Tue, 10 May 2016 15:08:51 -0400
Fixed changset 108.
1 /*
2 * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2016, Loongson Technology. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "nativeInst_mips.hpp"
31 #include "oops/instanceOop.hpp"
32 #include "oops/method.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/frame.inline.hpp"
37 #include "runtime/handles.inline.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubCodeGenerator.hpp"
40 #include "runtime/stubRoutines.hpp"
41 #include "runtime/thread.inline.hpp"
42 #include "utilities/top.hpp"
43 #ifdef COMPILER2
44 #include "opto/runtime.hpp"
45 #endif
48 // Declaration and definition of StubGenerator (no .hpp file).
49 // For a more detailed description of the stub routine structure
50 // see the comment in stubRoutines.hpp
52 #define __ _masm->
53 //#define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
54 //#define a__ ((Assembler*)_masm)->
56 //#ifdef PRODUCT
57 //#define BLOCK_COMMENT(str) /* nothing */
58 //#else
59 //#define BLOCK_COMMENT(str) __ block_comment(str)
60 //#endif
62 //#define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
63 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
65 // Stub Code definitions
67 static address handle_unsafe_access() {
68 JavaThread* thread = JavaThread::current();
69 address pc = thread->saved_exception_pc();
70 // pc is the instruction which we must emulate
71 // doing a no-op is fine: return garbage from the load
72 // therefore, compute npc
73 //address npc = Assembler::locate_next_instruction(pc);
74 address npc = (address)((unsigned long)pc + sizeof(unsigned long));
76 // request an async exception
77 thread->set_pending_unsafe_access_error();
79 // return address of next instruction to execute
80 return npc;
81 }
83 class StubGenerator: public StubCodeGenerator {
84 private:
86 // ABI mips n64
87 // This fig is not MIPS ABI. It is call Java from C ABI.
88 // Call stubs are used to call Java from C
89 //
90 // [ return_from_Java ]
91 // [ argument word n-1 ] <--- sp
92 // ...
93 // [ argument word 0 ]
94 // ...
95 //-10 [ S6 ]
96 // -9 [ S5 ]
97 // -8 [ S4 ]
98 // -7 [ S3 ]
99 // -6 [ S0 ]
100 // -5 [ TSR(S2) ]
101 // -4 [ LVP(S7) ]
102 // -3 [ BCP(S1) ]
103 // -2 [ saved fp ] <--- fp_after_call
104 // -1 [ return address ]
105 // 0 [ ptr. to call wrapper ] <--- a0 (old sp -->)fp
106 // 1 [ result ] <--- a1
107 // 2 [ result_type ] <--- a2
108 // 3 [ method ] <--- a3
109 // 4 [ entry_point ] <--- a4
110 // 5 [ parameters ] <--- a5
111 // 6 [ parameter_size ] <--- a6
112 // 7 [ thread ] <--- a7
114 //
115 // _LP64: n64 does not save paras in sp.
116 //
117 // [ return_from_Java ]
118 // [ argument word n-1 ] <--- sp
119 // ...
120 // [ argument word 0 ]
121 // ...
122 //-14 [ thread ]
123 //-13 [ result_type ] <--- a2
124 //-12 [ result ] <--- a1
125 //-11 [ ptr. to call wrapper ] <--- a0
126 //-10 [ S6 ]
127 // -9 [ S5 ]
128 // -8 [ S4 ]
129 // -7 [ S3 ]
130 // -6 [ S0 ]
131 // -5 [ TSR(S2) ]
132 // -4 [ LVP(S7) ]
133 // -3 [ BCP(S1) ]
134 // -2 [ saved fp ] <--- fp_after_call
135 // -1 [ return address ]
136 // 0 [ ] <--- old sp
137 /*
138 * 2014/01/16 Fu: Find a right place in the call_stub for GP.
139 * GP will point to the starting point of Interpreter::dispatch_table(itos).
140 * It should be saved/restored before/after Java calls.
141 *
142 */
143 enum call_stub_layout {
144 RA_off = -1,
145 FP_off = -2,
146 BCP_off = -3,
147 LVP_off = -4,
148 TSR_off = -5,
149 S1_off = -6,
150 S3_off = -7,
151 S4_off = -8,
152 S5_off = -9,
153 S6_off = -10,
154 result_off = -11,
155 result_type_off = -12,
156 thread_off = -13,
157 total_off = thread_off - 3,
158 GP_off = -16,
159 };
161 address generate_call_stub(address& return_address) {
163 StubCodeMark mark(this, "StubRoutines", "call_stub");
164 address start = __ pc();
166 // same as in generate_catch_exception()!
168 // stub code
169 // save ra and fp
170 __ sd(RA, SP, RA_off * wordSize);
171 __ sd(FP, SP, FP_off * wordSize);
172 __ sd(BCP, SP, BCP_off * wordSize);
173 __ sd(LVP, SP, LVP_off * wordSize);
174 __ sd(GP, SP, GP_off * wordSize);
175 __ sd(TSR, SP, TSR_off * wordSize);
176 __ sd(S1, SP, S1_off * wordSize);
177 __ sd(S3, SP, S3_off * wordSize);
178 __ sd(S4, SP, S4_off * wordSize);
179 __ sd(S5, SP, S5_off * wordSize);
180 __ sd(S6, SP, S6_off * wordSize);
183 __ li48(GP, (long)Interpreter::dispatch_table(itos));
185 // I think 14 is the max gap between argument and callee saved register
186 __ daddi(FP, SP, (-2) * wordSize);
187 __ daddi(SP, SP, total_off * wordSize);
188 //FIXME, aoqi. find a suitable place to save A1 & A2.
