false /*restoring*/);
2467 __ movptr(rax, len_param); // return length
2468 __ leave(); // required for proper stackwalking of RuntimeStub frame
2469 __ ret(0);
2470
2471 __ BIND(L_key_192_256);
2472 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2473 __ cmpl(rax, 52);
2474 __ jcc(Assembler::notEqual, L_key_256);
2475
2476 // 192-bit code follows here (could be changed to use more xmm registers)
2477 __ movl(pos, 0);
2478 __ align(OptoLoopAlignment);
2479 __ BIND(L_loopTop_192);
2480 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2481 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2482
2483 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2484 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2485 __ aesenc(xmm_result, as_XMMRegister(rnum));
2486 }
2487 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2488 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2489 }
2490 load_key(xmm_temp, key, 0xc0);
2491 __ aesenclast(xmm_result, xmm_temp);
2492
2493 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2494 // no need to store r to memory until we exit
2495 __ addptr(pos, AESBlockSize);
2496 __ subptr(len_reg, AESBlockSize);
2497 __ jcc(Assembler::notEqual, L_loopTop_192);
2498 __ jmp(L_exit);
2499
2500 __ BIND(L_key_256);
2501 // 256-bit code follows here (could be changed to use more xmm registers)
2502 __ movl(pos, 0);
2503 __ align(OptoLoopAlignment);
2504 __ BIND(L_loopTop_256);
2505 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2506 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2507
2508 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2509 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2510 __ aesenc(xmm_result, as_XMMRegister(rnum));
2511 }
2512 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2513 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2514 }
2515 load_key(xmm_temp, key, 0xe0);
2516 __ aesenclast(xmm_result, xmm_temp);
2517
2518 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2519 // no need to store r to memory until we exit
2520 __ addptr(pos, AESBlockSize);
2521 __ subptr(len_reg, AESBlockSize);
2522 __ jcc(Assembler::notEqual, L_loopTop_256);
2523 __ jmp(L_exit);
2524
2525 return start;
2526 }
2527
2528
2529 // CBC AES Decryption.
2530 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2531 //
2532 // Arguments:
2533 //
2534 // Inputs:
2535 // c_rarg0 - source byte array address
2536 // c_rarg1 - destination byte array address
2537 // c_rarg2 - K (key) in little endian int array
2538 // c_rarg3 - r vector byte array address
2539 // c_rarg4 - input length
2540 //
2541 // Output:
2542 // rax - input length
2543 //
2544
2545 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
2546 assert(UseAES, "need AES instructions and misaligned SSE support");
2547 __ align(CodeEntryAlignment);
2548 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2549 address start = __ pc();
2550
2551 const Register from = rsi; // source array address
2552 const Register to = rdx; // destination array address
2553 const Register key = rcx; // key array address
2554 const Register rvec = rdi; // r byte array initialized from initvector array address
2555 // and left with the results of the last encryption block
2556 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2557 const Register pos = rax;
2558
2559 const int PARALLEL_FACTOR = 4;
2560 const int ROUNDS[3] = { 10, 12, 14 }; //aes rounds for key128, key192, key256
2561
2562 Label L_exit;
2563 Label L_singleBlock_loopTop[3]; //128, 192, 256
2564 Label L_multiBlock_loopTop[3]; //128, 192, 256
2565
2566 const XMMRegister xmm_prev_block_cipher = xmm0; // holds cipher of previous block
2567 const XMMRegister xmm_key_shuf_mask = xmm1;
2568
2569 const XMMRegister xmm_key_tmp0 = xmm2;
2570 const XMMRegister xmm_key_tmp1 = xmm3;
2571
2572 // registers holding the six results in the parallelized loop
2573 const XMMRegister xmm_result0 = xmm4;
2574 const XMMRegister xmm_result1 = xmm5;
2575 const XMMRegister xmm_result2 = xmm6;
2576 const XMMRegister xmm_result3 = xmm7;
2577
2578 __ enter(); // required for proper stackwalking of RuntimeStub frame
2579 handleSOERegisters(true /*saving*/);
2580
2581 // load registers from incoming parameters
2582 const Address from_param(rbp, 8+0);
2583 const Address to_param (rbp, 8+4);
2584 const Address key_param (rbp, 8+8);
2585 const Address rvec_param (rbp, 8+12);
2586 const Address len_param (rbp, 8+16);
2587
2588 __ movptr(from , from_param);
2589 __ movptr(to , to_param);
2590 __ movptr(key , key_param);
2591 __ movptr(rvec , rvec_param);
2592 __ movptr(len_reg , len_param);
2593
2594 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2595 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
2596
2597 __ xorptr(pos, pos);
2598
2599 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2600 // rvec is reused
2601 __ movl(rvec, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2602 __ cmpl(rvec, 52);
2603 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
2604 __ cmpl(rvec, 60);
2605 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
2606
2607 #define DoFour(opc, src_reg) \
2608 __ opc(xmm_result0, src_reg); \
2609 __ opc(xmm_result1, src_reg); \
2610 __ opc(xmm_result2, src_reg); \
2611 __ opc(xmm_result3, src_reg); \
2612
2613 for (int k = 0; k < 3; ++k) {
2614 __ align(OptoLoopAlignment);
2615 __ BIND(L_multiBlock_loopTop[k]);
2616 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
2617 __ jcc(Assembler::less, L_singleBlock_loopTop[k]);
2618
2619 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
2620 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2621 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2622 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2623
2624 // the java expanded key ordering is rotated one position from what we want
2625 // so we start from 0x10 here and hit 0x00 last
2626 load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2627 DoFour(pxor, xmm_key_tmp0); //xor with first key
2628 // do the aes dec rounds
2629 for (int rnum = 1; rnum <= ROUNDS[k];) {
2630 //load two keys at a time
2631 //k1->0x20, ..., k9->0xa0, k10->0x00
2632 load_key(xmm_key_tmp1, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2633 load_key(xmm_key_tmp0, key, ((rnum + 2) % (ROUNDS[k] + 1)) * 0x10, xmm_key_shuf_mask); // hit 0x00 last!
2634 DoFour(aesdec, xmm_key_tmp1);
2635 rnum++;
2636 if (rnum != ROUNDS[k]) {
2637 DoFour(aesdec, xmm_key_tmp0);
2638 }
2639 else {
2640 DoFour(aesdeclast, xmm_key_tmp0);
2641 }
2642 rnum++;
2643 }
2644
2645 // for each result, xor with the r vector of previous cipher block
2646 __ pxor(xmm_result0, xmm_prev_block_cipher);
2647 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2648 __ pxor(xmm_result1, xmm_prev_block_cipher);
2649 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2650 __ pxor(xmm_result2, xmm_prev_block_cipher);
2651 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2652 __ pxor(xmm_result3, xmm_prev_block_cipher);
2653 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
2654
2655 // store 4 results into the next 64 bytes of output
2656 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2657 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2658 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2659 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2660
2661 __ addptr(pos, 4 * AESBlockSize);
2662 __ subptr(len_reg, 4 * AESBlockSize);
2663 __ jmp(L_multiBlock_loopTop[k]);
2664
2665 //singleBlock starts here
2666 __ align(OptoLoopAlignment);
2667 __ BIND(L_singleBlock_loopTop[k]);
2668 __ cmpptr(len_reg, 0); // any blocks left?
