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 "memory/universe.hpp"
32 #include "nativeInst_x86.hpp"
33 #include "oops/instanceOop.hpp"
34 #include "oops/method.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/methodHandles.hpp"
38 #include "runtime/frame.inline.hpp"
39 #include "runtime/handles.inline.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/stubCodeGenerator.hpp"
42 #include "runtime/stubRoutines.hpp"
43 #include "runtime/thread.inline.hpp"
44 #ifdef COMPILER2
45 #include "opto/runtime.hpp"
46 #endif
47
48 // Declaration and definition of StubGenerator (no .hpp file).
49 // For a more detailed description of the stub routine structure
50 // see the comment in stubRoutines.hpp
51
52 #define __ _masm->
53 #define a__ ((Assembler*)_masm)->
54
55 #ifdef PRODUCT
56 #define BLOCK_COMMENT(str) /* nothing */
57 #else
58 #define BLOCK_COMMENT(str) __ block_comment(str)
59 #endif
60
61 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
62
63 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
64 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
65
66 // -------------------------------------------------------------------------------------------------------------------------
67 // Stub Code definitions
68
69 class StubGenerator: public StubCodeGenerator {
70 private:
71
72 #ifdef PRODUCT
73 #define inc_counter_np(counter) ((void)0)
74 #else
75 void inc_counter_np_(int& counter) {
76 __ incrementl(ExternalAddress((address)&counter));
77 }
78 #define inc_counter_np(counter) \
79 BLOCK_COMMENT("inc_counter " #counter); \
80 inc_counter_np_(counter);
81 #endif //PRODUCT
82
83 void inc_copy_counter_np(BasicType t) {
84 #ifndef PRODUCT
85 switch (t) {
86 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
87 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
88 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
89 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
90 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
91 default: ShouldNotReachHere();
92 }
93 #endif //PRODUCT
94 }
95
96 //------------------------------------------------------------------------------------------------------------------------
97 // Call stubs are used to call Java from C
98 //
99 // [ return_from_Java ] <--- rsp
100 // [ argument word n ]
101 // ...
102 // -N [ argument word 1 ]
103 // -7 [ Possible padding for stack alignment ]
104 // -6 [ Possible padding for stack alignment ]
105 // -5 [ Possible padding for stack alignment ]
106 // -4 [ mxcsr save ] <--- rsp_after_call
107 // -3 [ saved rbx, ]
108 // -2 [ saved rsi ]
109 // -1 [ saved rdi ]
110 // 0 [ saved rbp, ] <--- rbp,
111 // 1 [ return address ]
112 // 2 [ ptr. to call wrapper ]
113 // 3 [ result ]
114 // 4 [ result_type ]
115 // 5 [ method ]
116 // 6 [ entry_point ]
117 // 7 [ parameters ]
118 // 8 [ parameter_size ]
119 // 9 [ thread ]
120
121
122 address generate_call_stub(address& return_address) {
123 StubCodeMark mark(this, "StubRoutines", "call_stub");
124 address start = __ pc();
125
126 // stub code parameters / addresses
127 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
128 bool sse_save = false;
129 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
130 const int locals_count_in_bytes (4*wordSize);
131 const Address mxcsr_save (rbp, -4 * wordSize);
132 const Address saved_rbx (rbp, -3 * wordSize);
133 const Address saved_rsi (rbp, -2 * wordSize);
134 const Address saved_rdi (rbp, -1 * wordSize);
135 const Address result (rbp, 3 * wordSize);
136 const Address result_type (rbp, 4 * wordSize);
137 const Address method (rbp, 5 * wordSize);
138 const Address entry_point (rbp, 6 * wordSize);
139 const Address parameters (rbp, 7 * wordSize);
140 const Address parameter_size(rbp, 8 * wordSize);
141 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
142 sse_save = UseSSE > 0;
143
144 // stub code
145 __ enter();
146 __ movptr(rcx, parameter_size); // parameter counter
147 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
148 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
149 __ subptr(rsp, rcx);
150 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
151
152 // save rdi, rsi, & rbx, according to C calling conventions
153 __ movptr(saved_rdi, rdi);
154 __ movptr(saved_rsi, rsi);
155 __ movptr(saved_rbx, rbx);
156
157 // save and initialize %mxcsr
158 if (sse_save) {
159 Label skip_ldmx;
160 __ stmxcsr(mxcsr_save);
161 __ movl(rax, mxcsr_save);
162 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
163 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
164 __ cmp32(rax, mxcsr_std);
165 __ jcc(Assembler::equal, skip_ldmx);
166 __ ldmxcsr(mxcsr_std);
167 __ bind(skip_ldmx);
168 }
169
170 // make sure the control word is correct.
171 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
172
173 #ifdef ASSERT
174 // make sure we have no pending exceptions
175 { Label L;
176 __ movptr(rcx, thread);
177 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
178 __ jcc(Assembler::equal, L);
179 __ stop("StubRoutines::call_stub: entered with pending exception");
180 __ bind(L);
181 }
182 #endif
183
184 // pass parameters if any
185 BLOCK_COMMENT("pass parameters if any");
186 Label parameters_done;
187 __ movl(rcx, parameter_size); // parameter counter
188 __ testl(rcx, rcx);
189 __ jcc(Assembler::zero, parameters_done);
190
191 // parameter passing loop
192
193 Label loop;
194 // Copy Java parameters in reverse order (receiver last)
195 // Note that the argument order is inverted in the process
196 // source is rdx[rcx: N-1..0]
197 // dest is rsp[rbx: 0..N-1]
198
199 __ movptr(rdx, parameters); // parameter pointer
200 __ xorptr(rbx, rbx);
201
202 __ BIND(loop);
203
204 // get parameter
205 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
206 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
207 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
208 __ increment(rbx);
209 __ decrement(rcx);
210 __ jcc(Assembler::notZero, loop);
211
212 // call Java function
213 __ BIND(parameters_done);
214 __ movptr(rbx, method); // get Method*
215 __ movptr(rax, entry_point); // get entry_point
216 __ mov(rsi, rsp); // set sender sp
217 BLOCK_COMMENT("call Java function");
218 __ call(rax);
219
220 BLOCK_COMMENT("call_stub_return_address:");
221 return_address = __ pc();
222
223 #ifdef COMPILER2
224 {
225 Label L_skip;
226 if (UseSSE >= 2) {
227 __ verify_FPU(0, "call_stub_return");
228 } else {
229 for (int i = 1; i < 8; i++) {
230 __ ffree(i);
231 }
232
233 // UseSSE <= 1 so double result should be left on TOS
234 __ movl(rsi, result_type);
235 __ cmpl(rsi, T_DOUBLE);
236 __ jcc(Assembler::equal, L_skip);
237 if (UseSSE == 0) {
238 // UseSSE == 0 so float result should be left on TOS
239 __ cmpl(rsi, T_FLOAT);
240 __ jcc(Assembler::equal, L_skip);
241 }
242 __ ffree(0);
243 }
244 __ BIND(L_skip);
245 }
246 #endif // COMPILER2
247
248 // store result depending on type
249 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
250 __ movptr(rdi, result);
251 Label is_long, is_float, is_double, exit;
252 __ movl(rsi, result_type);
253 __ cmpl(rsi, T_LONG);
254 __ jcc(Assembler::equal, is_long);
255 __ cmpl(rsi, T_FLOAT);
256 __ jcc(Assembler::equal, is_float);
257 __ cmpl(rsi, T_DOUBLE);
258 __ jcc(Assembler::equal, is_double);
259
260 // handle T_INT case
261 __ movl(Address(rdi, 0), rax);
262 __ BIND(exit);
263
264 // check that FPU stack is empty
265 __ verify_FPU(0, "generate_call_stub");
266
267 // pop parameters
268 __ lea(rsp, rsp_after_call);
269
270 // restore %mxcsr
271 if (sse_save) {
272 __ ldmxcsr(mxcsr_save);
273 }
274
275 // restore rdi, rsi and rbx,
276 __ movptr(rbx, saved_rbx);
277 __ movptr(rsi, saved_rsi);
278 __ movptr(rdi, saved_rdi);
279 __ addptr(rsp, 4*wordSize);
280
281 // return
282 __ pop(rbp);
283 __ ret(0);
284
285 // handle return types different from T_INT
286 __ BIND(is_long);
287 __ movl(Address(rdi, 0 * wordSize), rax);
288 __ movl(Address(rdi, 1 * wordSize), rdx);
289 __ jmp(exit);
290
291 __ BIND(is_float);
292 // interpreter uses xmm0 for return values
293 if (UseSSE >= 1) {
294 __ movflt(Address(rdi, 0), xmm0);
295 } else {
296 __ fstp_s(Address(rdi, 0));
297 }
298 __ jmp(exit);
299
300 __ BIND(is_double);
301 // interpreter uses xmm0 for return values
302 if (UseSSE >= 2) {
303 __ movdbl(Address(rdi, 0), xmm0);
304 } else {
305 __ fstp_d(Address(rdi, 0));
306 }
307 __ jmp(exit);
308
309 return start;
310 }
311
312
313 //------------------------------------------------------------------------------------------------------------------------
314 // Return point for a Java call if there's an exception thrown in Java code.
315 // The exception is caught and transformed into a pending exception stored in
316 // JavaThread that can be tested from within the VM.
317 //
318 // Note: Usually the parameters are removed by the callee. In case of an exception
319 // crossing an activation frame boundary, that is not the case if the callee
320 // is compiled code => need to setup the rsp.
321 //
322 // rax,: exception oop
323
324 address generate_catch_exception() {
325 StubCodeMark mark(this, "StubRoutines", "catch_exception");
326 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
327 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
328 address start = __ pc();
329
330 // get thread directly
331 __ movptr(rcx, thread);
332 #ifdef ASSERT
333 // verify that threads correspond
334 { Label L;
335 __ get_thread(rbx);
336 __ cmpptr(rbx, rcx);
337 __ jcc(Assembler::equal, L);
338 __ stop("StubRoutines::catch_exception: threads must correspond");
339 __ bind(L);
340 }
341 #endif
342 // set pending exception
343 __ verify_oop(rax);
344 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
345 __ lea(Address(rcx, Thread::exception_file_offset ()),
346 ExternalAddress((address)__FILE__));
347 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
348 // complete return to VM
349 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
350 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
351
352 return start;
353 }
354
355
356 //------------------------------------------------------------------------------------------------------------------------
357 // Continuation point for runtime calls returning with a pending exception.
358 // The pending exception check happened in the runtime or native call stub.
359 // The pending exception in Thread is converted into a Java-level exception.
360 //
361 // Contract with Java-level exception handlers:
362 // rax: exception
363 // rdx: throwing pc
364 //
365 // NOTE: At entry of this stub, exception-pc must be on stack !!
366
367 address generate_forward_exception() {
368 StubCodeMark mark(this, "StubRoutines", "forward exception");
369 address start = __ pc();
370 const Register thread = rcx;
371
372 // other registers used in this stub
373 const Register exception_oop = rax;
374 const Register handler_addr = rbx;
375 const Register exception_pc = rdx;
376
377 // Upon entry, the sp points to the return address returning into Java
378 // (interpreted or compiled) code; i.e., the return address becomes the
379 // throwing pc.
380 //
381 // Arguments pushed before the runtime call are still on the stack but
382 // the exception handler will reset the stack pointer -> ignore them.
383 // A potential result in registers can be ignored as well.
384
385 #ifdef ASSERT
386 // make sure this code is only executed if there is a pending exception
387 { Label L;
388 __ get_thread(thread);
389 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
390 __ jcc(Assembler::notEqual, L);
391 __ stop("StubRoutines::forward exception: no pending exception (1)");
392 __ bind(L);
393 }
394 #endif
395
396 // compute exception handler into rbx,
397 __ get_thread(thread);
398 __ movptr(exception_pc, Address(rsp, 0));
399 BLOCK_COMMENT("call exception_handler_for_return_address");
400 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
401 __ mov(handler_addr, rax);
402
403 // setup rax & rdx, remove return address & clear pending exception
404 __ get_thread(thread);
405 __ pop(exception_pc);
406 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
407 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
408
409 #ifdef ASSERT
410 // make sure exception is set
411 { Label L;
412 __ testptr(exception_oop, exception_oop);
413 __ jcc(Assembler::notEqual, L);
414 __ stop("StubRoutines::forward exception: no pending exception (2)");
415 __ bind(L);
416 }
417 #endif
418
419 // Verify that there is really a valid exception in RAX.
420 __ verify_oop(exception_oop);
421
422 // continue at exception handler (return address removed)
423 // rax: exception
424 // rbx: exception handler
425 // rdx: throwing pc
426 __ jmp(handler_addr);
427
428 return start;
429 }
430
431
432 //----------------------------------------------------------------------------------------------------
433 // Support for int32_t Atomic::xchg(volatile int32_t* dest, int32_t exchange_value)
434 //
435 // xchg exists as far back as 8086, lock needed for MP only
436 // Stack layout immediately after call:
437 //
438 // 0 [ret addr ] <--- rsp
439 // 1 [ ex ]
440 // 2 [ dest ]
441 //
442 // Result: *dest <- ex, return (old *dest)
443 //
444 // Note: win32 does not currently use this code
445
446 address generate_atomic_xchg() {
447 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
448 address start = __ pc();
449
450 __ push(rdx);
451 Address exchange(rsp, 2 * wordSize);
452 Address dest_addr(rsp, 3 * wordSize);
453 __ movl(rax, exchange);
454 __ movptr(rdx, dest_addr);
455 __ xchgl(rax, Address(rdx, 0));
456 __ pop(rdx);
457 __ ret(0);
458
459 return start;
460 }
461
462 //----------------------------------------------------------------------------------------------------
463 // Support for void verify_mxcsr()
464 //
465 // This routine is used with -Xcheck:jni to verify that native
466 // JNI code does not return to Java code without restoring the
467 // MXCSR register to our expected state.
468
469
470 address generate_verify_mxcsr() {
471 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
472 address start = __ pc();
473
474 const Address mxcsr_save(rsp, 0);
475
476 if (CheckJNICalls && UseSSE > 0 ) {
477 Label ok_ret;
478 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
479 __ push(rax);
480 __ subptr(rsp, wordSize); // allocate a temp location
481 __ stmxcsr(mxcsr_save);
482 __ movl(rax, mxcsr_save);
483 __ andl(rax, MXCSR_MASK);
484 __ cmp32(rax, mxcsr_std);
485 __ jcc(Assembler::equal, ok_ret);
486
487 __ warn("MXCSR changed by native JNI code.");
488
489 __ ldmxcsr(mxcsr_std);
490
491 __ bind(ok_ret);
492 __ addptr(rsp, wordSize);
493 __ pop(rax);
494 }
495
496 __ ret(0);
497
498 return start;
499 }
500
501
502 //---------------------------------------------------------------------------
503 // Support for void verify_fpu_cntrl_wrd()
504 //
505 // This routine is used with -Xcheck:jni to verify that native
506 // JNI code does not return to Java code without restoring the
507 // FP control word to our expected state.
508
509 address generate_verify_fpu_cntrl_wrd() {
510 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
511 address start = __ pc();
512
513 const Address fpu_cntrl_wrd_save(rsp, 0);
514
515 if (CheckJNICalls) {
516 Label ok_ret;
517 __ push(rax);
518 __ subptr(rsp, wordSize); // allocate a temp location
519 __ fnstcw(fpu_cntrl_wrd_save);
520 __ movl(rax, fpu_cntrl_wrd_save);
521 __ andl(rax, FPU_CNTRL_WRD_MASK);
522 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
523 __ cmp32(rax, fpu_std);
524 __ jcc(Assembler::equal, ok_ret);
525
526 __ warn("Floating point control word changed by native JNI code.");
527
528 __ fldcw(fpu_std);
529
530 __ bind(ok_ret);
531 __ addptr(rsp, wordSize);
532 __ pop(rax);
533 }
534
535 __ ret(0);
536
537 return start;
538 }
539
540 //---------------------------------------------------------------------------
541 // Wrapper for slow-case handling of double-to-integer conversion
542 // d2i or f2i fast case failed either because it is nan or because
543 // of under/overflow.
