1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "gc/shared/barrierSet.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "gc/shared/barrierSetNMethod.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "memory/universe.hpp"
33 #include "nativeInst_x86.hpp"
34 #include "oops/instanceOop.hpp"
35 #include "oops/method.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/methodHandles.hpp"
39 #include "runtime/frame.inline.hpp"
40 #include "runtime/handles.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubCodeGenerator.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/thread.inline.hpp"
45 #ifdef COMPILER2
46 #include "opto/runtime.hpp"
47 #endif
48
49 // Declaration and definition of StubGenerator (no .hpp file).
50 // For a more detailed description of the stub routine structure
51 // see the comment in stubRoutines.hpp
52
53 #define __ _masm->
54 #define a__ ((Assembler*)_masm)->
55
56 #ifdef PRODUCT
57 #define BLOCK_COMMENT(str) /* nothing */
58 #else
59 #define BLOCK_COMMENT(str) __ block_comment(str)
60 #endif
61
62 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
63
64 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
65 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
66
67 // -------------------------------------------------------------------------------------------------------------------------
68 // Stub Code definitions
69
70 class StubGenerator: public StubCodeGenerator {
71 private:
72
73 #ifdef PRODUCT
74 #define inc_counter_np(counter) ((void)0)
75 #else
76 void inc_counter_np_(int& counter) {
77 __ incrementl(ExternalAddress((address)&counter));
78 }
79 #define inc_counter_np(counter) \
80 BLOCK_COMMENT("inc_counter " #counter); \
81 inc_counter_np_(counter);
82 #endif //PRODUCT
83
84 void inc_copy_counter_np(BasicType t) {
85 #ifndef PRODUCT
86 switch (t) {
87 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
88 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
89 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
90 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
91 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
92 default: ShouldNotReachHere();
93 }
94 #endif //PRODUCT
95 }
96
97 //------------------------------------------------------------------------------------------------------------------------
98 // Call stubs are used to call Java from C
99 //
100 // [ return_from_Java ] <--- rsp
101 // [ argument word n ]
102 // ...
103 // -N [ argument word 1 ]
104 // -7 [ Possible padding for stack alignment ]
105 // -6 [ Possible padding for stack alignment ]
106 // -5 [ Possible padding for stack alignment ]
107 // -4 [ mxcsr save ] <--- rsp_after_call
108 // -3 [ saved rbx, ]
109 // -2 [ saved rsi ]
110 // -1 [ saved rdi ]
111 // 0 [ saved rbp, ] <--- rbp,
112 // 1 [ return address ]
113 // 2 [ ptr. to call wrapper ]
114 // 3 [ result ]
115 // 4 [ result_type ]
116 // 5 [ method ]
117 // 6 [ entry_point ]
118 // 7 [ parameters ]
119 // 8 [ parameter_size ]
120 // 9 [ thread ]
121
122
123 address generate_call_stub(address& return_address) {
124 StubCodeMark mark(this, "StubRoutines", "call_stub");
125 address start = __ pc();
126
127 // stub code parameters / addresses
128 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
129 bool sse_save = false;
130 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
131 const int locals_count_in_bytes (4*wordSize);
132 const Address mxcsr_save (rbp, -4 * wordSize);
133 const Address saved_rbx (rbp, -3 * wordSize);
134 const Address saved_rsi (rbp, -2 * wordSize);
135 const Address saved_rdi (rbp, -1 * wordSize);
136 const Address result (rbp, 3 * wordSize);
137 const Address result_type (rbp, 4 * wordSize);
138 const Address method (rbp, 5 * wordSize);
139 const Address entry_point (rbp, 6 * wordSize);
140 const Address parameters (rbp, 7 * wordSize);
141 const Address parameter_size(rbp, 8 * wordSize);
142 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
143 sse_save = UseSSE > 0;
144
145 // stub code
146 __ enter();
147 __ movptr(rcx, parameter_size); // parameter counter
148 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
149 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
150 __ subptr(rsp, rcx);
151 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
152
153 // save rdi, rsi, & rbx, according to C calling conventions
154 __ movptr(saved_rdi, rdi);
155 __ movptr(saved_rsi, rsi);
156 __ movptr(saved_rbx, rbx);
157
158 // save and initialize %mxcsr
159 if (sse_save) {
160 Label skip_ldmx;
161 __ stmxcsr(mxcsr_save);
162 __ movl(rax, mxcsr_save);
163 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
164 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
165 __ cmp32(rax, mxcsr_std);
166 __ jcc(Assembler::equal, skip_ldmx);
167 __ ldmxcsr(mxcsr_std);
168 __ bind(skip_ldmx);
169 }
170
171 // make sure the control word is correct.
172 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
173
174 #ifdef ASSERT
175 // make sure we have no pending exceptions
176 { Label L;
177 __ movptr(rcx, thread);
178 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
179 __ jcc(Assembler::equal, L);
180 __ stop("StubRoutines::call_stub: entered with pending exception");
181 __ bind(L);
182 }
183 #endif
184
185 // pass parameters if any
186 BLOCK_COMMENT("pass parameters if any");
187 Label parameters_done;
188 __ movl(rcx, parameter_size); // parameter counter
189 __ testl(rcx, rcx);
190 __ jcc(Assembler::zero, parameters_done);
191
192 // parameter passing loop
193
194 Label loop;
195 // Copy Java parameters in reverse order (receiver last)
196 // Note that the argument order is inverted in the process
197 // source is rdx[rcx: N-1..0]
198 // dest is rsp[rbx: 0..N-1]
199
200 __ movptr(rdx, parameters); // parameter pointer
201 __ xorptr(rbx, rbx);
202
203 __ BIND(loop);
204
205 // get parameter
206 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
207 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
208 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
209 __ increment(rbx);
210 __ decrement(rcx);
211 __ jcc(Assembler::notZero, loop);
212
213 // call Java function
214 __ BIND(parameters_done);
215 __ movptr(rbx, method); // get Method*
216 __ movptr(rax, entry_point); // get entry_point
217 __ mov(rsi, rsp); // set sender sp
218 BLOCK_COMMENT("call Java function");
219 __ call(rax);
220
221 BLOCK_COMMENT("call_stub_return_address:");
222 return_address = __ pc();
223
224 #ifdef COMPILER2
225 {
226 Label L_skip;
227 if (UseSSE >= 2) {
228 __ verify_FPU(0, "call_stub_return");
229 } else {
230 for (int i = 1; i < 8; i++) {
231 __ ffree(i);
232 }
233
234 // UseSSE <= 1 so double result should be left on TOS
235 __ movl(rsi, result_type);
236 __ cmpl(rsi, T_DOUBLE);
237 __ jcc(Assembler::equal, L_skip);
238 if (UseSSE == 0) {
239 // UseSSE == 0 so float result should be left on TOS
240 __ cmpl(rsi, T_FLOAT);
241 __ jcc(Assembler::equal, L_skip);
242 }
243 __ ffree(0);
244 }
245 __ BIND(L_skip);
246 }
247 #endif // COMPILER2
248
249 // store result depending on type
250 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
251 __ movptr(rdi, result);
252 Label is_long, is_float, is_double, exit;
253 __ movl(rsi, result_type);
254 __ cmpl(rsi, T_LONG);
255 __ jcc(Assembler::equal, is_long);
256 __ cmpl(rsi, T_FLOAT);
257 __ jcc(Assembler::equal, is_float);
258 __ cmpl(rsi, T_DOUBLE);
259 __ jcc(Assembler::equal, is_double);
260
261 // handle T_INT case
262 __ movl(Address(rdi, 0), rax);
263 __ BIND(exit);
264
265 // check that FPU stack is empty
266 __ verify_FPU(0, "generate_call_stub");
267
268 // pop parameters
269 __ lea(rsp, rsp_after_call);
270
271 // restore %mxcsr
272 if (sse_save) {
273 __ ldmxcsr(mxcsr_save);
274 }
275
276 // restore rdi, rsi and rbx,
277 __ movptr(rbx, saved_rbx);
278 __ movptr(rsi, saved_rsi);
279 __ movptr(rdi, saved_rdi);
280 __ addptr(rsp, 4*wordSize);
281
282 // return
283 __ pop(rbp);
284 __ ret(0);
285
286 // handle return types different from T_INT
287 __ BIND(is_long);
288 __ movl(Address(rdi, 0 * wordSize), rax);
289 __ movl(Address(rdi, 1 * wordSize), rdx);
290 __ jmp(exit);
291
292 __ BIND(is_float);
293 // interpreter uses xmm0 for return values
294 if (UseSSE >= 1) {
295 __ movflt(Address(rdi, 0), xmm0);
296 } else {
297 __ fstp_s(Address(rdi, 0));
298 }
299 __ jmp(exit);
300
301 __ BIND(is_double);
302 // interpreter uses xmm0 for return values
303 if (UseSSE >= 2) {
304 __ movdbl(Address(rdi, 0), xmm0);
305 } else {
306 __ fstp_d(Address(rdi, 0));
307 }
308 __ jmp(exit);
309
310 return start;
311 }
312
313
314 //------------------------------------------------------------------------------------------------------------------------
315 // Return point for a Java call if there's an exception thrown in Java code.
316 // The exception is caught and transformed into a pending exception stored in
317 // JavaThread that can be tested from within the VM.
318 //
319 // Note: Usually the parameters are removed by the callee. In case of an exception
320 // crossing an activation frame boundary, that is not the case if the callee
321 // is compiled code => need to setup the rsp.
322 //
323 // rax,: exception oop
324
325 address generate_catch_exception() {
326 StubCodeMark mark(this, "StubRoutines", "catch_exception");
327 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
328 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
329 address start = __ pc();
330
331 // get thread directly
332 __ movptr(rcx, thread);
333 #ifdef ASSERT
334 // verify that threads correspond
335 { Label L;
336 __ get_thread(rbx);
337 __ cmpptr(rbx, rcx);
338 __ jcc(Assembler::equal, L);
339 __ stop("StubRoutines::catch_exception: threads must correspond");
340 __ bind(L);
341 }
342 #endif
343 // set pending exception
344 __ verify_oop(rax);
345 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
346 __ lea(Address(rcx, Thread::exception_file_offset ()),
347 ExternalAddress((address)__FILE__));
348 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
349 // complete return to VM
350 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
351 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
352
353 return start;
354 }
355
356
357 //------------------------------------------------------------------------------------------------------------------------
358 // Continuation point for runtime calls returning with a pending exception.
359 // The pending exception check happened in the runtime or native call stub.
360 // The pending exception in Thread is converted into a Java-level exception.
361 //
362 // Contract with Java-level exception handlers:
363 // rax: exception
364 // rdx: throwing pc
365 //
366 // NOTE: At entry of this stub, exception-pc must be on stack !!
367
368 address generate_forward_exception() {
369 StubCodeMark mark(this, "StubRoutines", "forward exception");
370 address start = __ pc();
371 const Register thread = rcx;
372
373 // other registers used in this stub
374 const Register exception_oop = rax;
375 const Register handler_addr = rbx;
376 const Register exception_pc = rdx;
377
378 // Upon entry, the sp points to the return address returning into Java
379 // (interpreted or compiled) code; i.e., the return address becomes the
380 // throwing pc.
381 //
382 // Arguments pushed before the runtime call are still on the stack but
383 // the exception handler will reset the stack pointer -> ignore them.
384 // A potential result in registers can be ignored as well.
385
386 #ifdef ASSERT
387 // make sure this code is only executed if there is a pending exception
388 { Label L;
389 __ get_thread(thread);
390 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
391 __ jcc(Assembler::notEqual, L);
392 __ stop("StubRoutines::forward exception: no pending exception (1)");
393 __ bind(L);
394 }
395 #endif
396
397 // compute exception handler into rbx,
398 __ get_thread(thread);
399 __ movptr(exception_pc, Address(rsp, 0));
400 BLOCK_COMMENT("call exception_handler_for_return_address");
401 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
402 __ mov(handler_addr, rax);
403
404 // setup rax & rdx, remove return address & clear pending exception
405 __ get_thread(thread);
406 __ pop(exception_pc);
407 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
408 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
409
410 #ifdef ASSERT
411 // make sure exception is set
412 { Label L;
413 __ testptr(exception_oop, exception_oop);
414 __ jcc(Assembler::notEqual, L);
415 __ stop("StubRoutines::forward exception: no pending exception (2)");
416 __ bind(L);
417 }
418 #endif
419
420 // Verify that there is really a valid exception in RAX.
421 __ verify_oop(exception_oop);
422
423 // continue at exception handler (return address removed)
424 // rax: exception
425 // rbx: exception handler
426 // rdx: throwing pc
427 __ jmp(handler_addr);
428
429 return start;
430 }
431
432
433 //----------------------------------------------------------------------------------------------------
434 // Implementation of int32_t atomic_xchg(int32_t exchange_value, volatile int32_t* dest)
435 // used by Atomic::xchg(volatile int32_t* dest, int32_t exchange_value)
436 //
437 // xchg exists as far back as 8086, lock needed for MP only
438 // Stack layout immediately after call:
439 //
440 // 0 [ret addr ] <--- rsp
441 // 1 [ ex ]
442 // 2 [ dest ]
443 //
444 // Result: *dest <- ex, return (old *dest)
445 //
446 // Note: win32 does not currently use this code
447
448 address generate_atomic_xchg() {
449 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
450 address start = __ pc();
451
452 __ push(rdx);
453 Address exchange(rsp, 2 * wordSize);
454 Address dest_addr(rsp, 3 * wordSize);
455 __ movl(rax, exchange);
456 __ movptr(rdx, dest_addr);
457 __ xchgl(rax, Address(rdx, 0));
458 __ pop(rdx);
459 __ ret(0);
460
461 return start;
462 }
463
464 //----------------------------------------------------------------------------------------------------
465 // Support for void verify_mxcsr()
466 //
467 // This routine is used with -Xcheck:jni to verify that native
468 // JNI code does not return to Java code without restoring the
469 // MXCSR register to our expected state.
470
471
472 address generate_verify_mxcsr() {
473 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
474 address start = __ pc();
475
476 const Address mxcsr_save(rsp, 0);
477
478 if (CheckJNICalls && UseSSE > 0 ) {
479 Label ok_ret;
480 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
481 __ push(rax);
482 __ subptr(rsp, wordSize); // allocate a temp location
483 __ stmxcsr(mxcsr_save);
484 __ movl(rax, mxcsr_save);
485 __ andl(rax, MXCSR_MASK);
486 __ cmp32(rax, mxcsr_std);
487 __ jcc(Assembler::equal, ok_ret);
488
489 __ warn("MXCSR changed by native JNI code.");
490
491 __ ldmxcsr(mxcsr_std);
492
493 __ bind(ok_ret);
494 __ addptr(rsp, wordSize);
495 __ pop(rax);
496 }
497
498 __ ret(0);
499
500 return start;
501 }
502
503
504 //---------------------------------------------------------------------------
505 // Support for void verify_fpu_cntrl_wrd()
506 //
507 // This routine is used with -Xcheck:jni to verify that native
508 // JNI code does not return to Java code without restoring the
509 // FP control word to our expected state.
510
511 address generate_verify_fpu_cntrl_wrd() {
512 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
513 address start = __ pc();
514
515 const Address fpu_cntrl_wrd_save(rsp, 0);
516
517 if (CheckJNICalls) {
518 Label ok_ret;
519 __ push(rax);
520 __ subptr(rsp, wordSize); // allocate a temp location
521 __ fnstcw(fpu_cntrl_wrd_save);
522 __ movl(rax, fpu_cntrl_wrd_save);
523 __ andl(rax, FPU_CNTRL_WRD_MASK);
524 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
525 __ cmp32(rax, fpu_std);
526 __ jcc(Assembler::equal, ok_ret);
527
528 __ warn("Floating point control word changed by native JNI code.");
529
530 __ fldcw(fpu_std);
531
532 __ bind(ok_ret);
533 __ addptr(rsp, wordSize);
534 __ pop(rax);
535 }
536
537 __ ret(0);
538
539 return start;
540 }
541
542 //---------------------------------------------------------------------------
543 // Wrapper for slow-case handling of double-to-integer conversion
544 // d2i or f2i fast case failed either because it is nan or because
545 // of under/overflow.