2669 __ jcc(Assembler::equal, L_exit);
2670 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2671 __ movdqa(xmm_result1, xmm_result0);
2672
2673 load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2674 __ pxor(xmm_result0, xmm_key_tmp0);
2675 // do the aes dec rounds
2676 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
2677 // the java expanded key ordering is rotated one position from what we want
2678 load_key(xmm_key_tmp0, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2679 __ aesdec(xmm_result0, xmm_key_tmp0);
2680 }
2681 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
2682 __ aesdeclast(xmm_result0, xmm_key_tmp0);
2683 __ pxor(xmm_result0, xmm_prev_block_cipher); // xor with the current r vector
2684 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result0); // store into the next 16 bytes of output
2685 // no need to store r to memory until we exit
2686 __ movdqa(xmm_prev_block_cipher, xmm_result1); // set up next r vector with cipher input from this block
2687
2688 __ addptr(pos, AESBlockSize);
2689 __ subptr(len_reg, AESBlockSize);
2690 __ jmp(L_singleBlock_loopTop[k]);
2691 }//for 128/192/256
2692
2693 __ BIND(L_exit);
2694 __ movptr(rvec, rvec_param); // restore this since reused earlier
2695 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
2696 handleSOERegisters(false /*restoring*/);
2697 __ movptr(rax, len_param); // return length
2698 __ leave(); // required for proper stackwalking of RuntimeStub frame
2699 __ ret(0);
2700
2701 return start;
2702 }
2703
2704 // CTR AES crypt.
2705 // In 32-bit stub, parallelize 4 blocks at a time
2706 // Arguments:
2707 //
2708 // Inputs:
2709 // c_rarg0 - source byte array address
2710 // c_rarg1 - destination byte array address
2711 // c_rarg2 - K (key) in little endian int array
2712 // c_rarg3 - counter vector byte array address
2713 // c_rarg4 - input length
2714 //
2715 // Output:
2716 // rax - input length
2717 //
2718 address generate_counterMode_AESCrypt_Parallel() {
2719 assert(UseAES, "need AES instructions and misaligned SSE support");
2720 __ align(CodeEntryAlignment);
2721 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
2722 address start = __ pc();
2723 const Register from = rsi; // source array address
2724 const Register to = rdx; // destination array address
2725 const Register key = rcx; // key array address
2726 const Register counter = rdi; // counter byte array initialized from initvector array address
2727 // and updated with the incremented counter in the end
2728 const Register len_reg = rbx;
2729 const Register pos = rax;
2730
2731 __ enter(); // required for proper stackwalking of RuntimeStub frame
2732 handleSOERegisters(true /*saving*/); // save rbx, rsi, rdi
2733
2734 // load registers from incoming parameters
2735 const Address from_param(rbp, 8+0);
2736 const Address to_param (rbp, 8+4);
2737 const Address key_param (rbp, 8+8);
2738 const Address rvec_param (rbp, 8+12);
2739 const Address len_param (rbp, 8+16);
2740 const Address saved_counter_param(rbp, 8 + 20);
2741 const Address used_addr_param(rbp, 8 + 24);
2742
2743 __ movptr(from , from_param);
2744 __ movptr(to , to_param);
2745 __ movptr(len_reg , len_param);
2746
2747 // Use the partially used encrpyted counter from last invocation
2748 Label L_exit_preLoop, L_preLoop_start;
2749
2750 // Use the registers 'counter' and 'key' here in this preloop
2751 // to hold of last 2 params 'used' and 'saved_encCounter_start'
2752 Register used = counter;
2753 Register saved_encCounter_start = key;
2754 Register used_addr = saved_encCounter_start;
2755
2756 __ movptr(used_addr, used_addr_param);
2757 __ movptr(used, Address(used_addr, 0));
2758 __ movptr(saved_encCounter_start, saved_counter_param);
2759
2760 __ BIND(L_preLoop_start);
2761 __ cmpptr(used, 16);
2762 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
2763 __ cmpptr(len_reg, 0);
2764 __ jcc(Assembler::lessEqual, L_exit_preLoop);
2765 __ movb(rax, Address(saved_encCounter_start, used));
2766 __ xorb(rax, Address(from, 0));
2767 __ movb(Address(to, 0), rax);
2768 __ addptr(from, 1);
2769 __ addptr(to, 1);
2770 __ addptr(used, 1);
2771 __ subptr(len_reg, 1);
2772
2773 __ jmp(L_preLoop_start);
2774
2775 __ BIND(L_exit_preLoop);
2776 __ movptr(used_addr, used_addr_param);
2777 __ movptr(used_addr, used_addr_param);
2778 __ movl(Address(used_addr, 0), used);
2779
2780 // load the parameters 'key' and 'counter'
2781 __ movptr(key, key_param);
2782 __ movptr(counter, rvec_param);
2783
2784 // xmm register assignments for the loops below
2785 const XMMRegister xmm_curr_counter = xmm0;
2786 const XMMRegister xmm_counter_shuf_mask = xmm1; // need to be reloaded
2787 const XMMRegister xmm_key_shuf_mask = xmm2; // need to be reloaded
2788 const XMMRegister xmm_key = xmm3;
2789 const XMMRegister xmm_result0 = xmm4;
2790 const XMMRegister xmm_result1 = xmm5;
2791 const XMMRegister xmm_result2 = xmm6;
2792 const XMMRegister xmm_result3 = xmm7;
2793 const XMMRegister xmm_from0 = xmm1; //reuse XMM register
2794 const XMMRegister xmm_from1 = xmm2;
2795 const XMMRegister xmm_from2 = xmm3;
2796 const XMMRegister xmm_from3 = xmm4;
2797
2798 //for key_128, key_192, key_256
2799 const int rounds[3] = {10, 12, 14};
2800 Label L_singleBlockLoopTop[3];
2801 Label L_multiBlock_loopTop[3];
2802 Label L_key192_top, L_key256_top;
2803 Label L_incCounter[3][4]; // 3: different key length, 4: 4 blocks at a time
2804 Label L_incCounter_single[3]; //for single block, key128, key192, key256
2805 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];
2806 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];
2807
2808 Label L_exit;
2809 const int PARALLEL_FACTOR = 4; //because of the limited register number
2810
2811 // initialize counter with initial counter
2812 __ movdqu(xmm_curr_counter, Address(counter, 0x00));
2813 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2814 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled for increase
2815
2816 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
2817 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2818 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2819 __ cmpl(rax, 52);
2820 __ jcc(Assembler::equal, L_key192_top);
2821 __ cmpl(rax, 60);
2822 __ jcc(Assembler::equal, L_key256_top);
2823
2824 //key128 begins here
2825 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2826
2827 #define CTR_DoFour(opc, src_reg) \
2828 __ opc(xmm_result0, src_reg); \
2829 __ opc(xmm_result1, src_reg); \
2830 __ opc(xmm_result2, src_reg); \
2831 __ opc(xmm_result3, src_reg);
2832
2833 // k == 0 : generate code for key_128
2834 // k == 1 : generate code for key_192
2835 // k == 2 : generate code for key_256
2836 for (int k = 0; k < 3; ++k) {
2837 //multi blocks starts here
2838 __ align(OptoLoopAlignment);
2839 __ BIND(L_multiBlock_loopTop[k]);
2840 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
2841 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
2842
2843 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2844 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2845
2846 //load, then increase counters
2847 CTR_DoFour(movdqa, xmm_curr_counter);
2848 __ push(rbx);
2849 inc_counter(rbx, xmm_result1, 0x01, L_incCounter[k][0]);
2850 inc_counter(rbx, xmm_result2, 0x02, L_incCounter[k][1]);
2851 inc_counter(rbx, xmm_result3, 0x03, L_incCounter[k][2]);
2852 inc_counter(rbx, xmm_curr_counter, 0x04, L_incCounter[k][3]);
2853 __ pop (rbx);
2854
2855 load_key(xmm_key, key, 0x00, xmm_key_shuf_mask); // load Round 0 key. interleaving for better performance
2856
2857 CTR_DoFour(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
2858 CTR_DoFour(pxor, xmm_key); //PXOR with Round 0 key
2859
2860 for (int i = 1; i < rounds[k]; ++i) {
2861 load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2862 CTR_DoFour(aesenc, xmm_key);
2863 }
2864 load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2865 CTR_DoFour(aesenclast, xmm_key);
2866
2867 // get next PARALLEL_FACTOR blocks into xmm_from registers
2868 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2869 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2870 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2871
2872 // PXOR with input text
2873 __ pxor(xmm_result0, xmm_from0); //result0 is xmm4
2874 __ pxor(xmm_result1, xmm_from1);
2875 __ pxor(xmm_result2, xmm_from2);
2876
2877 // store PARALLEL_FACTOR results into the next 64 bytes of output
2878 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2879 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2880 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2881
2882 // do it here after xmm_result0 is saved, because xmm_from3 reuse the same register of xmm_result0.