544 // Input: FPU TOS: float value
545 // Output: rax, (rdx): integer (long) result
546
547 address generate_d2i_wrapper(BasicType t, address fcn) {
548 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
549 address start = __ pc();
550
551 // Capture info about frame layout
552 enum layout { FPUState_off = 0,
553 rbp_off = FPUStateSizeInWords,
554 rdi_off,
555 rsi_off,
556 rcx_off,
557 rbx_off,
558 saved_argument_off,
559 saved_argument_off2, // 2nd half of double
560 framesize
561 };
562
563 assert(FPUStateSizeInWords == 27, "update stack layout");
564
565 // Save outgoing argument to stack across push_FPU_state()
566 __ subptr(rsp, wordSize * 2);
567 __ fstp_d(Address(rsp, 0));
568
569 // Save CPU & FPU state
570 __ push(rbx);
571 __ push(rcx);
572 __ push(rsi);
573 __ push(rdi);
574 __ push(rbp);
575 __ push_FPU_state();
576
577 // push_FPU_state() resets the FP top of stack
578 // Load original double into FP top of stack
579 __ fld_d(Address(rsp, saved_argument_off * wordSize));
580 // Store double into stack as outgoing argument
581 __ subptr(rsp, wordSize*2);
582 __ fst_d(Address(rsp, 0));
583
584 // Prepare FPU for doing math in C-land
585 __ empty_FPU_stack();
586 // Call the C code to massage the double. Result in EAX
587 if (t == T_INT)
588 { BLOCK_COMMENT("SharedRuntime::d2i"); }
589 else if (t == T_LONG)
590 { BLOCK_COMMENT("SharedRuntime::d2l"); }
591 __ call_VM_leaf( fcn, 2 );
592
593 // Restore CPU & FPU state
594 __ pop_FPU_state();
595 __ pop(rbp);
596 __ pop(rdi);
597 __ pop(rsi);
598 __ pop(rcx);
599 __ pop(rbx);
600 __ addptr(rsp, wordSize * 2);
601
602 __ ret(0);
603
604 return start;
605 }
606 //---------------------------------------------------------------------------------------------------
607
608 address generate_vector_mask(const char *stub_name, int32_t mask) {
609 __ align(CodeEntryAlignment);
610 StubCodeMark mark(this, "StubRoutines", stub_name);
611 address start = __ pc();
612
613 for (int i = 0; i < 16; i++) {
614 __ emit_data(mask, relocInfo::none, 0);
615 }
616
617 return start;
618 }
619
620 address generate_vector_mask_long_double(const char *stub_name, int32_t maskhi, int32_t masklo) {
621 __ align(CodeEntryAlignment);
622 StubCodeMark mark(this, "StubRoutines", stub_name);
623 address start = __ pc();
624
625 for (int i = 0; i < 8; i++) {
626 __ emit_data(masklo, relocInfo::none, 0);
627 __ emit_data(maskhi, relocInfo::none, 0);
628 }
629
630 return start;
631 }
632
633 //----------------------------------------------------------------------------------------------------
634
635 address generate_vector_byte_perm_mask(const char *stub_name) {
636 __ align(CodeEntryAlignment);
637 StubCodeMark mark(this, "StubRoutines", stub_name);
638 address start = __ pc();
639
640 __ emit_data(0x00000001, relocInfo::none, 0);
641 __ emit_data(0x00000000, relocInfo::none, 0);
642 __ emit_data(0x00000003, relocInfo::none, 0);
643 __ emit_data(0x00000000, relocInfo::none, 0);
644 __ emit_data(0x00000005, relocInfo::none, 0);
645 __ emit_data(0x00000000, relocInfo::none, 0);
646 __ emit_data(0x00000007, relocInfo::none, 0);
647 __ emit_data(0x00000000, relocInfo::none, 0);
648 __ emit_data(0x00000000, relocInfo::none, 0);
649 __ emit_data(0x00000000, relocInfo::none, 0);
650 __ emit_data(0x00000002, relocInfo::none, 0);
651 __ emit_data(0x00000000, relocInfo::none, 0);
652 __ emit_data(0x00000004, relocInfo::none, 0);
653 __ emit_data(0x00000000, relocInfo::none, 0);
654 __ emit_data(0x00000006, relocInfo::none, 0);
655 __ emit_data(0x00000000, relocInfo::none, 0);
656
657 return start;
658 }
659
660 //----------------------------------------------------------------------------------------------------
661 // Non-destructive plausibility checks for oops
662
663 address generate_verify_oop() {
664 StubCodeMark mark(this, "StubRoutines", "verify_oop");
665 address start = __ pc();
666
667 // Incoming arguments on stack after saving rax,:
668 //
669 // [tos ]: saved rdx
670 // [tos + 1]: saved EFLAGS
671 // [tos + 2]: return address
672 // [tos + 3]: char* error message
673 // [tos + 4]: oop object to verify
674 // [tos + 5]: saved rax, - saved by caller and bashed
675
676 Label exit, error;
677 __ pushf();
678 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
679 __ push(rdx); // save rdx
680 // make sure object is 'reasonable'
681 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
682 __ testptr(rax, rax);
683 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
684
685 // Check if the oop is in the right area of memory
686 const int oop_mask = Universe::verify_oop_mask();
687 const int oop_bits = Universe::verify_oop_bits();
688 __ mov(rdx, rax);
689 __ andptr(rdx, oop_mask);
690 __ cmpptr(rdx, oop_bits);
691 __ jcc(Assembler::notZero, error);
692
693 // make sure klass is 'reasonable', which is not zero.
694 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
695 __ testptr(rax, rax);
696 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
697
698 // return if everything seems ok
699 __ bind(exit);
700 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
701 __ pop(rdx); // restore rdx
702 __ popf(); // restore EFLAGS
703 __ ret(3 * wordSize); // pop arguments
704
705 // handle errors
706 __ bind(error);
707 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
708 __ pop(rdx); // get saved rdx back
709 __ popf(); // get saved EFLAGS off stack -- will be ignored
710 __ pusha(); // push registers (eip = return address & msg are already pushed)
711 BLOCK_COMMENT("call MacroAssembler::debug");
712 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
713 __ hlt();
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 bool add_entry = (t != T_OBJECT && (!aligned || t == T_INT));
893 // UnsafeCopyMemory page error: continue after ucm
894 UnsafeCopyMemoryMark ucmm(this, add_entry, true);
895 __ subptr(to, from); // to --> to_from
896 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
897 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
898 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
899 // align source address at 4 bytes address boundary
900 if (t == T_BYTE) {
901 // One byte misalignment happens only for byte arrays
902 __ testl(from, 1);
903 __ jccb(Assembler::zero, L_skip_align1);
904 __ movb(rax, Address(from, 0));
905 __ movb(Address(from, to_from, Address::times_1, 0), rax);
906 __ increment(from);
907 __ decrement(count);
908 __ BIND(L_skip_align1);
909 }
910 // Two bytes misalignment happens only for byte and short (char) arrays
911 __ testl(from, 2);
912 __ jccb(Assembler::zero, L_skip_align2);
913 __ movw(rax, Address(from, 0));
914 __ movw(Address(from, to_from, Address::times_1, 0), rax);
915 __ addptr(from, 2);
916 __ subl(count, 1<<(shift-1));
917 __ BIND(L_skip_align2);
918 }
919 if (!VM_Version::supports_mmx()) {
920 __ mov(rax, count); // save 'count'
921 __ shrl(count, shift); // bytes count
922 __ addptr(to_from, from);// restore 'to'
923 __ rep_mov();
924 __ subptr(to_from, from);// restore 'to_from'
925 __ mov(count, rax); // restore 'count'
926 __ jmpb(L_copy_2_bytes); // all dwords were copied
927 } else {
928 if (!UseUnalignedLoadStores) {
929 // align to 8 bytes, we know we are 4 byte aligned to start
930 __ testptr(from, 4);
931 __ jccb(Assembler::zero, L_copy_64_bytes);
932 __ movl(rax, Address(from, 0));
933 __ movl(Address(from, to_from, Address::times_1, 0), rax);
934 __ addptr(from, 4);
935 __ subl(count, 1<<shift);
936 }
937 __ BIND(L_copy_64_bytes);
938 __ mov(rax, count);
939 __ shrl(rax, shift+1); // 8 bytes chunk count
940 //
941 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
942 //
943 if (UseXMMForArrayCopy) {
944 xmm_copy_forward(from, to_from, rax);
945 } else {
946 mmx_copy_forward(from, to_from, rax);
947 }
948 }
949 // copy tailing dword
950 __ BIND(L_copy_4_bytes);
951 __ testl(count, 1<<shift);
952 __ jccb(Assembler::zero, L_copy_2_bytes);
953 __ movl(rax, Address(from, 0));
954 __ movl(Address(from, to_from, Address::times_1, 0), rax);
955 if (t == T_BYTE || t == T_SHORT) {
956 __ addptr(from, 4);
957 __ BIND(L_copy_2_bytes);
958 // copy tailing word
959 __ testl(count, 1<<(shift-1));
960 __ jccb(Assembler::zero, L_copy_byte);
961 __ movw(rax, Address(from, 0));
962 __ movw(Address(from, to_from, Address::times_1, 0), rax);
963 if (t == T_BYTE) {
964 __ addptr(from, 2);
965 __ BIND(L_copy_byte);
966 // copy tailing byte
967 __ testl(count, 1);
968 __ jccb(Assembler::zero, L_exit);
969 __ movb(rax, Address(from, 0));
970 __ movb(Address(from, to_from, Address::times_1, 0), rax);
971 __ BIND(L_exit);
972 } else {
973 __ BIND(L_copy_byte);
974 }
975 } else {
976 __ BIND(L_copy_2_bytes);
977 }
978 }
979
980 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
981 __ emms();
982 }
983 __ movl(count, Address(rsp, 12+12)); // reread 'count'
984 bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
985
986 if (t == T_OBJECT) {
987 __ BIND(L_0_count);
988 }
989 inc_copy_counter_np(t);
990 __ pop(rdi);
991 __ pop(rsi);
992 __ leave(); // required for proper stackwalking of RuntimeStub frame
993 __ vzeroupper();
994 __ xorptr(rax, rax); // return 0
995 __ ret(0);
996 return start;
997 }
998
999
1000 address generate_fill(BasicType t, bool aligned, const char *name) {
1001 __ align(CodeEntryAlignment);
1002 StubCodeMark mark(this, "StubRoutines", name);
1003 address start = __ pc();
1004
1005 BLOCK_COMMENT("Entry:");
1006
1007 const Register to = rdi; // source array address
1008 const Register value = rdx; // value
1009 const Register count = rsi; // elements count
1010
1011 __ enter(); // required for proper stackwalking of RuntimeStub frame
1012 __ push(rsi);
1013 __ push(rdi);
1014 __ movptr(to , Address(rsp, 12+ 4));
1015 __ movl(value, Address(rsp, 12+ 8));
1016 __ movl(count, Address(rsp, 12+ 12));
1017
1018 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1019
1020 __ pop(rdi);
1021 __ pop(rsi);
1022 __ leave(); // required for proper stackwalking of RuntimeStub frame
1023 __ ret(0);
1024 return start;
1025 }
1026
1027 address generate_conjoint_copy(BasicType t, bool aligned,
1028 Address::ScaleFactor sf,
1029 address nooverlap_target,
1030 address* entry, const char *name,
1031 bool dest_uninitialized = false) {
1032 __ align(CodeEntryAlignment);
1033 StubCodeMark mark(this, "StubRoutines", name);
1034 address start = __ pc();
1035
1036 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1037 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1038
1039 int shift = Address::times_ptr - sf;
1040
1041 const Register src = rax; // source array address
1042 const Register dst = rdx; // destination array address
1043 const Register from = rsi; // source array address
1044 const Register to = rdi; // destination array address
1045 const Register count = rcx; // elements count
1046 const Register end = rax; // array end address
1047
1048 __ enter(); // required for proper stackwalking of RuntimeStub frame
1049 __ push(rsi);
1050 __ push(rdi);
1051 __ movptr(src , Address(rsp, 12+ 4)); // from
1052 __ movptr(dst , Address(rsp, 12+ 8)); // to
1053 __ movl2ptr(count, Address(rsp, 12+12)); // count
1054
1055 if (entry != NULL) {
1056 *entry = __ pc(); // Entry point from generic arraycopy stub.
1057 BLOCK_COMMENT("Entry:");
1058 }
1059
1060 // nooverlap_target expects arguments in rsi and rdi.
1061 __ mov(from, src);
1062 __ mov(to , dst);
1063
1064 // arrays overlap test: dispatch to disjoint stub if necessary.