189 /*
190 __ sd(A0, FP, frame::entry_frame_call_wrapper_offset * wordSize);
191 __ sd(A1, FP, 3 * wordSize);
192 __ sd(A2, FP, 4 * wordSize);
193 __ sd(A3, FP, 5 * wordSize);
194 __ sd(A4, FP, 6 * wordSize);
195 __ sd(A5, FP, 7 * wordSize);
196 __ sd(A6, FP, 8 * wordSize);
197 __ sd(A7, FP, 9 * wordSize);
198 */
199 __ sd(A0, FP, frame::entry_frame_call_wrapper_offset * wordSize);
200 __ sd(A1, FP, result_off * wordSize);
201 __ sd(A2, FP, result_type_off * wordSize);
202 __ sd(A7, FP, thread_off * wordSize);
204 #ifdef OPT_THREAD
205 //__ get_thread(TREG);
206 __ move(TREG, A7);
208 //__ ld(TREG, FP, thread_off * wordSize);
209 #endif
210 //add for compressedoops
211 __ reinit_heapbase();
213 #ifdef ASSERT
214 // make sure we have no pending exceptions
215 {
216 Label L;
217 __ ld(AT, A7, in_bytes(Thread::pending_exception_offset()));
218 __ beq(AT, R0, L);
219 __ delayed()->nop();
220 /* FIXME: I do not know how to realize stop in mips arch, do it in the future */
221 __ stop("StubRoutines::call_stub: entered with pending exception");
222 __ bind(L);
223 }
224 #endif
226 // pass parameters if any
227 // A5: parameter
228 // A6: parameter_size
229 // T0: parameter_size_tmp(--)
230 // T2: offset(++)
231 // T3: tmp
232 Label parameters_done;
233 // judge if the parameter_size equals 0
234 __ beq(A6, R0, parameters_done);
235 __ delayed()->nop();
236 __ dsll(AT, A6, Interpreter::logStackElementSize);
237 __ dsub(SP, SP, AT);
238 __ move(AT, -StackAlignmentInBytes);
239 __ andr(SP, SP , AT);
240 // Copy Java parameters in reverse order (receiver last)
241 // Note that the argument order is inverted in the process
242 // source is edx[ecx: N-1..0]
243 // dest is esp[ebx: 0..N-1]
244 Label loop;
245 __ move(T0, A6);
246 __ move(T2, R0);
247 __ bind(loop);
249 // get parameter
250 __ dsll(T3, T0, LogBytesPerWord);
251 __ dadd(T3, T3, A5);
252 __ ld(AT, T3, -wordSize);
253 __ dsll(T3, T2, LogBytesPerWord);
254 __ dadd(T3, T3, SP);
255 __ sd(AT, T3, Interpreter::expr_offset_in_bytes(0));
256 __ daddi(T2, T2, 1);
257 __ daddi(T0, T0, -1);
258 __ bne(T0, R0, loop);
259 __ delayed()->nop();
260 // advance to next parameter
262 // call Java function
263 __ bind(parameters_done);
265 // receiver in V0, methodOop in Rmethod
267 __ move(Rmethod, A3);
268 __ move(Rsender, SP); //set sender sp
269 __ jalr(A4);
270 __ delayed()->nop();
271 return_address = __ pc();
273 Label common_return;
274 __ bind(common_return);
276 // store result depending on type
277 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
278 __ ld(T0, FP, result_off * wordSize); // result --> T0
279 Label is_long, is_float, is_double, exit;
280 __ ld(T2, FP, result_type_off * wordSize); // result_type --> T2
281 __ daddi(T3, T2, (-1) * T_LONG);
282 __ beq(T3, R0, is_long);
283 __ delayed()->daddi(T3, T2, (-1) * T_FLOAT);
284 __ beq(T3, R0, is_float);
285 __ delayed()->daddi(T3, T2, (-1) * T_DOUBLE);
286 __ beq(T3, R0, is_double);
287 __ delayed()->nop();
289 // handle T_INT case
290 __ sd(V0, T0, 0 * wordSize);
291 __ bind(exit);
293 // restore
294 __ daddi(SP, FP, 2 * wordSize );
295 __ ld(RA, SP, RA_off * wordSize);
296 __ ld(FP, SP, FP_off * wordSize);
297 __ ld(BCP, SP, BCP_off * wordSize);
298 __ ld(LVP, SP, LVP_off * wordSize);
299 __ ld(GP, SP, GP_off * wordSize);
300 __ ld(TSR, SP, TSR_off * wordSize);
302 __ ld(S1, SP, S1_off * wordSize);
303 __ ld(S3, SP, S3_off * wordSize);
304 __ ld(S4, SP, S4_off * wordSize);
305 __ ld(S5, SP, S5_off * wordSize);
306 __ ld(S6, SP, S6_off * wordSize);
308 // return
309 __ jr(RA);
310 __ delayed()->nop();
312 // handle return types different from T_INT
313 __ bind(is_long);
314 __ sd(V0, T0, 0 * wordSize);
315 //__ sd(V1, T0, 1 * wordSize);
316 __ sd(R0, T0, 1 * wordSize);
317 __ b(exit);
318 __ delayed()->nop();
320 __ bind(is_float);
321 __ swc1(F0, T0, 0 * wordSize);
322 __ b(exit);
323 __ delayed()->nop();
325 __ bind(is_double);
326 __ sdc1(F0, T0, 0 * wordSize);
327 //__ sdc1(F1, T0, 1 * wordSize);
328 __ sd(R0, T0, 1 * wordSize);
329 __ b(exit);
330 __ delayed()->nop();
331 //FIXME, 1.6 mips version add operation of fpu here
332 StubRoutines::gs2::set_call_stub_compiled_return(__ pc());
333 __ b(common_return);
334 __ delayed()->nop();
335 return start;
336 }
338 // Return point for a Java call if there's an exception thrown in
339 // Java code. The exception is caught and transformed into a
340 // pending exception stored in JavaThread that can be tested from
341 // within the VM.
342 //
343 // Note: Usually the parameters are removed by the callee. In case
344 // of an exception crossing an activation frame boundary, that is
345 // not the case if the callee is compiled code => need to setup the
346 // rsp.