2883 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2884 __ pxor(xmm_result3, xmm_from3);
2885 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2886
2887 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
2888 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
2889 __ jmp(L_multiBlock_loopTop[k]);
2890
2891 // singleBlock starts here
2892 __ align(OptoLoopAlignment);
2893 __ BIND(L_singleBlockLoopTop[k]);
2894 __ cmpptr(len_reg, 0);
2895 __ jcc(Assembler::equal, L_exit);
2896 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2897 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2898 __ movdqa(xmm_result0, xmm_curr_counter);
2899 load_key(xmm_key, key, 0x00, xmm_key_shuf_mask);
2900 __ push(rbx);//rbx is used for increasing counter
2901 inc_counter(rbx, xmm_curr_counter, 0x01, L_incCounter_single[k]);
2902 __ pop (rbx);
2903 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
2904 __ pxor(xmm_result0, xmm_key);
2905 for (int i = 1; i < rounds[k]; i++) {
2906 load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2907 __ aesenc(xmm_result0, xmm_key);
2908 }
2909 load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2910 __ aesenclast(xmm_result0, xmm_key);
2911 __ cmpptr(len_reg, AESBlockSize);
2912 __ jcc(Assembler::less, L_processTail_insr[k]);
2913 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2914 __ pxor(xmm_result0, xmm_from0);
2915 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2916 __ addptr(pos, AESBlockSize);
2917 __ subptr(len_reg, AESBlockSize);
2918 __ jmp(L_singleBlockLoopTop[k]);
2919
2920 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
2921 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
2922 __ testptr(len_reg, 8);
2923 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
2924 __ subptr(pos,8);
2925 __ pinsrd(xmm_from0, Address(from, pos), 0);
2926 __ pinsrd(xmm_from0, Address(from, pos, Address::times_1, 4), 1);
2927 __ BIND(L_processTail_4_insr[k]);
2928 __ testptr(len_reg, 4);
2929 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
2930 __ subptr(pos,4);
2931 __ pslldq(xmm_from0, 4);
2932 __ pinsrd(xmm_from0, Address(from, pos), 0);
2933 __ BIND(L_processTail_2_insr[k]);
2934 __ testptr(len_reg, 2);
2935 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
2936 __ subptr(pos, 2);
2937 __ pslldq(xmm_from0, 2);
2938 __ pinsrw(xmm_from0, Address(from, pos), 0);
2939 __ BIND(L_processTail_1_insr[k]);
2940 __ testptr(len_reg, 1);
2941 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
2942 __ subptr(pos, 1);
2943 __ pslldq(xmm_from0, 1);
2944 __ pinsrb(xmm_from0, Address(from, pos), 0);
2945 __ BIND(L_processTail_exit_insr[k]);
2946
2947 __ movptr(saved_encCounter_start, saved_counter_param);
2948 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
2949 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
2950
2951 __ testptr(len_reg, 8);
2952 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
2953 __ pextrd(Address(to, pos), xmm_result0, 0);
2954 __ pextrd(Address(to, pos, Address::times_1, 4), xmm_result0, 1);
2955 __ psrldq(xmm_result0, 8);
2956 __ addptr(pos, 8);
2957 __ BIND(L_processTail_4_extr[k]);
2958 __ testptr(len_reg, 4);
2959 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
2960 __ pextrd(Address(to, pos), xmm_result0, 0);
2961 __ psrldq(xmm_result0, 4);
2962 __ addptr(pos, 4);
2963 __ BIND(L_processTail_2_extr[k]);
2964 __ testptr(len_reg, 2);
2965 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
2966 __ pextrb(Address(to, pos), xmm_result0, 0);
2967 __ pextrb(Address(to, pos, Address::times_1, 1), xmm_result0, 1);
2968 __ psrldq(xmm_result0, 2);
2969 __ addptr(pos, 2);
2970 __ BIND(L_processTail_1_extr[k]);
2971 __ testptr(len_reg, 1);
2972 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
2973 __ pextrb(Address(to, pos), xmm_result0, 0);
2974
2975 __ BIND(L_processTail_exit_extr[k]);
2976 __ movptr(used_addr, used_addr_param);
2977 __ movl(Address(used_addr, 0), len_reg);
2978 __ jmp(L_exit);
2979 }
2980
2981 __ BIND(L_exit);
2982 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2983 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
2984 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
2985 handleSOERegisters(false /*restoring*/);
2986 __ movptr(rax, len_param); // return length
2987 __ leave(); // required for proper stackwalking of RuntimeStub frame
2988 __ ret(0);
2989
2990 __ BIND (L_key192_top);
2991 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2992 __ jmp(L_multiBlock_loopTop[1]); //key192
2993
2994 __ BIND (L_key256_top);
2995 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2996 __ jmp(L_multiBlock_loopTop[2]); //key192
2997
2998 return start;
2999 }
3000
3001 address generate_upper_word_mask() {
3002 __ align(64);
3003 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3004 address start = __ pc();
3005 __ emit_data(0x00000000, relocInfo::none, 0);
3006 __ emit_data(0x00000000, relocInfo::none, 0);
3007 __ emit_data(0x00000000, relocInfo::none, 0);
3008 __ emit_data(0xFFFFFFFF, relocInfo::none, 0);
3009 return start;
3010 }
3011
3012 address generate_shuffle_byte_flip_mask() {
3013 __ align(64);
3014 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3015 address start = __ pc();
3016 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3017 __ emit_data(0x08090a0b, relocInfo::none, 0);
3018 __ emit_data(0x04050607, relocInfo::none, 0);
3019 __ emit_data(0x00010203, relocInfo::none, 0);
3020 return start;
3021 }
3022
3023 // ofs and limit are use for multi-block byte array.