1065 RuntimeAddress nooverlap(nooverlap_target);
1066 __ cmpptr(dst, src);
1067 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1068 __ jump_cc(Assembler::belowEqual, nooverlap);
1069 __ cmpptr(dst, end);
1070 __ jump_cc(Assembler::aboveEqual, nooverlap);
1071
1072 if (t == T_OBJECT) {
1073 __ testl(count, count);
1074 __ jcc(Assembler::zero, L_0_count);
1075 }
1076
1077 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
1078 if (dest_uninitialized) {
1079 decorators |= IS_DEST_UNINITIALIZED;
1080 }
1081 if (aligned) {
1082 decorators |= ARRAYCOPY_ALIGNED;
1083 }
1084
1085 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1086 bs->arraycopy_prologue(_masm, decorators, t, from, to, count);
1087
1088 {
1089 bool add_entry = (t != T_OBJECT && (!aligned || t == T_INT));
1090 // UnsafeCopyMemory page error: continue after ucm
1091 UnsafeCopyMemoryMark ucmm(this, add_entry, true);
1092 // copy from high to low
1093 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1094 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1095 if (t == T_BYTE || t == T_SHORT) {
1096 // Align the end of destination array at 4 bytes address boundary
1097 __ lea(end, Address(dst, count, sf, 0));
1098 if (t == T_BYTE) {
1099 // One byte misalignment happens only for byte arrays
1100 __ testl(end, 1);
1101 __ jccb(Assembler::zero, L_skip_align1);
1102 __ decrement(count);
1103 __ movb(rdx, Address(from, count, sf, 0));
1104 __ movb(Address(to, count, sf, 0), rdx);
1105 __ BIND(L_skip_align1);
1106 }
1107 // Two bytes misalignment happens only for byte and short (char) arrays
1108 __ testl(end, 2);
1109 __ jccb(Assembler::zero, L_skip_align2);
1110 __ subptr(count, 1<<(shift-1));
1111 __ movw(rdx, Address(from, count, sf, 0));
1112 __ movw(Address(to, count, sf, 0), rdx);
1113 __ BIND(L_skip_align2);
1114 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1115 __ jcc(Assembler::below, L_copy_4_bytes);
1116 }
1117
1118 if (!VM_Version::supports_mmx()) {
1119 __ std();
1120 __ mov(rax, count); // Save 'count'
1121 __ mov(rdx, to); // Save 'to'
1122 __ lea(rsi, Address(from, count, sf, -4));
1123 __ lea(rdi, Address(to , count, sf, -4));
1124 __ shrptr(count, shift); // bytes count
1125 __ rep_mov();
1126 __ cld();
1127 __ mov(count, rax); // restore 'count'
1128 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1129 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1130 __ mov(to, rdx); // restore 'to'
1131 __ jmpb(L_copy_2_bytes); // all dword were copied
1132 } else {
1133 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1134 __ testptr(end, 4);
1135 __ jccb(Assembler::zero, L_copy_8_bytes);
1136 __ subl(count, 1<<shift);
1137 __ movl(rdx, Address(from, count, sf, 0));
1138 __ movl(Address(to, count, sf, 0), rdx);
1139 __ jmpb(L_copy_8_bytes);
1140
1141 __ align(OptoLoopAlignment);
1142 // Move 8 bytes
1143 __ BIND(L_copy_8_bytes_loop);
1144 if (UseXMMForArrayCopy) {
1145 __ movq(xmm0, Address(from, count, sf, 0));
1146 __ movq(Address(to, count, sf, 0), xmm0);
1147 } else {
1148 __ movq(mmx0, Address(from, count, sf, 0));
1149 __ movq(Address(to, count, sf, 0), mmx0);
1150 }
1151 __ BIND(L_copy_8_bytes);
1152 __ subl(count, 2<<shift);
1153 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1154 __ addl(count, 2<<shift);
1155 if (!UseXMMForArrayCopy) {
1156 __ emms();
1157 }
1158 }
1159 __ BIND(L_copy_4_bytes);
1160 // copy prefix qword
1161 __ testl(count, 1<<shift);
1162 __ jccb(Assembler::zero, L_copy_2_bytes);
1163 __ movl(rdx, Address(from, count, sf, -4));
1164 __ movl(Address(to, count, sf, -4), rdx);
1165
1166 if (t == T_BYTE || t == T_SHORT) {
1167 __ subl(count, (1<<shift));
1168 __ BIND(L_copy_2_bytes);
1169 // copy prefix dword
1170 __ testl(count, 1<<(shift-1));
1171 __ jccb(Assembler::zero, L_copy_byte);
1172 __ movw(rdx, Address(from, count, sf, -2));
1173 __ movw(Address(to, count, sf, -2), rdx);
1174 if (t == T_BYTE) {
1175 __ subl(count, 1<<(shift-1));
1176 __ BIND(L_copy_byte);
1177 // copy prefix byte
1178 __ testl(count, 1);
1179 __ jccb(Assembler::zero, L_exit);
1180 __ movb(rdx, Address(from, 0));
1181 __ movb(Address(to, 0), rdx);
1182 __ BIND(L_exit);
1183 } else {
1184 __ BIND(L_copy_byte);
1185 }
1186 } else {
1187 __ BIND(L_copy_2_bytes);
1188 }
1189 }
1190
1191 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1192 __ emms();
1193 }
1194 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1195 bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
1196
1197 if (t == T_OBJECT) {
1198 __ BIND(L_0_count);
1199 }
1200 inc_copy_counter_np(t);
1201 __ pop(rdi);
1202 __ pop(rsi);
1203 __ leave(); // required for proper stackwalking of RuntimeStub frame
1204 __ xorptr(rax, rax); // return 0
1205 __ ret(0);
1206 return start;
1207 }
1208
1209
1210 address generate_disjoint_long_copy(address* entry, const char *name) {
1211 __ align(CodeEntryAlignment);
1212 StubCodeMark mark(this, "StubRoutines", name);
1213 address start = __ pc();
1214
1215 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1216 const Register from = rax; // source array address
1217 const Register to = rdx; // destination array address
1218 const Register count = rcx; // elements count
1219 const Register to_from = rdx; // (to - from)
1220
1221 __ enter(); // required for proper stackwalking of RuntimeStub frame
1222 __ movptr(from , Address(rsp, 8+0)); // from
1223 __ movptr(to , Address(rsp, 8+4)); // to
1224 __ movl2ptr(count, Address(rsp, 8+8)); // count
1225
1226 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1227 BLOCK_COMMENT("Entry:");
1228
1229 {
1230 // UnsafeCopyMemory page error: continue after ucm
1231 UnsafeCopyMemoryMark ucmm(this, true, true);
1232 __ subptr(to, from); // to --> to_from
1233 if (VM_Version::supports_mmx()) {
1234 if (UseXMMForArrayCopy) {
1235 xmm_copy_forward(from, to_from, count);
1236 } else {
1237 mmx_copy_forward(from, to_from, count);
1238 }
1239 } else {
1240 __ jmpb(L_copy_8_bytes);
1241 __ align(OptoLoopAlignment);
1242 __ BIND(L_copy_8_bytes_loop);
1243 __ fild_d(Address(from, 0));
1244 __ fistp_d(Address(from, to_from, Address::times_1));
1245 __ addptr(from, 8);
1246 __ BIND(L_copy_8_bytes);
1247 __ decrement(count);
1248 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1249 }
1250 }
1251 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1252 __ emms();
1253 }
1254 inc_copy_counter_np(T_LONG);
1255 __ leave(); // required for proper stackwalking of RuntimeStub frame
1256 __ vzeroupper();
1257 __ xorptr(rax, rax); // return 0
1258 __ ret(0);
1259 return start;
1260 }
1261
1262 address generate_conjoint_long_copy(address nooverlap_target,
1263 address* entry, const char *name) {
1264 __ align(CodeEntryAlignment);
1265 StubCodeMark mark(this, "StubRoutines", name);
1266 address start = __ pc();
1267
1268 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1269 const Register from = rax; // source array address
1270 const Register to = rdx; // destination array address
1271 const Register count = rcx; // elements count
1272 const Register end_from = rax; // source array end address
1273
1274 __ enter(); // required for proper stackwalking of RuntimeStub frame
1275 __ movptr(from , Address(rsp, 8+0)); // from
1276 __ movptr(to , Address(rsp, 8+4)); // to
1277 __ movl2ptr(count, Address(rsp, 8+8)); // count
1278
1279 *entry = __ pc(); // Entry point from generic arraycopy stub.
1280 BLOCK_COMMENT("Entry:");
1281
1282 // arrays overlap test
1283 __ cmpptr(to, from);
1284 RuntimeAddress nooverlap(nooverlap_target);
1285 __ jump_cc(Assembler::belowEqual, nooverlap);
1286 __ lea(end_from, Address(from, count, Address::times_8, 0));
1287 __ cmpptr(to, end_from);
1288 __ movptr(from, Address(rsp, 8)); // from
1289 __ jump_cc(Assembler::aboveEqual, nooverlap);
1290
1291 {
1292 // UnsafeCopyMemory page error: continue after ucm
1293 UnsafeCopyMemoryMark ucmm(this, true, true);
1294
1295 __ jmpb(L_copy_8_bytes);
1296
1297 __ align(OptoLoopAlignment);
1298 __ BIND(L_copy_8_bytes_loop);
1299 if (VM_Version::supports_mmx()) {
1300 if (UseXMMForArrayCopy) {
1301 __ movq(xmm0, Address(from, count, Address::times_8));
1302 __ movq(Address(to, count, Address::times_8), xmm0);
1303 } else {
1304 __ movq(mmx0, Address(from, count, Address::times_8));
1305 __ movq(Address(to, count, Address::times_8), mmx0);
1306 }
1307 } else {
1308 __ fild_d(Address(from, count, Address::times_8));
1309 __ fistp_d(Address(to, count, Address::times_8));
1310 }
1311 __ BIND(L_copy_8_bytes);
1312 __ decrement(count);
1313 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1314
1315 }
1316 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1317 __ emms();
1318 }
1319 inc_copy_counter_np(T_LONG);
1320 __ leave(); // required for proper stackwalking of RuntimeStub frame
1321 __ xorptr(rax, rax); // return 0
1322 __ ret(0);
1323 return start;
1324 }
1325
1326
1327 // Helper for generating a dynamic type check.
1328 // The sub_klass must be one of {rbx, rdx, rsi}.
1329 // The temp is killed.
1330 void generate_type_check(Register sub_klass,
1331 Address& super_check_offset_addr,
1332 Address& super_klass_addr,
1333 Register temp,
1334 Label* L_success, Label* L_failure) {
1335 BLOCK_COMMENT("type_check:");
1336
1337 Label L_fallthrough;
1338 #define LOCAL_JCC(assembler_con, label_ptr) \
1339 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1340 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1341
1342 // The following is a strange variation of the fast path which requires
1343 // one less register, because needed values are on the argument stack.
1344 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1345 // L_success, L_failure, NULL);
1346 assert_different_registers(sub_klass, temp);
1347
1348 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1349
1350 // if the pointers are equal, we are done (e.g., String[] elements)
1351 __ cmpptr(sub_klass, super_klass_addr);
1352 LOCAL_JCC(Assembler::equal, L_success);
1353
1354 // check the supertype display:
1355 __ movl2ptr(temp, super_check_offset_addr);
1356 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1357 __ movptr(temp, super_check_addr); // load displayed supertype
1358 __ cmpptr(temp, super_klass_addr); // test the super type
1359 LOCAL_JCC(Assembler::equal, L_success);
1360
1361 // if it was a primary super, we can just fail immediately
1362 __ cmpl(super_check_offset_addr, sc_offset);
1363 LOCAL_JCC(Assembler::notEqual, L_failure);
1364
1365 // The repne_scan instruction uses fixed registers, which will get spilled.
1366 // We happen to know this works best when super_klass is in rax.
1367 Register super_klass = temp;
1368 __ movptr(super_klass, super_klass_addr);
1369 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1370 L_success, L_failure);
1371
1372 __ bind(L_fallthrough);
1373
1374 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1375 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1376
1377 #undef LOCAL_JCC
1378 }
1379
1380 //
1381 // Generate checkcasting array copy stub
1382 //
1383 // Input:
1384 // 4(rsp) - source array address
1385 // 8(rsp) - destination array address
1386 // 12(rsp) - element count, can be zero
1387 // 16(rsp) - size_t ckoff (super_check_offset)
1388 // 20(rsp) - oop ckval (super_klass)
1389 //
1390 // Output:
1391 // rax, == 0 - success
1392 // rax, == -1^K - failure, where K is partial transfer count
1393 //
1394 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1395 __ align(CodeEntryAlignment);
1396 StubCodeMark mark(this, "StubRoutines", name);
1397 address start = __ pc();
1398
1399 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1400
1401 // register use:
1402 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1403 // rdi, rsi -- element access (oop, klass)
1404 // rbx, -- temp
1405 const Register from = rax; // source array address
1406 const Register to = rdx; // destination array address
1407 const Register length = rcx; // elements count
1408 const Register elem = rdi; // each oop copied
1409 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1410 const Register temp = rbx; // lone remaining temp
1411
1412 __ enter(); // required for proper stackwalking of RuntimeStub frame
1413
1414 __ push(rsi);
1415 __ push(rdi);
1416 __ push(rbx);
1417
1418 Address from_arg(rsp, 16+ 4); // from
1419 Address to_arg(rsp, 16+ 8); // to
1420 Address length_arg(rsp, 16+12); // elements count
1421 Address ckoff_arg(rsp, 16+16); // super_check_offset
1422 Address ckval_arg(rsp, 16+20); // super_klass
1423
1424 // Load up:
1425 __ movptr(from, from_arg);
1426 __ movptr(to, to_arg);
1427 __ movl2ptr(length, length_arg);
1428
1429 if (entry != NULL) {
1430 *entry = __ pc(); // Entry point from generic arraycopy stub.
1431 BLOCK_COMMENT("Entry:");
1432 }
1433
1434 //---------------------------------------------------------------
1435 // Assembler stub will be used for this call to arraycopy
1436 // if the two arrays are subtypes of Object[] but the
1437 // destination array type is not equal to or a supertype
1438 // of the source type. Each element must be separately
1439 // checked.
1440
1441 // Loop-invariant addresses. They are exclusive end pointers.
1442 Address end_from_addr(from, length, Address::times_ptr, 0);
1443 Address end_to_addr(to, length, Address::times_ptr, 0);
1444
1445 Register end_from = from; // re-use
1446 Register end_to = to; // re-use
1447 Register count = length; // re-use
1448
1449 // Loop-variant addresses. They assume post-incremented count < 0.
1450 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1451 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1452 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1453
1454 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST;
1455 if (dest_uninitialized) {
1456 decorators |= IS_DEST_UNINITIALIZED;
1457 }
1458
1459 BasicType type = T_OBJECT;
1460 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1461 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1462
1463 // Copy from low to high addresses, indexed from the end of each array.
1464 __ lea(end_from, end_from_addr);
1465 __ lea(end_to, end_to_addr);
1466 assert(length == count, ""); // else fix next line:
1467 __ negptr(count); // negate and test the length
1468 __ jccb(Assembler::notZero, L_load_element);
1469
1470 // Empty array: Nothing to do.
1471 __ xorptr(rax, rax); // return 0 on (trivial) success
1472 __ jmp(L_done);
1473
1474 // ======== begin loop ========
1475 // (Loop is rotated; its entry is L_load_element.)
1476 // Loop control:
1477 // for (count = -count; count != 0; count++)
1478 // Base pointers src, dst are biased by 8*count,to last element.
1479 __ align(OptoLoopAlignment);
1480
1481 __ BIND(L_store_element);
1482 __ movptr(to_element_addr, elem); // store the oop
1483 __ increment(count); // increment the count toward zero
1484 __ jccb(Assembler::zero, L_do_card_marks);
1485
1486 // ======== loop entry is here ========
1487 __ BIND(L_load_element);
1488 __ movptr(elem, from_element_addr); // load the oop
1489 __ testptr(elem, elem);
1490 __ jccb(Assembler::zero, L_store_element);
1491
1492 // (Could do a trick here: Remember last successful non-null
1493 // element stored and make a quick oop equality check on it.)
1494
1495 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1496 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1497 &L_store_element, NULL);
1498 // (On fall-through, we have failed the element type check.)
1499 // ======== end loop ========
1500
1501 // It was a real error; we must depend on the caller to finish the job.
1502 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1503 // Emit GC store barriers for the oops we have copied (length_arg + count),
1504 // and report their number to the caller.
1505 assert_different_registers(to, count, rax);
1506 Label L_post_barrier;
1507 __ addl(count, length_arg); // transfers = (length - remaining)
1508 __ movl2ptr(rax, count); // save the value
1509 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1510 __ jccb(Assembler::notZero, L_post_barrier);
1511 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1512
1513 // Come here on success only.
1514 __ BIND(L_do_card_marks);
1515 __ xorptr(rax, rax); // return 0 on success
1516 __ movl2ptr(count, length_arg);
1517
1518 __ BIND(L_post_barrier);
1519 __ movptr(to, to_arg); // reload
1520 bs->arraycopy_epilogue(_masm, decorators, type, from, to, count);
1521
1522 // Common exit point (success or failure).
1523 __ BIND(L_done);
1524 __ pop(rbx);
1525 __ pop(rdi);
1526 __ pop(rsi);
1527 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1528 __ leave(); // required for proper stackwalking of RuntimeStub frame
1529 __ ret(0);
1530
1531 return start;
1532 }
1533
1534 //
1535 // Generate 'unsafe' array copy stub
1536 // Though just as safe as the other stubs, it takes an unscaled
1537 // size_t argument instead of an element count.
1538 //
1539 // Input:
1540 // 4(rsp) - source array address
1541 // 8(rsp) - destination array address
1542 // 12(rsp) - byte count, can be zero
1543 //
1544 // Output:
1545 // rax, == 0 - success
1546 // rax, == -1 - need to call System.arraycopy
1547 //
1548 // Examines the alignment of the operands and dispatches
1549 // to a long, int, short, or byte copy loop.
1550 //
1551 address generate_unsafe_copy(const char *name,
1552 address byte_copy_entry,
1553 address short_copy_entry,
1554 address int_copy_entry,
1555 address long_copy_entry) {
1556
1557 Label L_long_aligned, L_int_aligned, L_short_aligned;
1558
1559 __ align(CodeEntryAlignment);
1560 StubCodeMark mark(this, "StubRoutines", name);
1561 address start = __ pc();
1562
1563 const Register from = rax; // source array address
1564 const Register to = rdx; // destination array address
1565 const Register count = rcx; // elements count
1566
1567 __ enter(); // required for proper stackwalking of RuntimeStub frame
1568 __ push(rsi);
1569 __ push(rdi);
1570 Address from_arg(rsp, 12+ 4); // from
1571 Address to_arg(rsp, 12+ 8); // to
1572 Address count_arg(rsp, 12+12); // byte count
1573
1574 // Load up:
1575 __ movptr(from , from_arg);
1576 __ movptr(to , to_arg);
1577 __ movl2ptr(count, count_arg);
1578
1579 // bump this on entry, not on exit:
1580 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1581
1582 const Register bits = rsi;
1583 __ mov(bits, from);
1584 __ orptr(bits, to);
1585 __ orptr(bits, count);
1586
1587 __ testl(bits, BytesPerLong-1);
1588 __ jccb(Assembler::zero, L_long_aligned);
1589
1590 __ testl(bits, BytesPerInt-1);
1591 __ jccb(Assembler::zero, L_int_aligned);
1592
1593 __ testl(bits, BytesPerShort-1);
1594 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1595
1596 __ BIND(L_short_aligned);
1597 __ shrptr(count, LogBytesPerShort); // size => short_count
1598 __ movl(count_arg, count); // update 'count'
1599 __ jump(RuntimeAddress(short_copy_entry));
1600
1601 __ BIND(L_int_aligned);
1602 __ shrptr(count, LogBytesPerInt); // size => int_count
1603 __ movl(count_arg, count); // update 'count'
1604 __ jump(RuntimeAddress(int_copy_entry));
1605
1606 __ BIND(L_long_aligned);
1607 __ shrptr(count, LogBytesPerLong); // size => qword_count
1608 __ movl(count_arg, count); // update 'count'
1609 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1610 __ pop(rsi);
1611 __ jump(RuntimeAddress(long_copy_entry));
1612
1613 return start;
1614 }
1615
1616
1617 // Perform range checks on the proposed arraycopy.