546 // Input: FPU TOS: float value
547 // Output: rax, (rdx): integer (long) result
548
549 address generate_d2i_wrapper(BasicType t, address fcn) {
550 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
551 address start = __ pc();
552
553 // Capture info about frame layout
554 enum layout { FPUState_off = 0,
555 rbp_off = FPUStateSizeInWords,
556 rdi_off,
557 rsi_off,
558 rcx_off,
559 rbx_off,
560 saved_argument_off,
561 saved_argument_off2, // 2nd half of double
562 framesize
563 };
564
565 assert(FPUStateSizeInWords == 27, "update stack layout");
566
567 // Save outgoing argument to stack across push_FPU_state()
568 __ subptr(rsp, wordSize * 2);
569 __ fstp_d(Address(rsp, 0));
570
571 // Save CPU & FPU state
572 __ push(rbx);
573 __ push(rcx);
574 __ push(rsi);
575 __ push(rdi);
576 __ push(rbp);
577 __ push_FPU_state();
578
579 // push_FPU_state() resets the FP top of stack
580 // Load original double into FP top of stack
581 __ fld_d(Address(rsp, saved_argument_off * wordSize));
582 // Store double into stack as outgoing argument
583 __ subptr(rsp, wordSize*2);
584 __ fst_d(Address(rsp, 0));
585
586 // Prepare FPU for doing math in C-land
587 __ empty_FPU_stack();
588 // Call the C code to massage the double. Result in EAX
589 if (t == T_INT)
590 { BLOCK_COMMENT("SharedRuntime::d2i"); }
591 else if (t == T_LONG)
592 { BLOCK_COMMENT("SharedRuntime::d2l"); }
593 __ call_VM_leaf( fcn, 2 );
594
595 // Restore CPU & FPU state
596 __ pop_FPU_state();
597 __ pop(rbp);
598 __ pop(rdi);
599 __ pop(rsi);
600 __ pop(rcx);
601 __ pop(rbx);
602 __ addptr(rsp, wordSize * 2);
603
604 __ ret(0);
605
606 return start;
607 }
608 //---------------------------------------------------------------------------------------------------
609
610 address generate_vector_mask(const char *stub_name, int32_t mask) {
611 __ align(CodeEntryAlignment);
612 StubCodeMark mark(this, "StubRoutines", stub_name);
613 address start = __ pc();
614
615 for (int i = 0; i < 16; i++) {
616 __ emit_data(mask, relocInfo::none, 0);
617 }
618
619 return start;
620 }
621
622 address generate_iota_indices(const char *stub_name) {
623 __ align(CodeEntryAlignment);
624 StubCodeMark mark(this, "StubRoutines", stub_name);
625 address start = __ pc();
626 __ emit_data(0x03020100, relocInfo::none, 0);
627 __ emit_data(0x07060504, relocInfo::none, 0);
628 __ emit_data(0x0B0A0908, relocInfo::none, 0);
629 __ emit_data(0x0F0E0D0C, relocInfo::none, 0);
630 __ emit_data(0x13121110, relocInfo::none, 0);
631 __ emit_data(0x17161514, relocInfo::none, 0);
632 __ emit_data(0x1B1A1918, relocInfo::none, 0);
633 __ emit_data(0x1F1E1D1C, relocInfo::none, 0);
634 __ emit_data(0x23222120, relocInfo::none, 0);
635 __ emit_data(0x27262524, relocInfo::none, 0);
636 __ emit_data(0x2B2A2928, relocInfo::none, 0);
637 __ emit_data(0x2F2E2D2C, relocInfo::none, 0);
638 __ emit_data(0x33323130, relocInfo::none, 0);
639 __ emit_data(0x37363534, relocInfo::none, 0);
640 __ emit_data(0x3B3A3938, relocInfo::none, 0);
641 __ emit_data(0x3F3E3D3C, relocInfo::none, 0);
642 return start;
643 }
644
645 address generate_vector_mask_long_double(const char *stub_name, int32_t maskhi, int32_t masklo) {
646 __ align(CodeEntryAlignment);
647 StubCodeMark mark(this, "StubRoutines", stub_name);
648 address start = __ pc();
649
650 for (int i = 0; i < 8; i++) {
651 __ emit_data(masklo, relocInfo::none, 0);
652 __ emit_data(maskhi, relocInfo::none, 0);
653 }
654
655 return start;
656 }
657
658 //----------------------------------------------------------------------------------------------------
659
660 address generate_vector_byte_perm_mask(const char *stub_name) {
661 __ align(CodeEntryAlignment);
662 StubCodeMark mark(this, "StubRoutines", stub_name);
663 address start = __ pc();
664
665 __ emit_data(0x00000001, relocInfo::none, 0);
666 __ emit_data(0x00000000, relocInfo::none, 0);
667 __ emit_data(0x00000003, relocInfo::none, 0);
668 __ emit_data(0x00000000, relocInfo::none, 0);
669 __ emit_data(0x00000005, relocInfo::none, 0);
670 __ emit_data(0x00000000, relocInfo::none, 0);
671 __ emit_data(0x00000007, relocInfo::none, 0);
672 __ emit_data(0x00000000, relocInfo::none, 0);
673 __ emit_data(0x00000000, relocInfo::none, 0);
674 __ emit_data(0x00000000, relocInfo::none, 0);
675 __ emit_data(0x00000002, relocInfo::none, 0);
676 __ emit_data(0x00000000, relocInfo::none, 0);
677 __ emit_data(0x00000004, relocInfo::none, 0);
678 __ emit_data(0x00000000, relocInfo::none, 0);
679 __ emit_data(0x00000006, relocInfo::none, 0);
680 __ emit_data(0x00000000, relocInfo::none, 0);
681
682 return start;
683 }
684
685 address generate_vector_custom_i32(const char *stub_name, Assembler::AvxVectorLen len,
686 int32_t val0, int32_t val1, int32_t val2, int32_t val3,
687 int32_t val4 = 0, int32_t val5 = 0, int32_t val6 = 0, int32_t val7 = 0,
688 int32_t val8 = 0, int32_t val9 = 0, int32_t val10 = 0, int32_t val11 = 0,
689 int32_t val12 = 0, int32_t val13 = 0, int32_t val14 = 0, int32_t val15 = 0) {
690 __ align(CodeEntryAlignment);
691 StubCodeMark mark(this, "StubRoutines", stub_name);
692 address start = __ pc();
693
694 assert(len != Assembler::AVX_NoVec, "vector len must be specified");
695 __ emit_data(val0, relocInfo::none, 0);
696 __ emit_data(val1, relocInfo::none, 0);
697 __ emit_data(val2, relocInfo::none, 0);
698 __ emit_data(val3, relocInfo::none, 0);
699 if (len >= Assembler::AVX_256bit) {
700 __ emit_data(val4, relocInfo::none, 0);
701 __ emit_data(val5, relocInfo::none, 0);
702 __ emit_data(val6, relocInfo::none, 0);
703 __ emit_data(val7, relocInfo::none, 0);
704 if (len >= Assembler::AVX_512bit) {
705 __ emit_data(val8, relocInfo::none, 0);
706 __ emit_data(val9, relocInfo::none, 0);
707 __ emit_data(val10, relocInfo::none, 0);
708 __ emit_data(val11, relocInfo::none, 0);
709 __ emit_data(val12, relocInfo::none, 0);
710 __ emit_data(val13, relocInfo::none, 0);
711 __ emit_data(val14, relocInfo::none, 0);
712 __ emit_data(val15, relocInfo::none, 0);
713 }
714 }
715
716 return start;
717 }
718
719 //----------------------------------------------------------------------------------------------------
720 // Non-destructive plausibility checks for oops
721
722 address generate_verify_oop() {
723 StubCodeMark mark(this, "StubRoutines", "verify_oop");
724 address start = __ pc();
725
726 // Incoming arguments on stack after saving rax,:
727 //
728 // [tos ]: saved rdx
729 // [tos + 1]: saved EFLAGS
730 // [tos + 2]: return address
731 // [tos + 3]: char* error message
732 // [tos + 4]: oop object to verify
733 // [tos + 5]: saved rax, - saved by caller and bashed
734
735 Label exit, error;
736 __ pushf();
737 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
738 __ push(rdx); // save rdx
739 // make sure object is 'reasonable'
740 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
741 __ testptr(rax, rax);
742 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
743
744 // Check if the oop is in the right area of memory
745 const int oop_mask = Universe::verify_oop_mask();
746 const int oop_bits = Universe::verify_oop_bits();
747 __ mov(rdx, rax);
748 __ andptr(rdx, oop_mask);
749 __ cmpptr(rdx, oop_bits);
750 __ jcc(Assembler::notZero, error);
751
752 // make sure klass is 'reasonable', which is not zero.
753 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
754 __ testptr(rax, rax);
755 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
756
757 // return if everything seems ok
758 __ bind(exit);
759 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
760 __ pop(rdx); // restore rdx
761 __ popf(); // restore EFLAGS
762 __ ret(3 * wordSize); // pop arguments
763
764 // handle errors
765 __ bind(error);
766 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
767 __ pop(rdx); // get saved rdx back
768 __ popf(); // get saved EFLAGS off stack -- will be ignored
769 __ pusha(); // push registers (eip = return address & msg are already pushed)
770 BLOCK_COMMENT("call MacroAssembler::debug");
771 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
772 __ hlt();
773 return start;
774 }
775
776
777 // Copy 64 bytes chunks
778 //
779 // Inputs:
780 // from - source array address
781 // to_from - destination array address - from
782 // qword_count - 8-bytes element count, negative
783 //
784 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
785 assert( UseSSE >= 2, "supported cpu only" );
786 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
787
788 // Copy 64-byte chunks
789 __ jmpb(L_copy_64_bytes);
790 __ align(OptoLoopAlignment);
791 __ BIND(L_copy_64_bytes_loop);
792
793 if (UseUnalignedLoadStores) {
794 if (UseAVX > 2) {
795 __ evmovdqul(xmm0, Address(from, 0), Assembler::AVX_512bit);
796 __ evmovdqul(Address(from, to_from, Address::times_1, 0), xmm0, Assembler::AVX_512bit);
797 } else if (UseAVX == 2) {
798 __ vmovdqu(xmm0, Address(from, 0));
799 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
800 __ vmovdqu(xmm1, Address(from, 32));
801 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
802 } else {
803 __ movdqu(xmm0, Address(from, 0));
804 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
805 __ movdqu(xmm1, Address(from, 16));
806 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
807 __ movdqu(xmm2, Address(from, 32));
808 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
809 __ movdqu(xmm3, Address(from, 48));
810 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
811 }
812 } else {
813 __ movq(xmm0, Address(from, 0));
814 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
815 __ movq(xmm1, Address(from, 8));
816 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
817 __ movq(xmm2, Address(from, 16));
818 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
819 __ movq(xmm3, Address(from, 24));
820 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
821 __ movq(xmm4, Address(from, 32));
822 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
823 __ movq(xmm5, Address(from, 40));
824 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
825 __ movq(xmm6, Address(from, 48));
826 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
827 __ movq(xmm7, Address(from, 56));
828 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
829 }
830
831 __ addl(from, 64);
832 __ BIND(L_copy_64_bytes);
833 __ subl(qword_count, 8);
834 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
835
836 if (UseUnalignedLoadStores && (UseAVX == 2)) {
837 // clean upper bits of YMM registers
838 __ vpxor(xmm0, xmm0);
839 __ vpxor(xmm1, xmm1);
840 }
841 __ addl(qword_count, 8);
842 __ jccb(Assembler::zero, L_exit);
843 //
844 // length is too short, just copy qwords
845 //
846 __ BIND(L_copy_8_bytes);
847 __ movq(xmm0, Address(from, 0));
848 __ movq(Address(from, to_from, Address::times_1), xmm0);
849 __ addl(from, 8);
850 __ decrement(qword_count);
851 __ jcc(Assembler::greater, L_copy_8_bytes);
852 __ BIND(L_exit);
853 }
854
855 address generate_disjoint_copy(BasicType t, bool aligned,
856 Address::ScaleFactor sf,
857 address* entry, const char *name,
858 bool dest_uninitialized = false) {
859 __ align(CodeEntryAlignment);
860 StubCodeMark mark(this, "StubRoutines", name);
861 address start = __ pc();
862
863 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
864 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
865
866 int shift = Address::times_ptr - sf;
867
868 const Register from = rsi; // source array address
869 const Register to = rdi; // destination array address
870 const Register count = rcx; // elements count
871 const Register to_from = to; // (to - from)
872 const Register saved_to = rdx; // saved destination array address
873
874 __ enter(); // required for proper stackwalking of RuntimeStub frame
875 __ push(rsi);
876 __ push(rdi);
877 __ movptr(from , Address(rsp, 12+ 4));
878 __ movptr(to , Address(rsp, 12+ 8));
879 __ movl(count, Address(rsp, 12+ 12));
880
881 if (entry != NULL) {
882 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
883 BLOCK_COMMENT("Entry:");
884 }
885
886 if (t == T_OBJECT) {
887 __ testl(count, count);
888 __ jcc(Assembler::zero, L_0_count);
889 }
890
891 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_DISJOINT;
892 if (dest_uninitialized) {
893 decorators |= IS_DEST_UNINITIALIZED;
894 }
895 if (aligned) {
896 decorators |= ARRAYCOPY_ALIGNED;
897 }
898
899 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
900 bs->arraycopy_prologue(_masm, decorators, t, from, to, count);
901 {
902 bool add_entry = (t != T_OBJECT && (!aligned || t == T_INT));
903 // UnsafeCopyMemory page error: continue after ucm
904 UnsafeCopyMemoryMark ucmm(this, add_entry, true);
905 __ subptr(to, from); // to --> to_from
906 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
907 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
908 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
909 // align source address at 4 bytes address boundary
910 if (t == T_BYTE) {
911 // One byte misalignment happens only for byte arrays
912 __ testl(from, 1);
913 __ jccb(Assembler::zero, L_skip_align1);
914 __ movb(rax, Address(from, 0));
915 __ movb(Address(from, to_from, Address::times_1, 0), rax);
916 __ increment(from);
917 __ decrement(count);
918 __ BIND(L_skip_align1);
919 }
920 // Two bytes misalignment happens only for byte and short (char) arrays
921 __ testl(from, 2);
922 __ jccb(Assembler::zero, L_skip_align2);
923 __ movw(rax, Address(from, 0));
924 __ movw(Address(from, to_from, Address::times_1, 0), rax);
925 __ addptr(from, 2);
926 __ subl(count, 1<<(shift-1));
927 __ BIND(L_skip_align2);
928 }
929 if (!UseXMMForArrayCopy) {
930 __ mov(rax, count); // save 'count'
931 __ shrl(count, shift); // bytes count
932 __ addptr(to_from, from);// restore 'to'
933 __ rep_mov();
934 __ subptr(to_from, from);// restore 'to_from'
935 __ mov(count, rax); // restore 'count'
936 __ jmpb(L_copy_2_bytes); // all dwords were copied
937 } else {
938 if (!UseUnalignedLoadStores) {
939 // align to 8 bytes, we know we are 4 byte aligned to start
940 __ testptr(from, 4);
941 __ jccb(Assembler::zero, L_copy_64_bytes);
942 __ movl(rax, Address(from, 0));
943 __ movl(Address(from, to_from, Address::times_1, 0), rax);
944 __ addptr(from, 4);
945 __ subl(count, 1<<shift);
946 }
947 __ BIND(L_copy_64_bytes);
948 __ mov(rax, count);
949 __ shrl(rax, shift+1); // 8 bytes chunk count
950 //
951 // Copy 8-byte chunks through XMM registers, 8 per iteration of the loop
952 //
953 xmm_copy_forward(from, to_from, rax);
954 }
955 // copy tailing dword
956 __ BIND(L_copy_4_bytes);
957 __ testl(count, 1<<shift);
958 __ jccb(Assembler::zero, L_copy_2_bytes);
959 __ movl(rax, Address(from, 0));
960 __ movl(Address(from, to_from, Address::times_1, 0), rax);
961 if (t == T_BYTE || t == T_SHORT) {
962 __ addptr(from, 4);
963 __ BIND(L_copy_2_bytes);
964 // copy tailing word
965 __ testl(count, 1<<(shift-1));
966 __ jccb(Assembler::zero, L_copy_byte);
967 __ movw(rax, Address(from, 0));
968 __ movw(Address(from, to_from, Address::times_1, 0), rax);
969 if (t == T_BYTE) {
970 __ addptr(from, 2);
971 __ BIND(L_copy_byte);
972 // copy tailing byte
973 __ testl(count, 1);
974 __ jccb(Assembler::zero, L_exit);
975 __ movb(rax, Address(from, 0));
976 __ movb(Address(from, to_from, Address::times_1, 0), rax);
977 __ BIND(L_exit);
978 } else {
979 __ BIND(L_copy_byte);
980 }
981 } else {
982 __ BIND(L_copy_2_bytes);
983 }
984 }
985
986 __ movl(count, Address(rsp, 12+12)); // reread 'count'
987 bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
988
989 if (t == T_OBJECT) {
990 __ BIND(L_0_count);
991 }
992 inc_copy_counter_np(t);
993 __ pop(rdi);
994 __ pop(rsi);
995 __ leave(); // required for proper stackwalking of RuntimeStub frame
996 __ vzeroupper();
997 __ xorptr(rax, rax); // return 0
998 __ ret(0);
999 return start;
1000 }
1001
1002
1003 address generate_fill(BasicType t, bool aligned, const char *name) {
1004 __ align(CodeEntryAlignment);
1005 StubCodeMark mark(this, "StubRoutines", name);
1006 address start = __ pc();
1007
1008 BLOCK_COMMENT("Entry:");
1009
1010 const Register to = rdi; // source array address
1011 const Register value = rdx; // value
1012 const Register count = rsi; // elements count
1013
1014 __ enter(); // required for proper stackwalking of RuntimeStub frame
1015 __ push(rsi);
1016 __ push(rdi);
1017 __ movptr(to , Address(rsp, 12+ 4));
1018 __ movl(value, Address(rsp, 12+ 8));
1019 __ movl(count, Address(rsp, 12+ 12));
1020
1021 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1022
1023 __ pop(rdi);
1024 __ pop(rsi);
1025 __ leave(); // required for proper stackwalking of RuntimeStub frame
1026 __ ret(0);
1027 return start;
1028 }
1029
1030 address generate_conjoint_copy(BasicType t, bool aligned,
1031 Address::ScaleFactor sf,
1032 address nooverlap_target,
1033 address* entry, const char *name,
1034 bool dest_uninitialized = false) {
1035 __ align(CodeEntryAlignment);
1036 StubCodeMark mark(this, "StubRoutines", name);
1037 address start = __ pc();
1038
1039 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1040 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1041
1042 int shift = Address::times_ptr - sf;
1043
1044 const Register src = rax; // source array address
1045 const Register dst = rdx; // destination array address
1046 const Register from = rsi; // source array address
1047 const Register to = rdi; // destination array address
1048 const Register count = rcx; // elements count
1049 const Register end = rax; // array end address
1050
1051 __ enter(); // required for proper stackwalking of RuntimeStub frame
1052 __ push(rsi);
1053 __ push(rdi);
1054 __ movptr(src , Address(rsp, 12+ 4)); // from
1055 __ movptr(dst , Address(rsp, 12+ 8)); // to
1056 __ movl2ptr(count, Address(rsp, 12+12)); // count
1057
1058 if (entry != NULL) {
1059 *entry = __ pc(); // Entry point from generic arraycopy stub.
1060 BLOCK_COMMENT("Entry:");
1061 }
1062
1063 // nooverlap_target expects arguments in rsi and rdi.
1064 __ mov(from, src);
1065 __ mov(to , dst);
1066
1067 // arrays overlap test: dispatch to disjoint stub if necessary.