3024 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3025 address generate_sha1_implCompress(bool multi_block, const char *name) {
3026 __ align(CodeEntryAlignment);
3027 StubCodeMark mark(this, "StubRoutines", name);
3028 address start = __ pc();
3029
3030 Register buf = rax;
3031 Register state = rdx;
3032 Register ofs = rcx;
3033 Register limit = rdi;
3034
3035 const Address buf_param(rbp, 8 + 0);
3036 const Address state_param(rbp, 8 + 4);
3037 const Address ofs_param(rbp, 8 + 8);
3038 const Address limit_param(rbp, 8 + 12);
3039
3040 const XMMRegister abcd = xmm0;
3041 const XMMRegister e0 = xmm1;
3042 const XMMRegister e1 = xmm2;
3043 const XMMRegister msg0 = xmm3;
3044
3045 const XMMRegister msg1 = xmm4;
3046 const XMMRegister msg2 = xmm5;
3047 const XMMRegister msg3 = xmm6;
3048 const XMMRegister shuf_mask = xmm7;
3049
3050 __ enter();
3051 __ subptr(rsp, 8 * wordSize);
3052 handleSOERegisters(true /*saving*/);
3053
3054 __ movptr(buf, buf_param);
3055 __ movptr(state, state_param);
3056 if (multi_block) {
3057 __ movptr(ofs, ofs_param);
3058 __ movptr(limit, limit_param);
3059 }
3060
3061 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3062 buf, state, ofs, limit, rsp, multi_block);
3063
3064 handleSOERegisters(false /*restoring*/);
3065 __ addptr(rsp, 8 * wordSize);
3066 __ leave();
3067 __ ret(0);
3068 return start;
3069 }
3070
3071 address generate_pshuffle_byte_flip_mask() {
3072 __ align(64);
3073 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3074 address start = __ pc();
3075 __ emit_data(0x00010203, relocInfo::none, 0);
3076 __ emit_data(0x04050607, relocInfo::none, 0);
3077 __ emit_data(0x08090a0b, relocInfo::none, 0);
3078 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3079 return start;
3080 }
3081
3082 // ofs and limit are use for multi-block byte array.
3083 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3084 address generate_sha256_implCompress(bool multi_block, const char *name) {
3085 __ align(CodeEntryAlignment);
3086 StubCodeMark mark(this, "StubRoutines", name);
3087 address start = __ pc();
3088
3089 Register buf = rbx;
3090 Register state = rsi;
3091 Register ofs = rdx;
3092 Register limit = rcx;
3093
3094 const Address buf_param(rbp, 8 + 0);
3095 const Address state_param(rbp, 8 + 4);
3096 const Address ofs_param(rbp, 8 + 8);
3097 const Address limit_param(rbp, 8 + 12);
3098
3099 const XMMRegister msg = xmm0;
3100 const XMMRegister state0 = xmm1;
3101 const XMMRegister state1 = xmm2;
3102 const XMMRegister msgtmp0 = xmm3;
3103
3104 const XMMRegister msgtmp1 = xmm4;
3105 const XMMRegister msgtmp2 = xmm5;
3106 const XMMRegister msgtmp3 = xmm6;
3107 const XMMRegister msgtmp4 = xmm7;
3108
3109 __ enter();
3110 __ subptr(rsp, 8 * wordSize);
3111 handleSOERegisters(true /*saving*/);
3112 __ movptr(buf, buf_param);
3113 __ movptr(state, state_param);
3114 if (multi_block) {
3115 __ movptr(ofs, ofs_param);
3116 __ movptr(limit, limit_param);
3117 }
3118
3119 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3120 buf, state, ofs, limit, rsp, multi_block);
3121
3122 handleSOERegisters(false);
3123 __ addptr(rsp, 8 * wordSize);
3124 __ leave();
3125 __ ret(0);
3126 return start;
3127 }
3128
3129 // byte swap x86 long
3130 address generate_ghash_long_swap_mask() {
3131 __ align(CodeEntryAlignment);
3132 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
3133 address start = __ pc();
3134 __ emit_data(0x0b0a0908, relocInfo::none, 0);
3135 __ emit_data(0x0f0e0d0c, relocInfo::none, 0);
3136 __ emit_data(0x03020100, relocInfo::none, 0);
3137 __ emit_data(0x07060504, relocInfo::none, 0);
3138
3139 return start;
3140 }
3141
3142 // byte swap x86 byte array
3143 address generate_ghash_byte_swap_mask() {
3144 __ align(CodeEntryAlignment);
3145 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
3146 address start = __ pc();
3147 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3148 __ emit_data(0x08090a0b, relocInfo::none, 0);
3149 __ emit_data(0x04050607, relocInfo::none, 0);
3150 __ emit_data(0x00010203, relocInfo::none, 0);
3151 return start;
3152 }
3153
3154 /* Single and multi-block ghash operations */
3155 address generate_ghash_processBlocks() {
3156 assert(UseGHASHIntrinsics, "need GHASH intrinsics and CLMUL support");
3157 __ align(CodeEntryAlignment);
3158 Label L_ghash_loop, L_exit;
3159 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
3160 address start = __ pc();
3161
3162 const Register state = rdi;
3163 const Register subkeyH = rsi;
3164 const Register data = rdx;
3165 const Register blocks = rcx;
3166
3167 const Address state_param(rbp, 8+0);
3168 const Address subkeyH_param(rbp, 8+4);
3169 const Address data_param(rbp, 8+8);
3170 const Address blocks_param(rbp, 8+12);
3171
3172 const XMMRegister xmm_temp0 = xmm0;
3173 const XMMRegister xmm_temp1 = xmm1;
3174 const XMMRegister xmm_temp2 = xmm2;
3175 const XMMRegister xmm_temp3 = xmm3;
3176 const XMMRegister xmm_temp4 = xmm4;
3177 const XMMRegister xmm_temp5 = xmm5;
3178 const XMMRegister xmm_temp6 = xmm6;
3179 const XMMRegister xmm_temp7 = xmm7;
3180
3181 __ enter();
3182 handleSOERegisters(true); // Save registers
3183
3184 __ movptr(state, state_param);
3185 __ movptr(subkeyH, subkeyH_param);
3186 __ movptr(data, data_param);
3187 __ movptr(blocks, blocks_param);
3188
3189 __ movdqu(xmm_temp0, Address(state, 0));
3190 __ pshufb(xmm_temp0, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3191
3192 __ movdqu(xmm_temp1, Address(subkeyH, 0));
3193 __ pshufb(xmm_temp1, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3194
3195 __ BIND(L_ghash_loop);
3196 __ movdqu(xmm_temp2, Address(data, 0));
3197 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
3198
3199 __ pxor(xmm_temp0, xmm_temp2);
3200
3201 //
3202 // Multiply with the hash key
3203 //
3204 __ movdqu(xmm_temp3, xmm_temp0);
3205 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
3206 __ movdqu(xmm_temp4, xmm_temp0);
3207 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
3208
3209 __ movdqu(xmm_temp5, xmm_temp0);
3210 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
3211 __ movdqu(xmm_temp6, xmm_temp0);
3212 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
3213
3214 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
3215
3216 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
3217 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
3218 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
3219 __ pxor(xmm_temp3, xmm_temp5);
3220 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
3221 // of the carry-less multiplication of
3222 // xmm0 by xmm1.