1618 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1619 void arraycopy_range_checks(Register src,
1620 Register src_pos,
1621 Register dst,
1622 Register dst_pos,
1623 Address& length,
1624 Label& L_failed) {
1625 BLOCK_COMMENT("arraycopy_range_checks:");
1626 const Register src_end = src_pos; // source array end position
1627 const Register dst_end = dst_pos; // destination array end position
1628 __ addl(src_end, length); // src_pos + length
1629 __ addl(dst_end, length); // dst_pos + length
1630
1631 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1632 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1633 __ jcc(Assembler::above, L_failed);
1634
1635 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1636 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1637 __ jcc(Assembler::above, L_failed);
1638
1639 BLOCK_COMMENT("arraycopy_range_checks done");
1640 }
1641
1642
1643 //
1644 // Generate generic array copy stubs
1645 //
1646 // Input:
1647 // 4(rsp) - src oop
1648 // 8(rsp) - src_pos
1649 // 12(rsp) - dst oop
1650 // 16(rsp) - dst_pos
1651 // 20(rsp) - element count
1652 //
1653 // Output:
1654 // rax, == 0 - success
1655 // rax, == -1^K - failure, where K is partial transfer count
1656 //
1657 address generate_generic_copy(const char *name,
1658 address entry_jbyte_arraycopy,
1659 address entry_jshort_arraycopy,
1660 address entry_jint_arraycopy,
1661 address entry_oop_arraycopy,
1662 address entry_jlong_arraycopy,
1663 address entry_checkcast_arraycopy) {
1664 Label L_failed, L_failed_0, L_objArray;
1665
1666 { int modulus = CodeEntryAlignment;
1667 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1668 int advance = target - (__ offset() % modulus);
1669 if (advance < 0) advance += modulus;
1670 if (advance > 0) __ nop(advance);
1671 }
1672 StubCodeMark mark(this, "StubRoutines", name);
1673
1674 // Short-hop target to L_failed. Makes for denser prologue code.
1675 __ BIND(L_failed_0);
1676 __ jmp(L_failed);
1677 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1678
1679 __ align(CodeEntryAlignment);
1680 address start = __ pc();
1681
1682 __ enter(); // required for proper stackwalking of RuntimeStub frame
1683 __ push(rsi);
1684 __ push(rdi);
1685
1686 // bump this on entry, not on exit:
1687 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1688
1689 // Input values
1690 Address SRC (rsp, 12+ 4);
1691 Address SRC_POS (rsp, 12+ 8);
1692 Address DST (rsp, 12+12);
1693 Address DST_POS (rsp, 12+16);
1694 Address LENGTH (rsp, 12+20);
1695
1696 //-----------------------------------------------------------------------
1697 // Assembler stub will be used for this call to arraycopy
1698 // if the following conditions are met:
1699 //
1700 // (1) src and dst must not be null.
1701 // (2) src_pos must not be negative.
1702 // (3) dst_pos must not be negative.
1703 // (4) length must not be negative.
1704 // (5) src klass and dst klass should be the same and not NULL.
1705 // (6) src and dst should be arrays.
1706 // (7) src_pos + length must not exceed length of src.
1707 // (8) dst_pos + length must not exceed length of dst.
1708 //
1709
1710 const Register src = rax; // source array oop
1711 const Register src_pos = rsi;
1712 const Register dst = rdx; // destination array oop
1713 const Register dst_pos = rdi;
1714 const Register length = rcx; // transfer count
1715
1716 // if (src == NULL) return -1;
1717 __ movptr(src, SRC); // src oop
1718 __ testptr(src, src);
1719 __ jccb(Assembler::zero, L_failed_0);
1720
1721 // if (src_pos < 0) return -1;
1722 __ movl2ptr(src_pos, SRC_POS); // src_pos
1723 __ testl(src_pos, src_pos);
1724 __ jccb(Assembler::negative, L_failed_0);
1725
1726 // if (dst == NULL) return -1;
1727 __ movptr(dst, DST); // dst oop
1728 __ testptr(dst, dst);
1729 __ jccb(Assembler::zero, L_failed_0);
1730
1731 // if (dst_pos < 0) return -1;
1732 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1733 __ testl(dst_pos, dst_pos);
1734 __ jccb(Assembler::negative, L_failed_0);
1735
1736 // if (length < 0) return -1;
1737 __ movl2ptr(length, LENGTH); // length
1738 __ testl(length, length);
1739 __ jccb(Assembler::negative, L_failed_0);
1740
1741 // if (src->klass() == NULL) return -1;
1742 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1743 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1744 const Register rcx_src_klass = rcx; // array klass
1745 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1746
1747 #ifdef ASSERT
1748 // assert(src->klass() != NULL);
1749 BLOCK_COMMENT("assert klasses not null");
1750 { Label L1, L2;
1751 __ testptr(rcx_src_klass, rcx_src_klass);
1752 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1753 __ bind(L1);
1754 __ stop("broken null klass");
1755 __ bind(L2);
1756 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1757 __ jccb(Assembler::equal, L1); // this would be broken also
1758 BLOCK_COMMENT("assert done");
1759 }
1760 #endif //ASSERT
1761
1762 // Load layout helper (32-bits)
1763 //
1764 // |array_tag| | header_size | element_type | |log2_element_size|
1765 // 32 30 24 16 8 2 0
1766 //
1767 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1768 //
1769
1770 int lh_offset = in_bytes(Klass::layout_helper_offset());
1771 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1772
1773 // Handle objArrays completely differently...
1774 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1775 __ cmpl(src_klass_lh_addr, objArray_lh);
1776 __ jcc(Assembler::equal, L_objArray);
1777
1778 // if (src->klass() != dst->klass()) return -1;
1779 __ cmpptr(rcx_src_klass, dst_klass_addr);
1780 __ jccb(Assembler::notEqual, L_failed_0);
1781
1782 const Register rcx_lh = rcx; // layout helper
1783 assert(rcx_lh == rcx_src_klass, "known alias");
1784 __ movl(rcx_lh, src_klass_lh_addr);
1785
1786 // if (!src->is_Array()) return -1;
1787 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1788 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1789
1790 // At this point, it is known to be a typeArray (array_tag 0x3).
1791 #ifdef ASSERT
1792 { Label L;
1793 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1794 __ jcc(Assembler::greaterEqual, L); // signed cmp
1795 __ stop("must be a primitive array");
1796 __ bind(L);
1797 }
1798 #endif
1799
1800 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1801 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1802
1803 // TypeArrayKlass
1804 //
1805 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1806 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1807 //
1808 const Register rsi_offset = rsi; // array offset
1809 const Register src_array = src; // src array offset
1810 const Register dst_array = dst; // dst array offset
1811 const Register rdi_elsize = rdi; // log2 element size
1812
1813 __ mov(rsi_offset, rcx_lh);
1814 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1815 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1816 __ addptr(src_array, rsi_offset); // src array offset
1817 __ addptr(dst_array, rsi_offset); // dst array offset
1818 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1819
1820 // next registers should be set before the jump to corresponding stub
1821 const Register from = src; // source array address
1822 const Register to = dst; // destination array address
1823 const Register count = rcx; // elements count
1824 // some of them should be duplicated on stack
1825 #define FROM Address(rsp, 12+ 4)
1826 #define TO Address(rsp, 12+ 8) // Not used now
1827 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1828
1829 BLOCK_COMMENT("scale indexes to element size");
1830 __ movl2ptr(rsi, SRC_POS); // src_pos
1831 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1832 assert(src_array == from, "");
1833 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1834 __ movl2ptr(rdi, DST_POS); // dst_pos
1835 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1836 assert(dst_array == to, "");
1837 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1838 __ movptr(FROM, from); // src_addr
1839 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1840 __ movl2ptr(count, LENGTH); // elements count
1841
1842 BLOCK_COMMENT("choose copy loop based on element size");
1843 __ cmpl(rdi_elsize, 0);
1844
1845 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1846 __ cmpl(rdi_elsize, LogBytesPerShort);
1847 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1848 __ cmpl(rdi_elsize, LogBytesPerInt);
1849 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1850 #ifdef ASSERT
1851 __ cmpl(rdi_elsize, LogBytesPerLong);
1852 __ jccb(Assembler::notEqual, L_failed);
1853 #endif
1854 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1855 __ pop(rsi);
1856 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1857
1858 __ BIND(L_failed);
1859 __ xorptr(rax, rax);
1860 __ notptr(rax); // return -1
1861 __ pop(rdi);
1862 __ pop(rsi);
1863 __ leave(); // required for proper stackwalking of RuntimeStub frame
1864 __ ret(0);
1865
1866 // ObjArrayKlass
1867 __ BIND(L_objArray);
1868 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1869
1870 Label L_plain_copy, L_checkcast_copy;
1871 // test array classes for subtyping
1872 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1873 __ jccb(Assembler::notEqual, L_checkcast_copy);
1874
1875 // Identically typed arrays can be copied without element-wise checks.
1876 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1877 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1878
1879 __ BIND(L_plain_copy);
1880 __ movl2ptr(count, LENGTH); // elements count
1881 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1882 __ lea(from, Address(src, src_pos, Address::times_ptr,
1883 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1884 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1885 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1886 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1887 __ movptr(FROM, from); // src_addr
1888 __ movptr(TO, to); // dst_addr
1889 __ movl(COUNT, count); // count
1890 __ jump(RuntimeAddress(entry_oop_arraycopy));
1891
1892 __ BIND(L_checkcast_copy);
1893 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1894 {
1895 // Handy offsets:
1896 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1897 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1898
1899 Register rsi_dst_klass = rsi;
1900 Register rdi_temp = rdi;
1901 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1902 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1903 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1904
1905 // Before looking at dst.length, make sure dst is also an objArray.
1906 __ movptr(rsi_dst_klass, dst_klass_addr);
1907 __ cmpl(dst_klass_lh_addr, objArray_lh);
1908 __ jccb(Assembler::notEqual, L_failed);
1909
1910 // It is safe to examine both src.length and dst.length.
1911 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1912 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1913 // (Now src_pos and dst_pos are killed, but not src and dst.)
1914
1915 // We'll need this temp (don't forget to pop it after the type check).
1916 __ push(rbx);
1917 Register rbx_src_klass = rbx;
1918
1919 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1920 __ movptr(rsi_dst_klass, dst_klass_addr);
1921 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1922 Label L_fail_array_check;
1923 generate_type_check(rbx_src_klass,
1924 super_check_offset_addr, dst_klass_addr,
1925 rdi_temp, NULL, &L_fail_array_check);
1926 // (On fall-through, we have passed the array type check.)
1927 __ pop(rbx);
1928 __ jmp(L_plain_copy);
1929
1930 __ BIND(L_fail_array_check);
1931 // Reshuffle arguments so we can call checkcast_arraycopy:
1932
1933 // match initial saves for checkcast_arraycopy
1934 // push(rsi); // already done; see above
1935 // push(rdi); // already done; see above
1936 // push(rbx); // already done; see above
1937
1938 // Marshal outgoing arguments now, freeing registers.
1939 Address from_arg(rsp, 16+ 4); // from
1940 Address to_arg(rsp, 16+ 8); // to
1941 Address length_arg(rsp, 16+12); // elements count
1942 Address ckoff_arg(rsp, 16+16); // super_check_offset
1943 Address ckval_arg(rsp, 16+20); // super_klass
1944
1945 Address SRC_POS_arg(rsp, 16+ 8);
1946 Address DST_POS_arg(rsp, 16+16);
1947 Address LENGTH_arg(rsp, 16+20);
1948 // push rbx, changed the incoming offsets (why not just use rbp,??)
1949 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1950
1951 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1952 __ movl2ptr(length, LENGTH_arg); // reload elements count
1953 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1954 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1955
1956 __ movptr(ckval_arg, rbx); // destination element type
1957 __ movl(rbx, Address(rbx, sco_offset));
1958 __ movl(ckoff_arg, rbx); // corresponding class check offset
1959
1960 __ movl(length_arg, length); // outgoing length argument
1961
1962 __ lea(from, Address(src, src_pos, Address::times_ptr,
1963 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1964 __ movptr(from_arg, from);
1965
1966 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1967 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1968 __ movptr(to_arg, to);
1969 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1970 }
1971
1972 return start;
1973 }
1974
1975 void generate_arraycopy_stubs() {
1976 address entry;
1977 address entry_jbyte_arraycopy;
1978 address entry_jshort_arraycopy;
1979 address entry_jint_arraycopy;
1980 address entry_oop_arraycopy;
1981 address entry_jlong_arraycopy;
1982 address entry_checkcast_arraycopy;
1983
1984 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1985 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1986 "arrayof_jbyte_disjoint_arraycopy");
1987 StubRoutines::_arrayof_jbyte_arraycopy =
1988 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1989 NULL, "arrayof_jbyte_arraycopy");
1990 StubRoutines::_jbyte_disjoint_arraycopy =
1991 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1992 "jbyte_disjoint_arraycopy");
1993 StubRoutines::_jbyte_arraycopy =
1994 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1995 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1996
1997 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1998 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
1999 "arrayof_jshort_disjoint_arraycopy");
2000 StubRoutines::_arrayof_jshort_arraycopy =
2001 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2002 NULL, "arrayof_jshort_arraycopy");
2003 StubRoutines::_jshort_disjoint_arraycopy =
2004 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2005 "jshort_disjoint_arraycopy");
2006 StubRoutines::_jshort_arraycopy =
2007 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2008 &entry_jshort_arraycopy, "jshort_arraycopy");
2009
2010 // Next arrays are always aligned on 4 bytes at least.