347 //
348 // rax: exception oop
350 address generate_catch_exception() {
351 StubCodeMark mark(this, "StubRoutines", "catch_exception");
352 address start = __ pc();
354 Register thread = TREG;
356 // get thread directly
357 #ifndef OPT_THREAD
358 __ ld(thread, FP, thread_off * wordSize);
359 #endif
361 #ifdef ASSERT
362 // verify that threads correspond
363 { Label L;
364 __ get_thread(T8);
365 __ beq(T8, thread, L);
366 __ delayed()->nop();
367 __ stop("StubRoutines::catch_exception: threads must correspond");
368 __ bind(L);
369 }
370 #endif
371 // set pending exception
372 __ verify_oop(V0);
373 __ sd(V0, thread, in_bytes(Thread::pending_exception_offset()));
374 __ li(AT, (long)__FILE__);
375 __ sd(AT, thread, in_bytes(Thread::exception_file_offset ()));
376 __ li(AT, (long)__LINE__);
377 __ sd(AT, thread, in_bytes(Thread::exception_line_offset ()));
379 // complete return to VM
380 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
381 __ jmp(StubRoutines::_call_stub_return_address, relocInfo::none);
382 __ delayed()->nop();
384 return start;
385 }
387 // Continuation point for runtime calls returning with a pending
388 // exception. The pending exception check happened in the runtime
389 // or native call stub. The pending exception in Thread is
390 // converted into a Java-level exception.
391 //
392 // Contract with Java-level exception handlers:
393 // rax: exception
394 // rdx: throwing pc
395 //
396 // NOTE: At entry of this stub, exception-pc must be on stack !!
398 address generate_forward_exception() {
399 StubCodeMark mark(this, "StubRoutines", "forward exception");
400 //Register thread = TREG;
401 Register thread = TREG;
402 address start = __ pc();
404 // Upon entry, the sp points to the return address returning into Java
405 // (interpreted or compiled) code; i.e., the return address becomes the
406 // throwing pc.
407 //
408 // Arguments pushed before the runtime call are still on the stack but
409 // the exception handler will reset the stack pointer -> ignore them.
410 // A potential result in registers can be ignored as well.
412 #ifdef ASSERT
413 // make sure this code is only executed if there is a pending exception
414 #ifndef OPT_THREAD
415 __ get_thread(thread);
416 #endif
417 { Label L;
418 __ ld(AT, thread, in_bytes(Thread::pending_exception_offset()));
419 __ bne(AT, R0, L);
420 __ delayed()->nop();
421 __ stop("StubRoutines::forward exception: no pending exception (1)");
422 __ bind(L);
423 }
424 #endif
426 // compute exception handler into T9
427 __ ld(A1, SP, 0);
428 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, A1);
429 __ move(T9, V0);
430 __ pop(V1);
432 #ifndef OPT_THREAD
433 __ get_thread(thread);
434 #endif
435 __ ld(V0, thread, in_bytes(Thread::pending_exception_offset()));
436 __ sd(R0, thread, in_bytes(Thread::pending_exception_offset()));
438 #ifdef ASSERT
439 // make sure exception is set
440 { Label L;
441 __ bne(V0, R0, L);
442 __ delayed()->nop();
443 __ stop("StubRoutines::forward exception: no pending exception (2)");
444 __ bind(L);
445 }
446 #endif
448 // continue at exception handler (return address removed)
449 // V0: exception
450 // T9: exception handler
451 // V1: throwing pc
452 __ verify_oop(V0);
453 __ jr(T9);
454 __ delayed()->nop();
456 return start;
457 }
459 // Support for intptr_t get_previous_fp()
460 //
461 // This routine is used to find the previous frame pointer for the
462 // caller (current_frame_guess). This is used as part of debugging
463 // ps() is seemingly lost trying to find frames.
464 // This code assumes that caller current_frame_guess) has a frame.
465 address generate_get_previous_fp() {
466 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
467 const Address old_fp (FP, 0);
468 const Address older_fp (V0, 0);
469 address start = __ pc();
470 __ enter();
471 __ lw(V0, old_fp); // callers fp
472 __ lw(V0, older_fp); // the frame for ps()
473 __ leave();
474 __ jr(RA);
475 __ delayed()->nop();
476 return start;
477 }
478 // The following routine generates a subroutine to throw an
479 // asynchronous UnknownError when an unsafe access gets a fault that
480 // could not be reasonably prevented by the programmer. (Example:
481 // SIGBUS/OBJERR.)
482 address generate_handler_for_unsafe_access() {
483 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
484 address start = __ pc();
485 __ pushad(); // push registers
486 // Address next_pc(esp, RegisterImpl::number_of_registers * BytesPerWord);
487 __ call(CAST_FROM_FN_PTR(address, handle_unsafe_access), relocInfo::runtime_call_type);
488 __ delayed()->nop();
489 __ sw(V0, SP, RegisterImpl::number_of_registers * BytesPerWord);
490 __ popad();
491 __ jr(RA);
492 __ delayed()->nop();
493 return start;
494 }
496 // Non-destructive plausibility checks for oops
497 //
498 // Arguments:
499 // all args on stack!