3223
3224 // We shift the result of the multiplication by one bit position
3225 // to the left to cope for the fact that the bits are reversed.
3226 __ movdqu(xmm_temp7, xmm_temp3);
3227 __ movdqu(xmm_temp4, xmm_temp6);
3228 __ pslld (xmm_temp3, 1);
3229 __ pslld(xmm_temp6, 1);
3230 __ psrld(xmm_temp7, 31);
3231 __ psrld(xmm_temp4, 31);
3232 __ movdqu(xmm_temp5, xmm_temp7);
3233 __ pslldq(xmm_temp4, 4);
3234 __ pslldq(xmm_temp7, 4);
3235 __ psrldq(xmm_temp5, 12);
3236 __ por(xmm_temp3, xmm_temp7);
3237 __ por(xmm_temp6, xmm_temp4);
3238 __ por(xmm_temp6, xmm_temp5);
3239
3240 //
3241 // First phase of the reduction
3242 //
3243 // Move xmm3 into xmm4, xmm5, xmm7 in order to perform the shifts
3244 // independently.
3245 __ movdqu(xmm_temp7, xmm_temp3);
3246 __ movdqu(xmm_temp4, xmm_temp3);
3247 __ movdqu(xmm_temp5, xmm_temp3);
3248 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
3249 __ pslld(xmm_temp4, 30); // packed right shift shifting << 30
3250 __ pslld(xmm_temp5, 25); // packed right shift shifting << 25
3251 __ pxor(xmm_temp7, xmm_temp4); // xor the shifted versions
3252 __ pxor(xmm_temp7, xmm_temp5);
3253 __ movdqu(xmm_temp4, xmm_temp7);
3254 __ pslldq(xmm_temp7, 12);
3255 __ psrldq(xmm_temp4, 4);
3256 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
3257
3258 //
3259 // Second phase of the reduction
3260 //
3261 // Make 3 copies of xmm3 in xmm2, xmm5, xmm7 for doing these
3262 // shift operations.
3263 __ movdqu(xmm_temp2, xmm_temp3);
3264 __ movdqu(xmm_temp7, xmm_temp3);
3265 __ movdqu(xmm_temp5, xmm_temp3);
3266 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
3267 __ psrld(xmm_temp7, 2); // packed left shifting >> 2
3268 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
3269 __ pxor(xmm_temp2, xmm_temp7); // xor the shifted versions
3270 __ pxor(xmm_temp2, xmm_temp5);
3271 __ pxor(xmm_temp2, xmm_temp4);
3272 __ pxor(xmm_temp3, xmm_temp2);
3273 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
3274
3275 __ decrement(blocks);
3276 __ jcc(Assembler::zero, L_exit);
3277 __ movdqu(xmm_temp0, xmm_temp6);
3278 __ addptr(data, 16);
3279 __ jmp(L_ghash_loop);
3280
3281 __ BIND(L_exit);
3282 // Byte swap 16-byte result
3283 __ pshufb(xmm_temp6, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3284 __ movdqu(Address(state, 0), xmm_temp6); // store the result
3285
3286 handleSOERegisters(false); // restore registers
3287 __ leave();
3288 __ ret(0);
3289 return start;
3290 }
3291
3292 /**
3293 * Arguments:
3294 *
3295 * Inputs:
3296 * rsp(4) - int crc
3297 * rsp(8) - byte* buf
3298 * rsp(12) - int length
3299 *
3300 * Ouput:
3301 * rax - int crc result
3302 */
3303 address generate_updateBytesCRC32() {
3304 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3305
3306 __ align(CodeEntryAlignment);
3307 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
3308
3309 address start = __ pc();
3310
3311 const Register crc = rdx; // crc
3312 const Register buf = rsi; // source java byte array address
3313 const Register len = rcx; // length
3314 const Register table = rdi; // crc_table address (reuse register)
3315 const Register tmp = rbx;
3316 assert_different_registers(crc, buf, len, table, tmp, rax);
3317
3318 BLOCK_COMMENT("Entry:");
3319 __ enter(); // required for proper stackwalking of RuntimeStub frame
3320 __ push(rsi);
3321 __ push(rdi);
3322 __ push(rbx);
3323
3324 Address crc_arg(rbp, 8 + 0);
3325 Address buf_arg(rbp, 8 + 4);
3326 Address len_arg(rbp, 8 + 8);
3327
3328 // Load up:
3329 __ movl(crc, crc_arg);
3330 __ movptr(buf, buf_arg);
3331 __ movl(len, len_arg);
3332
3333 __ kernel_crc32(crc, buf, len, table, tmp);
3334
3335 __ movl(rax, crc);
3336 __ pop(rbx);
3337 __ pop(rdi);
3338 __ pop(rsi);
3339 __ vzeroupper();
3340 __ leave(); // required for proper stackwalking of RuntimeStub frame
3341 __ ret(0);
3342
3343 return start;
3344 }
3345
3346 /**
3347 * Arguments:
3348 *
3349 * Inputs:
3350 * rsp(4) - int crc
3351 * rsp(8) - byte* buf
3352 * rsp(12) - int length
3353 * rsp(16) - table_start - optional (present only when doing a library_calll,
3354 * not used by x86 algorithm)
3355 *
3356 * Ouput:
3357 * rax - int crc result
3358 */
3359 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
3360 assert(UseCRC32CIntrinsics, "need SSE4_2");
3361 __ align(CodeEntryAlignment);
3362 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
3363 address start = __ pc();
3364 const Register crc = rax; // crc
3365 const Register buf = rcx; // source java byte array address
3366 const Register len = rdx; // length
3367 const Register d = rbx;
3368 const Register g = rsi;
3369 const Register h = rdi;
3370 const Register empty = 0; // will never be used, in order not
3371 // to change a signature for crc32c_IPL_Alg2_Alt2
3372 // between 64/32 I'm just keeping it here
3373 assert_different_registers(crc, buf, len, d, g, h);
3374
3375 BLOCK_COMMENT("Entry:");
3376 __ enter(); // required for proper stackwalking of RuntimeStub frame
3377 Address crc_arg(rsp, 4 + 4 + 0); // ESP+4 +
3378 // we need to add additional 4 because __ enter
3379 // have just pushed ebp on a stack
3380 Address buf_arg(rsp, 4 + 4 + 4);
3381 Address len_arg(rsp, 4 + 4 + 8);
3382 // Load up:
3383 __ movl(crc, crc_arg);
3384 __ movl(buf, buf_arg);
3385 __ movl(len, len_arg);
3386 __ push(d);
3387 __ push(g);
3388 __ push(h);
3389 __ crc32c_ipl_alg2_alt2(crc, buf, len,
3390 d, g, h,
3391 empty, empty, empty,
3392 xmm0, xmm1, xmm2,
3393 is_pclmulqdq_supported);
3394 __ pop(h);
3395 __ pop(g);
3396 __ pop(d);
3397 __ vzeroupper();
3398 __ leave(); // required for proper stackwalking of RuntimeStub frame
3399 __ ret(0);
3400
3401 return start;
3402 }
3403
3404 address generate_libmExp() {
3405 StubCodeMark mark(this, "StubRoutines", "libmExp");
3406
3407 address start = __ pc();
3408
3409 const XMMRegister x0 = xmm0;
3410 const XMMRegister x1 = xmm1;
3411 const XMMRegister x2 = xmm2;
3412 const XMMRegister x3 = xmm3;
3413
3414 const XMMRegister x4 = xmm4;