2011 StubRoutines::_jint_disjoint_arraycopy =
2012 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2013 "jint_disjoint_arraycopy");
2014 StubRoutines::_jint_arraycopy =
2015 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2016 &entry_jint_arraycopy, "jint_arraycopy");
2017
2018 StubRoutines::_oop_disjoint_arraycopy =
2019 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2020 "oop_disjoint_arraycopy");
2021 StubRoutines::_oop_arraycopy =
2022 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2023 &entry_oop_arraycopy, "oop_arraycopy");
2024
2025 StubRoutines::_oop_disjoint_arraycopy_uninit =
2026 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2027 "oop_disjoint_arraycopy_uninit",
2028 /*dest_uninitialized*/true);
2029 StubRoutines::_oop_arraycopy_uninit =
2030 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2031 NULL, "oop_arraycopy_uninit",
2032 /*dest_uninitialized*/true);
2033
2034 StubRoutines::_jlong_disjoint_arraycopy =
2035 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2036 StubRoutines::_jlong_arraycopy =
2037 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2038 "jlong_arraycopy");
2039
2040 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2041 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2042 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2043 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2044 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2045 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2046
2047 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2048 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2049 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2050 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2051
2052 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2053 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2054 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2055 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2056
2057 StubRoutines::_checkcast_arraycopy =
2058 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2059 StubRoutines::_checkcast_arraycopy_uninit =
2060 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2061
2062 StubRoutines::_unsafe_arraycopy =
2063 generate_unsafe_copy("unsafe_arraycopy",
2064 entry_jbyte_arraycopy,
2065 entry_jshort_arraycopy,
2066 entry_jint_arraycopy,
2067 entry_jlong_arraycopy);
2068
2069 StubRoutines::_generic_arraycopy =
2070 generate_generic_copy("generic_arraycopy",
2071 entry_jbyte_arraycopy,
2072 entry_jshort_arraycopy,
2073 entry_jint_arraycopy,
2074 entry_oop_arraycopy,
2075 entry_jlong_arraycopy,
2076 entry_checkcast_arraycopy);
2077 }
2078
2079 // AES intrinsic stubs
2080 enum {AESBlockSize = 16};
2081
2082 address generate_key_shuffle_mask() {
2083 __ align(16);
2084 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2085 address start = __ pc();
2086 __ emit_data(0x00010203, relocInfo::none, 0 );
2087 __ emit_data(0x04050607, relocInfo::none, 0 );
2088 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2089 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2090 return start;
2091 }
2092
2093 address generate_counter_shuffle_mask() {
2094 __ align(16);
2095 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2096 address start = __ pc();
2097 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
2098 __ emit_data(0x08090a0b, relocInfo::none, 0);
2099 __ emit_data(0x04050607, relocInfo::none, 0);
2100 __ emit_data(0x00010203, relocInfo::none, 0);
2101 return start;
2102 }
2103
2104 // Utility routine for loading a 128-bit key word in little endian format
2105 // can optionally specify that the shuffle mask is already in an xmmregister
2106 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2107 __ movdqu(xmmdst, Address(key, offset));
2108 if (xmm_shuf_mask != NULL) {
2109 __ pshufb(xmmdst, xmm_shuf_mask);
2110 } else {
2111 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2112 }
2113 }
2114
2115 // aesenc using specified key+offset
2116 // can optionally specify that the shuffle mask is already in an xmmregister
2117 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2118 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2119 __ aesenc(xmmdst, xmmtmp);
2120 }
2121
2122 // aesdec using specified key+offset
2123 // can optionally specify that the shuffle mask is already in an xmmregister
2124 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2125 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2126 __ aesdec(xmmdst, xmmtmp);
2127 }
2128
2129 // Utility routine for increase 128bit counter (iv in CTR mode)
2130 // XMM_128bit, D3, D2, D1, D0
2131 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
2132 __ pextrd(reg, xmmdst, 0x0);
2133 __ addl(reg, inc_delta);
2134 __ pinsrd(xmmdst, reg, 0x0);
2135 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2136
2137 __ pextrd(reg, xmmdst, 0x01); // Carry-> D1
2138 __ addl(reg, 0x01);
2139 __ pinsrd(xmmdst, reg, 0x01);
2140 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2141
2142 __ pextrd(reg, xmmdst, 0x02); // Carry-> D2
2143 __ addl(reg, 0x01);
2144 __ pinsrd(xmmdst, reg, 0x02);
2145 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2146
2147 __ pextrd(reg, xmmdst, 0x03); // Carry -> D3
2148 __ addl(reg, 0x01);
2149 __ pinsrd(xmmdst, reg, 0x03);
2150
2151 __ BIND(next_block); // next instruction
2152 }
2153
2154
2155 // Arguments:
2156 //
2157 // Inputs:
2158 // c_rarg0 - source byte array address
2159 // c_rarg1 - destination byte array address
2160 // c_rarg2 - K (key) in little endian int array
2161 //
2162 address generate_aescrypt_encryptBlock() {
2163 assert(UseAES, "need AES instructions and misaligned SSE support");
2164 __ align(CodeEntryAlignment);
2165 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2166 Label L_doLast;
2167 address start = __ pc();
2168
2169 const Register from = rdx; // source array address
2170 const Register to = rdx; // destination array address
2171 const Register key = rcx; // key array address
2172 const Register keylen = rax;
2173 const Address from_param(rbp, 8+0);
2174 const Address to_param (rbp, 8+4);
2175 const Address key_param (rbp, 8+8);
2176
2177 const XMMRegister xmm_result = xmm0;
2178 const XMMRegister xmm_key_shuf_mask = xmm1;
2179 const XMMRegister xmm_temp1 = xmm2;
2180 const XMMRegister xmm_temp2 = xmm3;
2181 const XMMRegister xmm_temp3 = xmm4;
2182 const XMMRegister xmm_temp4 = xmm5;
2183
2184 __ enter(); // required for proper stackwalking of RuntimeStub frame
2185
2186 __ movptr(from, from_param);
2187 __ movptr(key, key_param);
2188
2189 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2190 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2191
2192 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2193 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2194 __ movptr(to, to_param);
2195
2196 // For encryption, the java expanded key ordering is just what we need
2197
2198 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2199 __ pxor(xmm_result, xmm_temp1);
2200
2201 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2202 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2203 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2204 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2205
2206 __ aesenc(xmm_result, xmm_temp1);
2207 __ aesenc(xmm_result, xmm_temp2);
2208 __ aesenc(xmm_result, xmm_temp3);
2209 __ aesenc(xmm_result, xmm_temp4);
2210
2211 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2212 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2213 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2214 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2215
2216 __ aesenc(xmm_result, xmm_temp1);
2217 __ aesenc(xmm_result, xmm_temp2);
2218 __ aesenc(xmm_result, xmm_temp3);
2219 __ aesenc(xmm_result, xmm_temp4);
2220
2221 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2222 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2223
2224 __ cmpl(keylen, 44);
2225 __ jccb(Assembler::equal, L_doLast);
2226
2227 __ aesenc(xmm_result, xmm_temp1);
2228 __ aesenc(xmm_result, xmm_temp2);
2229
2230 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2231 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2232
2233 __ cmpl(keylen, 52);
2234 __ jccb(Assembler::equal, L_doLast);
2235
2236 __ aesenc(xmm_result, xmm_temp1);
2237 __ aesenc(xmm_result, xmm_temp2);
2238
2239 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2240 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2241
2242 __ BIND(L_doLast);
2243 __ aesenc(xmm_result, xmm_temp1);
2244 __ aesenclast(xmm_result, xmm_temp2);
2245 __ movdqu(Address(to, 0), xmm_result); // store the result
2246 __ xorptr(rax, rax); // return 0
2247 __ leave(); // required for proper stackwalking of RuntimeStub frame
2248 __ ret(0);
2249
2250 return start;
2251 }
2252
2253
2254 // Arguments:
2255 //
2256 // Inputs:
2257 // c_rarg0 - source byte array address
2258 // c_rarg1 - destination byte array address
2259 // c_rarg2 - K (key) in little endian int array
2260 //
2261 address generate_aescrypt_decryptBlock() {
2262 assert(UseAES, "need AES instructions and misaligned SSE support");
2263 __ align(CodeEntryAlignment);
2264 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2265 Label L_doLast;
2266 address start = __ pc();
2267
2268 const Register from = rdx; // source array address
2269 const Register to = rdx; // destination array address
2270 const Register key = rcx; // key array address
2271 const Register keylen = rax;
2272 const Address from_param(rbp, 8+0);
2273 const Address to_param (rbp, 8+4);
2274 const Address key_param (rbp, 8+8);
2275
2276 const XMMRegister xmm_result = xmm0;
2277 const XMMRegister xmm_key_shuf_mask = xmm1;
2278 const XMMRegister xmm_temp1 = xmm2;
2279 const XMMRegister xmm_temp2 = xmm3;
2280 const XMMRegister xmm_temp3 = xmm4;
2281 const XMMRegister xmm_temp4 = xmm5;
2282
2283 __ enter(); // required for proper stackwalking of RuntimeStub frame
2284
2285 __ movptr(from, from_param);
2286 __ movptr(key, key_param);
2287
2288 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2289 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2290
2291 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2292 __ movdqu(xmm_result, Address(from, 0));
2293 __ movptr(to, to_param);
2294
2295 // for decryption java expanded key ordering is rotated one position from what we want
2296 // so we start from 0x10 here and hit 0x00 last
2297 // we don't know if the key is aligned, hence not using load-execute form
2298 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2299 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2300 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2301 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2302
2303 __ pxor (xmm_result, xmm_temp1);
2304 __ aesdec(xmm_result, xmm_temp2);
2305 __ aesdec(xmm_result, xmm_temp3);
2306 __ aesdec(xmm_result, xmm_temp4);
2307
2308 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2309 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2310 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2311 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2312
2313 __ aesdec(xmm_result, xmm_temp1);
2314 __ aesdec(xmm_result, xmm_temp2);
2315 __ aesdec(xmm_result, xmm_temp3);
2316 __ aesdec(xmm_result, xmm_temp4);
2317
2318 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2319 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2320 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2321
2322 __ cmpl(keylen, 44);
2323 __ jccb(Assembler::equal, L_doLast);
2324
2325 __ aesdec(xmm_result, xmm_temp1);
2326 __ aesdec(xmm_result, xmm_temp2);
2327
2328 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2329 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2330
2331 __ cmpl(keylen, 52);
2332 __ jccb(Assembler::equal, L_doLast);
2333
2334 __ aesdec(xmm_result, xmm_temp1);
2335 __ aesdec(xmm_result, xmm_temp2);
2336
2337 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2338 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2339
2340 __ BIND(L_doLast);
2341 __ aesdec(xmm_result, xmm_temp1);
2342 __ aesdec(xmm_result, xmm_temp2);
2343
2344 // for decryption the aesdeclast operation is always on key+0x00
2345 __ aesdeclast(xmm_result, xmm_temp3);
2346 __ movdqu(Address(to, 0), xmm_result); // store the result
2347 __ xorptr(rax, rax); // return 0
2348 __ leave(); // required for proper stackwalking of RuntimeStub frame
2349 __ ret(0);
2350
2351 return start;
2352 }
2353
2354 void handleSOERegisters(bool saving) {
2355 const int saveFrameSizeInBytes = 4 * wordSize;
2356 const Address saved_rbx (rbp, -3 * wordSize);
2357 const Address saved_rsi (rbp, -2 * wordSize);
2358 const Address saved_rdi (rbp, -1 * wordSize);
2359
2360 if (saving) {
2361 __ subptr(rsp, saveFrameSizeInBytes);
2362 __ movptr(saved_rsi, rsi);
2363 __ movptr(saved_rdi, rdi);
2364 __ movptr(saved_rbx, rbx);
2365 } else {
2366 // restoring
2367 __ movptr(rsi, saved_rsi);
2368 __ movptr(rdi, saved_rdi);
2369 __ movptr(rbx, saved_rbx);
2370 }
2371 }
2372
2373 // Arguments:
2374 //
2375 // Inputs:
2376 // c_rarg0 - source byte array address
2377 // c_rarg1 - destination byte array address
2378 // c_rarg2 - K (key) in little endian int array
2379 // c_rarg3 - r vector byte array address
2380 // c_rarg4 - input length
2381 //
2382 // Output:
2383 // rax - input length
2384 //
2385 address generate_cipherBlockChaining_encryptAESCrypt() {
2386 assert(UseAES, "need AES instructions and misaligned SSE support");
2387 __ align(CodeEntryAlignment);
2388 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2389 address start = __ pc();
2390
2391 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2392 const Register from = rsi; // source array address
2393 const Register to = rdx; // destination array address
2394 const Register key = rcx; // key array address
2395 const Register rvec = rdi; // r byte array initialized from initvector array address
2396 // and left with the results of the last encryption block
2397 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2398 const Register pos = rax;
2399
2400 // xmm register assignments for the loops below
2401 const XMMRegister xmm_result = xmm0;
2402 const XMMRegister xmm_temp = xmm1;
2403 // first 6 keys preloaded into xmm2-xmm7
2404 const int XMM_REG_NUM_KEY_FIRST = 2;
2405 const int XMM_REG_NUM_KEY_LAST = 7;
2406 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2407
2408 __ enter(); // required for proper stackwalking of RuntimeStub frame
2409 handleSOERegisters(true /*saving*/);
2410
2411 // load registers from incoming parameters
2412 const Address from_param(rbp, 8+0);
2413 const Address to_param (rbp, 8+4);
2414 const Address key_param (rbp, 8+8);
2415 const Address rvec_param (rbp, 8+12);
2416 const Address len_param (rbp, 8+16);
2417 __ movptr(from , from_param);
2418 __ movptr(to , to_param);
2419 __ movptr(key , key_param);
2420 __ movptr(rvec , rvec_param);
2421 __ movptr(len_reg , len_param);
2422
2423 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2424 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2425 // load up xmm regs 2 thru 7 with keys 0-5
2426 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2427 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2428 offset += 0x10;
2429 }
2430
2431 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2432
2433 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2434 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2435 __ cmpl(rax, 44);
2436 __ jcc(Assembler::notEqual, L_key_192_256);
2437
2438 // 128 bit code follows here
2439 __ movl(pos, 0);
2440 __ align(OptoLoopAlignment);
2441 __ BIND(L_loopTop_128);
2442 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2443 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2444
2445 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2446 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2447 __ aesenc(xmm_result, as_XMMRegister(rnum));
2448 }
2449 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2450 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2451 }
2452 load_key(xmm_temp, key, 0xa0);
2453 __ aesenclast(xmm_result, xmm_temp);
2454
2455 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2456 // no need to store r to memory until we exit
2457 __ addptr(pos, AESBlockSize);
2458 __ subptr(len_reg, AESBlockSize);
2459 __ jcc(Assembler::notEqual, L_loopTop_128);
2460
2461 __ BIND(L_exit);
2462 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2463
2464 handleSOERegisters(false /*restoring*/);
2465 __ movptr(rax, len_param); // return length
2466 __ leave(); // required for proper stackwalking of RuntimeStub frame
2467 __ ret(0);
2468
2469 __ BIND(L_key_192_256);
2470 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2471 __ cmpl(rax, 52);
2472 __ jcc(Assembler::notEqual, L_key_256);
2473
2474 // 192-bit code follows here (could be changed to use more xmm registers)
2475 __ movl(pos, 0);
2476 __ align(OptoLoopAlignment);
2477 __ BIND(L_loopTop_192);
2478 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2479 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2480
2481 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2482 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2483 __ aesenc(xmm_result, as_XMMRegister(rnum));
2484 }
2485 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2486 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2487 }
2488 load_key(xmm_temp, key, 0xc0);
2489 __ aesenclast(xmm_result, xmm_temp);
2490
2491 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2492 // no need to store r to memory until we exit
2493 __ addptr(pos, AESBlockSize);
2494 __ subptr(len_reg, AESBlockSize);
2495 __ jcc(Assembler::notEqual, L_loopTop_192);
2496 __ jmp(L_exit);
2497
2498 __ BIND(L_key_256);
2499 // 256-bit code follows here (could be changed to use more xmm registers)
2500 __ movl(pos, 0);
2501 __ align(OptoLoopAlignment);
2502 __ BIND(L_loopTop_256);
2503 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2504 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2505
2506 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2507 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2508 __ aesenc(xmm_result, as_XMMRegister(rnum));
2509 }
2510 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2511 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2512 }
2513 load_key(xmm_temp, key, 0xe0);
2514 __ aesenclast(xmm_result, xmm_temp);
2515
2516 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2517 // no need to store r to memory until we exit
2518 __ addptr(pos, AESBlockSize);
2519 __ subptr(len_reg, AESBlockSize);
2520 __ jcc(Assembler::notEqual, L_loopTop_256);
2521 __ jmp(L_exit);
2522
2523 return start;
2524 }
2525
2526
2527 // CBC AES Decryption.