1068 RuntimeAddress nooverlap(nooverlap_target);
1069 __ cmpptr(dst, src);
1070 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1071 __ jump_cc(Assembler::belowEqual, nooverlap);
1072 __ cmpptr(dst, end);
1073 __ jump_cc(Assembler::aboveEqual, nooverlap);
1074
1075 if (t == T_OBJECT) {
1076 __ testl(count, count);
1077 __ jcc(Assembler::zero, L_0_count);
1078 }
1079
1080 DecoratorSet decorators = IN_HEAP | IS_ARRAY;
1081 if (dest_uninitialized) {
1082 decorators |= IS_DEST_UNINITIALIZED;
1083 }
1084 if (aligned) {
1085 decorators |= ARRAYCOPY_ALIGNED;
1086 }
1087
1088 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1089 bs->arraycopy_prologue(_masm, decorators, t, from, to, count);
1090
1091 {
1092 bool add_entry = (t != T_OBJECT && (!aligned || t == T_INT));
1093 // UnsafeCopyMemory page error: continue after ucm
1094 UnsafeCopyMemoryMark ucmm(this, add_entry, true);
1095 // copy from high to low
1096 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1097 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1098 if (t == T_BYTE || t == T_SHORT) {
1099 // Align the end of destination array at 4 bytes address boundary
1100 __ lea(end, Address(dst, count, sf, 0));
1101 if (t == T_BYTE) {
1102 // One byte misalignment happens only for byte arrays
1103 __ testl(end, 1);
1104 __ jccb(Assembler::zero, L_skip_align1);
1105 __ decrement(count);
1106 __ movb(rdx, Address(from, count, sf, 0));
1107 __ movb(Address(to, count, sf, 0), rdx);
1108 __ BIND(L_skip_align1);
1109 }
1110 // Two bytes misalignment happens only for byte and short (char) arrays
1111 __ testl(end, 2);
1112 __ jccb(Assembler::zero, L_skip_align2);
1113 __ subptr(count, 1<<(shift-1));
1114 __ movw(rdx, Address(from, count, sf, 0));
1115 __ movw(Address(to, count, sf, 0), rdx);
1116 __ BIND(L_skip_align2);
1117 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1118 __ jcc(Assembler::below, L_copy_4_bytes);
1119 }
1120
1121 if (!UseXMMForArrayCopy) {
1122 __ std();
1123 __ mov(rax, count); // Save 'count'
1124 __ mov(rdx, to); // Save 'to'
1125 __ lea(rsi, Address(from, count, sf, -4));
1126 __ lea(rdi, Address(to , count, sf, -4));
1127 __ shrptr(count, shift); // bytes count
1128 __ rep_mov();
1129 __ cld();
1130 __ mov(count, rax); // restore 'count'
1131 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1132 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1133 __ mov(to, rdx); // restore 'to'
1134 __ jmpb(L_copy_2_bytes); // all dword were copied
1135 } else {
1136 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1137 __ testptr(end, 4);
1138 __ jccb(Assembler::zero, L_copy_8_bytes);
1139 __ subl(count, 1<<shift);
1140 __ movl(rdx, Address(from, count, sf, 0));
1141 __ movl(Address(to, count, sf, 0), rdx);
1142 __ jmpb(L_copy_8_bytes);
1143
1144 __ align(OptoLoopAlignment);
1145 // Move 8 bytes
1146 __ BIND(L_copy_8_bytes_loop);
1147 __ movq(xmm0, Address(from, count, sf, 0));
1148 __ movq(Address(to, count, sf, 0), xmm0);
1149 __ BIND(L_copy_8_bytes);
1150 __ subl(count, 2<<shift);
1151 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1152 __ addl(count, 2<<shift);
1153 }
1154 __ BIND(L_copy_4_bytes);
1155 // copy prefix qword
1156 __ testl(count, 1<<shift);
1157 __ jccb(Assembler::zero, L_copy_2_bytes);
1158 __ movl(rdx, Address(from, count, sf, -4));
1159 __ movl(Address(to, count, sf, -4), rdx);
1160
1161 if (t == T_BYTE || t == T_SHORT) {
1162 __ subl(count, (1<<shift));
1163 __ BIND(L_copy_2_bytes);
1164 // copy prefix dword
1165 __ testl(count, 1<<(shift-1));
1166 __ jccb(Assembler::zero, L_copy_byte);
1167 __ movw(rdx, Address(from, count, sf, -2));
1168 __ movw(Address(to, count, sf, -2), rdx);
1169 if (t == T_BYTE) {
1170 __ subl(count, 1<<(shift-1));
1171 __ BIND(L_copy_byte);
1172 // copy prefix byte
1173 __ testl(count, 1);
1174 __ jccb(Assembler::zero, L_exit);
1175 __ movb(rdx, Address(from, 0));
1176 __ movb(Address(to, 0), rdx);
1177 __ BIND(L_exit);
1178 } else {
1179 __ BIND(L_copy_byte);
1180 }
1181 } else {
1182 __ BIND(L_copy_2_bytes);
1183 }
1184 }
1185
1186 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1187 bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
1188
1189 if (t == T_OBJECT) {
1190 __ BIND(L_0_count);
1191 }
1192 inc_copy_counter_np(t);
1193 __ pop(rdi);
1194 __ pop(rsi);
1195 __ leave(); // required for proper stackwalking of RuntimeStub frame
1196 __ xorptr(rax, rax); // return 0
1197 __ ret(0);
1198 return start;
1199 }
1200
1201
1202 address generate_disjoint_long_copy(address* entry, const char *name) {
1203 __ align(CodeEntryAlignment);
1204 StubCodeMark mark(this, "StubRoutines", name);
1205 address start = __ pc();
1206
1207 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1208 const Register from = rax; // source array address
1209 const Register to = rdx; // destination array address
1210 const Register count = rcx; // elements count
1211 const Register to_from = rdx; // (to - from)
1212
1213 __ enter(); // required for proper stackwalking of RuntimeStub frame
1214 __ movptr(from , Address(rsp, 8+0)); // from
1215 __ movptr(to , Address(rsp, 8+4)); // to
1216 __ movl2ptr(count, Address(rsp, 8+8)); // count
1217
1218 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1219 BLOCK_COMMENT("Entry:");
1220
1221 {
1222 // UnsafeCopyMemory page error: continue after ucm
1223 UnsafeCopyMemoryMark ucmm(this, true, true);
1224 __ subptr(to, from); // to --> to_from
1225 if (UseXMMForArrayCopy) {
1226 xmm_copy_forward(from, to_from, count);
1227 } else {
1228 __ jmpb(L_copy_8_bytes);
1229 __ align(OptoLoopAlignment);
1230 __ BIND(L_copy_8_bytes_loop);
1231 __ fild_d(Address(from, 0));
1232 __ fistp_d(Address(from, to_from, Address::times_1));
1233 __ addptr(from, 8);
1234 __ BIND(L_copy_8_bytes);
1235 __ decrement(count);
1236 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1237 }
1238 }
1239 inc_copy_counter_np(T_LONG);
1240 __ leave(); // required for proper stackwalking of RuntimeStub frame
1241 __ vzeroupper();
1242 __ xorptr(rax, rax); // return 0
1243 __ ret(0);
1244 return start;
1245 }
1246
1247 address generate_conjoint_long_copy(address nooverlap_target,
1248 address* entry, const char *name) {
1249 __ align(CodeEntryAlignment);
1250 StubCodeMark mark(this, "StubRoutines", name);
1251 address start = __ pc();
1252
1253 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1254 const Register from = rax; // source array address
1255 const Register to = rdx; // destination array address
1256 const Register count = rcx; // elements count
1257 const Register end_from = rax; // source array end address
1258
1259 __ enter(); // required for proper stackwalking of RuntimeStub frame
1260 __ movptr(from , Address(rsp, 8+0)); // from
1261 __ movptr(to , Address(rsp, 8+4)); // to
1262 __ movl2ptr(count, Address(rsp, 8+8)); // count
1263
1264 *entry = __ pc(); // Entry point from generic arraycopy stub.
1265 BLOCK_COMMENT("Entry:");
1266
1267 // arrays overlap test
1268 __ cmpptr(to, from);
1269 RuntimeAddress nooverlap(nooverlap_target);
1270 __ jump_cc(Assembler::belowEqual, nooverlap);
1271 __ lea(end_from, Address(from, count, Address::times_8, 0));
1272 __ cmpptr(to, end_from);
1273 __ movptr(from, Address(rsp, 8)); // from
1274 __ jump_cc(Assembler::aboveEqual, nooverlap);
1275
1276 {
1277 // UnsafeCopyMemory page error: continue after ucm
1278 UnsafeCopyMemoryMark ucmm(this, true, true);
1279
1280 __ jmpb(L_copy_8_bytes);
1281
1282 __ align(OptoLoopAlignment);
1283 __ BIND(L_copy_8_bytes_loop);
1284 if (UseXMMForArrayCopy) {
1285 __ movq(xmm0, Address(from, count, Address::times_8));
1286 __ movq(Address(to, count, Address::times_8), xmm0);
1287 } else {
1288 __ fild_d(Address(from, count, Address::times_8));
1289 __ fistp_d(Address(to, count, Address::times_8));
1290 }
1291 __ BIND(L_copy_8_bytes);
1292 __ decrement(count);
1293 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1294
1295 }
1296 inc_copy_counter_np(T_LONG);
1297 __ leave(); // required for proper stackwalking of RuntimeStub frame
1298 __ xorptr(rax, rax); // return 0
1299 __ ret(0);
1300 return start;
1301 }
1302
1303
1304 // Helper for generating a dynamic type check.
1305 // The sub_klass must be one of {rbx, rdx, rsi}.
1306 // The temp is killed.
1307 void generate_type_check(Register sub_klass,
1308 Address& super_check_offset_addr,
1309 Address& super_klass_addr,
1310 Register temp,
1311 Label* L_success, Label* L_failure) {
1312 BLOCK_COMMENT("type_check:");
1313
1314 Label L_fallthrough;
1315 #define LOCAL_JCC(assembler_con, label_ptr) \
1316 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1317 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1318
1319 // The following is a strange variation of the fast path which requires
1320 // one less register, because needed values are on the argument stack.
1321 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1322 // L_success, L_failure, NULL);
1323 assert_different_registers(sub_klass, temp);
1324
1325 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1326
1327 // if the pointers are equal, we are done (e.g., String[] elements)
1328 __ cmpptr(sub_klass, super_klass_addr);
1329 LOCAL_JCC(Assembler::equal, L_success);
1330
1331 // check the supertype display:
1332 __ movl2ptr(temp, super_check_offset_addr);
1333 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1334 __ movptr(temp, super_check_addr); // load displayed supertype
1335 __ cmpptr(temp, super_klass_addr); // test the super type
1336 LOCAL_JCC(Assembler::equal, L_success);
1337
1338 // if it was a primary super, we can just fail immediately
1339 __ cmpl(super_check_offset_addr, sc_offset);
1340 LOCAL_JCC(Assembler::notEqual, L_failure);
1341
1342 // The repne_scan instruction uses fixed registers, which will get spilled.
1343 // We happen to know this works best when super_klass is in rax.
1344 Register super_klass = temp;
1345 __ movptr(super_klass, super_klass_addr);
1346 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1347 L_success, L_failure);
1348
1349 __ bind(L_fallthrough);
1350
1351 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1352 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1353
1354 #undef LOCAL_JCC
1355 }
1356
1357 //
1358 // Generate checkcasting array copy stub
1359 //
1360 // Input:
1361 // 4(rsp) - source array address
1362 // 8(rsp) - destination array address
1363 // 12(rsp) - element count, can be zero
1364 // 16(rsp) - size_t ckoff (super_check_offset)
1365 // 20(rsp) - oop ckval (super_klass)
1366 //
1367 // Output:
1368 // rax, == 0 - success
1369 // rax, == -1^K - failure, where K is partial transfer count
1370 //
1371 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1372 __ align(CodeEntryAlignment);
1373 StubCodeMark mark(this, "StubRoutines", name);
1374 address start = __ pc();
1375
1376 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1377
1378 // register use:
1379 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1380 // rdi, rsi -- element access (oop, klass)
1381 // rbx, -- temp
1382 const Register from = rax; // source array address
1383 const Register to = rdx; // destination array address
1384 const Register length = rcx; // elements count
1385 const Register elem = rdi; // each oop copied
1386 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1387 const Register temp = rbx; // lone remaining temp
1388
1389 __ enter(); // required for proper stackwalking of RuntimeStub frame
1390
1391 __ push(rsi);
1392 __ push(rdi);
1393 __ push(rbx);
1394
1395 Address from_arg(rsp, 16+ 4); // from
1396 Address to_arg(rsp, 16+ 8); // to
1397 Address length_arg(rsp, 16+12); // elements count
1398 Address ckoff_arg(rsp, 16+16); // super_check_offset
1399 Address ckval_arg(rsp, 16+20); // super_klass
1400
1401 // Load up:
1402 __ movptr(from, from_arg);
1403 __ movptr(to, to_arg);
1404 __ movl2ptr(length, length_arg);
1405
1406 if (entry != NULL) {
1407 *entry = __ pc(); // Entry point from generic arraycopy stub.
1408 BLOCK_COMMENT("Entry:");
1409 }
1410
1411 //---------------------------------------------------------------
1412 // Assembler stub will be used for this call to arraycopy
1413 // if the two arrays are subtypes of Object[] but the
1414 // destination array type is not equal to or a supertype
1415 // of the source type. Each element must be separately
1416 // checked.
1417
1418 // Loop-invariant addresses. They are exclusive end pointers.
1419 Address end_from_addr(from, length, Address::times_ptr, 0);
1420 Address end_to_addr(to, length, Address::times_ptr, 0);
1421
1422 Register end_from = from; // re-use
1423 Register end_to = to; // re-use
1424 Register count = length; // re-use
1425
1426 // Loop-variant addresses. They assume post-incremented count < 0.
1427 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1428 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1429 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1430
1431 DecoratorSet decorators = IN_HEAP | IS_ARRAY | ARRAYCOPY_CHECKCAST;
1432 if (dest_uninitialized) {
1433 decorators |= IS_DEST_UNINITIALIZED;
1434 }
1435
1436 BasicType type = T_OBJECT;
1437 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
1438 bs->arraycopy_prologue(_masm, decorators, type, from, to, count);
1439
1440 // Copy from low to high addresses, indexed from the end of each array.
1441 __ lea(end_from, end_from_addr);
1442 __ lea(end_to, end_to_addr);
1443 assert(length == count, ""); // else fix next line:
1444 __ negptr(count); // negate and test the length
1445 __ jccb(Assembler::notZero, L_load_element);
1446
1447 // Empty array: Nothing to do.
1448 __ xorptr(rax, rax); // return 0 on (trivial) success
1449 __ jmp(L_done);
1450
1451 // ======== begin loop ========
1452 // (Loop is rotated; its entry is L_load_element.)
1453 // Loop control:
1454 // for (count = -count; count != 0; count++)
1455 // Base pointers src, dst are biased by 8*count,to last element.
1456 __ align(OptoLoopAlignment);
1457
1458 __ BIND(L_store_element);
1459 __ movptr(to_element_addr, elem); // store the oop
1460 __ increment(count); // increment the count toward zero
1461 __ jccb(Assembler::zero, L_do_card_marks);
1462
1463 // ======== loop entry is here ========
1464 __ BIND(L_load_element);
1465 __ movptr(elem, from_element_addr); // load the oop
1466 __ testptr(elem, elem);
1467 __ jccb(Assembler::zero, L_store_element);
1468
1469 // (Could do a trick here: Remember last successful non-null
1470 // element stored and make a quick oop equality check on it.)
1471
1472 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1473 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1474 &L_store_element, NULL);
1475 // (On fall-through, we have failed the element type check.)
1476 // ======== end loop ========
1477
1478 // It was a real error; we must depend on the caller to finish the job.
1479 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1480 // Emit GC store barriers for the oops we have copied (length_arg + count),
1481 // and report their number to the caller.
1482 assert_different_registers(to, count, rax);
1483 Label L_post_barrier;
1484 __ addl(count, length_arg); // transfers = (length - remaining)
1485 __ movl2ptr(rax, count); // save the value
1486 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1487 __ jccb(Assembler::notZero, L_post_barrier);
1488 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1489
1490 // Come here on success only.
1491 __ BIND(L_do_card_marks);
1492 __ xorptr(rax, rax); // return 0 on success
1493 __ movl2ptr(count, length_arg);
1494
1495 __ BIND(L_post_barrier);
1496 __ movptr(to, to_arg); // reload
1497 bs->arraycopy_epilogue(_masm, decorators, type, from, to, count);
1498
1499 // Common exit point (success or failure).
1500 __ BIND(L_done);
1501 __ pop(rbx);
1502 __ pop(rdi);
1503 __ pop(rsi);
1504 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1505 __ leave(); // required for proper stackwalking of RuntimeStub frame
1506 __ ret(0);
1507
1508 return start;
1509 }
1510
1511 //
1512 // Generate 'unsafe' array copy stub
1513 // Though just as safe as the other stubs, it takes an unscaled
1514 // size_t argument instead of an element count.
1515 //
1516 // Input:
1517 // 4(rsp) - source array address
1518 // 8(rsp) - destination array address
1519 // 12(rsp) - byte count, can be zero
1520 //
1521 // Output:
1522 // rax, == 0 - success
1523 // rax, == -1 - need to call System.arraycopy
1524 //
1525 // Examines the alignment of the operands and dispatches
1526 // to a long, int, short, or byte copy loop.
1527 //
1528 address generate_unsafe_copy(const char *name,
1529 address byte_copy_entry,
1530 address short_copy_entry,
1531 address int_copy_entry,
1532 address long_copy_entry) {
1533
1534 Label L_long_aligned, L_int_aligned, L_short_aligned;
1535
1536 __ align(CodeEntryAlignment);
1537 StubCodeMark mark(this, "StubRoutines", name);
1538 address start = __ pc();
1539
1540 const Register from = rax; // source array address
1541 const Register to = rdx; // destination array address
1542 const Register count = rcx; // elements count
1543
1544 __ enter(); // required for proper stackwalking of RuntimeStub frame
1545 __ push(rsi);
1546 __ push(rdi);
1547 Address from_arg(rsp, 12+ 4); // from
1548 Address to_arg(rsp, 12+ 8); // to
1549 Address count_arg(rsp, 12+12); // byte count
1550
1551 // Load up:
1552 __ movptr(from , from_arg);
1553 __ movptr(to , to_arg);
1554 __ movl2ptr(count, count_arg);
1555
1556 // bump this on entry, not on exit:
1557 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1558
1559 const Register bits = rsi;
1560 __ mov(bits, from);
1561 __ orptr(bits, to);
1562 __ orptr(bits, count);
1563
1564 __ testl(bits, BytesPerLong-1);
1565 __ jccb(Assembler::zero, L_long_aligned);
1566
1567 __ testl(bits, BytesPerInt-1);
1568 __ jccb(Assembler::zero, L_int_aligned);
1569
1570 __ testl(bits, BytesPerShort-1);
1571 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1572
1573 __ BIND(L_short_aligned);
1574 __ shrptr(count, LogBytesPerShort); // size => short_count
1575 __ movl(count_arg, count); // update 'count'
1576 __ jump(RuntimeAddress(short_copy_entry));
1577
1578 __ BIND(L_int_aligned);
1579 __ shrptr(count, LogBytesPerInt); // size => int_count
1580 __ movl(count_arg, count); // update 'count'
1581 __ jump(RuntimeAddress(int_copy_entry));
1582
1583 __ BIND(L_long_aligned);
1584 __ shrptr(count, LogBytesPerLong); // size => qword_count
1585 __ movl(count_arg, count); // update 'count'
1586 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1587 __ pop(rsi);
1588 __ jump(RuntimeAddress(long_copy_entry));
1589
1590 return start;
1591 }
1592
1593
1594 // Perform range checks on the proposed arraycopy.
1595 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1596 void arraycopy_range_checks(Register src,
1597 Register src_pos,
1598 Register dst,
1599 Register dst_pos,
1600 Address& length,
1601 Label& L_failed) {
1602 BLOCK_COMMENT("arraycopy_range_checks:");
1603 const Register src_end = src_pos; // source array end position
1604 const Register dst_end = dst_pos; // destination array end position
1605 __ addl(src_end, length); // src_pos + length
1606 __ addl(dst_end, length); // dst_pos + length
1607
1608 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1609 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1610 __ jcc(Assembler::above, L_failed);
1611
1612 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1613 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1614 __ jcc(Assembler::above, L_failed);
1615
1616 BLOCK_COMMENT("arraycopy_range_checks done");
1617 }
1618
1619
1620 //
1621 // Generate generic array copy stubs
1622 //
1623 // Input:
1624 // 4(rsp) - src oop
1625 // 8(rsp) - src_pos
1626 // 12(rsp) - dst oop
1627 // 16(rsp) - dst_pos
1628 // 20(rsp) - element count
1629 //
1630 // Output:
1631 // rax, == 0 - success
1632 // rax, == -1^K - failure, where K is partial transfer count
1633 //
1634 address generate_generic_copy(const char *name,
1635 address entry_jbyte_arraycopy,
1636 address entry_jshort_arraycopy,
1637 address entry_jint_arraycopy,
1638 address entry_oop_arraycopy,
1639 address entry_jlong_arraycopy,
1640 address entry_checkcast_arraycopy) {
1641 Label L_failed, L_failed_0, L_objArray;
1642
1643 { int modulus = CodeEntryAlignment;
1644 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1645 int advance = target - (__ offset() % modulus);
1646 if (advance < 0) advance += modulus;
1647 if (advance > 0) __ nop(advance);
1648 }
1649 StubCodeMark mark(this, "StubRoutines", name);
1650
1651 // Short-hop target to L_failed. Makes for denser prologue code.