3415 const XMMRegister x5 = xmm5;
3416 const XMMRegister x6 = xmm6;
3417 const XMMRegister x7 = xmm7;
3418
3419 const Register tmp = rbx;
3420
3421 BLOCK_COMMENT("Entry:");
3422 __ enter(); // required for proper stackwalking of RuntimeStub frame
3423 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3424 __ leave(); // required for proper stackwalking of RuntimeStub frame
3425 __ ret(0);
3426
3427 return start;
3428
3429 }
3430
3431 address generate_libmLog() {
3432 StubCodeMark mark(this, "StubRoutines", "libmLog");
3433
3434 address start = __ pc();
3435
3436 const XMMRegister x0 = xmm0;
3437 const XMMRegister x1 = xmm1;
3438 const XMMRegister x2 = xmm2;
3439 const XMMRegister x3 = xmm3;
3440
3441 const XMMRegister x4 = xmm4;
3442 const XMMRegister x5 = xmm5;
3443 const XMMRegister x6 = xmm6;
3444 const XMMRegister x7 = xmm7;
3445
3446 const Register tmp = rbx;
3447
3448 BLOCK_COMMENT("Entry:");
3449 __ enter(); // required for proper stackwalking of RuntimeStub frame
3450 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3451 __ leave(); // required for proper stackwalking of RuntimeStub frame
3452 __ ret(0);
3453
3454 return start;
3455
3456 }
3457
3458 address generate_libmLog10() {
3459 StubCodeMark mark(this, "StubRoutines", "libmLog10");
3460
3461 address start = __ pc();
3462
3463 const XMMRegister x0 = xmm0;
3464 const XMMRegister x1 = xmm1;
3465 const XMMRegister x2 = xmm2;
3466 const XMMRegister x3 = xmm3;
3467
3468 const XMMRegister x4 = xmm4;
3469 const XMMRegister x5 = xmm5;
3470 const XMMRegister x6 = xmm6;
3471 const XMMRegister x7 = xmm7;
3472
3473 const Register tmp = rbx;
3474
3475 BLOCK_COMMENT("Entry:");
3476 __ enter(); // required for proper stackwalking of RuntimeStub frame
3477 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3478 __ leave(); // required for proper stackwalking of RuntimeStub frame
3479 __ ret(0);
3480
3481 return start;
3482
3483 }
3484
3485 address generate_libmPow() {
3486 StubCodeMark mark(this, "StubRoutines", "libmPow");
3487
3488 address start = __ pc();
3489
3490 const XMMRegister x0 = xmm0;
3491 const XMMRegister x1 = xmm1;
3492 const XMMRegister x2 = xmm2;
3493 const XMMRegister x3 = xmm3;
3494
3495 const XMMRegister x4 = xmm4;
3496 const XMMRegister x5 = xmm5;
3497 const XMMRegister x6 = xmm6;
3498 const XMMRegister x7 = xmm7;
3499
3500 const Register tmp = rbx;
3501
3502 BLOCK_COMMENT("Entry:");
3503 __ enter(); // required for proper stackwalking of RuntimeStub frame
3504 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3505 __ leave(); // required for proper stackwalking of RuntimeStub frame
3506 __ ret(0);
3507
3508 return start;
3509
3510 }
3511
3512 address generate_libm_reduce_pi04l() {
3513 StubCodeMark mark(this, "StubRoutines", "libm_reduce_pi04l");
3514
3515 address start = __ pc();
3516
3517 BLOCK_COMMENT("Entry:");
3518 __ libm_reduce_pi04l(rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3519
3520 return start;
3521
3522 }
3523
3524 address generate_libm_sin_cos_huge() {
3525 StubCodeMark mark(this, "StubRoutines", "libm_sin_cos_huge");
3526
3527 address start = __ pc();
3528
3529 const XMMRegister x0 = xmm0;
3530 const XMMRegister x1 = xmm1;
3531
3532 BLOCK_COMMENT("Entry:");
3533 __ libm_sincos_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3534
3535 return start;
3536
3537 }
3538
3539 address generate_libmSin() {
3540 StubCodeMark mark(this, "StubRoutines", "libmSin");
3541
3542 address start = __ pc();
3543
3544 const XMMRegister x0 = xmm0;
3545 const XMMRegister x1 = xmm1;
3546 const XMMRegister x2 = xmm2;
3547 const XMMRegister x3 = xmm3;
3548
3549 const XMMRegister x4 = xmm4;
3550 const XMMRegister x5 = xmm5;
3551 const XMMRegister x6 = xmm6;
3552 const XMMRegister x7 = xmm7;
3553
3554 BLOCK_COMMENT("Entry:");
3555 __ enter(); // required for proper stackwalking of RuntimeStub frame
3556 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rdx);
3557 __ leave(); // required for proper stackwalking of RuntimeStub frame
3558 __ ret(0);
3559
3560 return start;
3561
3562 }
3563
3564 address generate_libmCos() {
3565 StubCodeMark mark(this, "StubRoutines", "libmCos");
3566
3567 address start = __ pc();
3568
3569 const XMMRegister x0 = xmm0;
3570 const XMMRegister x1 = xmm1;
3571 const XMMRegister x2 = xmm2;
3572 const XMMRegister x3 = xmm3;
3573
3574 const XMMRegister x4 = xmm4;
3575 const XMMRegister x5 = xmm5;
3576 const XMMRegister x6 = xmm6;
3577 const XMMRegister x7 = xmm7;
3578
3579 const Register tmp = rbx;
3580
3581 BLOCK_COMMENT("Entry:");
3582 __ enter(); // required for proper stackwalking of RuntimeStub frame
3583 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3584 __ leave(); // required for proper stackwalking of RuntimeStub frame
3585 __ ret(0);
3586
3587 return start;
3588
3589 }
3590
3591 address generate_libm_tan_cot_huge() {
3592 StubCodeMark mark(this, "StubRoutines", "libm_tan_cot_huge");
3593
3594 address start = __ pc();
3595
3596 const XMMRegister x0 = xmm0;
3597 const XMMRegister x1 = xmm1;
3598
3599 BLOCK_COMMENT("Entry:");
3600 __ libm_tancot_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3601
3602 return start;
3603
3604 }
3605
3606 address generate_libmTan() {
3607 StubCodeMark mark(this, "StubRoutines", "libmTan");
3608
3609 address start = __ pc();
3610
3611 const XMMRegister x0 = xmm0;
3612 const XMMRegister x1 = xmm1;
3613 const XMMRegister x2 = xmm2;
3614 const XMMRegister x3 = xmm3;
3615
3616 const XMMRegister x4 = xmm4;
3617 const XMMRegister x5 = xmm5;
3618 const XMMRegister x6 = xmm6;
3619 const XMMRegister x7 = xmm7;
3620
3621 const Register tmp = rbx;
3622
3623 BLOCK_COMMENT("Entry:");
3624 __ enter(); // required for proper stackwalking of RuntimeStub frame
3625 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3626 __ leave(); // required for proper stackwalking of RuntimeStub frame
3627 __ ret(0);
3628
3629 return start;
3630
3631 }
3632
3633 // Safefetch stubs.