2528 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2529 //
2530 // Arguments:
2531 //
2532 // Inputs:
2533 // c_rarg0 - source byte array address
2534 // c_rarg1 - destination byte array address
2535 // c_rarg2 - K (key) in little endian int array
2536 // c_rarg3 - r vector byte array address
2537 // c_rarg4 - input length
2538 //
2539 // Output:
2540 // rax - input length
2541 //
2542
2543 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
2544 assert(UseAES, "need AES instructions and misaligned SSE support");
2545 __ align(CodeEntryAlignment);
2546 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2547 address start = __ pc();
2548
2549 const Register from = rsi; // source array address
2550 const Register to = rdx; // destination array address
2551 const Register key = rcx; // key array address
2552 const Register rvec = rdi; // r byte array initialized from initvector array address
2553 // and left with the results of the last encryption block
2554 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2555 const Register pos = rax;
2556
2557 const int PARALLEL_FACTOR = 4;
2558 const int ROUNDS[3] = { 10, 12, 14 }; //aes rounds for key128, key192, key256
2559
2560 Label L_exit;
2561 Label L_singleBlock_loopTop[3]; //128, 192, 256
2562 Label L_multiBlock_loopTop[3]; //128, 192, 256
2563
2564 const XMMRegister xmm_prev_block_cipher = xmm0; // holds cipher of previous block
2565 const XMMRegister xmm_key_shuf_mask = xmm1;
2566
2567 const XMMRegister xmm_key_tmp0 = xmm2;
2568 const XMMRegister xmm_key_tmp1 = xmm3;
2569
2570 // registers holding the six results in the parallelized loop
2571 const XMMRegister xmm_result0 = xmm4;
2572 const XMMRegister xmm_result1 = xmm5;
2573 const XMMRegister xmm_result2 = xmm6;
2574 const XMMRegister xmm_result3 = xmm7;
2575
2576 __ enter(); // required for proper stackwalking of RuntimeStub frame
2577 handleSOERegisters(true /*saving*/);
2578
2579 // load registers from incoming parameters
2580 const Address from_param(rbp, 8+0);
2581 const Address to_param (rbp, 8+4);
2582 const Address key_param (rbp, 8+8);
2583 const Address rvec_param (rbp, 8+12);
2584 const Address len_param (rbp, 8+16);
2585
2586 __ movptr(from , from_param);
2587 __ movptr(to , to_param);
2588 __ movptr(key , key_param);
2589 __ movptr(rvec , rvec_param);
2590 __ movptr(len_reg , len_param);
2591
2592 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2593 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
2594
2595 __ xorptr(pos, pos);
2596
2597 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2598 // rvec is reused
2599 __ movl(rvec, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2600 __ cmpl(rvec, 52);
2601 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
2602 __ cmpl(rvec, 60);
2603 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
2604
2605 #define DoFour(opc, src_reg) \
2606 __ opc(xmm_result0, src_reg); \
2607 __ opc(xmm_result1, src_reg); \
2608 __ opc(xmm_result2, src_reg); \
2609 __ opc(xmm_result3, src_reg); \
2610
2611 for (int k = 0; k < 3; ++k) {
2612 __ align(OptoLoopAlignment);
2613 __ BIND(L_multiBlock_loopTop[k]);
2614 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
2615 __ jcc(Assembler::less, L_singleBlock_loopTop[k]);
2616
2617 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
2618 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2619 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2620 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2621
2622 // the java expanded key ordering is rotated one position from what we want
2623 // so we start from 0x10 here and hit 0x00 last
2624 load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2625 DoFour(pxor, xmm_key_tmp0); //xor with first key
2626 // do the aes dec rounds
2627 for (int rnum = 1; rnum <= ROUNDS[k];) {
2628 //load two keys at a time
2629 //k1->0x20, ..., k9->0xa0, k10->0x00
2630 load_key(xmm_key_tmp1, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2631 load_key(xmm_key_tmp0, key, ((rnum + 2) % (ROUNDS[k] + 1)) * 0x10, xmm_key_shuf_mask); // hit 0x00 last!
2632 DoFour(aesdec, xmm_key_tmp1);
2633 rnum++;
2634 if (rnum != ROUNDS[k]) {
2635 DoFour(aesdec, xmm_key_tmp0);
2636 }
2637 else {
2638 DoFour(aesdeclast, xmm_key_tmp0);
2639 }
2640 rnum++;
2641 }
2642
2643 // for each result, xor with the r vector of previous cipher block
2644 __ pxor(xmm_result0, xmm_prev_block_cipher);
2645 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2646 __ pxor(xmm_result1, xmm_prev_block_cipher);
2647 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2648 __ pxor(xmm_result2, xmm_prev_block_cipher);
2649 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2650 __ pxor(xmm_result3, xmm_prev_block_cipher);
2651 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
2652
2653 // store 4 results into the next 64 bytes of output
2654 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2655 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2656 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2657 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2658
2659 __ addptr(pos, 4 * AESBlockSize);
2660 __ subptr(len_reg, 4 * AESBlockSize);
2661 __ jmp(L_multiBlock_loopTop[k]);
2662
2663 //singleBlock starts here
2664 __ align(OptoLoopAlignment);
2665 __ BIND(L_singleBlock_loopTop[k]);
2666 __ cmpptr(len_reg, 0); // any blocks left?
2667 __ jcc(Assembler::equal, L_exit);
2668 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2669 __ movdqa(xmm_result1, xmm_result0);
2670
2671 load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2672 __ pxor(xmm_result0, xmm_key_tmp0);
2673 // do the aes dec rounds
2674 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
2675 // the java expanded key ordering is rotated one position from what we want
2676 load_key(xmm_key_tmp0, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2677 __ aesdec(xmm_result0, xmm_key_tmp0);
2678 }
2679 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
2680 __ aesdeclast(xmm_result0, xmm_key_tmp0);
2681 __ pxor(xmm_result0, xmm_prev_block_cipher); // xor with the current r vector
2682 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result0); // store into the next 16 bytes of output
2683 // no need to store r to memory until we exit
2684 __ movdqa(xmm_prev_block_cipher, xmm_result1); // set up next r vector with cipher input from this block
2685
2686 __ addptr(pos, AESBlockSize);
2687 __ subptr(len_reg, AESBlockSize);
2688 __ jmp(L_singleBlock_loopTop[k]);
2689 }//for 128/192/256
2690
2691 __ BIND(L_exit);
2692 __ movptr(rvec, rvec_param); // restore this since reused earlier
2693 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
2694 handleSOERegisters(false /*restoring*/);
2695 __ movptr(rax, len_param); // return length
2696 __ leave(); // required for proper stackwalking of RuntimeStub frame
2697 __ ret(0);
2698
2699 return start;
2700 }
2701
2702 // CTR AES crypt.
2703 // In 32-bit stub, parallelize 4 blocks at a time
2704 // Arguments:
2705 //
2706 // Inputs:
2707 // c_rarg0 - source byte array address
2708 // c_rarg1 - destination byte array address
2709 // c_rarg2 - K (key) in little endian int array
2710 // c_rarg3 - counter vector byte array address
2711 // c_rarg4 - input length
2712 //
2713 // Output:
2714 // rax - input length
2715 //
2716 address generate_counterMode_AESCrypt_Parallel() {
2717 assert(UseAES, "need AES instructions and misaligned SSE support");
2718 __ align(CodeEntryAlignment);
2719 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
2720 address start = __ pc();
2721 const Register from = rsi; // source array address
2722 const Register to = rdx; // destination array address
2723 const Register key = rcx; // key array address
2724 const Register counter = rdi; // counter byte array initialized from initvector array address
2725 // and updated with the incremented counter in the end
2726 const Register len_reg = rbx;
2727 const Register pos = rax;
2728
2729 __ enter(); // required for proper stackwalking of RuntimeStub frame
2730 handleSOERegisters(true /*saving*/); // save rbx, rsi, rdi
2731
2732 // load registers from incoming parameters
2733 const Address from_param(rbp, 8+0);
2734 const Address to_param (rbp, 8+4);
2735 const Address key_param (rbp, 8+8);
2736 const Address rvec_param (rbp, 8+12);
2737 const Address len_param (rbp, 8+16);
2738 const Address saved_counter_param(rbp, 8 + 20);
2739 const Address used_addr_param(rbp, 8 + 24);
2740
2741 __ movptr(from , from_param);
2742 __ movptr(to , to_param);
2743 __ movptr(len_reg , len_param);
2744
2745 // Use the partially used encrpyted counter from last invocation
2746 Label L_exit_preLoop, L_preLoop_start;
2747
2748 // Use the registers 'counter' and 'key' here in this preloop
2749 // to hold of last 2 params 'used' and 'saved_encCounter_start'
2750 Register used = counter;
2751 Register saved_encCounter_start = key;
2752 Register used_addr = saved_encCounter_start;
2753
2754 __ movptr(used_addr, used_addr_param);
2755 __ movptr(used, Address(used_addr, 0));
2756 __ movptr(saved_encCounter_start, saved_counter_param);
2757
2758 __ BIND(L_preLoop_start);
2759 __ cmpptr(used, 16);
2760 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
2761 __ cmpptr(len_reg, 0);
2762 __ jcc(Assembler::lessEqual, L_exit_preLoop);
2763 __ movb(rax, Address(saved_encCounter_start, used));
2764 __ xorb(rax, Address(from, 0));
2765 __ movb(Address(to, 0), rax);
2766 __ addptr(from, 1);
2767 __ addptr(to, 1);
2768 __ addptr(used, 1);
2769 __ subptr(len_reg, 1);
2770
2771 __ jmp(L_preLoop_start);
2772
2773 __ BIND(L_exit_preLoop);
2774 __ movptr(used_addr, used_addr_param);
2775 __ movptr(used_addr, used_addr_param);
2776 __ movl(Address(used_addr, 0), used);
2777
2778 // load the parameters 'key' and 'counter'
2779 __ movptr(key, key_param);
2780 __ movptr(counter, rvec_param);
2781
2782 // xmm register assignments for the loops below
2783 const XMMRegister xmm_curr_counter = xmm0;
2784 const XMMRegister xmm_counter_shuf_mask = xmm1; // need to be reloaded
2785 const XMMRegister xmm_key_shuf_mask = xmm2; // need to be reloaded
2786 const XMMRegister xmm_key = xmm3;
2787 const XMMRegister xmm_result0 = xmm4;
2788 const XMMRegister xmm_result1 = xmm5;
2789 const XMMRegister xmm_result2 = xmm6;
2790 const XMMRegister xmm_result3 = xmm7;
2791 const XMMRegister xmm_from0 = xmm1; //reuse XMM register
2792 const XMMRegister xmm_from1 = xmm2;
2793 const XMMRegister xmm_from2 = xmm3;
2794 const XMMRegister xmm_from3 = xmm4;
2795
2796 //for key_128, key_192, key_256
2797 const int rounds[3] = {10, 12, 14};
2798 Label L_singleBlockLoopTop[3];
2799 Label L_multiBlock_loopTop[3];
2800 Label L_key192_top, L_key256_top;
2801 Label L_incCounter[3][4]; // 3: different key length, 4: 4 blocks at a time
2802 Label L_incCounter_single[3]; //for single block, key128, key192, key256
2803 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];
2804 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];
2805
2806 Label L_exit;
2807 const int PARALLEL_FACTOR = 4; //because of the limited register number
2808
2809 // initialize counter with initial counter
2810 __ movdqu(xmm_curr_counter, Address(counter, 0x00));
2811 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2812 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled for increase
2813
2814 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
2815 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2816 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2817 __ cmpl(rax, 52);
2818 __ jcc(Assembler::equal, L_key192_top);
2819 __ cmpl(rax, 60);
2820 __ jcc(Assembler::equal, L_key256_top);
2821
2822 //key128 begins here
2823 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2824
2825 #define CTR_DoFour(opc, src_reg) \
2826 __ opc(xmm_result0, src_reg); \
2827 __ opc(xmm_result1, src_reg); \
2828 __ opc(xmm_result2, src_reg); \
2829 __ opc(xmm_result3, src_reg);
2830
2831 // k == 0 : generate code for key_128
2832 // k == 1 : generate code for key_192
2833 // k == 2 : generate code for key_256
2834 for (int k = 0; k < 3; ++k) {
2835 //multi blocks starts here
2836 __ align(OptoLoopAlignment);
2837 __ BIND(L_multiBlock_loopTop[k]);
2838 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
2839 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
2840
2841 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2842 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2843
2844 //load, then increase counters
2845 CTR_DoFour(movdqa, xmm_curr_counter);
2846 __ push(rbx);
2847 inc_counter(rbx, xmm_result1, 0x01, L_incCounter[k][0]);
2848 inc_counter(rbx, xmm_result2, 0x02, L_incCounter[k][1]);
2849 inc_counter(rbx, xmm_result3, 0x03, L_incCounter[k][2]);
2850 inc_counter(rbx, xmm_curr_counter, 0x04, L_incCounter[k][3]);
2851 __ pop (rbx);
2852
2853 load_key(xmm_key, key, 0x00, xmm_key_shuf_mask); // load Round 0 key. interleaving for better performance
2854
2855 CTR_DoFour(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
2856 CTR_DoFour(pxor, xmm_key); //PXOR with Round 0 key
2857
2858 for (int i = 1; i < rounds[k]; ++i) {
2859 load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2860 CTR_DoFour(aesenc, xmm_key);
2861 }
2862 load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2863 CTR_DoFour(aesenclast, xmm_key);
2864
2865 // get next PARALLEL_FACTOR blocks into xmm_from registers
2866 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2867 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2868 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2869
2870 // PXOR with input text
2871 __ pxor(xmm_result0, xmm_from0); //result0 is xmm4
2872 __ pxor(xmm_result1, xmm_from1);
2873 __ pxor(xmm_result2, xmm_from2);
2874
2875 // store PARALLEL_FACTOR results into the next 64 bytes of output
2876 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2877 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2878 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2879
2880 // do it here after xmm_result0 is saved, because xmm_from3 reuse the same register of xmm_result0.
2881 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2882 __ pxor(xmm_result3, xmm_from3);
2883 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2884
2885 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
2886 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
2887 __ jmp(L_multiBlock_loopTop[k]);
2888
2889 // singleBlock starts here
2890 __ align(OptoLoopAlignment);
2891 __ BIND(L_singleBlockLoopTop[k]);
2892 __ cmpptr(len_reg, 0);
2893 __ jcc(Assembler::equal, L_exit);
2894 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2895 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2896 __ movdqa(xmm_result0, xmm_curr_counter);
2897 load_key(xmm_key, key, 0x00, xmm_key_shuf_mask);
2898 __ push(rbx);//rbx is used for increasing counter
2899 inc_counter(rbx, xmm_curr_counter, 0x01, L_incCounter_single[k]);
2900 __ pop (rbx);
2901 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
2902 __ pxor(xmm_result0, xmm_key);
2903 for (int i = 1; i < rounds[k]; i++) {
2904 load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2905 __ aesenc(xmm_result0, xmm_key);
2906 }
2907 load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2908 __ aesenclast(xmm_result0, xmm_key);
2909 __ cmpptr(len_reg, AESBlockSize);
2910 __ jcc(Assembler::less, L_processTail_insr[k]);
2911 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2912 __ pxor(xmm_result0, xmm_from0);
2913 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2914 __ addptr(pos, AESBlockSize);
2915 __ subptr(len_reg, AESBlockSize);
2916 __ jmp(L_singleBlockLoopTop[k]);
2917
2918 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
2919 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
2920 __ testptr(len_reg, 8);
2921 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
2922 __ subptr(pos,8);
2923 __ pinsrd(xmm_from0, Address(from, pos), 0);
2924 __ pinsrd(xmm_from0, Address(from, pos, Address::times_1, 4), 1);
2925 __ BIND(L_processTail_4_insr[k]);
2926 __ testptr(len_reg, 4);
2927 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
2928 __ subptr(pos,4);
2929 __ pslldq(xmm_from0, 4);
2930 __ pinsrd(xmm_from0, Address(from, pos), 0);
2931 __ BIND(L_processTail_2_insr[k]);
2932 __ testptr(len_reg, 2);
2933 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
2934 __ subptr(pos, 2);
2935 __ pslldq(xmm_from0, 2);
2936 __ pinsrw(xmm_from0, Address(from, pos), 0);
2937 __ BIND(L_processTail_1_insr[k]);
2938 __ testptr(len_reg, 1);
2939 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
2940 __ subptr(pos, 1);
2941 __ pslldq(xmm_from0, 1);
2942 __ pinsrb(xmm_from0, Address(from, pos), 0);
2943 __ BIND(L_processTail_exit_insr[k]);
2944
2945 __ movptr(saved_encCounter_start, saved_counter_param);
2946 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
2947 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
2948
2949 __ testptr(len_reg, 8);
2950 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
2951 __ pextrd(Address(to, pos), xmm_result0, 0);
2952 __ pextrd(Address(to, pos, Address::times_1, 4), xmm_result0, 1);
2953 __ psrldq(xmm_result0, 8);
2954 __ addptr(pos, 8);
2955 __ BIND(L_processTail_4_extr[k]);
2956 __ testptr(len_reg, 4);
2957 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
2958 __ pextrd(Address(to, pos), xmm_result0, 0);
2959 __ psrldq(xmm_result0, 4);
2960 __ addptr(pos, 4);
2961 __ BIND(L_processTail_2_extr[k]);
2962 __ testptr(len_reg, 2);
2963 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
2964 __ pextrb(Address(to, pos), xmm_result0, 0);
2965 __ pextrb(Address(to, pos, Address::times_1, 1), xmm_result0, 1);
2966 __ psrldq(xmm_result0, 2);
2967 __ addptr(pos, 2);
2968 __ BIND(L_processTail_1_extr[k]);
2969 __ testptr(len_reg, 1);
2970 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
2971 __ pextrb(Address(to, pos), xmm_result0, 0);
2972
2973 __ BIND(L_processTail_exit_extr[k]);
2974 __ movptr(used_addr, used_addr_param);
2975 __ movl(Address(used_addr, 0), len_reg);
2976 __ jmp(L_exit);
2977 }
2978
2979 __ BIND(L_exit);
2980 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2981 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
2982 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
2983 handleSOERegisters(false /*restoring*/);
2984 __ movptr(rax, len_param); // return length
2985 __ leave(); // required for proper stackwalking of RuntimeStub frame
2986 __ ret(0);
2987
2988 __ BIND (L_key192_top);
2989 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2990 __ jmp(L_multiBlock_loopTop[1]); //key192
2991
2992 __ BIND (L_key256_top);
2993 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2994 __ jmp(L_multiBlock_loopTop[2]); //key192
2995
2996 return start;
2997 }
2998
2999 address generate_upper_word_mask() {
3000 __ align(64);
3001 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
3002 address start = __ pc();
3003 __ emit_data(0x00000000, relocInfo::none, 0);
3004 __ emit_data(0x00000000, relocInfo::none, 0);
3005 __ emit_data(0x00000000, relocInfo::none, 0);
3006 __ emit_data(0xFFFFFFFF, relocInfo::none, 0);
3007 return start;
3008 }
3009
3010 address generate_shuffle_byte_flip_mask() {
3011 __ align(64);
3012 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
3013 address start = __ pc();
3014 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3015 __ emit_data(0x08090a0b, relocInfo::none, 0);
3016 __ emit_data(0x04050607, relocInfo::none, 0);
3017 __ emit_data(0x00010203, relocInfo::none, 0);
3018 return start;
3019 }
3020
3021 // ofs and limit are use for multi-block byte array.