1652 __ BIND(L_failed_0);
1653 __ jmp(L_failed);
1654 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1655
1656 __ align(CodeEntryAlignment);
1657 address start = __ pc();
1658
1659 __ enter(); // required for proper stackwalking of RuntimeStub frame
1660 __ push(rsi);
1661 __ push(rdi);
1662
1663 // bump this on entry, not on exit:
1664 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1665
1666 // Input values
1667 Address SRC (rsp, 12+ 4);
1668 Address SRC_POS (rsp, 12+ 8);
1669 Address DST (rsp, 12+12);
1670 Address DST_POS (rsp, 12+16);
1671 Address LENGTH (rsp, 12+20);
1672
1673 //-----------------------------------------------------------------------
1674 // Assembler stub will be used for this call to arraycopy
1675 // if the following conditions are met:
1676 //
1677 // (1) src and dst must not be null.
1678 // (2) src_pos must not be negative.
1679 // (3) dst_pos must not be negative.
1680 // (4) length must not be negative.
1681 // (5) src klass and dst klass should be the same and not NULL.
1682 // (6) src and dst should be arrays.
1683 // (7) src_pos + length must not exceed length of src.
1684 // (8) dst_pos + length must not exceed length of dst.
1685 //
1686
1687 const Register src = rax; // source array oop
1688 const Register src_pos = rsi;
1689 const Register dst = rdx; // destination array oop
1690 const Register dst_pos = rdi;
1691 const Register length = rcx; // transfer count
1692
1693 // if (src == NULL) return -1;
1694 __ movptr(src, SRC); // src oop
1695 __ testptr(src, src);
1696 __ jccb(Assembler::zero, L_failed_0);
1697
1698 // if (src_pos < 0) return -1;
1699 __ movl2ptr(src_pos, SRC_POS); // src_pos
1700 __ testl(src_pos, src_pos);
1701 __ jccb(Assembler::negative, L_failed_0);
1702
1703 // if (dst == NULL) return -1;
1704 __ movptr(dst, DST); // dst oop
1705 __ testptr(dst, dst);
1706 __ jccb(Assembler::zero, L_failed_0);
1707
1708 // if (dst_pos < 0) return -1;
1709 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1710 __ testl(dst_pos, dst_pos);
1711 __ jccb(Assembler::negative, L_failed_0);
1712
1713 // if (length < 0) return -1;
1714 __ movl2ptr(length, LENGTH); // length
1715 __ testl(length, length);
1716 __ jccb(Assembler::negative, L_failed_0);
1717
1718 // if (src->klass() == NULL) return -1;
1719 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1720 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1721 const Register rcx_src_klass = rcx; // array klass
1722 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1723
1724 #ifdef ASSERT
1725 // assert(src->klass() != NULL);
1726 BLOCK_COMMENT("assert klasses not null");
1727 { Label L1, L2;
1728 __ testptr(rcx_src_klass, rcx_src_klass);
1729 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1730 __ bind(L1);
1731 __ stop("broken null klass");
1732 __ bind(L2);
1733 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1734 __ jccb(Assembler::equal, L1); // this would be broken also
1735 BLOCK_COMMENT("assert done");
1736 }
1737 #endif //ASSERT
1738
1739 // Load layout helper (32-bits)
1740 //
1741 // |array_tag| | header_size | element_type | |log2_element_size|
1742 // 32 30 24 16 8 2 0
1743 //
1744 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1745 //
1746
1747 int lh_offset = in_bytes(Klass::layout_helper_offset());
1748 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1749
1750 // Handle objArrays completely differently...
1751 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1752 __ cmpl(src_klass_lh_addr, objArray_lh);
1753 __ jcc(Assembler::equal, L_objArray);
1754
1755 // if (src->klass() != dst->klass()) return -1;
1756 __ cmpptr(rcx_src_klass, dst_klass_addr);
1757 __ jccb(Assembler::notEqual, L_failed_0);
1758
1759 const Register rcx_lh = rcx; // layout helper
1760 assert(rcx_lh == rcx_src_klass, "known alias");
1761 __ movl(rcx_lh, src_klass_lh_addr);
1762
1763 // if (!src->is_Array()) return -1;
1764 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1765 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1766
1767 // At this point, it is known to be a typeArray (array_tag 0x3).
1768 #ifdef ASSERT
1769 { Label L;
1770 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1771 __ jcc(Assembler::greaterEqual, L); // signed cmp
1772 __ stop("must be a primitive array");
1773 __ bind(L);
1774 }
1775 #endif
1776
1777 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1778 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1779
1780 // TypeArrayKlass
1781 //
1782 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1783 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1784 //
1785 const Register rsi_offset = rsi; // array offset
1786 const Register src_array = src; // src array offset
1787 const Register dst_array = dst; // dst array offset
1788 const Register rdi_elsize = rdi; // log2 element size
1789
1790 __ mov(rsi_offset, rcx_lh);
1791 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1792 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1793 __ addptr(src_array, rsi_offset); // src array offset
1794 __ addptr(dst_array, rsi_offset); // dst array offset
1795 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1796
1797 // next registers should be set before the jump to corresponding stub
1798 const Register from = src; // source array address
1799 const Register to = dst; // destination array address
1800 const Register count = rcx; // elements count
1801 // some of them should be duplicated on stack
1802 #define FROM Address(rsp, 12+ 4)
1803 #define TO Address(rsp, 12+ 8) // Not used now
1804 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1805
1806 BLOCK_COMMENT("scale indexes to element size");
1807 __ movl2ptr(rsi, SRC_POS); // src_pos
1808 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1809 assert(src_array == from, "");
1810 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1811 __ movl2ptr(rdi, DST_POS); // dst_pos
1812 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1813 assert(dst_array == to, "");
1814 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1815 __ movptr(FROM, from); // src_addr
1816 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1817 __ movl2ptr(count, LENGTH); // elements count
1818
1819 BLOCK_COMMENT("choose copy loop based on element size");
1820 __ cmpl(rdi_elsize, 0);
1821
1822 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1823 __ cmpl(rdi_elsize, LogBytesPerShort);
1824 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1825 __ cmpl(rdi_elsize, LogBytesPerInt);
1826 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1827 #ifdef ASSERT
1828 __ cmpl(rdi_elsize, LogBytesPerLong);
1829 __ jccb(Assembler::notEqual, L_failed);
1830 #endif
1831 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1832 __ pop(rsi);
1833 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1834
1835 __ BIND(L_failed);
1836 __ xorptr(rax, rax);
1837 __ notptr(rax); // return -1
1838 __ pop(rdi);
1839 __ pop(rsi);
1840 __ leave(); // required for proper stackwalking of RuntimeStub frame
1841 __ ret(0);
1842
1843 // ObjArrayKlass
1844 __ BIND(L_objArray);
1845 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1846
1847 Label L_plain_copy, L_checkcast_copy;
1848 // test array classes for subtyping
1849 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1850 __ jccb(Assembler::notEqual, L_checkcast_copy);
1851
1852 // Identically typed arrays can be copied without element-wise checks.
1853 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1854 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1855
1856 __ BIND(L_plain_copy);
1857 __ movl2ptr(count, LENGTH); // elements count
1858 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1859 __ lea(from, Address(src, src_pos, Address::times_ptr,
1860 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1861 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1862 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1863 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1864 __ movptr(FROM, from); // src_addr
1865 __ movptr(TO, to); // dst_addr
1866 __ movl(COUNT, count); // count
1867 __ jump(RuntimeAddress(entry_oop_arraycopy));
1868
1869 __ BIND(L_checkcast_copy);
1870 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1871 {
1872 // Handy offsets:
1873 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1874 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1875
1876 Register rsi_dst_klass = rsi;
1877 Register rdi_temp = rdi;
1878 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1879 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1880 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1881
1882 // Before looking at dst.length, make sure dst is also an objArray.
1883 __ movptr(rsi_dst_klass, dst_klass_addr);
1884 __ cmpl(dst_klass_lh_addr, objArray_lh);
1885 __ jccb(Assembler::notEqual, L_failed);
1886
1887 // It is safe to examine both src.length and dst.length.
1888 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1889 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1890 // (Now src_pos and dst_pos are killed, but not src and dst.)
1891
1892 // We'll need this temp (don't forget to pop it after the type check).
1893 __ push(rbx);
1894 Register rbx_src_klass = rbx;
1895
1896 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1897 __ movptr(rsi_dst_klass, dst_klass_addr);
1898 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1899 Label L_fail_array_check;
1900 generate_type_check(rbx_src_klass,
1901 super_check_offset_addr, dst_klass_addr,
1902 rdi_temp, NULL, &L_fail_array_check);
1903 // (On fall-through, we have passed the array type check.)
1904 __ pop(rbx);
1905 __ jmp(L_plain_copy);
1906
1907 __ BIND(L_fail_array_check);
1908 // Reshuffle arguments so we can call checkcast_arraycopy:
1909
1910 // match initial saves for checkcast_arraycopy
1911 // push(rsi); // already done; see above
1912 // push(rdi); // already done; see above
1913 // push(rbx); // already done; see above
1914
1915 // Marshal outgoing arguments now, freeing registers.
1916 Address from_arg(rsp, 16+ 4); // from
1917 Address to_arg(rsp, 16+ 8); // to
1918 Address length_arg(rsp, 16+12); // elements count
1919 Address ckoff_arg(rsp, 16+16); // super_check_offset
1920 Address ckval_arg(rsp, 16+20); // super_klass
1921
1922 Address SRC_POS_arg(rsp, 16+ 8);
1923 Address DST_POS_arg(rsp, 16+16);
1924 Address LENGTH_arg(rsp, 16+20);
1925 // push rbx, changed the incoming offsets (why not just use rbp,??)
1926 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1927
1928 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1929 __ movl2ptr(length, LENGTH_arg); // reload elements count
1930 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1931 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1932
1933 __ movptr(ckval_arg, rbx); // destination element type
1934 __ movl(rbx, Address(rbx, sco_offset));
1935 __ movl(ckoff_arg, rbx); // corresponding class check offset
1936
1937 __ movl(length_arg, length); // outgoing length argument
1938
1939 __ lea(from, Address(src, src_pos, Address::times_ptr,
1940 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1941 __ movptr(from_arg, from);
1942
1943 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1944 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1945 __ movptr(to_arg, to);
1946 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1947 }
1948
1949 return start;
1950 }
1951
1952 void generate_arraycopy_stubs() {
1953 address entry;
1954 address entry_jbyte_arraycopy;
1955 address entry_jshort_arraycopy;
1956 address entry_jint_arraycopy;
1957 address entry_oop_arraycopy;
1958 address entry_jlong_arraycopy;
1959 address entry_checkcast_arraycopy;
1960
1961 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1962 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1963 "arrayof_jbyte_disjoint_arraycopy");
1964 StubRoutines::_arrayof_jbyte_arraycopy =
1965 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1966 NULL, "arrayof_jbyte_arraycopy");
1967 StubRoutines::_jbyte_disjoint_arraycopy =
1968 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1969 "jbyte_disjoint_arraycopy");
1970 StubRoutines::_jbyte_arraycopy =
1971 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1972 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1973
1974 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1975 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
1976 "arrayof_jshort_disjoint_arraycopy");
1977 StubRoutines::_arrayof_jshort_arraycopy =
1978 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
1979 NULL, "arrayof_jshort_arraycopy");
1980 StubRoutines::_jshort_disjoint_arraycopy =
1981 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
1982 "jshort_disjoint_arraycopy");
1983 StubRoutines::_jshort_arraycopy =
1984 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
1985 &entry_jshort_arraycopy, "jshort_arraycopy");
1986
1987 // Next arrays are always aligned on 4 bytes at least.
1988 StubRoutines::_jint_disjoint_arraycopy =
1989 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
1990 "jint_disjoint_arraycopy");
1991 StubRoutines::_jint_arraycopy =
1992 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
1993 &entry_jint_arraycopy, "jint_arraycopy");
1994
1995 StubRoutines::_oop_disjoint_arraycopy =
1996 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
1997 "oop_disjoint_arraycopy");
1998 StubRoutines::_oop_arraycopy =
1999 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2000 &entry_oop_arraycopy, "oop_arraycopy");
2001
2002 StubRoutines::_oop_disjoint_arraycopy_uninit =
2003 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2004 "oop_disjoint_arraycopy_uninit",
2005 /*dest_uninitialized*/true);
2006 StubRoutines::_oop_arraycopy_uninit =
2007 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2008 NULL, "oop_arraycopy_uninit",
2009 /*dest_uninitialized*/true);
2010
2011 StubRoutines::_jlong_disjoint_arraycopy =
2012 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2013 StubRoutines::_jlong_arraycopy =
2014 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2015 "jlong_arraycopy");
2016
2017 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2018 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2019 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2020 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2021 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2022 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2023
2024 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2025 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2026 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2027 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2028
2029 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2030 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2031 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2032 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2033
2034 StubRoutines::_checkcast_arraycopy =
2035 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2036 StubRoutines::_checkcast_arraycopy_uninit =
2037 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2038
2039 StubRoutines::_unsafe_arraycopy =
2040 generate_unsafe_copy("unsafe_arraycopy",
2041 entry_jbyte_arraycopy,
2042 entry_jshort_arraycopy,
2043 entry_jint_arraycopy,
2044 entry_jlong_arraycopy);
2045
2046 StubRoutines::_generic_arraycopy =
2047 generate_generic_copy("generic_arraycopy",
2048 entry_jbyte_arraycopy,
2049 entry_jshort_arraycopy,
2050 entry_jint_arraycopy,
2051 entry_oop_arraycopy,
2052 entry_jlong_arraycopy,
2053 entry_checkcast_arraycopy);
2054 }
2055
2056 // AES intrinsic stubs
2057 enum {AESBlockSize = 16};
2058
2059 address generate_key_shuffle_mask() {
2060 __ align(16);
2061 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2062 address start = __ pc();
2063 __ emit_data(0x00010203, relocInfo::none, 0 );
2064 __ emit_data(0x04050607, relocInfo::none, 0 );
2065 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2066 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2067 return start;
2068 }
2069
2070 address generate_counter_shuffle_mask() {
2071 __ align(16);
2072 StubCodeMark mark(this, "StubRoutines", "counter_shuffle_mask");
2073 address start = __ pc();
2074 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
2075 __ emit_data(0x08090a0b, relocInfo::none, 0);
2076 __ emit_data(0x04050607, relocInfo::none, 0);
2077 __ emit_data(0x00010203, relocInfo::none, 0);
2078 return start;
2079 }
2080
2081 // Utility routine for loading a 128-bit key word in little endian format
2082 // can optionally specify that the shuffle mask is already in an xmmregister
2083 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2084 __ movdqu(xmmdst, Address(key, offset));
2085 if (xmm_shuf_mask != NULL) {
2086 __ pshufb(xmmdst, xmm_shuf_mask);
2087 } else {
2088 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2089 }
2090 }
2091
2092 // aesenc using specified key+offset
2093 // can optionally specify that the shuffle mask is already in an xmmregister
2094 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2095 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2096 __ aesenc(xmmdst, xmmtmp);
2097 }
2098
2099 // aesdec using specified key+offset
2100 // can optionally specify that the shuffle mask is already in an xmmregister
2101 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2102 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2103 __ aesdec(xmmdst, xmmtmp);
2104 }
2105
2106 // Utility routine for increase 128bit counter (iv in CTR mode)
2107 // XMM_128bit, D3, D2, D1, D0
2108 void inc_counter(Register reg, XMMRegister xmmdst, int inc_delta, Label& next_block) {
2109 __ pextrd(reg, xmmdst, 0x0);
2110 __ addl(reg, inc_delta);
2111 __ pinsrd(xmmdst, reg, 0x0);
2112 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2113
2114 __ pextrd(reg, xmmdst, 0x01); // Carry-> D1
2115 __ addl(reg, 0x01);
2116 __ pinsrd(xmmdst, reg, 0x01);
2117 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2118
2119 __ pextrd(reg, xmmdst, 0x02); // Carry-> D2
2120 __ addl(reg, 0x01);
2121 __ pinsrd(xmmdst, reg, 0x02);
2122 __ jcc(Assembler::carryClear, next_block); // jump if no carry
2123
2124 __ pextrd(reg, xmmdst, 0x03); // Carry -> D3
2125 __ addl(reg, 0x01);
2126 __ pinsrd(xmmdst, reg, 0x03);
2127
2128 __ BIND(next_block); // next instruction
2129 }
2130
2131
2132 // Arguments:
2133 //
2134 // Inputs:
2135 // c_rarg0 - source byte array address
2136 // c_rarg1 - destination byte array address
2137 // c_rarg2 - K (key) in little endian int array
2138 //
2139 address generate_aescrypt_encryptBlock() {
2140 assert(UseAES, "need AES instructions and misaligned SSE support");
2141 __ align(CodeEntryAlignment);
2142 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2143 Label L_doLast;
2144 address start = __ pc();
2145
2146 const Register from = rdx; // source array address
2147 const Register to = rdx; // destination array address
2148 const Register key = rcx; // key array address
2149 const Register keylen = rax;
2150 const Address from_param(rbp, 8+0);
2151 const Address to_param (rbp, 8+4);
2152 const Address key_param (rbp, 8+8);
2153
2154 const XMMRegister xmm_result = xmm0;
2155 const XMMRegister xmm_key_shuf_mask = xmm1;
2156 const XMMRegister xmm_temp1 = xmm2;
2157 const XMMRegister xmm_temp2 = xmm3;
2158 const XMMRegister xmm_temp3 = xmm4;
2159 const XMMRegister xmm_temp4 = xmm5;
2160
2161 __ enter(); // required for proper stackwalking of RuntimeStub frame
2162
