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