500 //
501 // Stack after saving c_rarg3:
502 // [tos + 0]: saved c_rarg3
503 // [tos + 1]: saved c_rarg2
504 // [tos + 2]: saved r12 (several TemplateTable methods use it)
505 // [tos + 3]: saved flags
506 // [tos + 4]: return address
507 // * [tos + 5]: error message (char*)
508 // * [tos + 6]: object to verify (oop)
509 // * [tos + 7]: saved rax - saved by caller and bashed
510 // * = popped on exit
511 address generate_verify_oop() {
512 StubCodeMark mark(this, "StubRoutines", "verify_oop");
513 address start = __ pc();
514 __ reinit_heapbase();
515 __ verify_oop_subroutine();
516 address end = __ pc();
517 return start;
518 }
520 //
521 // Generate overlap test for array copy stubs
522 //
523 // Input:
524 // A0 - array1
525 // A1 - array2
526 // A2 - element count
527 //
528 // Note: this code can only use %eax, %ecx, and %edx
529 //
531 // use T9 as temp
532 void array_overlap_test(address no_overlap_target, int log2_elem_size) {
533 int elem_size = 1 << log2_elem_size;
534 Address::ScaleFactor sf = Address::times_1;
536 switch (log2_elem_size) {
537 case 0: sf = Address::times_1; break;
538 case 1: sf = Address::times_2; break;
539 case 2: sf = Address::times_4; break;
540 case 3: sf = Address::times_8; break;
541 }
543 __ dsll(AT, A2, sf);
544 __ dadd(AT, AT, A0);
545 __ lea(T9, Address(AT, -elem_size));
546 __ dsub(AT, A1, A0);
547 __ blez(AT, no_overlap_target);
548 __ delayed()->nop();
549 __ dsub(AT, A1, T9);
550 __ bgtz(AT, no_overlap_target);
551 __ delayed()->nop();
553 // 2016/05/10 aoqi: If A0 = 0xf... and A1 = 0x0..., than goto no_overlap_target
554 Label L;
555 __ bgez(A0, L);
556 __ delayed()->nop();
557 __ bgtz(A1, no_overlap_target);
558 __ delayed()->nop();
559 __ bind(L);
561 }
563 //
564 // Generate store check for array
565 //
566 // Input:
567 // %edi - starting address
568 // %ecx - element count
569 //
570 // The 2 input registers are overwritten
571 //
573 //
574 // Generate store check for array
575 //
576 // Input:
577 // T0 - starting address(edi)
578 // T1 - element count (ecx)
579 //
580 // The 2 input registers are overwritten
581 //
583 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
585 void array_store_check() {
586 BarrierSet* bs = Universe::heap()->barrier_set();
587 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind");
588 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
589 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
590 Label l_0;
592 __ dsll(AT, T1, TIMES_OOP);
593 __ dadd(AT, T0, AT);
594 __ daddiu(T1, AT, - BytesPerHeapOop);
596 __ shr(T0, CardTableModRefBS::card_shift);
597 __ shr(T1, CardTableModRefBS::card_shift);
599 __ dsub(T1, T1, T0); // end --> cards count
600 __ bind(l_0);
602 __ li48(AT, (long)ct->byte_map_base);
603 __ dadd(AT, AT, T0);
604 __ dadd(AT, AT, T1);
605 __ sb(R0, AT, 0);
606 //__ daddi(T1, T1, -4);
607 __ daddi(T1, T1, - 1);
608 __ bgez(T1, l_0);
609 __ delayed()->nop();
610 }
612 // Arguments:
613 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
614 // ignored
615 // name - stub name string
616 //
617 // Inputs:
618 // c_rarg0 - source array address
619 // c_rarg1 - destination array address
620 // c_rarg2 - element count, treated as ssize_t, can be zero
621 //
622 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
623 // we let the hardware handle it. The one to eight bytes within words,
624 // dwords or qwords that span cache line boundaries will still be loaded
625 // and stored atomically.
626 //
627 // Side Effects:
628 // disjoint_byte_copy_entry is set to the no-overlap entry point
629 // used by generate_conjoint_byte_copy().
630 //
631 address generate_disjoint_byte_copy(bool aligned, const char *name) {
632 StubCodeMark mark(this, "StubRoutines", name);
633 __ align(CodeEntryAlignment);
634 address start = __ pc();
635 Label l_0, l_1, l_2, l_3, l_4, l_5, l_6;
637 __ push(T3);
638 __ push(T0);
639 __ push(T1);
640 __ push(T8);
641 __ move(T3, A0);
642 __ move(T0, A1);
643 __ move(T1, A2);
644 __ move(T8, T1); // original count in T1
645 __ daddi(AT, T1, -3);
646 __ blez(AT, l_4);
647 __ delayed()->nop();
648 if (!aligned) {
649 //TODO: copy 8 bytes at one time
650 // 2016/5/8 Jin: only when src and dest has the same alignment can we do lw/sw */
651 __ andi(AT, T3, 3);
652 __ andi(T9, T0, 3);
653 __ bne(AT, T9, l_5);
654 __ delayed()->nop();
656 // align source address at dword address boundary
657 __ move(T1, 4);
658 __ sub(T1, T1, T3);
659 __ andi(T1, T1, 3);
660 __ beq(T1, R0, l_1);
661 __ delayed()->nop();
662 __ sub(T8,T8,T1);
663 __ bind(l_0);
664 __ lb(AT, T3, 0);
665 __ sb(AT, T0, 0);
666 __ addi(T3, T3, 1);
667 __ addi(T0, T0, 1);
668 __ addi(T1 ,T1, -1);
669 __ bne(T1, R0, l_0);
670 __ delayed()->nop();
671 __ bind(l_1);
672 __ move(T1, T8);
673 }
674 __ shr(T1, 2);
675 __ beq(T1, R0, l_4); // no dwords to move
676 __ delayed()->nop();
677 // copy aligned dwords
678 __ bind(l_2);
679 __ align(16);
680 __ bind(l_3);
681 __ lw(AT, T3, 0);
682 __ sw(AT, T0, 0 );
683 __ addi(T3, T3, 4);
684 __ addi(T0, T0, 4);
685 __ addi(T1, T1, -1);
686 __ bne(T1, R0, l_3);
687 __ delayed()->nop();
688 __ bind(l_4);
689 __ move(T1, T8);
690 __ andi(T1, T1, 3);
691 __ beq(T1, R0, l_6);
692 __ delayed()->nop();
693 // copy suffix
694 __ bind(l_5);
695 __ lb(AT, T3, 0);
696 __ sb(AT, T0, 0);
697 __ addi(T3, T3, 1);
698 __ addi(T0, T0, 1);
699 __ addi(T1, T1, -1);
700 __ bne(T1, R0, l_5 );
701 __ delayed()->nop();
702 __ bind(l_6);
703 __ pop(T8);
704 __ pop(T1);
705 __ pop(T0);
706 __ pop(T3);
707 __ jr(RA);
708 __ delayed()->nop();
709 return start;
710 }
712 // Arguments:
713 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
714 // ignored
715 // name - stub name string
716 //
717 // Inputs:
718 // A0 - source array address
719 // A1 - destination array address
720 // A2 - element count, treated as ssize_t, can be zero
721 //
722 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
723 // we let the hardware handle it. The one to eight bytes within words,
724 // dwords or qwords that span cache line boundaries will still be loaded
725 // and stored atomically.