2163 __ movptr(from, from_param);
2164 __ movptr(key, key_param);
2165
2166 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2167 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2168
2169 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2170 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2171 __ movptr(to, to_param);
2172
2173 // For encryption, the java expanded key ordering is just what we need
2174
2175 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2176 __ pxor(xmm_result, xmm_temp1);
2177
2178 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2179 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2180 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2181 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2182
2183 __ aesenc(xmm_result, xmm_temp1);
2184 __ aesenc(xmm_result, xmm_temp2);
2185 __ aesenc(xmm_result, xmm_temp3);
2186 __ aesenc(xmm_result, xmm_temp4);
2187
2188 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2189 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2190 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2191 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2192
2193 __ aesenc(xmm_result, xmm_temp1);
2194 __ aesenc(xmm_result, xmm_temp2);
2195 __ aesenc(xmm_result, xmm_temp3);
2196 __ aesenc(xmm_result, xmm_temp4);
2197
2198 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2199 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2200
2201 __ cmpl(keylen, 44);
2202 __ jccb(Assembler::equal, L_doLast);
2203
2204 __ aesenc(xmm_result, xmm_temp1);
2205 __ aesenc(xmm_result, xmm_temp2);
2206
2207 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2208 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2209
2210 __ cmpl(keylen, 52);
2211 __ jccb(Assembler::equal, L_doLast);
2212
2213 __ aesenc(xmm_result, xmm_temp1);
2214 __ aesenc(xmm_result, xmm_temp2);
2215
2216 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2217 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2218
2219 __ BIND(L_doLast);
2220 __ aesenc(xmm_result, xmm_temp1);
2221 __ aesenclast(xmm_result, xmm_temp2);
2222 __ movdqu(Address(to, 0), xmm_result); // store the result
2223 __ xorptr(rax, rax); // return 0
2224 __ leave(); // required for proper stackwalking of RuntimeStub frame
2225 __ ret(0);
2226
2227 return start;
2228 }
2229
2230
2231 // Arguments:
2232 //
2233 // Inputs:
2234 // c_rarg0 - source byte array address
2235 // c_rarg1 - destination byte array address
2236 // c_rarg2 - K (key) in little endian int array
2237 //
2238 address generate_aescrypt_decryptBlock() {
2239 assert(UseAES, "need AES instructions and misaligned SSE support");
2240 __ align(CodeEntryAlignment);
2241 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2242 Label L_doLast;
2243 address start = __ pc();
2244
2245 const Register from = rdx; // source array address
2246 const Register to = rdx; // destination array address
2247 const Register key = rcx; // key array address
2248 const Register keylen = rax;
2249 const Address from_param(rbp, 8+0);
2250 const Address to_param (rbp, 8+4);
2251 const Address key_param (rbp, 8+8);
2252
2253 const XMMRegister xmm_result = xmm0;
2254 const XMMRegister xmm_key_shuf_mask = xmm1;
2255 const XMMRegister xmm_temp1 = xmm2;
2256 const XMMRegister xmm_temp2 = xmm3;
2257 const XMMRegister xmm_temp3 = xmm4;
2258 const XMMRegister xmm_temp4 = xmm5;
2259
2260 __ enter(); // required for proper stackwalking of RuntimeStub frame
2261
2262 __ movptr(from, from_param);
2263 __ movptr(key, key_param);
2264
2265 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2266 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2267
2268 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2269 __ movdqu(xmm_result, Address(from, 0));
2270 __ movptr(to, to_param);
2271
2272 // for decryption java expanded key ordering is rotated one position from what we want
2273 // so we start from 0x10 here and hit 0x00 last
2274 // we don't know if the key is aligned, hence not using load-execute form
2275 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2276 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2277 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2278 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2279
2280 __ pxor (xmm_result, xmm_temp1);
2281 __ aesdec(xmm_result, xmm_temp2);
2282 __ aesdec(xmm_result, xmm_temp3);
2283 __ aesdec(xmm_result, xmm_temp4);
2284
2285 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2286 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2287 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2288 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2289
2290 __ aesdec(xmm_result, xmm_temp1);
2291 __ aesdec(xmm_result, xmm_temp2);
2292 __ aesdec(xmm_result, xmm_temp3);
2293 __ aesdec(xmm_result, xmm_temp4);
2294
2295 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2296 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2297 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2298
2299 __ cmpl(keylen, 44);
2300 __ jccb(Assembler::equal, L_doLast);
2301
2302 __ aesdec(xmm_result, xmm_temp1);
2303 __ aesdec(xmm_result, xmm_temp2);
2304
2305 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2306 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2307
2308 __ cmpl(keylen, 52);
2309 __ jccb(Assembler::equal, L_doLast);
2310
2311 __ aesdec(xmm_result, xmm_temp1);
2312 __ aesdec(xmm_result, xmm_temp2);
2313
2314 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2315 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2316
2317 __ BIND(L_doLast);
2318 __ aesdec(xmm_result, xmm_temp1);
2319 __ aesdec(xmm_result, xmm_temp2);
2320
2321 // for decryption the aesdeclast operation is always on key+0x00
2322 __ aesdeclast(xmm_result, xmm_temp3);
2323 __ movdqu(Address(to, 0), xmm_result); // store the result
2324 __ xorptr(rax, rax); // return 0
2325 __ leave(); // required for proper stackwalking of RuntimeStub frame
2326 __ ret(0);
2327
2328 return start;
2329 }
2330
2331 void handleSOERegisters(bool saving) {
2332 const int saveFrameSizeInBytes = 4 * wordSize;
2333 const Address saved_rbx (rbp, -3 * wordSize);
2334 const Address saved_rsi (rbp, -2 * wordSize);
2335 const Address saved_rdi (rbp, -1 * wordSize);
2336
2337 if (saving) {
2338 __ subptr(rsp, saveFrameSizeInBytes);
2339 __ movptr(saved_rsi, rsi);
2340 __ movptr(saved_rdi, rdi);
2341 __ movptr(saved_rbx, rbx);
2342 } else {
2343 // restoring
2344 __ movptr(rsi, saved_rsi);
2345 __ movptr(rdi, saved_rdi);
2346 __ movptr(rbx, saved_rbx);
2347 }
2348 }
2349
2350 // Arguments:
2351 //
2352 // Inputs:
2353 // c_rarg0 - source byte array address
2354 // c_rarg1 - destination byte array address
2355 // c_rarg2 - K (key) in little endian int array
2356 // c_rarg3 - r vector byte array address
2357 // c_rarg4 - input length
2358 //
2359 // Output:
2360 // rax - input length
2361 //
2362 address generate_cipherBlockChaining_encryptAESCrypt() {
2363 assert(UseAES, "need AES instructions and misaligned SSE support");
2364 __ align(CodeEntryAlignment);
2365 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2366 address start = __ pc();
2367
2368 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2369 const Register from = rsi; // source array address
2370 const Register to = rdx; // destination array address
2371 const Register key = rcx; // key array address
2372 const Register rvec = rdi; // r byte array initialized from initvector array address
2373 // and left with the results of the last encryption block
2374 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2375 const Register pos = rax;
2376
2377 // xmm register assignments for the loops below
2378 const XMMRegister xmm_result = xmm0;
2379 const XMMRegister xmm_temp = xmm1;
2380 // first 6 keys preloaded into xmm2-xmm7
2381 const int XMM_REG_NUM_KEY_FIRST = 2;
2382 const int XMM_REG_NUM_KEY_LAST = 7;
2383 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2384
2385 __ enter(); // required for proper stackwalking of RuntimeStub frame
2386 handleSOERegisters(true /*saving*/);
2387
2388 // load registers from incoming parameters
2389 const Address from_param(rbp, 8+0);
2390 const Address to_param (rbp, 8+4);
2391 const Address key_param (rbp, 8+8);
2392 const Address rvec_param (rbp, 8+12);
2393 const Address len_param (rbp, 8+16);
2394 __ movptr(from , from_param);
2395 __ movptr(to , to_param);
2396 __ movptr(key , key_param);
2397 __ movptr(rvec , rvec_param);
2398 __ movptr(len_reg , len_param);
2399
2400 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2401 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2402 // load up xmm regs 2 thru 7 with keys 0-5
2403 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2404 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2405 offset += 0x10;
2406 }
2407
2408 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2409
2410 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2411 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2412 __ cmpl(rax, 44);
2413 __ jcc(Assembler::notEqual, L_key_192_256);
2414
2415 // 128 bit code follows here
2416 __ movl(pos, 0);
2417 __ align(OptoLoopAlignment);
2418 __ BIND(L_loopTop_128);
2419 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2420 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2421
2422 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2423 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2424 __ aesenc(xmm_result, as_XMMRegister(rnum));
2425 }
2426 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2427 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2428 }
2429 load_key(xmm_temp, key, 0xa0);
2430 __ aesenclast(xmm_result, xmm_temp);
2431
2432 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2433 // no need to store r to memory until we exit
2434 __ addptr(pos, AESBlockSize);
2435 __ subptr(len_reg, AESBlockSize);
2436 __ jcc(Assembler::notEqual, L_loopTop_128);
2437
2438 __ BIND(L_exit);
2439 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2440
2441 handleSOERegisters(false /*restoring*/);
2442 __ movptr(rax, len_param); // return length
2443 __ leave(); // required for proper stackwalking of RuntimeStub frame
2444 __ ret(0);
2445
2446 __ BIND(L_key_192_256);
2447 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2448 __ cmpl(rax, 52);
2449 __ jcc(Assembler::notEqual, L_key_256);
2450
2451 // 192-bit code follows here (could be changed to use more xmm registers)
2452 __ movl(pos, 0);
2453 __ align(OptoLoopAlignment);
2454 __ BIND(L_loopTop_192);
2455 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2456 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2457
2458 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2459 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2460 __ aesenc(xmm_result, as_XMMRegister(rnum));
2461 }
2462 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2463 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2464 }
2465 load_key(xmm_temp, key, 0xc0);
2466 __ aesenclast(xmm_result, xmm_temp);
2467
2468 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2469 // no need to store r to memory until we exit
2470 __ addptr(pos, AESBlockSize);
2471 __ subptr(len_reg, AESBlockSize);
2472 __ jcc(Assembler::notEqual, L_loopTop_192);
2473 __ jmp(L_exit);
2474
2475 __ BIND(L_key_256);
2476 // 256-bit code follows here (could be changed to use more xmm registers)
2477 __ movl(pos, 0);
2478 __ align(OptoLoopAlignment);
2479 __ BIND(L_loopTop_256);
2480 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2481 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2482
2483 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2484 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2485 __ aesenc(xmm_result, as_XMMRegister(rnum));
2486 }
2487 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2488 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2489 }
2490 load_key(xmm_temp, key, 0xe0);
2491 __ aesenclast(xmm_result, xmm_temp);
2492
2493 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2494 // no need to store r to memory until we exit
2495 __ addptr(pos, AESBlockSize);
2496 __ subptr(len_reg, AESBlockSize);
2497 __ jcc(Assembler::notEqual, L_loopTop_256);
2498 __ jmp(L_exit);
2499
2500 return start;
2501 }
2502
2503
2504 // CBC AES Decryption.
2505 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2506 //
2507 // Arguments:
2508 //
2509 // Inputs:
2510 // c_rarg0 - source byte array address
2511 // c_rarg1 - destination byte array address
2512 // c_rarg2 - K (key) in little endian int array
2513 // c_rarg3 - r vector byte array address
2514 // c_rarg4 - input length
2515 //
2516 // Output:
2517 // rax - input length
2518 //
2519
2520 address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
2521 assert(UseAES, "need AES instructions and misaligned SSE support");
2522 __ align(CodeEntryAlignment);
2523 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2524 address start = __ pc();
2525
2526 const Register from = rsi; // source array address
2527 const Register to = rdx; // destination array address
2528 const Register key = rcx; // key array address
2529 const Register rvec = rdi; // r byte array initialized from initvector array address
2530 // and left with the results of the last encryption block
2531 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2532 const Register pos = rax;
2533
2534 const int PARALLEL_FACTOR = 4;
2535 const int ROUNDS[3] = { 10, 12, 14 }; //aes rounds for key128, key192, key256
2536
2537 Label L_exit;
2538 Label L_singleBlock_loopTop[3]; //128, 192, 256
2539 Label L_multiBlock_loopTop[3]; //128, 192, 256
2540
2541 const XMMRegister xmm_prev_block_cipher = xmm0; // holds cipher of previous block
2542 const XMMRegister xmm_key_shuf_mask = xmm1;
2543
2544 const XMMRegister xmm_key_tmp0 = xmm2;
2545 const XMMRegister xmm_key_tmp1 = xmm3;
2546
2547 // registers holding the six results in the parallelized loop
2548 const XMMRegister xmm_result0 = xmm4;
2549 const XMMRegister xmm_result1 = xmm5;
2550 const XMMRegister xmm_result2 = xmm6;
2551 const XMMRegister xmm_result3 = xmm7;
2552
2553 __ enter(); // required for proper stackwalking of RuntimeStub frame
2554 handleSOERegisters(true /*saving*/);
2555
2556 // load registers from incoming parameters
2557 const Address from_param(rbp, 8+0);
2558 const Address to_param (rbp, 8+4);
2559 const Address key_param (rbp, 8+8);
2560 const Address rvec_param (rbp, 8+12);
2561 const Address len_param (rbp, 8+16);
2562
2563 __ movptr(from , from_param);
2564 __ movptr(to , to_param);
2565 __ movptr(key , key_param);
2566 __ movptr(rvec , rvec_param);
2567 __ movptr(len_reg , len_param);
2568
2569 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2570 __ movdqu(xmm_prev_block_cipher, Address(rvec, 0x00)); // initialize with initial rvec
2571
2572 __ xorptr(pos, pos);
2573
2574 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2575 // rvec is reused
2576 __ movl(rvec, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2577 __ cmpl(rvec, 52);
2578 __ jcc(Assembler::equal, L_multiBlock_loopTop[1]);
2579 __ cmpl(rvec, 60);
2580 __ jcc(Assembler::equal, L_multiBlock_loopTop[2]);
2581
2582 #define DoFour(opc, src_reg) \
2583 __ opc(xmm_result0, src_reg); \
2584 __ opc(xmm_result1, src_reg); \
2585 __ opc(xmm_result2, src_reg); \
2586 __ opc(xmm_result3, src_reg); \
2587
2588 for (int k = 0; k < 3; ++k) {
2589 __ align(OptoLoopAlignment);
2590 __ BIND(L_multiBlock_loopTop[k]);
2591 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least 4 blocks left
2592 __ jcc(Assembler::less, L_singleBlock_loopTop[k]);
2593
2594 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0 * AESBlockSize)); // get next 4 blocks into xmmresult registers
2595 __ movdqu(xmm_result1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2596 __ movdqu(xmm_result2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2597 __ movdqu(xmm_result3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2598
2599 // the java expanded key ordering is rotated one position from what we want
2600 // so we start from 0x10 here and hit 0x00 last
2601 load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2602 DoFour(pxor, xmm_key_tmp0); //xor with first key
2603 // do the aes dec rounds
2604 for (int rnum = 1; rnum <= ROUNDS[k];) {
2605 //load two keys at a time
2606 //k1->0x20, ..., k9->0xa0, k10->0x00
2607 load_key(xmm_key_tmp1, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2608 load_key(xmm_key_tmp0, key, ((rnum + 2) % (ROUNDS[k] + 1)) * 0x10, xmm_key_shuf_mask); // hit 0x00 last!
2609 DoFour(aesdec, xmm_key_tmp1);
2610 rnum++;
2611 if (rnum != ROUNDS[k]) {
2612 DoFour(aesdec, xmm_key_tmp0);
2613 }
2614 else {
2615 DoFour(aesdeclast, xmm_key_tmp0);
2616 }
2617 rnum++;
2618 }
2619
2620 // for each result, xor with the r vector of previous cipher block
2621 __ pxor(xmm_result0, xmm_prev_block_cipher);
2622 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2623 __ pxor(xmm_result1, xmm_prev_block_cipher);
2624 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2625 __ pxor(xmm_result2, xmm_prev_block_cipher);
2626 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2627 __ pxor(xmm_result3, xmm_prev_block_cipher);
2628 __ movdqu(xmm_prev_block_cipher, Address(from, pos, Address::times_1, 3 * AESBlockSize)); // this will carry over to next set of blocks
2629
2630 // store 4 results into the next 64 bytes of output
2631 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2632 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2633 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2634 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2635
2636 __ addptr(pos, 4 * AESBlockSize);
2637 __ subptr(len_reg, 4 * AESBlockSize);
2638 __ jmp(L_multiBlock_loopTop[k]);
2639
2640 //singleBlock starts here
2641 __ align(OptoLoopAlignment);
2642 __ BIND(L_singleBlock_loopTop[k]);
2643 __ cmpptr(len_reg, 0); // any blocks left?
2644 __ jcc(Assembler::equal, L_exit);
2645 __ movdqu(xmm_result0, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2646 __ movdqa(xmm_result1, xmm_result0);
2647
2648 load_key(xmm_key_tmp0, key, 0x10, xmm_key_shuf_mask);
2649 __ pxor(xmm_result0, xmm_key_tmp0);
2650 // do the aes dec rounds
2651 for (int rnum = 1; rnum < ROUNDS[k]; rnum++) {
2652 // the java expanded key ordering is rotated one position from what we want
2653 load_key(xmm_key_tmp0, key, (rnum + 1) * 0x10, xmm_key_shuf_mask);
2654 __ aesdec(xmm_result0, xmm_key_tmp0);
2655 }
2656 load_key(xmm_key_tmp0, key, 0x00, xmm_key_shuf_mask);
2657 __ aesdeclast(xmm_result0, xmm_key_tmp0);
2658 __ pxor(xmm_result0, xmm_prev_block_cipher); // xor with the current r vector
2659 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result0); // store into the next 16 bytes of output
2660 // no need to store r to memory until we exit
2661 __ movdqa(xmm_prev_block_cipher, xmm_result1); // set up next r vector with cipher input from this block
2662
2663 __ addptr(pos, AESBlockSize);
2664 __ subptr(len_reg, AESBlockSize);
2665 __ jmp(L_singleBlock_loopTop[k]);
2666 }//for 128/192/256
2667
2668 __ BIND(L_exit);
2669 __ movptr(rvec, rvec_param); // restore this since reused earlier
2670 __ movdqu(Address(rvec, 0), xmm_prev_block_cipher); // final value of r stored in rvec of CipherBlockChaining object
2671 handleSOERegisters(false /*restoring*/);
2672 __ movptr(rax, len_param); // return length
2673 __ leave(); // required for proper stackwalking of RuntimeStub frame
2674 __ ret(0);
2675
2676 return start;
2677 }
2678
2679 // CTR AES crypt.