3022 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3023 address generate_sha1_implCompress(bool multi_block, const char *name) {
3024 __ align(CodeEntryAlignment);
3025 StubCodeMark mark(this, "StubRoutines", name);
3026 address start = __ pc();
3027
3028 Register buf = rax;
3029 Register state = rdx;
3030 Register ofs = rcx;
3031 Register limit = rdi;
3032
3033 const Address buf_param(rbp, 8 + 0);
3034 const Address state_param(rbp, 8 + 4);
3035 const Address ofs_param(rbp, 8 + 8);
3036 const Address limit_param(rbp, 8 + 12);
3037
3038 const XMMRegister abcd = xmm0;
3039 const XMMRegister e0 = xmm1;
3040 const XMMRegister e1 = xmm2;
3041 const XMMRegister msg0 = xmm3;
3042
3043 const XMMRegister msg1 = xmm4;
3044 const XMMRegister msg2 = xmm5;
3045 const XMMRegister msg3 = xmm6;
3046 const XMMRegister shuf_mask = xmm7;
3047
3048 __ enter();
3049 __ subptr(rsp, 8 * wordSize);
3050 handleSOERegisters(true /*saving*/);
3051
3052 __ movptr(buf, buf_param);
3053 __ movptr(state, state_param);
3054 if (multi_block) {
3055 __ movptr(ofs, ofs_param);
3056 __ movptr(limit, limit_param);
3057 }
3058
3059 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3060 buf, state, ofs, limit, rsp, multi_block);
3061
3062 handleSOERegisters(false /*restoring*/);
3063 __ addptr(rsp, 8 * wordSize);
3064 __ leave();
3065 __ ret(0);
3066 return start;
3067 }
3068
3069 address generate_pshuffle_byte_flip_mask() {
3070 __ align(64);
3071 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3072 address start = __ pc();
3073 __ emit_data(0x00010203, relocInfo::none, 0);
3074 __ emit_data(0x04050607, relocInfo::none, 0);
3075 __ emit_data(0x08090a0b, relocInfo::none, 0);
3076 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3077 return start;
3078 }
3079
3080 // ofs and limit are use for multi-block byte array.
3081 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3082 address generate_sha256_implCompress(bool multi_block, const char *name) {
3083 __ align(CodeEntryAlignment);
3084 StubCodeMark mark(this, "StubRoutines", name);
3085 address start = __ pc();
3086
3087 Register buf = rbx;
3088 Register state = rsi;
3089 Register ofs = rdx;
3090 Register limit = rcx;
3091
3092 const Address buf_param(rbp, 8 + 0);
3093 const Address state_param(rbp, 8 + 4);
3094 const Address ofs_param(rbp, 8 + 8);
3095 const Address limit_param(rbp, 8 + 12);
3096
3097 const XMMRegister msg = xmm0;
3098 const XMMRegister state0 = xmm1;
3099 const XMMRegister state1 = xmm2;
3100 const XMMRegister msgtmp0 = xmm3;
3101
3102 const XMMRegister msgtmp1 = xmm4;
3103 const XMMRegister msgtmp2 = xmm5;
3104 const XMMRegister msgtmp3 = xmm6;
3105 const XMMRegister msgtmp4 = xmm7;
3106
3107 __ enter();
3108 __ subptr(rsp, 8 * wordSize);
3109 handleSOERegisters(true /*saving*/);
3110 __ movptr(buf, buf_param);
3111 __ movptr(state, state_param);
3112 if (multi_block) {
3113 __ movptr(ofs, ofs_param);
3114 __ movptr(limit, limit_param);
3115 }
3116
3117 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3118 buf, state, ofs, limit, rsp, multi_block);
3119
3120 handleSOERegisters(false);
3121 __ addptr(rsp, 8 * wordSize);
3122 __ leave();
3123 __ ret(0);
3124 return start;
3125 }
3126
3127 // byte swap x86 long
3128 address generate_ghash_long_swap_mask() {
3129 __ align(CodeEntryAlignment);
3130 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
3131 address start = __ pc();
3132 __ emit_data(0x0b0a0908, relocInfo::none, 0);
3133 __ emit_data(0x0f0e0d0c, relocInfo::none, 0);
3134 __ emit_data(0x03020100, relocInfo::none, 0);
3135 __ emit_data(0x07060504, relocInfo::none, 0);
3136
3137 return start;
3138 }
3139
3140 // byte swap x86 byte array
3141 address generate_ghash_byte_swap_mask() {
3142 __ align(CodeEntryAlignment);
3143 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
3144 address start = __ pc();
3145 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3146 __ emit_data(0x08090a0b, relocInfo::none, 0);
3147 __ emit_data(0x04050607, relocInfo::none, 0);
3148 __ emit_data(0x00010203, relocInfo::none, 0);
3149 return start;
3150 }
3151
3152 /* Single and multi-block ghash operations */
3153 address generate_ghash_processBlocks() {
3154 assert(UseGHASHIntrinsics, "need GHASH intrinsics and CLMUL support");
3155 __ align(CodeEntryAlignment);
3156 Label L_ghash_loop, L_exit;
3157 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
3158 address start = __ pc();
3159
3160 const Register state = rdi;
3161 const Register subkeyH = rsi;
3162 const Register data = rdx;
3163 const Register blocks = rcx;
3164
3165 const Address state_param(rbp, 8+0);
3166 const Address subkeyH_param(rbp, 8+4);
3167 const Address data_param(rbp, 8+8);
3168 const Address blocks_param(rbp, 8+12);
3169
3170 const XMMRegister xmm_temp0 = xmm0;
3171 const XMMRegister xmm_temp1 = xmm1;
3172 const XMMRegister xmm_temp2 = xmm2;
3173 const XMMRegister xmm_temp3 = xmm3;
3174 const XMMRegister xmm_temp4 = xmm4;
3175 const XMMRegister xmm_temp5 = xmm5;
3176 const XMMRegister xmm_temp6 = xmm6;
3177 const XMMRegister xmm_temp7 = xmm7;
3178
3179 __ enter();
3180 handleSOERegisters(true); // Save registers
3181
3182 __ movptr(state, state_param);
3183 __ movptr(subkeyH, subkeyH_param);
3184 __ movptr(data, data_param);
3185 __ movptr(blocks, blocks_param);
3186
3187 __ movdqu(xmm_temp0, Address(state, 0));
3188 __ pshufb(xmm_temp0, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3189
3190 __ movdqu(xmm_temp1, Address(subkeyH, 0));
3191 __ pshufb(xmm_temp1, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3192
3193 __ BIND(L_ghash_loop);
3194 __ movdqu(xmm_temp2, Address(data, 0));
3195 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
3196
3197 __ pxor(xmm_temp0, xmm_temp2);
3198
3199 //
3200 // Multiply with the hash key
3201 //
3202 __ movdqu(xmm_temp3, xmm_temp0);
3203 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
3204 __ movdqu(xmm_temp4, xmm_temp0);
3205 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
3206
3207 __ movdqu(xmm_temp5, xmm_temp0);
3208 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
3209 __ movdqu(xmm_temp6, xmm_temp0);
3210 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
3211
3212 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
3213
3214 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
3215 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
3216 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
3217 __ pxor(xmm_temp3, xmm_temp5);
3218 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
3219 // of the carry-less multiplication of
3220 // xmm0 by xmm1.
3221
3222 // We shift the result of the multiplication by one bit position
3223 // to the left to cope for the fact that the bits are reversed.
3224 __ movdqu(xmm_temp7, xmm_temp3);
3225 __ movdqu(xmm_temp4, xmm_temp6);
3226 __ pslld (xmm_temp3, 1);
3227 __ pslld(xmm_temp6, 1);
3228 __ psrld(xmm_temp7, 31);
3229 __ psrld(xmm_temp4, 31);
3230 __ movdqu(xmm_temp5, xmm_temp7);
3231 __ pslldq(xmm_temp4, 4);
3232 __ pslldq(xmm_temp7, 4);
3233 __ psrldq(xmm_temp5, 12);
3234 __ por(xmm_temp3, xmm_temp7);
3235 __ por(xmm_temp6, xmm_temp4);
3236 __ por(xmm_temp6, xmm_temp5);
3237
3238 //
3239 // First phase of the reduction
3240 //
3241 // Move xmm3 into xmm4, xmm5, xmm7 in order to perform the shifts
3242 // independently.
3243 __ movdqu(xmm_temp7, xmm_temp3);
3244 __ movdqu(xmm_temp4, xmm_temp3);
3245 __ movdqu(xmm_temp5, xmm_temp3);
3246 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
3247 __ pslld(xmm_temp4, 30); // packed right shift shifting << 30
3248 __ pslld(xmm_temp5, 25); // packed right shift shifting << 25
3249 __ pxor(xmm_temp7, xmm_temp4); // xor the shifted versions
3250 __ pxor(xmm_temp7, xmm_temp5);
3251 __ movdqu(xmm_temp4, xmm_temp7);
3252 __ pslldq(xmm_temp7, 12);
3253 __ psrldq(xmm_temp4, 4);
3254 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
3255
3256 //
3257 // Second phase of the reduction
3258 //
3259 // Make 3 copies of xmm3 in xmm2, xmm5, xmm7 for doing these
3260 // shift operations.
3261 __ movdqu(xmm_temp2, xmm_temp3);
3262 __ movdqu(xmm_temp7, xmm_temp3);
3263 __ movdqu(xmm_temp5, xmm_temp3);
3264 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
3265 __ psrld(xmm_temp7, 2); // packed left shifting >> 2
3266 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
3267 __ pxor(xmm_temp2, xmm_temp7); // xor the shifted versions
3268 __ pxor(xmm_temp2, xmm_temp5);
3269 __ pxor(xmm_temp2, xmm_temp4);
3270 __ pxor(xmm_temp3, xmm_temp2);
3271 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
3272
3273 __ decrement(blocks);
3274 __ jcc(Assembler::zero, L_exit);
3275 __ movdqu(xmm_temp0, xmm_temp6);
3276 __ addptr(data, 16);
3277 __ jmp(L_ghash_loop);
3278
3279 __ BIND(L_exit);
3280 // Byte swap 16-byte result
3281 __ pshufb(xmm_temp6, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3282 __ movdqu(Address(state, 0), xmm_temp6); // store the result
3283
3284 handleSOERegisters(false); // restore registers
3285 __ leave();
3286 __ ret(0);
3287 return start;
3288 }
3289
3290 /**
3291 * Arguments:
3292 *
3293 * Inputs:
3294 * rsp(4) - int crc
3295 * rsp(8) - byte* buf
3296 * rsp(12) - int length
3297 *
3298 * Ouput:
3299 * rax - int crc result
3300 */
3301 address generate_updateBytesCRC32() {
3302 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3303
3304 __ align(CodeEntryAlignment);
3305 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
3306
3307 address start = __ pc();
3308
3309 const Register crc = rdx; // crc
3310 const Register buf = rsi; // source java byte array address
3311 const Register len = rcx; // length
3312 const Register table = rdi; // crc_table address (reuse register)
3313 const Register tmp = rbx;
3314 assert_different_registers(crc, buf, len, table, tmp, rax);
3315
3316 BLOCK_COMMENT("Entry:");
3317 __ enter(); // required for proper stackwalking of RuntimeStub frame
3318 __ push(rsi);
3319 __ push(rdi);
3320 __ push(rbx);
3321
3322 Address crc_arg(rbp, 8 + 0);
3323 Address buf_arg(rbp, 8 + 4);
3324 Address len_arg(rbp, 8 + 8);
3325
3326 // Load up:
3327 __ movl(crc, crc_arg);
3328 __ movptr(buf, buf_arg);
3329 __ movl(len, len_arg);
3330
3331 __ kernel_crc32(crc, buf, len, table, tmp);
3332
3333 __ movl(rax, crc);
3334 __ pop(rbx);
3335 __ pop(rdi);
3336 __ pop(rsi);
3337 __ vzeroupper();
3338 __ leave(); // required for proper stackwalking of RuntimeStub frame
3339 __ ret(0);
3340
3341 return start;
3342 }
3343
3344 /**
3345 * Arguments:
3346 *
3347 * Inputs:
3348 * rsp(4) - int crc
3349 * rsp(8) - byte* buf
3350 * rsp(12) - int length
3351 * rsp(16) - table_start - optional (present only when doing a library_calll,
3352 * not used by x86 algorithm)
3353 *
3354 * Ouput:
3355 * rax - int crc result
3356 */
3357 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
3358 assert(UseCRC32CIntrinsics, "need SSE4_2");
3359 __ align(CodeEntryAlignment);
3360 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
3361 address start = __ pc();
3362 const Register crc = rax; // crc
3363 const Register buf = rcx; // source java byte array address
3364 const Register len = rdx; // length
3365 const Register d = rbx;
3366 const Register g = rsi;
3367 const Register h = rdi;
3368 const Register empty = 0; // will never be used, in order not
3369 // to change a signature for crc32c_IPL_Alg2_Alt2
3370 // between 64/32 I'm just keeping it here
3371 assert_different_registers(crc, buf, len, d, g, h);
3372
3373 BLOCK_COMMENT("Entry:");
3374 __ enter(); // required for proper stackwalking of RuntimeStub frame
3375 Address crc_arg(rsp, 4 + 4 + 0); // ESP+4 +
3376 // we need to add additional 4 because __ enter
3377 // have just pushed ebp on a stack
3378 Address buf_arg(rsp, 4 + 4 + 4);
3379 Address len_arg(rsp, 4 + 4 + 8);
3380 // Load up:
3381 __ movl(crc, crc_arg);
3382 __ movl(buf, buf_arg);
3383 __ movl(len, len_arg);
3384 __ push(d);
3385 __ push(g);
3386 __ push(h);
3387 __ crc32c_ipl_alg2_alt2(crc, buf, len,
3388 d, g, h,
3389 empty, empty, empty,
3390 xmm0, xmm1, xmm2,
3391 is_pclmulqdq_supported);
3392 __ pop(h);
3393 __ pop(g);
3394 __ pop(d);
3395 __ vzeroupper();
3396 __ leave(); // required for proper stackwalking of RuntimeStub frame
3397 __ ret(0);
3398
3399 return start;
3400 }
3401
3402 address generate_libmExp() {
3403 StubCodeMark mark(this, "StubRoutines", "libmExp");
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_exp(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_libmLog() {
3430 StubCodeMark mark(this, "StubRoutines", "libmLog");
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_log(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_libmLog10() {
3457 StubCodeMark mark(this, "StubRoutines", "libmLog10");
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_log10(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_libmPow() {
3484 StubCodeMark mark(this, "StubRoutines", "libmPow");
3485
3486 address start = __ pc();
3487
3488 const XMMRegister x0 = xmm0;
3489 const XMMRegister x1 = xmm1;
3490 const XMMRegister x2 = xmm2;
3491 const XMMRegister x3 = xmm3;
3492
3493 const XMMRegister x4 = xmm4;
3494 const XMMRegister x5 = xmm5;
3495 const XMMRegister x6 = xmm6;
3496 const XMMRegister x7 = xmm7;
3497
3498 const Register tmp = rbx;
3499
3500 BLOCK_COMMENT("Entry:");
3501 __ enter(); // required for proper stackwalking of RuntimeStub frame
3502 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3503 __ leave(); // required for proper stackwalking of RuntimeStub frame
3504 __ ret(0);
3505
3506 return start;
3507
3508 }
3509
3510 address generate_libm_reduce_pi04l() {
3511 StubCodeMark mark(this, "StubRoutines", "libm_reduce_pi04l");
3512
3513 address start = __ pc();
3514
3515 BLOCK_COMMENT("Entry:");
3516 __ libm_reduce_pi04l(rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3517
3518 return start;
3519
3520 }
3521
3522 address generate_libm_sin_cos_huge() {
3523 StubCodeMark mark(this, "StubRoutines", "libm_sin_cos_huge");
3524
3525 address start = __ pc();
3526
3527 const XMMRegister x0 = xmm0;
3528 const XMMRegister x1 = xmm1;
3529
3530 BLOCK_COMMENT("Entry:");
3531 __ libm_sincos_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3532
3533 return start;
3534
3535 }
3536
3537 address generate_libmSin() {
3538 StubCodeMark mark(this, "StubRoutines", "libmSin");
3539
3540 address start = __ pc();
3541
3542 const XMMRegister x0 = xmm0;
3543 const XMMRegister x1 = xmm1;
3544 const XMMRegister x2 = xmm2;
3545 const XMMRegister x3 = xmm3;
3546
3547 const XMMRegister x4 = xmm4;
3548 const XMMRegister x5 = xmm5;
3549 const XMMRegister x6 = xmm6;
3550 const XMMRegister x7 = xmm7;
3551
3552 BLOCK_COMMENT("Entry:");
3553 __ enter(); // required for proper stackwalking of RuntimeStub frame
3554 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rdx);
3555 __ leave(); // required for proper stackwalking of RuntimeStub frame
3556 __ ret(0);
3557
3558 return start;
3559
3560 }
3561
3562 address generate_libmCos() {
3563 StubCodeMark mark(this, "StubRoutines", "libmCos");
3564
3565 address start = __ pc();
3566
3567 const XMMRegister x0 = xmm0;
3568 const XMMRegister x1 = xmm1;
3569 const XMMRegister x2 = xmm2;
3570 const XMMRegister x3 = xmm3;
3571
3572 const XMMRegister x4 = xmm4;
3573 const XMMRegister x5 = xmm5;
3574 const XMMRegister x6 = xmm6;
3575 const XMMRegister x7 = xmm7;
3576
3577 const Register tmp = rbx;
3578
3579 BLOCK_COMMENT("Entry:");
3580 __ enter(); // required for proper stackwalking of RuntimeStub frame
3581 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3582 __ leave(); // required for proper stackwalking of RuntimeStub frame
3583 __ ret(0);