726 //
727 address generate_conjoint_byte_copy(bool aligned, const char *name) {
728 __ align(CodeEntryAlignment);
729 StubCodeMark mark(this, "StubRoutines", name);
730 address start = __ pc();
732 Label l_copy_4_bytes_loop, l_copy_suffix, l_copy_suffix_loop, l_exit;
733 Label l_copy_byte, l_from_unaligned, l_unaligned, l_4_bytes_aligned;
735 address nooverlap_target = aligned ?
736 StubRoutines::arrayof_jbyte_disjoint_arraycopy() :
737 StubRoutines::jbyte_disjoint_arraycopy();
739 array_overlap_test(nooverlap_target, 0);
741 const Register from = A0; // source array address
742 const Register to = A1; // destination array address
743 const Register count = A2; // elements count
744 const Register end_from = T3; // source array end address
745 const Register end_to = T0; // destination array end address
746 const Register end_count = T1; // destination array end address
748 __ push(end_from);
749 __ push(end_to);
750 __ push(end_count);
751 __ push(T8);
753 // copy from high to low
754 __ move(end_count, count);
755 __ dadd(end_from, from, end_count);
756 __ dadd(end_to, to, end_count);
758 // 2016/05/08 aoqi: If end_from and end_to has differante alignment, unaligned copy is performed.
759 __ andi(AT, end_from, 3);
760 __ andi(T8, end_to, 3);
761 __ bne(AT, T8, l_copy_byte);
762 __ delayed()->nop();
764 // First deal with the unaligned data at the top.
765 __ bind(l_unaligned);
766 __ beq(end_count, R0, l_exit);
767 __ delayed()->nop();
769 __ andi(AT, end_from, 3);
770 __ bne(AT, R0, l_from_unaligned);
771 __ delayed()->nop();
773 __ andi(AT, end_to, 3);
774 __ beq(AT, R0, l_4_bytes_aligned);
775 __ delayed()->nop();
777 __ bind(l_from_unaligned);
778 __ lb(AT, end_from, -1);
779 __ sb(AT, end_to, -1);
780 __ daddi(end_from, end_from, -1);
781 __ daddi(end_to, end_to, -1);
782 __ daddi(end_count, end_count, -1);
783 __ b(l_unaligned);
784 __ delayed()->nop();
786 // now end_to, end_from point to 4-byte aligned high-ends
787 // end_count contains byte count that is not copied.
788 // copy 4 bytes at a time
789 __ bind(l_4_bytes_aligned);
791 __ move(T8, end_count);
792 __ daddi(AT, end_count, -3);
793 __ blez(AT, l_copy_suffix);
794 __ delayed()->nop();
796 //__ andi(T8, T8, 3);
797 __ lea(end_from, Address(end_from, -4));
798 __ lea(end_to, Address(end_to, -4));
800 __ dsrl(end_count, end_count, 2);
801 __ align(16);
802 __ bind(l_copy_4_bytes_loop); //l_copy_4_bytes
803 __ lw(AT, end_from, 0);
804 __ sw(AT, end_to, 0);
805 __ addi(end_from, end_from, -4);
806 __ addi(end_to, end_to, -4);
807 __ addi(end_count, end_count, -1);
808 __ bne(end_count, R0, l_copy_4_bytes_loop);
809 __ delayed()->nop();
811 __ b(l_copy_suffix);
812 __ delayed()->nop();
813 // copy dwords aligned or not with repeat move
814 // l_copy_suffix
815 // copy suffix (0-3 bytes)
816 __ bind(l_copy_suffix);
817 __ andi(T8, T8, 3);
818 __ beq(T8, R0, l_exit);
819 __ delayed()->nop();
820 __ addi(end_from, end_from, 3);
821 __ addi(end_to, end_to, 3);
822 __ bind(l_copy_suffix_loop);
823 __ lb(AT, end_from, 0);
824 __ sb(AT, end_to, 0);
825 __ addi(end_from, end_from, -1);
826 __ addi(end_to, end_to, -1);
827 __ addi(T8, T8, -1);
828 __ bne(T8, R0, l_copy_suffix_loop);
829 __ delayed()->nop();
831 __ bind(l_copy_byte);
832 __ beq(end_count, R0, l_exit);
833 __ delayed()->nop();
834 __ lb(AT, end_from, -1);
835 __ sb(AT, end_to, -1);
836 __ daddi(end_from, end_from, -1);
837 __ daddi(end_to, end_to, -1);
838 __ daddi(end_count, end_count, -1);
839 __ b(l_copy_byte);
840 __ delayed()->nop();
842 __ bind(l_exit);
843 __ pop(T8);
844 __ pop(end_count);
845 __ pop(end_to);
846 __ pop(end_from);
847 __ jr(RA);
848 __ delayed()->nop();
849 return start;
850 }
852 // Arguments:
853 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
854 // ignored
855 // name - stub name string
856 //
857 // Inputs:
858 // c_rarg0 - source array address
859 // c_rarg1 - destination array address
860 // c_rarg2 - element count, treated as ssize_t, can be zero
861 //
862 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
863 // let the hardware handle it. The two or four words within dwords
864 // or qwords that span cache line boundaries will still be loaded
865 // and stored atomically.
866 //
867 // Side Effects:
868 // disjoint_short_copy_entry is set to the no-overlap entry point
869 // used by generate_conjoint_short_copy().