2680 // In 32-bit stub, parallelize 4 blocks at a time
2681 // Arguments:
2682 //
2683 // Inputs:
2684 // c_rarg0 - source byte array address
2685 // c_rarg1 - destination byte array address
2686 // c_rarg2 - K (key) in little endian int array
2687 // c_rarg3 - counter vector byte array address
2688 // c_rarg4 - input length
2689 //
2690 // Output:
2691 // rax - input length
2692 //
2693 address generate_counterMode_AESCrypt_Parallel() {
2694 assert(UseAES, "need AES instructions and misaligned SSE support");
2695 __ align(CodeEntryAlignment);
2696 StubCodeMark mark(this, "StubRoutines", "counterMode_AESCrypt");
2697 address start = __ pc();
2698 const Register from = rsi; // source array address
2699 const Register to = rdx; // destination array address
2700 const Register key = rcx; // key array address
2701 const Register counter = rdi; // counter byte array initialized from initvector array address
2702 // and updated with the incremented counter in the end
2703 const Register len_reg = rbx;
2704 const Register pos = rax;
2705
2706 __ enter(); // required for proper stackwalking of RuntimeStub frame
2707 handleSOERegisters(true /*saving*/); // save rbx, rsi, rdi
2708
2709 // load registers from incoming parameters
2710 const Address from_param(rbp, 8+0);
2711 const Address to_param (rbp, 8+4);
2712 const Address key_param (rbp, 8+8);
2713 const Address rvec_param (rbp, 8+12);
2714 const Address len_param (rbp, 8+16);
2715 const Address saved_counter_param(rbp, 8 + 20);
2716 const Address used_addr_param(rbp, 8 + 24);
2717
2718 __ movptr(from , from_param);
2719 __ movptr(to , to_param);
2720 __ movptr(len_reg , len_param);
2721
2722 // Use the partially used encrpyted counter from last invocation
2723 Label L_exit_preLoop, L_preLoop_start;
2724
2725 // Use the registers 'counter' and 'key' here in this preloop
2726 // to hold of last 2 params 'used' and 'saved_encCounter_start'
2727 Register used = counter;
2728 Register saved_encCounter_start = key;
2729 Register used_addr = saved_encCounter_start;
2730
2731 __ movptr(used_addr, used_addr_param);
2732 __ movptr(used, Address(used_addr, 0));
2733 __ movptr(saved_encCounter_start, saved_counter_param);
2734
2735 __ BIND(L_preLoop_start);
2736 __ cmpptr(used, 16);
2737 __ jcc(Assembler::aboveEqual, L_exit_preLoop);
2738 __ cmpptr(len_reg, 0);
2739 __ jcc(Assembler::lessEqual, L_exit_preLoop);
2740 __ movb(rax, Address(saved_encCounter_start, used));
2741 __ xorb(rax, Address(from, 0));
2742 __ movb(Address(to, 0), rax);
2743 __ addptr(from, 1);
2744 __ addptr(to, 1);
2745 __ addptr(used, 1);
2746 __ subptr(len_reg, 1);
2747
2748 __ jmp(L_preLoop_start);
2749
2750 __ BIND(L_exit_preLoop);
2751 __ movptr(used_addr, used_addr_param);
2752 __ movptr(used_addr, used_addr_param);
2753 __ movl(Address(used_addr, 0), used);
2754
2755 // load the parameters 'key' and 'counter'
2756 __ movptr(key, key_param);
2757 __ movptr(counter, rvec_param);
2758
2759 // xmm register assignments for the loops below
2760 const XMMRegister xmm_curr_counter = xmm0;
2761 const XMMRegister xmm_counter_shuf_mask = xmm1; // need to be reloaded
2762 const XMMRegister xmm_key_shuf_mask = xmm2; // need to be reloaded
2763 const XMMRegister xmm_key = xmm3;
2764 const XMMRegister xmm_result0 = xmm4;
2765 const XMMRegister xmm_result1 = xmm5;
2766 const XMMRegister xmm_result2 = xmm6;
2767 const XMMRegister xmm_result3 = xmm7;
2768 const XMMRegister xmm_from0 = xmm1; //reuse XMM register
2769 const XMMRegister xmm_from1 = xmm2;
2770 const XMMRegister xmm_from2 = xmm3;
2771 const XMMRegister xmm_from3 = xmm4;
2772
2773 //for key_128, key_192, key_256
2774 const int rounds[3] = {10, 12, 14};
2775 Label L_singleBlockLoopTop[3];
2776 Label L_multiBlock_loopTop[3];
2777 Label L_key192_top, L_key256_top;
2778 Label L_incCounter[3][4]; // 3: different key length, 4: 4 blocks at a time
2779 Label L_incCounter_single[3]; //for single block, key128, key192, key256
2780 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];
2781 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];
2782
2783 Label L_exit;
2784 const int PARALLEL_FACTOR = 4; //because of the limited register number
2785
2786 // initialize counter with initial counter
2787 __ movdqu(xmm_curr_counter, Address(counter, 0x00));
2788 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2789 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled for increase
2790
2791 // key length could be only {11, 13, 15} * 4 = {44, 52, 60}
2792 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2793 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2794 __ cmpl(rax, 52);
2795 __ jcc(Assembler::equal, L_key192_top);
2796 __ cmpl(rax, 60);
2797 __ jcc(Assembler::equal, L_key256_top);
2798
2799 //key128 begins here
2800 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2801
2802 #define CTR_DoFour(opc, src_reg) \
2803 __ opc(xmm_result0, src_reg); \
2804 __ opc(xmm_result1, src_reg); \
2805 __ opc(xmm_result2, src_reg); \
2806 __ opc(xmm_result3, src_reg);
2807
2808 // k == 0 : generate code for key_128
2809 // k == 1 : generate code for key_192
2810 // k == 2 : generate code for key_256
2811 for (int k = 0; k < 3; ++k) {
2812 //multi blocks starts here
2813 __ align(OptoLoopAlignment);
2814 __ BIND(L_multiBlock_loopTop[k]);
2815 __ cmpptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // see if at least PARALLEL_FACTOR blocks left
2816 __ jcc(Assembler::less, L_singleBlockLoopTop[k]);
2817
2818 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2819 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2820
2821 //load, then increase counters
2822 CTR_DoFour(movdqa, xmm_curr_counter);
2823 __ push(rbx);
2824 inc_counter(rbx, xmm_result1, 0x01, L_incCounter[k][0]);
2825 inc_counter(rbx, xmm_result2, 0x02, L_incCounter[k][1]);
2826 inc_counter(rbx, xmm_result3, 0x03, L_incCounter[k][2]);
2827 inc_counter(rbx, xmm_curr_counter, 0x04, L_incCounter[k][3]);
2828 __ pop (rbx);
2829
2830 load_key(xmm_key, key, 0x00, xmm_key_shuf_mask); // load Round 0 key. interleaving for better performance
2831
2832 CTR_DoFour(pshufb, xmm_counter_shuf_mask); // after increased, shuffled counters back for PXOR
2833 CTR_DoFour(pxor, xmm_key); //PXOR with Round 0 key
2834
2835 for (int i = 1; i < rounds[k]; ++i) {
2836 load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2837 CTR_DoFour(aesenc, xmm_key);
2838 }
2839 load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2840 CTR_DoFour(aesenclast, xmm_key);
2841
2842 // get next PARALLEL_FACTOR blocks into xmm_from registers
2843 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2844 __ movdqu(xmm_from1, Address(from, pos, Address::times_1, 1 * AESBlockSize));
2845 __ movdqu(xmm_from2, Address(from, pos, Address::times_1, 2 * AESBlockSize));
2846
2847 // PXOR with input text
2848 __ pxor(xmm_result0, xmm_from0); //result0 is xmm4
2849 __ pxor(xmm_result1, xmm_from1);
2850 __ pxor(xmm_result2, xmm_from2);
2851
2852 // store PARALLEL_FACTOR results into the next 64 bytes of output
2853 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2854 __ movdqu(Address(to, pos, Address::times_1, 1 * AESBlockSize), xmm_result1);
2855 __ movdqu(Address(to, pos, Address::times_1, 2 * AESBlockSize), xmm_result2);
2856
2857 // do it here after xmm_result0 is saved, because xmm_from3 reuse the same register of xmm_result0.
2858 __ movdqu(xmm_from3, Address(from, pos, Address::times_1, 3 * AESBlockSize));
2859 __ pxor(xmm_result3, xmm_from3);
2860 __ movdqu(Address(to, pos, Address::times_1, 3 * AESBlockSize), xmm_result3);
2861
2862 __ addptr(pos, PARALLEL_FACTOR * AESBlockSize); // increase the length of crypt text
2863 __ subptr(len_reg, PARALLEL_FACTOR * AESBlockSize); // decrease the remaining length
2864 __ jmp(L_multiBlock_loopTop[k]);
2865
2866 // singleBlock starts here
2867 __ align(OptoLoopAlignment);
2868 __ BIND(L_singleBlockLoopTop[k]);
2869 __ cmpptr(len_reg, 0);
2870 __ jcc(Assembler::equal, L_exit);
2871 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2872 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2873 __ movdqa(xmm_result0, xmm_curr_counter);
2874 load_key(xmm_key, key, 0x00, xmm_key_shuf_mask);
2875 __ push(rbx);//rbx is used for increasing counter
2876 inc_counter(rbx, xmm_curr_counter, 0x01, L_incCounter_single[k]);
2877 __ pop (rbx);
2878 __ pshufb(xmm_result0, xmm_counter_shuf_mask);
2879 __ pxor(xmm_result0, xmm_key);
2880 for (int i = 1; i < rounds[k]; i++) {
2881 load_key(xmm_key, key, (0x10 * i), xmm_key_shuf_mask);
2882 __ aesenc(xmm_result0, xmm_key);
2883 }
2884 load_key(xmm_key, key, (0x10 * rounds[k]), xmm_key_shuf_mask);
2885 __ aesenclast(xmm_result0, xmm_key);
2886 __ cmpptr(len_reg, AESBlockSize);
2887 __ jcc(Assembler::less, L_processTail_insr[k]);
2888 __ movdqu(xmm_from0, Address(from, pos, Address::times_1, 0 * AESBlockSize));
2889 __ pxor(xmm_result0, xmm_from0);
2890 __ movdqu(Address(to, pos, Address::times_1, 0 * AESBlockSize), xmm_result0);
2891 __ addptr(pos, AESBlockSize);
2892 __ subptr(len_reg, AESBlockSize);
2893 __ jmp(L_singleBlockLoopTop[k]);
2894
2895 __ BIND(L_processTail_insr[k]); // Process the tail part of the input array
2896 __ addptr(pos, len_reg); // 1. Insert bytes from src array into xmm_from0 register
2897 __ testptr(len_reg, 8);
2898 __ jcc(Assembler::zero, L_processTail_4_insr[k]);
2899 __ subptr(pos,8);
2900 __ pinsrd(xmm_from0, Address(from, pos), 0);
2901 __ pinsrd(xmm_from0, Address(from, pos, Address::times_1, 4), 1);
2902 __ BIND(L_processTail_4_insr[k]);
2903 __ testptr(len_reg, 4);
2904 __ jcc(Assembler::zero, L_processTail_2_insr[k]);
2905 __ subptr(pos,4);
2906 __ pslldq(xmm_from0, 4);
2907 __ pinsrd(xmm_from0, Address(from, pos), 0);
2908 __ BIND(L_processTail_2_insr[k]);
2909 __ testptr(len_reg, 2);
2910 __ jcc(Assembler::zero, L_processTail_1_insr[k]);
2911 __ subptr(pos, 2);
2912 __ pslldq(xmm_from0, 2);
2913 __ pinsrw(xmm_from0, Address(from, pos), 0);
2914 __ BIND(L_processTail_1_insr[k]);
2915 __ testptr(len_reg, 1);
2916 __ jcc(Assembler::zero, L_processTail_exit_insr[k]);
2917 __ subptr(pos, 1);
2918 __ pslldq(xmm_from0, 1);
2919 __ pinsrb(xmm_from0, Address(from, pos), 0);
2920 __ BIND(L_processTail_exit_insr[k]);
2921
2922 __ movptr(saved_encCounter_start, saved_counter_param);
2923 __ movdqu(Address(saved_encCounter_start, 0), xmm_result0); // 2. Perform pxor of the encrypted counter and plaintext Bytes.
2924 __ pxor(xmm_result0, xmm_from0); // Also the encrypted counter is saved for next invocation.
2925
2926 __ testptr(len_reg, 8);
2927 __ jcc(Assembler::zero, L_processTail_4_extr[k]); // 3. Extract bytes from xmm_result0 into the dest. array
2928 __ pextrd(Address(to, pos), xmm_result0, 0);
2929 __ pextrd(Address(to, pos, Address::times_1, 4), xmm_result0, 1);
2930 __ psrldq(xmm_result0, 8);
2931 __ addptr(pos, 8);
2932 __ BIND(L_processTail_4_extr[k]);
2933 __ testptr(len_reg, 4);
2934 __ jcc(Assembler::zero, L_processTail_2_extr[k]);
2935 __ pextrd(Address(to, pos), xmm_result0, 0);
2936 __ psrldq(xmm_result0, 4);
2937 __ addptr(pos, 4);
2938 __ BIND(L_processTail_2_extr[k]);
2939 __ testptr(len_reg, 2);
2940 __ jcc(Assembler::zero, L_processTail_1_extr[k]);
2941 __ pextrb(Address(to, pos), xmm_result0, 0);
2942 __ pextrb(Address(to, pos, Address::times_1, 1), xmm_result0, 1);
2943 __ psrldq(xmm_result0, 2);
2944 __ addptr(pos, 2);
2945 __ BIND(L_processTail_1_extr[k]);
2946 __ testptr(len_reg, 1);
2947 __ jcc(Assembler::zero, L_processTail_exit_extr[k]);
2948 __ pextrb(Address(to, pos), xmm_result0, 0);
2949
2950 __ BIND(L_processTail_exit_extr[k]);
2951 __ movptr(used_addr, used_addr_param);
2952 __ movl(Address(used_addr, 0), len_reg);
2953 __ jmp(L_exit);
2954 }
2955
2956 __ BIND(L_exit);
2957 __ movdqu(xmm_counter_shuf_mask, ExternalAddress(StubRoutines::x86::counter_shuffle_mask_addr()));
2958 __ pshufb(xmm_curr_counter, xmm_counter_shuf_mask); //counter is shuffled back.
2959 __ movdqu(Address(counter, 0), xmm_curr_counter); //save counter back
2960 handleSOERegisters(false /*restoring*/);
2961 __ movptr(rax, len_param); // return length
2962 __ leave(); // required for proper stackwalking of RuntimeStub frame
2963 __ ret(0);
2964
2965 __ BIND (L_key192_top);
2966 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2967 __ jmp(L_multiBlock_loopTop[1]); //key192
2968
2969 __ BIND (L_key256_top);
2970 __ movptr(pos, 0); // init pos before L_multiBlock_loopTop
2971 __ jmp(L_multiBlock_loopTop[2]); //key192
2972
2973 return start;
2974 }
2975
2976 address generate_upper_word_mask() {
2977 __ align(64);
2978 StubCodeMark mark(this, "StubRoutines", "upper_word_mask");
2979 address start = __ pc();
2980 __ emit_data(0x00000000, relocInfo::none, 0);
2981 __ emit_data(0x00000000, relocInfo::none, 0);
2982 __ emit_data(0x00000000, relocInfo::none, 0);
2983 __ emit_data(0xFFFFFFFF, relocInfo::none, 0);
2984 return start;
2985 }
2986
2987 address generate_shuffle_byte_flip_mask() {
2988 __ align(64);
2989 StubCodeMark mark(this, "StubRoutines", "shuffle_byte_flip_mask");
2990 address start = __ pc();
2991 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
2992 __ emit_data(0x08090a0b, relocInfo::none, 0);
2993 __ emit_data(0x04050607, relocInfo::none, 0);
2994 __ emit_data(0x00010203, relocInfo::none, 0);
2995 return start;
2996 }
2997
2998 // ofs and limit are use for multi-block byte array.
2999 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3000 address generate_sha1_implCompress(bool multi_block, const char *name) {
3001 __ align(CodeEntryAlignment);
3002 StubCodeMark mark(this, "StubRoutines", name);
3003 address start = __ pc();
3004
3005 Register buf = rax;
3006 Register state = rdx;
3007 Register ofs = rcx;
3008 Register limit = rdi;
3009
3010 const Address buf_param(rbp, 8 + 0);
3011 const Address state_param(rbp, 8 + 4);
3012 const Address ofs_param(rbp, 8 + 8);
3013 const Address limit_param(rbp, 8 + 12);
3014
3015 const XMMRegister abcd = xmm0;
3016 const XMMRegister e0 = xmm1;
3017 const XMMRegister e1 = xmm2;
3018 const XMMRegister msg0 = xmm3;
3019
3020 const XMMRegister msg1 = xmm4;
3021 const XMMRegister msg2 = xmm5;
3022 const XMMRegister msg3 = xmm6;
3023 const XMMRegister shuf_mask = xmm7;
3024
3025 __ enter();
3026 __ subptr(rsp, 8 * wordSize);
3027 handleSOERegisters(true /*saving*/);
3028
3029 __ movptr(buf, buf_param);
3030 __ movptr(state, state_param);
3031 if (multi_block) {
3032 __ movptr(ofs, ofs_param);
3033 __ movptr(limit, limit_param);
3034 }
3035
3036 __ fast_sha1(abcd, e0, e1, msg0, msg1, msg2, msg3, shuf_mask,
3037 buf, state, ofs, limit, rsp, multi_block);
3038
3039 handleSOERegisters(false /*restoring*/);
3040 __ addptr(rsp, 8 * wordSize);
3041 __ leave();
3042 __ ret(0);
3043 return start;
3044 }
3045
3046 address generate_pshuffle_byte_flip_mask() {
3047 __ align(64);
3048 StubCodeMark mark(this, "StubRoutines", "pshuffle_byte_flip_mask");
3049 address start = __ pc();
3050 __ emit_data(0x00010203, relocInfo::none, 0);
3051 __ emit_data(0x04050607, relocInfo::none, 0);
3052 __ emit_data(0x08090a0b, relocInfo::none, 0);
3053 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3054 return start;
3055 }
3056
3057 // ofs and limit are use for multi-block byte array.