3634 void generate_safefetch(const char* name, int size, address* entry,
3635 address* fault_pc, address* continuation_pc) {
3636 // safefetch signatures:
3637 // int SafeFetch32(int* adr, int errValue);
3638 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3639
3640 StubCodeMark mark(this, "StubRoutines", name);
3641
3642 // Entry point, pc or function descriptor.
3643 *entry = __ pc();
3644
3645 __ movl(rax, Address(rsp, 0x8));
3646 __ movl(rcx, Address(rsp, 0x4));
3647 // Load *adr into eax, may fault.
3648 *fault_pc = __ pc();
3649 switch (size) {
3650 case 4:
3651 // int32_t
3652 __ movl(rax, Address(rcx, 0));
3653 break;
3654 case 8:
3655 // int64_t
3656 Unimplemented();
3657 break;
3658 default:
3659 ShouldNotReachHere();
3660 }
3661
3662 // Return errValue or *adr.
3663 *continuation_pc = __ pc();
3664 __ ret(0);
3665 }
3666
3667 address generate_method_entry_barrier() {
3668 __ align(CodeEntryAlignment);
3669 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
3670
3671 Label deoptimize_label;
3672
3673 address start = __ pc();
3674
3675 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
3676
3677 BLOCK_COMMENT("Entry:");
3678 __ enter(); // save rbp
3679
3680 // save rbx, because we want to use that value.
3681 // We could do without it but then we depend on the number of slots used by pusha
3682 __ push(rbx);
3683
3684 __ lea(rbx, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for rbx - this should be the return address
3685
3686 __ pusha();
3687
3688 // xmm0 and xmm1 may be used for passing float/double arguments
3689 const int xmm_size = wordSize * 2;
3690 const int xmm_spill_size = xmm_size * 2;
3691 __ subptr(rsp, xmm_spill_size);
3692 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
3693 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
3694
3695 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), rbx);
3696
3697 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
3698 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
3699 __ addptr(rsp, xmm_spill_size);
3700
3701 __ cmpl(rax, 1); // 1 means deoptimize
3702 __ jcc(Assembler::equal, deoptimize_label);
3703
3704 __ popa();
3705 __ pop(rbx);
3706
3707 __ leave();
3708
3709 __ addptr(rsp, 1 * wordSize); // cookie
3710 __ ret(0);
3711
3712 __ BIND(deoptimize_label);
3713
3714 __ popa();
3715 __ pop(rbx);
3716
3717 __ leave();
3718
3719 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
3720 // here while still having a correct stack is valuable
3721 __ testptr(rsp, Address(rsp, 0));
3722
3723 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
3724 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
3725
3726 return start;
3727 }
3728
3729 public:
3730 // Information about frame layout at time of blocking runtime call.
3731 // Note that we only have to preserve callee-saved registers since
3732 // the compilers are responsible for supplying a continuation point
3733 // if they expect all registers to be preserved.
3734 enum layout {
3735 thread_off, // last_java_sp
3736 arg1_off,
3737 arg2_off,
3738 rbp_off, // callee saved register
3739 ret_pc,
3740 framesize
3741 };
3742
3743 private:
3744
3745 #undef __
3746 #define __ masm->
3747
3748 //------------------------------------------------------------------------------------------------------------------------
3749 // Continuation point for throwing of implicit exceptions that are not handled in
3750 // the current activation. Fabricates an exception oop and initiates normal
3751 // exception dispatching in this frame.
3752 //
3753 // Previously the compiler (c2) allowed for callee save registers on Java calls.
3754 // This is no longer true after adapter frames were removed but could possibly
3755 // be brought back in the future if the interpreter code was reworked and it
3756 // was deemed worthwhile. The comment below was left to describe what must
3757 // happen here if callee saves were resurrected. As it stands now this stub
3758 // could actually be a vanilla BufferBlob and have now oopMap at all.
3759 // Since it doesn't make much difference we've chosen to leave it the
3760 // way it was in the callee save days and keep the comment.
3761
3762 // If we need to preserve callee-saved values we need a callee-saved oop map and
3763 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
3764 // If the compiler needs all registers to be preserved between the fault
3765 // point and the exception handler then it must assume responsibility for that in
3766 // AbstractCompiler::continuation_for_implicit_null_exception or
3767 // continuation_for_implicit_division_by_zero_exception. All other implicit
3768 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
3769 // either at call sites or otherwise assume that stack unwinding will be initiated,
3770 // so caller saved registers were assumed volatile in the compiler.
3771 address generate_throw_exception(const char* name, address runtime_entry,
3772 Register arg1 = noreg, Register arg2 = noreg) {
3773
3774 int insts_size = 256;
3775 int locs_size = 32;
3776
3777 CodeBuffer code(name, insts_size, locs_size);
3778 OopMapSet* oop_maps = new OopMapSet();
3779 MacroAssembler* masm = new MacroAssembler(&code);
3780
3781 address start = __ pc();
3782
3783 // This is an inlined and slightly modified version of call_VM
3784 // which has the ability to fetch the return PC out of
3785 // thread-local storage and also sets up last_Java_sp slightly
3786 // differently than the real call_VM
3787 Register java_thread = rbx;
3788 __ get_thread(java_thread);
3789
3790 __ enter(); // required for proper stackwalking of RuntimeStub frame
3791
3792 // pc and rbp, already pushed
3793 __ subptr(rsp, (framesize-2) * wordSize); // prolog
3794
3795 // Frame is now completed as far as size and linkage.
3796
3797 int frame_complete = __ pc() - start;
3798
3799 // push java thread (becomes first argument of C function)
3800 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
3801 if (arg1 != noreg) {
3802 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
3803 }
3804 if (arg2 != noreg) {
3805 assert(arg1 != noreg, "missing reg arg");
3806 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
3807 }
3808
3809 // Set up last_Java_sp and last_Java_fp
3810 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
3811
3812 // Call runtime
3813 BLOCK_COMMENT("call runtime_entry");
3814 __ call(RuntimeAddress(runtime_entry));
3815 // Generate oop map
3816 OopMap* map = new OopMap(framesize, 0);
3817 oop_maps->add_gc_map(__ pc() - start, map);
3818
3819 // restore the thread (cannot use the pushed argument since arguments
3820 // may be overwritten by C code generated by an optimizing compiler);
3821 // however can use the register value directly if it is callee saved.