3584
3585 return start;
3586
3587 }
3588
3589 address generate_libm_tan_cot_huge() {
3590 StubCodeMark mark(this, "StubRoutines", "libm_tan_cot_huge");
3591
3592 address start = __ pc();
3593
3594 const XMMRegister x0 = xmm0;
3595 const XMMRegister x1 = xmm1;
3596
3597 BLOCK_COMMENT("Entry:");
3598 __ libm_tancot_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3599
3600 return start;
3601
3602 }
3603
3604 address generate_libmTan() {
3605 StubCodeMark mark(this, "StubRoutines", "libmTan");
3606
3607 address start = __ pc();
3608
3609 const XMMRegister x0 = xmm0;
3610 const XMMRegister x1 = xmm1;
3611 const XMMRegister x2 = xmm2;
3612 const XMMRegister x3 = xmm3;
3613
3614 const XMMRegister x4 = xmm4;
3615 const XMMRegister x5 = xmm5;
3616 const XMMRegister x6 = xmm6;
3617 const XMMRegister x7 = xmm7;
3618
3619 const Register tmp = rbx;
3620
3621 BLOCK_COMMENT("Entry:");
3622 __ enter(); // required for proper stackwalking of RuntimeStub frame
3623 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3624 __ leave(); // required for proper stackwalking of RuntimeStub frame
3625 __ ret(0);
3626
3627 return start;
3628
3629 }
3630
3631 // Safefetch stubs.
3632 void generate_safefetch(const char* name, int size, address* entry,
3633 address* fault_pc, address* continuation_pc) {
3634 // safefetch signatures:
3635 // int SafeFetch32(int* adr, int errValue);
3636 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3637
3638 StubCodeMark mark(this, "StubRoutines", name);
3639
3640 // Entry point, pc or function descriptor.
3641 *entry = __ pc();
3642
3643 __ movl(rax, Address(rsp, 0x8));
3644 __ movl(rcx, Address(rsp, 0x4));
3645 // Load *adr into eax, may fault.
3646 *fault_pc = __ pc();
3647 switch (size) {
3648 case 4:
3649 // int32_t
3650 __ movl(rax, Address(rcx, 0));
3651 break;
3652 case 8:
3653 // int64_t
3654 Unimplemented();
3655 break;
3656 default:
3657 ShouldNotReachHere();
3658 }
3659
3660 // Return errValue or *adr.
3661 *continuation_pc = __ pc();
3662 __ ret(0);
3663 }
3664
3665 public:
3666 // Information about frame layout at time of blocking runtime call.
3667 // Note that we only have to preserve callee-saved registers since
3668 // the compilers are responsible for supplying a continuation point
3669 // if they expect all registers to be preserved.
3670 enum layout {
3671 thread_off, // last_java_sp
3672 arg1_off,
3673 arg2_off,
3674 rbp_off, // callee saved register
3675 ret_pc,
3676 framesize
3677 };
3678
3679 private:
3680
3681 #undef __
3682 #define __ masm->
3683
3684 //------------------------------------------------------------------------------------------------------------------------
3685 // Continuation point for throwing of implicit exceptions that are not handled in
3686 // the current activation. Fabricates an exception oop and initiates normal
3687 // exception dispatching in this frame.
3688 //
3689 // Previously the compiler (c2) allowed for callee save registers on Java calls.
3690 // This is no longer true after adapter frames were removed but could possibly
3691 // be brought back in the future if the interpreter code was reworked and it
3692 // was deemed worthwhile. The comment below was left to describe what must
3693 // happen here if callee saves were resurrected. As it stands now this stub
3694 // could actually be a vanilla BufferBlob and have now oopMap at all.
3695 // Since it doesn't make much difference we've chosen to leave it the
3696 // way it was in the callee save days and keep the comment.
3697
3698 // If we need to preserve callee-saved values we need a callee-saved oop map and
3699 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
3700 // If the compiler needs all registers to be preserved between the fault
3701 // point and the exception handler then it must assume responsibility for that in
3702 // AbstractCompiler::continuation_for_implicit_null_exception or
3703 // continuation_for_implicit_division_by_zero_exception. All other implicit
3704 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
3705 // either at call sites or otherwise assume that stack unwinding will be initiated,
3706 // so caller saved registers were assumed volatile in the compiler.
3707 address generate_throw_exception(const char* name, address runtime_entry,
3708 Register arg1 = noreg, Register arg2 = noreg) {
3709
3710 int insts_size = 256;
3711 int locs_size = 32;
3712
3713 CodeBuffer code(name, insts_size, locs_size);
3714 OopMapSet* oop_maps = new OopMapSet();
3715 MacroAssembler* masm = new MacroAssembler(&code);
3716
3717 address start = __ pc();
3718
3719 // This is an inlined and slightly modified version of call_VM
3720 // which has the ability to fetch the return PC out of
3721 // thread-local storage and also sets up last_Java_sp slightly
3722 // differently than the real call_VM
3723 Register java_thread = rbx;
3724 __ get_thread(java_thread);
3725
3726 __ enter(); // required for proper stackwalking of RuntimeStub frame
3727
3728 // pc and rbp, already pushed
3729 __ subptr(rsp, (framesize-2) * wordSize); // prolog
3730
3731 // Frame is now completed as far as size and linkage.
3732
3733 int frame_complete = __ pc() - start;
3734
3735 // push java thread (becomes first argument of C function)
3736 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
3737 if (arg1 != noreg) {
3738 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
3739 }
3740 if (arg2 != noreg) {
3741 assert(arg1 != noreg, "missing reg arg");
3742 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
3743 }
3744
3745 // Set up last_Java_sp and last_Java_fp
3746 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
3747
3748 // Call runtime
3749 BLOCK_COMMENT("call runtime_entry");
3750 __ call(RuntimeAddress(runtime_entry));
3751 // Generate oop map
3752 OopMap* map = new OopMap(framesize, 0);
3753 oop_maps->add_gc_map(__ pc() - start, map);
3754
3755 // restore the thread (cannot use the pushed argument since arguments
3756 // may be overwritten by C code generated by an optimizing compiler);
3757 // however can use the register value directly if it is callee saved.
3758 __ get_thread(java_thread);
3759
3760 __ reset_last_Java_frame(java_thread, true);
3761
3762 __ leave(); // required for proper stackwalking of RuntimeStub frame
3763
3764 // check for pending exceptions
3765 #ifdef ASSERT
3766 Label L;
3767 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3768 __ jcc(Assembler::notEqual, L);
3769 __ should_not_reach_here();
3770 __ bind(L);
3771 #endif /* ASSERT */
3772 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3773
3774
3775 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
3776 return stub->entry_point();
3777 }
3778
3779
3780 void create_control_words() {
3781 // Round to nearest, 53-bit mode, exceptions masked
3782 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
3783 // Round to zero, 53-bit mode, exception mased
3784 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
3785 // Round to nearest, 24-bit mode, exceptions masked
3786 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
3787 // Round to nearest, 64-bit mode, exceptions masked
3788 StubRoutines::_mxcsr_std = 0x1F80;
3789 // Note: the following two constants are 80-bit values
3790 // layout is critical for correct loading by FPU.
3791 // Bias for strict fp multiply/divide
3792 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
3793 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
3794 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
3795 // Un-Bias for strict fp multiply/divide
3796 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
3797 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
3798 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
3799 }
3800
3801 //---------------------------------------------------------------------------
3802 // Initialization
3803
3804 void generate_initial() {
3805 // Generates all stubs and initializes the entry points
3806
3807 //------------------------------------------------------------------------------------------------------------------------
3808 // entry points that exist in all platforms
3809 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
3810 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
3811 StubRoutines::_forward_exception_entry = generate_forward_exception();
3812
3813 StubRoutines::_call_stub_entry =
3814 generate_call_stub(StubRoutines::_call_stub_return_address);
3815 // is referenced by megamorphic call
3816 StubRoutines::_catch_exception_entry = generate_catch_exception();
3817
3818 // These are currently used by Solaris/Intel
3819 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3820
3821 // platform dependent
3822 create_control_words();
3823
3824 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3825 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
3826 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
3827 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
3828 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
3829 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
3830
3831 // Build this early so it's available for the interpreter
3832 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception",
3833 CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
3834 StubRoutines::_throw_delayed_StackOverflowError_entry = generate_throw_exception("delayed StackOverflowError throw_exception",
3835 CAST_FROM_FN_PTR(address, SharedRuntime::throw_delayed_StackOverflowError));
3836
3837 if (UseCRC32Intrinsics) {
3838 // set table address before stub generation which use it
3839 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3840 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3841 }
3842
3843 if (UseCRC32CIntrinsics) {
3844 bool supports_clmul = VM_Version::supports_clmul();
3845 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
3846 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
3847 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
3848 }
3849 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
3850 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3851 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3852 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3853 StubRoutines::x86::_L_2il0floatpacket_0_adr = (address)StubRoutines::x86::_L_2il0floatpacket_0;
3854 StubRoutines::x86::_Pi4Inv_adr = (address)StubRoutines::x86::_Pi4Inv;
3855 StubRoutines::x86::_Pi4x3_adr = (address)StubRoutines::x86::_Pi4x3;
3856 StubRoutines::x86::_Pi4x4_adr = (address)StubRoutines::x86::_Pi4x4;
3857 StubRoutines::x86::_ones_adr = (address)StubRoutines::x86::_ones;
3858 }
3859 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3860 StubRoutines::_dexp = generate_libmExp();
3861 }
3862 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3863 StubRoutines::_dlog = generate_libmLog();
3864 }
3865 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3866 StubRoutines::_dlog10 = generate_libmLog10();
3867 }
3868 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3869 StubRoutines::_dpow = generate_libmPow();
3870 }
3871 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3872 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3873 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3874 StubRoutines::_dlibm_reduce_pi04l = generate_libm_reduce_pi04l();
3875 }
3876 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3877 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3878 StubRoutines::_dlibm_sin_cos_huge = generate_libm_sin_cos_huge();
3879 }
3880 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3881 StubRoutines::_dsin = generate_libmSin();
3882 }
3883 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3884 StubRoutines::_dcos = generate_libmCos();
3885 }
3886 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3887 StubRoutines::_dlibm_tan_cot_huge = generate_libm_tan_cot_huge();
3888 StubRoutines::_dtan = generate_libmTan();
3889 }
3890 }
3891 }
3892
3893 void generate_all() {
3894 // Generates all stubs and initializes the entry points
3895
3896 // These entry points require SharedInfo::stack0 to be set up in non-core builds
3897 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
3898 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
3899 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
3900 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
3901
3902 //------------------------------------------------------------------------------------------------------------------------
3903 // entry points that are platform specific
3904
3905 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF);
3906 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x80000000);
3907 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask_long_double("vector_double_sign_mask", 0x7FFFFFFF, 0xFFFFFFFF);
3908 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask_long_double("vector_double_sign_flip", 0x80000000, 0x00000000);
3909 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff);
3910 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
3911 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask_long_double("vector_long_sign_mask", 0x80000000, 0x00000000);
3912
3913 // support for verify_oop (must happen after universe_init)
3914 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3915
3916 // arraycopy stubs used by compilers
3917 generate_arraycopy_stubs();
3918
3919 // don't bother generating these AES intrinsic stubs unless global flag is set
3920 if (UseAESIntrinsics) {
3921 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
3922
3923 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3924 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3925 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3926 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
3927 }
3928
3929 if (UseAESCTRIntrinsics) {
3930 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
3931 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
3932 }
3933
3934 if (UseSHA1Intrinsics) {
3935 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
3936 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
3937 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
3938 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
3939 }
3940 if (UseSHA256Intrinsics) {
3941 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
3942 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
3943 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
3944 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
3945 }
3946
3947 // Generate GHASH intrinsics code
3948 if (UseGHASHIntrinsics) {
3949 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
3950 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
3951 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
3952 }
3953
3954 // Safefetch stubs.
3955 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3956 &StubRoutines::_safefetch32_fault_pc,
3957 &StubRoutines::_safefetch32_continuation_pc);
3958 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
3959 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
3960 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3961 }
3962
3963
3964 public:
3965 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3966 if (all) {
3967 generate_all();
3968 } else {
3969 generate_initial();
3970 }
3971 }
3972 }; // end class declaration
3973
3974 #define UCM_TABLE_MAX_ENTRIES 8
3975 void StubGenerator_generate(CodeBuffer* code, bool all) {
3976 if (UnsafeCopyMemory::_table == NULL) {
3977 UnsafeCopyMemory::create_table(UCM_TABLE_MAX_ENTRIES);
3978 }
3979 StubGenerator g(code, all);
3980 }