3058 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
3059 address generate_sha256_implCompress(bool multi_block, const char *name) {
3060 __ align(CodeEntryAlignment);
3061 StubCodeMark mark(this, "StubRoutines", name);
3062 address start = __ pc();
3063
3064 Register buf = rbx;
3065 Register state = rsi;
3066 Register ofs = rdx;
3067 Register limit = rcx;
3068
3069 const Address buf_param(rbp, 8 + 0);
3070 const Address state_param(rbp, 8 + 4);
3071 const Address ofs_param(rbp, 8 + 8);
3072 const Address limit_param(rbp, 8 + 12);
3073
3074 const XMMRegister msg = xmm0;
3075 const XMMRegister state0 = xmm1;
3076 const XMMRegister state1 = xmm2;
3077 const XMMRegister msgtmp0 = xmm3;
3078
3079 const XMMRegister msgtmp1 = xmm4;
3080 const XMMRegister msgtmp2 = xmm5;
3081 const XMMRegister msgtmp3 = xmm6;
3082 const XMMRegister msgtmp4 = xmm7;
3083
3084 __ enter();
3085 __ subptr(rsp, 8 * wordSize);
3086 handleSOERegisters(true /*saving*/);
3087 __ movptr(buf, buf_param);
3088 __ movptr(state, state_param);
3089 if (multi_block) {
3090 __ movptr(ofs, ofs_param);
3091 __ movptr(limit, limit_param);
3092 }
3093
3094 __ fast_sha256(msg, state0, state1, msgtmp0, msgtmp1, msgtmp2, msgtmp3, msgtmp4,
3095 buf, state, ofs, limit, rsp, multi_block);
3096
3097 handleSOERegisters(false);
3098 __ addptr(rsp, 8 * wordSize);
3099 __ leave();
3100 __ ret(0);
3101 return start;
3102 }
3103
3104 // byte swap x86 long
3105 address generate_ghash_long_swap_mask() {
3106 __ align(CodeEntryAlignment);
3107 StubCodeMark mark(this, "StubRoutines", "ghash_long_swap_mask");
3108 address start = __ pc();
3109 __ emit_data(0x0b0a0908, relocInfo::none, 0);
3110 __ emit_data(0x0f0e0d0c, relocInfo::none, 0);
3111 __ emit_data(0x03020100, relocInfo::none, 0);
3112 __ emit_data(0x07060504, relocInfo::none, 0);
3113
3114 return start;
3115 }
3116
3117 // byte swap x86 byte array
3118 address generate_ghash_byte_swap_mask() {
3119 __ align(CodeEntryAlignment);
3120 StubCodeMark mark(this, "StubRoutines", "ghash_byte_swap_mask");
3121 address start = __ pc();
3122 __ emit_data(0x0c0d0e0f, relocInfo::none, 0);
3123 __ emit_data(0x08090a0b, relocInfo::none, 0);
3124 __ emit_data(0x04050607, relocInfo::none, 0);
3125 __ emit_data(0x00010203, relocInfo::none, 0);
3126 return start;
3127 }
3128
3129 /* Single and multi-block ghash operations */
3130 address generate_ghash_processBlocks() {
3131 assert(UseGHASHIntrinsics, "need GHASH intrinsics and CLMUL support");
3132 __ align(CodeEntryAlignment);
3133 Label L_ghash_loop, L_exit;
3134 StubCodeMark mark(this, "StubRoutines", "ghash_processBlocks");
3135 address start = __ pc();
3136
3137 const Register state = rdi;
3138 const Register subkeyH = rsi;
3139 const Register data = rdx;
3140 const Register blocks = rcx;
3141
3142 const Address state_param(rbp, 8+0);
3143 const Address subkeyH_param(rbp, 8+4);
3144 const Address data_param(rbp, 8+8);
3145 const Address blocks_param(rbp, 8+12);
3146
3147 const XMMRegister xmm_temp0 = xmm0;
3148 const XMMRegister xmm_temp1 = xmm1;
3149 const XMMRegister xmm_temp2 = xmm2;
3150 const XMMRegister xmm_temp3 = xmm3;
3151 const XMMRegister xmm_temp4 = xmm4;
3152 const XMMRegister xmm_temp5 = xmm5;
3153 const XMMRegister xmm_temp6 = xmm6;
3154 const XMMRegister xmm_temp7 = xmm7;
3155
3156 __ enter();
3157 handleSOERegisters(true); // Save registers
3158
3159 __ movptr(state, state_param);
3160 __ movptr(subkeyH, subkeyH_param);
3161 __ movptr(data, data_param);
3162 __ movptr(blocks, blocks_param);
3163
3164 __ movdqu(xmm_temp0, Address(state, 0));
3165 __ pshufb(xmm_temp0, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3166
3167 __ movdqu(xmm_temp1, Address(subkeyH, 0));
3168 __ pshufb(xmm_temp1, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3169
3170 __ BIND(L_ghash_loop);
3171 __ movdqu(xmm_temp2, Address(data, 0));
3172 __ pshufb(xmm_temp2, ExternalAddress(StubRoutines::x86::ghash_byte_swap_mask_addr()));
3173
3174 __ pxor(xmm_temp0, xmm_temp2);
3175
3176 //
3177 // Multiply with the hash key
3178 //
3179 __ movdqu(xmm_temp3, xmm_temp0);
3180 __ pclmulqdq(xmm_temp3, xmm_temp1, 0); // xmm3 holds a0*b0
3181 __ movdqu(xmm_temp4, xmm_temp0);
3182 __ pclmulqdq(xmm_temp4, xmm_temp1, 16); // xmm4 holds a0*b1
3183
3184 __ movdqu(xmm_temp5, xmm_temp0);
3185 __ pclmulqdq(xmm_temp5, xmm_temp1, 1); // xmm5 holds a1*b0
3186 __ movdqu(xmm_temp6, xmm_temp0);
3187 __ pclmulqdq(xmm_temp6, xmm_temp1, 17); // xmm6 holds a1*b1
3188
3189 __ pxor(xmm_temp4, xmm_temp5); // xmm4 holds a0*b1 + a1*b0
3190
3191 __ movdqu(xmm_temp5, xmm_temp4); // move the contents of xmm4 to xmm5
3192 __ psrldq(xmm_temp4, 8); // shift by xmm4 64 bits to the right
3193 __ pslldq(xmm_temp5, 8); // shift by xmm5 64 bits to the left
3194 __ pxor(xmm_temp3, xmm_temp5);
3195 __ pxor(xmm_temp6, xmm_temp4); // Register pair <xmm6:xmm3> holds the result
3196 // of the carry-less multiplication of
3197 // xmm0 by xmm1.
3198
3199 // We shift the result of the multiplication by one bit position
3200 // to the left to cope for the fact that the bits are reversed.
3201 __ movdqu(xmm_temp7, xmm_temp3);
3202 __ movdqu(xmm_temp4, xmm_temp6);
3203 __ pslld (xmm_temp3, 1);
3204 __ pslld(xmm_temp6, 1);
3205 __ psrld(xmm_temp7, 31);
3206 __ psrld(xmm_temp4, 31);
3207 __ movdqu(xmm_temp5, xmm_temp7);
3208 __ pslldq(xmm_temp4, 4);
3209 __ pslldq(xmm_temp7, 4);
3210 __ psrldq(xmm_temp5, 12);
3211 __ por(xmm_temp3, xmm_temp7);
3212 __ por(xmm_temp6, xmm_temp4);
3213 __ por(xmm_temp6, xmm_temp5);
3214
3215 //
3216 // First phase of the reduction
3217 //
3218 // Move xmm3 into xmm4, xmm5, xmm7 in order to perform the shifts
3219 // independently.
3220 __ movdqu(xmm_temp7, xmm_temp3);
3221 __ movdqu(xmm_temp4, xmm_temp3);
3222 __ movdqu(xmm_temp5, xmm_temp3);
3223 __ pslld(xmm_temp7, 31); // packed right shift shifting << 31
3224 __ pslld(xmm_temp4, 30); // packed right shift shifting << 30
3225 __ pslld(xmm_temp5, 25); // packed right shift shifting << 25
3226 __ pxor(xmm_temp7, xmm_temp4); // xor the shifted versions
3227 __ pxor(xmm_temp7, xmm_temp5);
3228 __ movdqu(xmm_temp4, xmm_temp7);
3229 __ pslldq(xmm_temp7, 12);
3230 __ psrldq(xmm_temp4, 4);
3231 __ pxor(xmm_temp3, xmm_temp7); // first phase of the reduction complete
3232
3233 //
3234 // Second phase of the reduction
3235 //
3236 // Make 3 copies of xmm3 in xmm2, xmm5, xmm7 for doing these
3237 // shift operations.
3238 __ movdqu(xmm_temp2, xmm_temp3);
3239 __ movdqu(xmm_temp7, xmm_temp3);
3240 __ movdqu(xmm_temp5, xmm_temp3);
3241 __ psrld(xmm_temp2, 1); // packed left shifting >> 1
3242 __ psrld(xmm_temp7, 2); // packed left shifting >> 2
3243 __ psrld(xmm_temp5, 7); // packed left shifting >> 7
3244 __ pxor(xmm_temp2, xmm_temp7); // xor the shifted versions
3245 __ pxor(xmm_temp2, xmm_temp5);
3246 __ pxor(xmm_temp2, xmm_temp4);
3247 __ pxor(xmm_temp3, xmm_temp2);
3248 __ pxor(xmm_temp6, xmm_temp3); // the result is in xmm6
3249
3250 __ decrement(blocks);
3251 __ jcc(Assembler::zero, L_exit);
3252 __ movdqu(xmm_temp0, xmm_temp6);
3253 __ addptr(data, 16);
3254 __ jmp(L_ghash_loop);
3255
3256 __ BIND(L_exit);
3257 // Byte swap 16-byte result
3258 __ pshufb(xmm_temp6, ExternalAddress(StubRoutines::x86::ghash_long_swap_mask_addr()));
3259 __ movdqu(Address(state, 0), xmm_temp6); // store the result
3260
3261 handleSOERegisters(false); // restore registers
3262 __ leave();
3263 __ ret(0);
3264 return start;
3265 }
3266
3267 /**
3268 * Arguments:
3269 *
3270 * Inputs:
3271 * rsp(4) - int crc
3272 * rsp(8) - byte* buf
3273 * rsp(12) - int length
3274 *
3275 * Ouput:
3276 * rax - int crc result
3277 */
3278 address generate_updateBytesCRC32() {
3279 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
3280
3281 __ align(CodeEntryAlignment);
3282 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
3283
3284 address start = __ pc();
3285
3286 const Register crc = rdx; // crc
3287 const Register buf = rsi; // source java byte array address
3288 const Register len = rcx; // length
3289 const Register table = rdi; // crc_table address (reuse register)
3290 const Register tmp = rbx;
3291 assert_different_registers(crc, buf, len, table, tmp, rax);
3292
3293 BLOCK_COMMENT("Entry:");
3294 __ enter(); // required for proper stackwalking of RuntimeStub frame
3295 __ push(rsi);
3296 __ push(rdi);
3297 __ push(rbx);
3298
3299 Address crc_arg(rbp, 8 + 0);
3300 Address buf_arg(rbp, 8 + 4);
3301 Address len_arg(rbp, 8 + 8);
3302
3303 // Load up:
3304 __ movl(crc, crc_arg);
3305 __ movptr(buf, buf_arg);
3306 __ movl(len, len_arg);
3307
3308 __ kernel_crc32(crc, buf, len, table, tmp);
3309
3310 __ movl(rax, crc);
3311 __ pop(rbx);
3312 __ pop(rdi);
3313 __ pop(rsi);
3314 __ vzeroupper();
3315 __ leave(); // required for proper stackwalking of RuntimeStub frame
3316 __ ret(0);
3317
3318 return start;
3319 }
3320
3321 /**
3322 * Arguments:
3323 *
3324 * Inputs:
3325 * rsp(4) - int crc
3326 * rsp(8) - byte* buf
3327 * rsp(12) - int length
3328 * rsp(16) - table_start - optional (present only when doing a library_calll,
3329 * not used by x86 algorithm)
3330 *
3331 * Ouput:
3332 * rax - int crc result
3333 */
3334 address generate_updateBytesCRC32C(bool is_pclmulqdq_supported) {
3335 assert(UseCRC32CIntrinsics, "need SSE4_2");
3336 __ align(CodeEntryAlignment);
3337 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32C");
3338 address start = __ pc();
3339 const Register crc = rax; // crc
3340 const Register buf = rcx; // source java byte array address
3341 const Register len = rdx; // length
3342 const Register d = rbx;
3343 const Register g = rsi;
3344 const Register h = rdi;
3345 const Register empty = 0; // will never be used, in order not
3346 // to change a signature for crc32c_IPL_Alg2_Alt2
3347 // between 64/32 I'm just keeping it here
3348 assert_different_registers(crc, buf, len, d, g, h);
3349
3350 BLOCK_COMMENT("Entry:");
3351 __ enter(); // required for proper stackwalking of RuntimeStub frame
3352 Address crc_arg(rsp, 4 + 4 + 0); // ESP+4 +
3353 // we need to add additional 4 because __ enter
3354 // have just pushed ebp on a stack
3355 Address buf_arg(rsp, 4 + 4 + 4);
3356 Address len_arg(rsp, 4 + 4 + 8);
3357 // Load up:
3358 __ movl(crc, crc_arg);
3359 __ movl(buf, buf_arg);
3360 __ movl(len, len_arg);
3361 __ push(d);
3362 __ push(g);
3363 __ push(h);
3364 __ crc32c_ipl_alg2_alt2(crc, buf, len,
3365 d, g, h,
3366 empty, empty, empty,
3367 xmm0, xmm1, xmm2,
3368 is_pclmulqdq_supported);
3369 __ pop(h);
3370 __ pop(g);
3371 __ pop(d);
3372 __ vzeroupper();
3373 __ leave(); // required for proper stackwalking of RuntimeStub frame
3374 __ ret(0);
3375
3376 return start;
3377 }
3378
3379 address generate_libmExp() {
3380 StubCodeMark mark(this, "StubRoutines", "libmExp");
3381
3382 address start = __ pc();
3383
3384 const XMMRegister x0 = xmm0;
3385 const XMMRegister x1 = xmm1;
3386 const XMMRegister x2 = xmm2;
3387 const XMMRegister x3 = xmm3;
3388
3389 const XMMRegister x4 = xmm4;
3390 const XMMRegister x5 = xmm5;
3391 const XMMRegister x6 = xmm6;
3392 const XMMRegister x7 = xmm7;
3393
3394 const Register tmp = rbx;
3395
3396 BLOCK_COMMENT("Entry:");
3397 __ enter(); // required for proper stackwalking of RuntimeStub frame
3398 __ fast_exp(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3399 __ leave(); // required for proper stackwalking of RuntimeStub frame
3400 __ ret(0);
3401
3402 return start;
3403
3404 }
3405
3406 address generate_libmLog() {
3407 StubCodeMark mark(this, "StubRoutines", "libmLog");
3408
3409 address start = __ pc();
3410
3411 const XMMRegister x0 = xmm0;
3412 const XMMRegister x1 = xmm1;
3413 const XMMRegister x2 = xmm2;
3414 const XMMRegister x3 = xmm3;
3415
3416 const XMMRegister x4 = xmm4;
3417 const XMMRegister x5 = xmm5;
3418 const XMMRegister x6 = xmm6;
3419 const XMMRegister x7 = xmm7;
3420
3421 const Register tmp = rbx;
3422
3423 BLOCK_COMMENT("Entry:");
3424 __ enter(); // required for proper stackwalking of RuntimeStub frame
3425 __ fast_log(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3426 __ leave(); // required for proper stackwalking of RuntimeStub frame
3427 __ ret(0);
3428
3429 return start;
3430
3431 }
3432
3433 address generate_libmLog10() {
3434 StubCodeMark mark(this, "StubRoutines", "libmLog10");
3435
3436 address start = __ pc();
3437
3438 const XMMRegister x0 = xmm0;
3439 const XMMRegister x1 = xmm1;
3440 const XMMRegister x2 = xmm2;
3441 const XMMRegister x3 = xmm3;
3442
3443 const XMMRegister x4 = xmm4;
3444 const XMMRegister x5 = xmm5;
3445 const XMMRegister x6 = xmm6;
3446 const XMMRegister x7 = xmm7;
3447
3448 const Register tmp = rbx;
3449
3450 BLOCK_COMMENT("Entry:");
3451 __ enter(); // required for proper stackwalking of RuntimeStub frame
3452 __ fast_log10(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3453 __ leave(); // required for proper stackwalking of RuntimeStub frame
3454 __ ret(0);
3455
3456 return start;
3457
3458 }
3459
3460 address generate_libmPow() {
3461 StubCodeMark mark(this, "StubRoutines", "libmPow");
3462
3463 address start = __ pc();
3464
3465 const XMMRegister x0 = xmm0;
3466 const XMMRegister x1 = xmm1;
3467 const XMMRegister x2 = xmm2;
3468 const XMMRegister x3 = xmm3;
3469
3470 const XMMRegister x4 = xmm4;
3471 const XMMRegister x5 = xmm5;
3472 const XMMRegister x6 = xmm6;
3473 const XMMRegister x7 = xmm7;
3474
3475 const Register tmp = rbx;
3476
3477 BLOCK_COMMENT("Entry:");
3478 __ enter(); // required for proper stackwalking of RuntimeStub frame
3479 __ fast_pow(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3480 __ leave(); // required for proper stackwalking of RuntimeStub frame
3481 __ ret(0);
3482
3483 return start;
3484
3485 }
3486
3487 address generate_libm_reduce_pi04l() {
3488 StubCodeMark mark(this, "StubRoutines", "libm_reduce_pi04l");
3489
3490 address start = __ pc();
3491
3492 BLOCK_COMMENT("Entry:");
3493 __ libm_reduce_pi04l(rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3494
3495 return start;
3496
3497 }
3498
3499 address generate_libm_sin_cos_huge() {
3500 StubCodeMark mark(this, "StubRoutines", "libm_sin_cos_huge");
3501
3502 address start = __ pc();
3503
3504 const XMMRegister x0 = xmm0;
3505 const XMMRegister x1 = xmm1;
3506
3507 BLOCK_COMMENT("Entry:");
3508 __ libm_sincos_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3509
3510 return start;
3511
3512 }
3513
3514 address generate_libmSin() {
3515 StubCodeMark mark(this, "StubRoutines", "libmSin");
3516
3517 address start = __ pc();
3518
3519 const XMMRegister x0 = xmm0;
3520 const XMMRegister x1 = xmm1;
3521 const XMMRegister x2 = xmm2;
3522 const XMMRegister x3 = xmm3;
3523
3524 const XMMRegister x4 = xmm4;
3525 const XMMRegister x5 = xmm5;
3526 const XMMRegister x6 = xmm6;
3527 const XMMRegister x7 = xmm7;
3528
3529 BLOCK_COMMENT("Entry:");
3530 __ enter(); // required for proper stackwalking of RuntimeStub frame
3531 __ fast_sin(x0, x1, x2, x3, x4, x5, x6, x7, rax, rbx, rdx);
3532 __ leave(); // required for proper stackwalking of RuntimeStub frame
3533 __ ret(0);
3534
3535 return start;
3536
3537 }
3538
3539 address generate_libmCos() {
3540 StubCodeMark mark(this, "StubRoutines", "libmCos");
3541
3542 address start = __ pc();
3543
3544 const XMMRegister x0 = xmm0;
3545 const XMMRegister x1 = xmm1;
3546 const XMMRegister x2 = xmm2;
3547 const XMMRegister x3 = xmm3;
3548
3549 const XMMRegister x4 = xmm4;
3550 const XMMRegister x5 = xmm5;
3551 const XMMRegister x6 = xmm6;
3552 const XMMRegister x7 = xmm7;
3553
3554 const Register tmp = rbx;
3555
3556 BLOCK_COMMENT("Entry:");
3557 __ enter(); // required for proper stackwalking of RuntimeStub frame
3558 __ fast_cos(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3559 __ leave(); // required for proper stackwalking of RuntimeStub frame
3560 __ ret(0);
3561
3562 return start;
3563
3564 }
3565
3566 address generate_libm_tan_cot_huge() {
3567 StubCodeMark mark(this, "StubRoutines", "libm_tan_cot_huge");
3568
3569 address start = __ pc();
3570
3571 const XMMRegister x0 = xmm0;
3572 const XMMRegister x1 = xmm1;
3573
3574 BLOCK_COMMENT("Entry:");
3575 __ libm_tancot_huge(x0, x1, rax, rcx, rdx, rbx, rsi, rdi, rbp, rsp);
3576
3577 return start;
3578
3579 }
3580
3581 address generate_libmTan() {
3582 StubCodeMark mark(this, "StubRoutines", "libmTan");
3583
3584 address start = __ pc();
3585
3586 const XMMRegister x0 = xmm0;
3587 const XMMRegister x1 = xmm1;
3588 const XMMRegister x2 = xmm2;
3589 const XMMRegister x3 = xmm3;
3590
3591 const XMMRegister x4 = xmm4;
3592 const XMMRegister x5 = xmm5;
3593 const XMMRegister x6 = xmm6;
3594 const XMMRegister x7 = xmm7;
3595
3596 const Register tmp = rbx;
3597
3598 BLOCK_COMMENT("Entry:");
3599 __ enter(); // required for proper stackwalking of RuntimeStub frame
3600 __ fast_tan(x0, x1, x2, x3, x4, x5, x6, x7, rax, rcx, rdx, tmp);
3601 __ leave(); // required for proper stackwalking of RuntimeStub frame
3602 __ ret(0);
3603
3604 return start;
3605
3606 }
3607
3608 // Safefetch stubs.