870 //
871 address generate_disjoint_short_copy(bool aligned, const char *name) {
872 Label l_1, l_2, l_3, l_4, l_5, l_6, l_7, l_8;
873 StubCodeMark mark(this, "StubRoutines", name);
874 __ align(CodeEntryAlignment);
875 address start = __ pc();
877 __ push(T3);
878 __ push(T0);
879 __ push(T1);
880 __ push(T8);
881 __ move(T1, A2);
882 __ move(T3, A0);
883 __ move(T0, A1);
885 if (!aligned) {
886 __ beq(T1, R0, l_5);
887 __ delayed()->nop();
888 // align source address at dword address boundary
889 __ move(T8, T3); // original from
890 __ andi(T8, T8, 3); // either 0 or 2
891 __ beq(T8, R0, l_1); // no prefix
892 __ delayed()->nop();
893 // copy prefix
894 __ lh(AT, T3, 0);
895 __ sh(AT, T0, 0);
896 __ add(T3, T3, T8);
897 __ add(T0, T0, T8);
898 __ addi(T1, T1, -1);
899 __ bind(l_1);
900 }
901 __ move(T8, T1); // word count less prefix
902 __ sra(T1, T1, 1);
903 __ beq(T1, R0, l_4);
904 __ delayed()->nop();
905 // copy aligned dwords
906 __ bind(l_2);
907 __ align(16);
908 __ bind(l_3);
909 __ lw(AT, T3, 0);
910 __ sw(AT, T0, 0 );
911 __ addi(T3, T3, 4);
912 __ addi(T0, T0, 4);
913 __ addi(T1, T1, -1);
914 __ bne(T1, R0, l_3);
915 __ delayed()->nop();
916 __ bind(l_4);
917 __ andi(T8, T8, 1);
918 __ beq(T8, R0, l_5);
919 __ delayed()->nop();
920 // copy suffix
921 __ lh(AT, T3, 0);
922 __ sh(AT, T0, 0);
923 __ bind(l_5);
924 __ pop(T8);
925 __ pop(T1);
926 __ pop(T0);
927 __ pop(T3);
928 __ jr(RA);
929 __ delayed()->nop();
930 return start;
931 }
933 // Arguments:
934 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
935 // ignored
936 // name - stub name string
937 //
938 // Inputs:
939 // c_rarg0 - source array address
940 // c_rarg1 - destination array address
941 // c_rarg2 - element count, treated as ssize_t, can be zero
942 //
943 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
944 // let the hardware handle it. The two or four words within dwords
945 // or qwords that span cache line boundaries will still be loaded
946 // and stored atomically.
947 //
948 address generate_conjoint_short_copy(bool aligned, const char *name) {
949 Label l_1, l_2, l_3, l_4, l_5;
950 StubCodeMark mark(this, "StubRoutines", name);
951 __ align(CodeEntryAlignment);
952 address start = __ pc();
953 address nooverlap_target = aligned ?
954 StubRoutines::arrayof_jshort_disjoint_arraycopy() :
955 StubRoutines::jshort_disjoint_arraycopy();
957 array_overlap_test(nooverlap_target, 1);
959 __ push(T3);
960 __ push(T0);
961 __ push(T1);
962 __ push(T8);
964 /*
965 __ pushl(esi);
966 __ movl(ecx, Address(esp, 4+12)); // count
967 __ pushl(edi);
968 __ movl(esi, Address(esp, 8+ 4)); // from
969 __ movl(edi, Address(esp, 8+ 8)); // to
970 */
971 __ move(T1, A2);
972 __ move(T3, A0);
973 __ move(T0, A1);
976 // copy dwords from high to low
977 // __ leal(esi, Address(esi, ecx, Address::times_2, -4)); // from + count*2 - 4
978 __ sll(AT, T1, Address::times_2);
979 __ add(AT, T3, AT);
980 __ lea(T3, Address( AT, -4));
981 //__ std();
982 //__ leal(edi, Address(edi, ecx, Address::times_2, -4)); // to + count*2 - 4
983 __ sll(AT,T1 , Address::times_2);
984 __ add(AT, T0, AT);
985 __ lea(T0, Address( AT, -4));
986 // __ movl(eax, ecx);
987 __ move(T8, T1);
988 __ bind(l_1);
989 // __ sarl(ecx, 1); // dword count
990 __ sra(T1,T1, 1);
991 //__ jcc(Assembler::equal, l_4); // no dwords to move
992 __ beq(T1, R0, l_4);
993 __ delayed()->nop();
994 /* __ cmpl(ecx, 32);
995 __ jcc(Assembler::above, l_3); // > 32 dwords
996 // copy dwords with loop
997 __ subl(edi, esi);
998 */ __ align(16);
999 __ bind(l_2);
1000 //__ movl(edx, Address(esi));
1001 __ lw(AT, T3, 0);
1002 //__ movl(Address(edi, esi, Address::times_1), edx);
1003 __ sw(AT, T0, 0);
1004 //__ subl(esi, 4);
1005 __ addi(T3, T3, -4);
1006 __ addi(T0, T0, -4);
1007 //__ decl(ecx);
1008 __ addi(T1, T1, -1);
1009 // __ jcc(Assembler::notEqual, l_2);
1010 __ bne(T1, R0, l_2);
1011 __ delayed()->nop();
1012 // __ addl(edi, esi);
1013 // __ jmp(l_4);
1014 __ b(l_4);
1015 __ delayed()->nop();
1016 // copy dwords with repeat move
1017 __ bind(l_3);
1018 // __ rep_movl();
1019 __ bind(l_4);
1020 // __ andl(eax, 1); // suffix count
1021 __ andi(T8, T8, 1); // suffix count
1022 //__ jcc(Assembler::equal, l_5); // no suffix
1023 __ beq(T8, R0, l_5 );
1024 __ delayed()->nop();
1025 // copy suffix
1026 // __ movw(edx, Address(esi, 2));
1027 __ lh(AT, T3, 2);
1028 // __ movw(Address(edi, 2), edx);
1029 __ sh(AT, T0, 2);
1030 __ bind(l_5);
1031 // __ cld();
1032 // __ popl(edi);
1033 // __ popl(esi);
1034 // __ ret(0);
1035 __ pop(T8);
1036 __ pop(T1);
1037 __ pop(T0);
1038 __ pop(T3);
1039 __ jr(RA);
1040 __ delayed()->nop();
1041 return start;
1042 }
1044 // Arguments:
1045 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1046 // ignored
1047 // is_oop - true => oop array, so generate store check code
1048 // name - stub name string
1049 //
1050 // Inputs:
1051 // c_rarg0 - source array address
1052 // c_rarg1 - destination array address
1053 // c_rarg2 - element count, treated as ssize_t, can be zero
1054 //
1055 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1056 // the hardware handle it. The two dwords within qwords that span
1057 // cache line boundaries will still be loaded and stored atomicly.
1058 //
1059 // Side Effects:
1060 // disjoint_int_copy_entry is set to the no-overlap entry point
1061 // used by generate_conjoint_int_oop_copy().