3822 __ get_thread(java_thread);
3823
3824 __ reset_last_Java_frame(java_thread, true);
3825
3826 __ leave(); // required for proper stackwalking of RuntimeStub frame
3827
3828 // check for pending exceptions
3829 #ifdef ASSERT
3830 Label L;
3831 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3832 __ jcc(Assembler::notEqual, L);
3833 __ should_not_reach_here();
3834 __ bind(L);
3835 #endif /* ASSERT */
3836 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3837
3838
3839 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
3840 return stub->entry_point();
3841 }
3842
3843
3844 void create_control_words() {
3845 // Round to nearest, 53-bit mode, exceptions masked
3846 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
3847 // Round to zero, 53-bit mode, exception mased
3848 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
3849 // Round to nearest, 24-bit mode, exceptions masked
3850 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
3851 // Round to nearest, 64-bit mode, exceptions masked
3852 StubRoutines::_mxcsr_std = 0x1F80;
3853 // Note: the following two constants are 80-bit values
3854 // layout is critical for correct loading by FPU.
3855 // Bias for strict fp multiply/divide
3856 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
3857 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
3858 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
3859 // Un-Bias for strict fp multiply/divide
3860 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
3861 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
3862 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
3863 }
3864
3865 //---------------------------------------------------------------------------
3866 // Initialization
3867
3868 void generate_initial() {
3869 // Generates all stubs and initializes the entry points
3870
3871 //------------------------------------------------------------------------------------------------------------------------
3872 // entry points that exist in all platforms
3873 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
3874 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
3875 StubRoutines::_forward_exception_entry = generate_forward_exception();
3876
3877 StubRoutines::_call_stub_entry =
3878 generate_call_stub(StubRoutines::_call_stub_return_address);
3879 // is referenced by megamorphic call
3880 StubRoutines::_catch_exception_entry = generate_catch_exception();
3881
3882 // These are currently used by Solaris/Intel
3883 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3884
3885 // platform dependent
3886 create_control_words();
3887
3888 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3889 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
3890 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
3891 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
3892 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
3893 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
3894
3895 // Build this early so it's available for the interpreter
3896 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception",
3897 CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
3898 StubRoutines::_throw_delayed_StackOverflowError_entry = generate_throw_exception("delayed StackOverflowError throw_exception",
3899 CAST_FROM_FN_PTR(address, SharedRuntime::throw_delayed_StackOverflowError));
3900
3901 if (UseCRC32Intrinsics) {
3902 // set table address before stub generation which use it
3903 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3904 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3905 }
3906
3907 if (UseCRC32CIntrinsics) {
3908 bool supports_clmul = VM_Version::supports_clmul();
3909 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
3910 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
3911 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
3912 }
3913 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
3914 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3915 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3916 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3917 StubRoutines::x86::_L_2il0floatpacket_0_adr = (address)StubRoutines::x86::_L_2il0floatpacket_0;
3918 StubRoutines::x86::_Pi4Inv_adr = (address)StubRoutines::x86::_Pi4Inv;
3919 StubRoutines::x86::_Pi4x3_adr = (address)StubRoutines::x86::_Pi4x3;
3920 StubRoutines::x86::_Pi4x4_adr = (address)StubRoutines::x86::_Pi4x4;
3921 StubRoutines::x86::_ones_adr = (address)StubRoutines::x86::_ones;
3922 }
3923 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3924 StubRoutines::_dexp = generate_libmExp();
3925 }
3926 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3927 StubRoutines::_dlog = generate_libmLog();
3928 }
3929 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3930 StubRoutines::_dlog10 = generate_libmLog10();
3931 }
3932 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3933 StubRoutines::_dpow = generate_libmPow();
3934 }
3935 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3936 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3937 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3938 StubRoutines::_dlibm_reduce_pi04l = generate_libm_reduce_pi04l();
3939 }
3940 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3941 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3942 StubRoutines::_dlibm_sin_cos_huge = generate_libm_sin_cos_huge();
3943 }
3944 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3945 StubRoutines::_dsin = generate_libmSin();
3946 }
3947 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3948 StubRoutines::_dcos = generate_libmCos();
3949 }
3950 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3951 StubRoutines::_dlibm_tan_cot_huge = generate_libm_tan_cot_huge();
3952 StubRoutines::_dtan = generate_libmTan();
3953 }
3954 }
3955 }
3956
3957 void generate_all() {
3958 // Generates all stubs and initializes the entry points
3959
3960 // These entry points require SharedInfo::stack0 to be set up in non-core builds
3961 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
3962 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
3963 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
3964 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
3965
3966 //------------------------------------------------------------------------------------------------------------------------
3967 // entry points that are platform specific
3968
3969 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF);
3970 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x80000000);
3971 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask_long_double("vector_double_sign_mask", 0x7FFFFFFF, 0xFFFFFFFF);
3972 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask_long_double("vector_double_sign_flip", 0x80000000, 0x00000000);
3973 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff);
3974 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
3975 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask_long_double("vector_long_sign_mask", 0x80000000, 0x00000000);
3976
3977 // support for verify_oop (must happen after universe_init)
3978 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3979
3980 // arraycopy stubs used by compilers
3981 generate_arraycopy_stubs();
3982
3983 // don't bother generating these AES intrinsic stubs unless global flag is set
3984 if (UseAESIntrinsics) {
3985 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
3986
3987 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3988 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3989 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3990 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
3991 }
3992
3993 if (UseAESCTRIntrinsics) {
3994 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
3995 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
3996 }
3997
3998 if (UseSHA1Intrinsics) {
3999 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
4000 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
4001 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
4002 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
4003 }
4004 if (UseSHA256Intrinsics) {
4005 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
4006 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
4007 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
4008 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
4009 }
4010
4011 // Generate GHASH intrinsics code
4012 if (UseGHASHIntrinsics) {
4013 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
4014 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
4015 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
4016 }
4017
4018 // Safefetch stubs.
4019 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
4020 &StubRoutines::_safefetch32_fault_pc,
4021 &StubRoutines::_safefetch32_continuation_pc);
4022 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
4023 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
4024 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
4025
4026 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
4027 if (bs_nm != NULL) {
4028 StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
4029 }
4030 }
4031
4032
4033 public:
4034 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
4035 if (all) {
4036 generate_all();
4037 } else {
4038 generate_initial();
4039 }
4040 }
4041 }; // end class declaration
4042
4043 #define UCM_TABLE_MAX_ENTRIES 8
4044 void StubGenerator_generate(CodeBuffer* code, bool all) {
4045 if (UnsafeCopyMemory::_table == NULL) {
4046 UnsafeCopyMemory::create_table(UCM_TABLE_MAX_ENTRIES);
4047 }
4048 StubGenerator g(code, all);
4049 }