3609 void generate_safefetch(const char* name, int size, address* entry,
3610 address* fault_pc, address* continuation_pc) {
3611 // safefetch signatures:
3612 // int SafeFetch32(int* adr, int errValue);
3613 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
3614
3615 StubCodeMark mark(this, "StubRoutines", name);
3616
3617 // Entry point, pc or function descriptor.
3618 *entry = __ pc();
3619
3620 __ movl(rax, Address(rsp, 0x8));
3621 __ movl(rcx, Address(rsp, 0x4));
3622 // Load *adr into eax, may fault.
3623 *fault_pc = __ pc();
3624 switch (size) {
3625 case 4:
3626 // int32_t
3627 __ movl(rax, Address(rcx, 0));
3628 break;
3629 case 8:
3630 // int64_t
3631 Unimplemented();
3632 break;
3633 default:
3634 ShouldNotReachHere();
3635 }
3636
3637 // Return errValue or *adr.
3638 *continuation_pc = __ pc();
3639 __ ret(0);
3640 }
3641
3642 address generate_method_entry_barrier() {
3643 __ align(CodeEntryAlignment);
3644 StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
3645
3646 Label deoptimize_label;
3647
3648 address start = __ pc();
3649
3650 __ push(-1); // cookie, this is used for writing the new rsp when deoptimizing
3651
3652 BLOCK_COMMENT("Entry:");
3653 __ enter(); // save rbp
3654
3655 // save rbx, because we want to use that value.
3656 // We could do without it but then we depend on the number of slots used by pusha
3657 __ push(rbx);
3658
3659 __ lea(rbx, Address(rsp, wordSize * 3)); // 1 for cookie, 1 for rbp, 1 for rbx - this should be the return address
3660
3661 __ pusha();
3662
3663 // xmm0 and xmm1 may be used for passing float/double arguments
3664 const int xmm_size = wordSize * 2;
3665 const int xmm_spill_size = xmm_size * 2;
3666 __ subptr(rsp, xmm_spill_size);
3667 __ movdqu(Address(rsp, xmm_size * 1), xmm1);
3668 __ movdqu(Address(rsp, xmm_size * 0), xmm0);
3669
3670 __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(address*)>(BarrierSetNMethod::nmethod_stub_entry_barrier)), rbx);
3671
3672 __ movdqu(xmm0, Address(rsp, xmm_size * 0));
3673 __ movdqu(xmm1, Address(rsp, xmm_size * 1));
3674 __ addptr(rsp, xmm_spill_size);
3675
3676 __ cmpl(rax, 1); // 1 means deoptimize
3677 __ jcc(Assembler::equal, deoptimize_label);
3678
3679 __ popa();
3680 __ pop(rbx);
3681
3682 __ leave();
3683
3684 __ addptr(rsp, 1 * wordSize); // cookie
3685 __ ret(0);
3686
3687 __ BIND(deoptimize_label);
3688
3689 __ popa();
3690 __ pop(rbx);
3691
3692 __ leave();
3693
3694 // this can be taken out, but is good for verification purposes. getting a SIGSEGV
3695 // here while still having a correct stack is valuable
3696 __ testptr(rsp, Address(rsp, 0));
3697
3698 __ movptr(rsp, Address(rsp, 0)); // new rsp was written in the barrier
3699 __ jmp(Address(rsp, -1 * wordSize)); // jmp target should be callers verified_entry_point
3700
3701 return start;
3702 }
3703
3704 public:
3705 // Information about frame layout at time of blocking runtime call.
3706 // Note that we only have to preserve callee-saved registers since
3707 // the compilers are responsible for supplying a continuation point
3708 // if they expect all registers to be preserved.
3709 enum layout {
3710 thread_off, // last_java_sp
3711 arg1_off,
3712 arg2_off,
3713 rbp_off, // callee saved register
3714 ret_pc,
3715 framesize
3716 };
3717
3718 private:
3719
3720 #undef __
3721 #define __ masm->
3722
3723 //------------------------------------------------------------------------------------------------------------------------
3724 // Continuation point for throwing of implicit exceptions that are not handled in
3725 // the current activation. Fabricates an exception oop and initiates normal
3726 // exception dispatching in this frame.
3727 //
3728 // Previously the compiler (c2) allowed for callee save registers on Java calls.
3729 // This is no longer true after adapter frames were removed but could possibly
3730 // be brought back in the future if the interpreter code was reworked and it
3731 // was deemed worthwhile. The comment below was left to describe what must
3732 // happen here if callee saves were resurrected. As it stands now this stub
3733 // could actually be a vanilla BufferBlob and have now oopMap at all.
3734 // Since it doesn't make much difference we've chosen to leave it the
3735 // way it was in the callee save days and keep the comment.
3736
3737 // If we need to preserve callee-saved values we need a callee-saved oop map and
3738 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
3739 // If the compiler needs all registers to be preserved between the fault
3740 // point and the exception handler then it must assume responsibility for that in
3741 // AbstractCompiler::continuation_for_implicit_null_exception or
3742 // continuation_for_implicit_division_by_zero_exception. All other implicit
3743 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
3744 // either at call sites or otherwise assume that stack unwinding will be initiated,
3745 // so caller saved registers were assumed volatile in the compiler.
3746 address generate_throw_exception(const char* name, address runtime_entry,
3747 Register arg1 = noreg, Register arg2 = noreg) {
3748
3749 int insts_size = 256;
3750 int locs_size = 32;
3751
3752 CodeBuffer code(name, insts_size, locs_size);
3753 OopMapSet* oop_maps = new OopMapSet();
3754 MacroAssembler* masm = new MacroAssembler(&code);
3755
3756 address start = __ pc();
3757
3758 // This is an inlined and slightly modified version of call_VM
3759 // which has the ability to fetch the return PC out of
3760 // thread-local storage and also sets up last_Java_sp slightly
3761 // differently than the real call_VM
3762 Register java_thread = rbx;
3763 __ get_thread(java_thread);
3764
3765 __ enter(); // required for proper stackwalking of RuntimeStub frame
3766
3767 // pc and rbp, already pushed
3768 __ subptr(rsp, (framesize-2) * wordSize); // prolog
3769
3770 // Frame is now completed as far as size and linkage.
3771
3772 int frame_complete = __ pc() - start;
3773
3774 // push java thread (becomes first argument of C function)
3775 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
3776 if (arg1 != noreg) {
3777 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
3778 }
3779 if (arg2 != noreg) {
3780 assert(arg1 != noreg, "missing reg arg");
3781 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
3782 }
3783
3784 // Set up last_Java_sp and last_Java_fp
3785 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
3786
3787 // Call runtime
3788 BLOCK_COMMENT("call runtime_entry");
3789 __ call(RuntimeAddress(runtime_entry));
3790 // Generate oop map
3791 OopMap* map = new OopMap(framesize, 0);
3792 oop_maps->add_gc_map(__ pc() - start, map);
3793
3794 // restore the thread (cannot use the pushed argument since arguments
3795 // may be overwritten by C code generated by an optimizing compiler);
3796 // however can use the register value directly if it is callee saved.
3797 __ get_thread(java_thread);
3798
3799 __ reset_last_Java_frame(java_thread, true);
3800
3801 __ leave(); // required for proper stackwalking of RuntimeStub frame
3802
3803 // check for pending exceptions
3804 #ifdef ASSERT
3805 Label L;
3806 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
3807 __ jcc(Assembler::notEqual, L);
3808 __ should_not_reach_here();
3809 __ bind(L);
3810 #endif /* ASSERT */
3811 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3812
3813
3814 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
3815 return stub->entry_point();
3816 }
3817
3818
3819 void create_control_words() {
3820 // Round to nearest, 53-bit mode, exceptions masked
3821 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
3822 // Round to zero, 53-bit mode, exception mased
3823 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
3824 // Round to nearest, 24-bit mode, exceptions masked
3825 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
3826 // Round to nearest, 64-bit mode, exceptions masked
3827 StubRoutines::_mxcsr_std = 0x1F80;
3828 // Note: the following two constants are 80-bit values
3829 // layout is critical for correct loading by FPU.
3830 // Bias for strict fp multiply/divide
3831 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
3832 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
3833 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
3834 // Un-Bias for strict fp multiply/divide
3835 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
3836 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
3837 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
3838 }
3839
3840 //---------------------------------------------------------------------------
3841 // Initialization
3842
3843 void generate_initial() {
3844 // Generates all stubs and initializes the entry points
3845
3846 //------------------------------------------------------------------------------------------------------------------------
3847 // entry points that exist in all platforms
3848 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
3849 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
3850 StubRoutines::_forward_exception_entry = generate_forward_exception();
3851
3852 StubRoutines::_call_stub_entry =
3853 generate_call_stub(StubRoutines::_call_stub_return_address);
3854 // is referenced by megamorphic call
3855 StubRoutines::_catch_exception_entry = generate_catch_exception();
3856
3857 // These are currently used by Solaris/Intel
3858 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3859
3860 // platform dependent
3861 create_control_words();
3862
3863 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3864 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
3865 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
3866 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
3867 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
3868 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
3869
3870 // Build this early so it's available for the interpreter
3871 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception",
3872 CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
3873 StubRoutines::_throw_delayed_StackOverflowError_entry = generate_throw_exception("delayed StackOverflowError throw_exception",
3874 CAST_FROM_FN_PTR(address, SharedRuntime::throw_delayed_StackOverflowError));
3875
3876 if (UseCRC32Intrinsics) {
3877 // set table address before stub generation which use it
3878 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
3879 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
3880 }
3881
3882 if (UseCRC32CIntrinsics) {
3883 bool supports_clmul = VM_Version::supports_clmul();
3884 StubRoutines::x86::generate_CRC32C_table(supports_clmul);
3885 StubRoutines::_crc32c_table_addr = (address)StubRoutines::x86::_crc32c_table;
3886 StubRoutines::_updateBytesCRC32C = generate_updateBytesCRC32C(supports_clmul);
3887 }
3888 if (VM_Version::supports_sse2() && UseLibmIntrinsic && InlineIntrinsics) {
3889 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3890 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3891 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3892 StubRoutines::x86::_L_2il0floatpacket_0_adr = (address)StubRoutines::x86::_L_2il0floatpacket_0;
3893 StubRoutines::x86::_Pi4Inv_adr = (address)StubRoutines::x86::_Pi4Inv;
3894 StubRoutines::x86::_Pi4x3_adr = (address)StubRoutines::x86::_Pi4x3;
3895 StubRoutines::x86::_Pi4x4_adr = (address)StubRoutines::x86::_Pi4x4;
3896 StubRoutines::x86::_ones_adr = (address)StubRoutines::x86::_ones;
3897 }
3898 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dexp)) {
3899 StubRoutines::_dexp = generate_libmExp();
3900 }
3901 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog)) {
3902 StubRoutines::_dlog = generate_libmLog();
3903 }
3904 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dlog10)) {
3905 StubRoutines::_dlog10 = generate_libmLog10();
3906 }
3907 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dpow)) {
3908 StubRoutines::_dpow = generate_libmPow();
3909 }
3910 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3911 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos) ||
3912 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3913 StubRoutines::_dlibm_reduce_pi04l = generate_libm_reduce_pi04l();
3914 }
3915 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin) ||
3916 vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3917 StubRoutines::_dlibm_sin_cos_huge = generate_libm_sin_cos_huge();
3918 }
3919 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dsin)) {
3920 StubRoutines::_dsin = generate_libmSin();
3921 }
3922 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dcos)) {
3923 StubRoutines::_dcos = generate_libmCos();
3924 }
3925 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_dtan)) {
3926 StubRoutines::_dlibm_tan_cot_huge = generate_libm_tan_cot_huge();
3927 StubRoutines::_dtan = generate_libmTan();
3928 }
3929 }
3930
3931 // Safefetch stubs.
3932 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3933 &StubRoutines::_safefetch32_fault_pc,
3934 &StubRoutines::_safefetch32_continuation_pc);
3935 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
3936 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
3937 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3938 }
3939
3940 void generate_all() {
3941 // Generates all stubs and initializes the entry points
3942
3943 // These entry points require SharedInfo::stack0 to be set up in non-core builds
3944 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
3945 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
3946 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
3947 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
3948
3949 //------------------------------------------------------------------------------------------------------------------------
3950 // entry points that are platform specific
3951
3952 StubRoutines::x86::_vector_float_sign_mask = generate_vector_mask("vector_float_sign_mask", 0x7FFFFFFF);
3953 StubRoutines::x86::_vector_float_sign_flip = generate_vector_mask("vector_float_sign_flip", 0x80000000);
3954 StubRoutines::x86::_vector_double_sign_mask = generate_vector_mask_long_double("vector_double_sign_mask", 0x7FFFFFFF, 0xFFFFFFFF);
3955 StubRoutines::x86::_vector_double_sign_flip = generate_vector_mask_long_double("vector_double_sign_flip", 0x80000000, 0x00000000);
3956 StubRoutines::x86::_vector_short_to_byte_mask = generate_vector_mask("vector_short_to_byte_mask", 0x00ff00ff);
3957 StubRoutines::x86::_vector_int_to_byte_mask = generate_vector_mask("vector_int_to_byte_mask", 0x000000ff);
3958 StubRoutines::x86::_vector_int_to_short_mask = generate_vector_mask("vector_int_to_short_mask", 0x0000ffff);
3959 StubRoutines::x86::_vector_32_bit_mask = generate_vector_custom_i32("vector_32_bit_mask", Assembler::AVX_512bit,
3960 0xFFFFFFFF, 0, 0, 0);
3961 StubRoutines::x86::_vector_64_bit_mask = generate_vector_custom_i32("vector_64_bit_mask", Assembler::AVX_512bit,
3962 0xFFFFFFFF, 0xFFFFFFFF, 0, 0);
3963 StubRoutines::x86::_vector_int_shuffle_mask = generate_vector_mask("vector_int_shuffle_mask", 0x03020100);
3964 StubRoutines::x86::_vector_short_shuffle_mask = generate_vector_mask("vector_short_shuffle_mask", 0x01000100);
3965 StubRoutines::x86::_vector_long_shuffle_mask = generate_vector_mask_long_double("vector_long_shuffle_mask", 0x00000001, 0x0);
3966 StubRoutines::x86::_vector_byte_perm_mask = generate_vector_byte_perm_mask("vector_byte_perm_mask");
3967 StubRoutines::x86::_vector_long_sign_mask = generate_vector_mask_long_double("vector_long_sign_mask", 0x80000000, 0x00000000);
3968 StubRoutines::x86::_vector_all_bits_set = generate_vector_mask("vector_all_bits_set", 0xFFFFFFFF);
3969 StubRoutines::x86::_vector_iota_indices = generate_iota_indices("iota_indices");
3970
3971 // support for verify_oop (must happen after universe_init)
3972 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3973
3974 // arraycopy stubs used by compilers
3975 generate_arraycopy_stubs();
3976
3977 // don't bother generating these AES intrinsic stubs unless global flag is set
3978 if (UseAESIntrinsics) {
3979 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
3980
3981 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3982 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3983 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3984 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt_Parallel();
3985 }
3986
3987 if (UseAESCTRIntrinsics) {
3988 StubRoutines::x86::_counter_shuffle_mask_addr = generate_counter_shuffle_mask();
3989 StubRoutines::_counterMode_AESCrypt = generate_counterMode_AESCrypt_Parallel();
3990 }
3991
3992 if (UseSHA1Intrinsics) {
3993 StubRoutines::x86::_upper_word_mask_addr = generate_upper_word_mask();
3994 StubRoutines::x86::_shuffle_byte_flip_mask_addr = generate_shuffle_byte_flip_mask();
3995 StubRoutines::_sha1_implCompress = generate_sha1_implCompress(false, "sha1_implCompress");
3996 StubRoutines::_sha1_implCompressMB = generate_sha1_implCompress(true, "sha1_implCompressMB");
3997 }
3998 if (UseSHA256Intrinsics) {
3999 StubRoutines::x86::_k256_adr = (address)StubRoutines::x86::_k256;
4000 StubRoutines::x86::_pshuffle_byte_flip_mask_addr = generate_pshuffle_byte_flip_mask();
4001 StubRoutines::_sha256_implCompress = generate_sha256_implCompress(false, "sha256_implCompress");
4002 StubRoutines::_sha256_implCompressMB = generate_sha256_implCompress(true, "sha256_implCompressMB");
4003 }
4004
4005 // Generate GHASH intrinsics code
4006 if (UseGHASHIntrinsics) {
4007 StubRoutines::x86::_ghash_long_swap_mask_addr = generate_ghash_long_swap_mask();
4008 StubRoutines::x86::_ghash_byte_swap_mask_addr = generate_ghash_byte_swap_mask();
4009 StubRoutines::_ghash_processBlocks = generate_ghash_processBlocks();
4010 }
4011
4012 BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
4013 if (bs_nm != NULL) {
4014 StubRoutines::x86::_method_entry_barrier = generate_method_entry_barrier();
4015 }
4016 }
4017
4018
4019 public:
4020 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
4021 if (all) {
4022 generate_all();
4023 } else {
4024 generate_initial();
4025 }
4026 }
4027 }; // end class declaration
4028
4029 #define UCM_TABLE_MAX_ENTRIES 8
4030 void StubGenerator_generate(CodeBuffer* code, bool all) {
4031 if (UnsafeCopyMemory::_table == NULL) {
4032 UnsafeCopyMemory::create_table(UCM_TABLE_MAX_ENTRIES);
4033 }
4034 StubGenerator g(code, all);
4035 }