1062 //
1063 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1064 Label l_2, l_3, l_4, l_stchk;
1065 StubCodeMark mark(this, "StubRoutines", name);
1066 __ align(CodeEntryAlignment);
1067 address start = __ pc();
1068 /*
1069 __ pushl(esi);
1070 __ movl(ecx, Address(esp, 4+12)); // count
1071 __ pushl(edi);
1072 __ movl(esi, Address(esp, 8+ 4)); // from
1073 __ movl(edi, Address(esp, 8+ 8)); // to
1074 */
1075 __ push(T3);
1076 __ push(T0);
1077 __ push(T1);
1078 __ push(T8);
1079 __ move(T1, A2);
1080 __ move(T3, A0);
1081 __ move(T0, A1);
1083 // __ cmpl(ecx, 32);
1084 // __ jcc(Assembler::belowEqual, l_2); // <= 32 dwords
1085 // __ rep_movl();
1086 __ b(l_2);
1087 __ delayed()->nop();
1088 if (is_oop) {
1089 // __ jmp(l_stchk);
1090 __ b(l_stchk);
1091 __ delayed()->nop();
1092 }
1093 // __ popl(edi);
1094 // __ popl(esi);
1095 // __ ret(0);
1096 __ pop(T8);
1097 __ pop(T1);
1098 __ pop(T0);
1099 __ pop(T3);
1100 __ jr(RA);
1101 __ delayed()->nop();
1103 __ bind(l_2);
1104 // __ subl(edi, esi);
1105 // __ testl(ecx, ecx);
1106 // __ jcc(Assembler::zero, l_4);
1107 __ beq(T1, R0, l_4);
1108 __ delayed()->nop();
1109 __ align(16);
1110 __ bind(l_3);
1111 //__ movl(edx, Address(esi));
1112 __ lw(AT, T3, 0);
1113 // __ movl(Address(edi, esi, Address::times_1), edx);
1114 __ sw(AT, T0, 0);
1115 // __ addl(esi, 4);
1116 __ addi(T3, T3, 4);
1117 __ addi(T0, T0, 4);
1118 // __ decl(ecx);
1119 __ addi(T1, T1, -1);
1120 // __ jcc(Assembler::notEqual, l_3);
1121 __ bne(T1, R0, l_3);
1122 __ delayed()->nop();
1123 if (is_oop) {
1124 __ bind(l_stchk);
1125 // __ movl(edi, Address(esp, 8+ 8));
1126 // __ movl(ecx, Address(esp, 8+ 12));
1127 __ move(T0, A1);
1128 __ move(T1, A2);
1129 array_store_check();
1130 }
1131 __ bind(l_4);
1132 // __ popl(edi);
1133 // __ popl(esi);
1134 // __ ret(0);
1135 __ pop(T8);
1136 __ pop(T1);
1137 __ pop(T0);
1138 __ pop(T3);
1139 __ jr(RA);
1140 __ delayed()->nop();
1141 return start;
1142 }
1144 // Arguments:
1145 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1146 // ignored
1147 // is_oop - true => oop array, so generate store check code
1148 // name - stub name string
1149 //
1150 // Inputs:
1151 // c_rarg0 - source array address
1152 // c_rarg1 - destination array address
1153 // c_rarg2 - element count, treated as ssize_t, can be zero
1154 //
1155 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1156 // the hardware handle it. The two dwords within qwords that span
1157 // cache line boundaries will still be loaded and stored atomicly.
1158 //
1159 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, const char *name) {
1160 Label l_2, l_3, l_4, l_stchk;
1161 StubCodeMark mark(this, "StubRoutines", name);
1162 __ align(CodeEntryAlignment);
1163 address start = __ pc();
1164 address nooverlap_target;
1166 if (is_oop) {
1167 nooverlap_target = aligned ?
1168 StubRoutines::arrayof_oop_disjoint_arraycopy() :
1169 StubRoutines::oop_disjoint_arraycopy();
1170 }else {
1171 nooverlap_target = aligned ?
1172 StubRoutines::arrayof_jint_disjoint_arraycopy() :
1173 StubRoutines::jint_disjoint_arraycopy();
1174 }
1176 array_overlap_test(nooverlap_target, 2);
1178 __ push(T3);
1179 __ push(T0);
1180 __ push(T1);
1181 __ push(T8);
1183 /*
1184 __ pushl(esi);
1185 __ movl(ecx, Address(esp, 4+12)); // count
1186 __ pushl(edi);
1187 __ movl(esi, Address(esp, 8+ 4)); // from
1188 __ movl(edi, Address(esp, 8+ 8)); // to
1189 */
1190 __ move(T1, A2);
1191 __ move(T3, A0);
1192 __ move(T0, A1);
1194 //__ leal(esi, Address(esi, ecx, Address::times_4, -4)); // from + count*4 - 4
1195 __ sll(AT, T1, Address::times_4);
1196 __ add(AT, T3, AT);
1197 __ lea(T3 , Address(AT, -4));
1198 //__ std();
1199 //__ leal(edi, Address(edi, ecx, Address::times_4, -4)); // to + count*4 - 4
1200 __ sll(AT, T1, Address::times_4);
1201 __ add(AT, T0, AT);
1202 __ lea(T0 , Address(AT, -4));
1204 // __ cmpl(ecx, 32);
1205 // __ jcc(Assembler::above, l_3); // > 32 dwords
1206 // __ testl(ecx, ecx);
1207 //__ jcc(Assembler::zero, l_4);
1208 __ beq(T1, R0, l_4);
1209 __ delayed()->nop();
1210 // __ subl(edi, esi);
1211 __ align(16);
1212 __ bind(l_2);
1213 // __ movl(edx, Address(esi));
1214 __ lw(AT, T3, 0);
1215 // __ movl(Address(esi, edi, Address::times_1), edx);
1216 __ sw(AT, T0, 0);
1217 // __ subl(esi, 4);
1218 __ addi(T3, T3, -4);
1219 __ addi(T0, T0, -4);
1220 // __ decl(ecx);
1221 __ addi(T1, T1, -1);
1222 //__ jcc(Assembler::notEqual, l_2);
1223 __ bne(T1, R0, l_2);
1224 __ delayed()->nop();
1225 if (is_oop) {
1226 // __ jmp(l_stchk);
1227 __ b( l_stchk);
1228 __ delayed()->nop();
1229 }
1230 __ bind(l_4);
1231 // __ cld();
1232 // __ popl(edi);
1233 // __ popl(esi);
1234 // __ ret(0);
1235 __ pop(T8);
1236 __ pop(T1);
1237 __ pop(T0);
1238 __ pop(T3);
1239 __ jr(RA);
1240 __ delayed()->nop();
1241 __ bind(l_3);
1242 // __ rep_movl();
1243 if (is_oop) {
1244 __ bind(l_stchk);
1245 // __ movl(edi, Address(esp, 8+ 8));
1246 __ move(T0, A1);
1247 // __ movl(ecx, Address(esp, 8+ 12));
1248 __ move(T1, A2);
1249 array_store_check();
1250 }
1251 // __ cld();
1252 // __ popl(edi);
1253 // __ popl(esi);
1254 // __ ret(0);
1255 __ pop(T8);
1256 __ pop(T1);
1257 __ pop(T0);
1258 __ pop(T3);
1259 __ jr(RA);
1260 __ delayed()->nop();
1261 return start;
1262 }
1264 // Arguments:
1265 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1266 // ignored
1267 // is_oop - true => oop array, so generate store check code
1268 // name - stub name string
1269 //
1270 // Inputs:
1271 // c_rarg0 - source array address
1272 // c_rarg1 - destination array address
1273 // c_rarg2 - element count, treated as ssize_t, can be zero
1274 //
1275 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1276 // the hardware handle it. The two dwords within qwords that span
1277 // cache line boundaries will still be loaded and stored atomicly.
1278 //
1279 // Side Effects:
1280 // disjoint_int_copy_entry is set to the no-overlap entry point
1281 // used by generate_conjoint_int_oop_copy().
1282 //
1283 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1284 Label l_2, l_3, l_4, l_stchk;
1285 StubCodeMark mark(this, "StubRoutines", name);
1286 __ align(CodeEntryAlignment);
1287 address start = __ pc();
1288 __ push(T3);
1289 __ push(T0);
1290 __ push(T1);
1291 __ push(T8);
1292 __ move(T1, A2);
1293 __ move(T3, A0);
1294 __ move(T0, A1);
1296 // __ cmpl(ecx, 32);
1297 // __ jcc(Assembler::belowEqual, l_2); // <= 32 dwords
1298 // __ rep_movl();
1299 __ b(l_2);
1300 __ delayed()->nop();
1301 if (is_oop) {
1302 // __ jmp(l_stchk);
1303 __ b(l_stchk);
1304 __ delayed()->nop();
1305 }
1306 // __ popl(edi);
1307 // __ popl(esi);
1308 // __ ret(0);
1309 __ pop(T8);
1310 __ pop(T1);
1311 __ pop(T0);
1312 __ pop(T3);
1313 __ jr(RA);
1314 __ delayed()->nop();
1316 __ bind(l_2);
1317 // __ subl(edi, esi);
1318 // __ testl(ecx, ecx);
1319 // __ jcc(Assembler::zero, l_4);
1320 __ beq(T1, R0, l_4);
1321 __ delayed()->nop();
1322 __ align(16);
1323 __ bind(l_3);
1324 //__ movl(edx, Address(esi));
1325 __ ld(AT, T3, 0);
1326 // __ movl(Address(edi, esi, Address::times_1), edx);
1327 __ sd(AT, T0, 0);
1328 // __ addl(esi, 4);
1329 __ addi(T3, T3, 8);
1330 __ addi(T0, T0, 8);
1331 // __ decl(ecx);
1332 __ addi(T1, T1, -1);
1333 // __ jcc(Assembler::notEqual, l_3);
1334 __ bne(T1, R0, l_3);
1335 __ delayed()->nop();
1336 if (is_oop) {
1337 __ bind(l_stchk);
1338 // __ movl(edi, Address(esp, 8+ 8));
1339 // __ movl(ecx, Address(esp, 8+ 12));
1340 __ move(T0, A1);
1341 __ move(T1, A2);
1342 array_store_check();
1343 }
1344 __ bind(l_4);
1345 // __ popl(edi);
1346 // __ popl(esi);
1347 // __ ret(0);
1348 __ pop(T8);
1349 __ pop(T1);
1350 __ pop(T0);
1351 __ pop(T3);
1352 __ jr(RA);
1353 __ delayed()->nop();
1354 return start;
1355 }
1357 // Arguments:
1358 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1359 // ignored
1360 // is_oop - true => oop array, so generate store check code
1361 // name - stub name string
1362 //
1363 // Inputs:
1364 // c_rarg0 - source array address
1365 // c_rarg1 - destination array address
1366 // c_rarg2 - element count, treated as ssize_t, can be zero
1367 //
1368 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1369 // the hardware handle it. The two dwords within qwords that span
1370 // cache line boundaries will still be loaded and stored atomicly.
1371 //
1372 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1373 Label l_2, l_3, l_4, l_stchk;
1374 StubCodeMark mark(this, "StubRoutines", name);
1375 __ align(CodeEntryAlignment);
1376 address start = __ pc();
1377 address nooverlap_target;
1379 if (is_oop) {
1380 nooverlap_target = aligned ?
1381 StubRoutines::arrayof_oop_disjoint_arraycopy() :
1382 StubRoutines::oop_disjoint_arraycopy();
1383 }else {
1384 nooverlap_target = aligned ?
1385 StubRoutines::arrayof_jlong_disjoint_arraycopy() :
1386 StubRoutines::jlong_disjoint_arraycopy();
1387 }
1389 array_overlap_test(nooverlap_target, 3);
1391 __ push(T3);
1392 __ push(T0);
1393 __ push(T1);
1394 __ push(T8);
1396 __ move(T1, A2);
1397 __ move(T3, A0);
1398 __ move(T0, A1);
1400 //__ leal(esi, Address(esi, ecx, Address::times_4, -4)); // from + count*4 - 4
1401 __ sll(AT, T1, Address::times_8);
1402 __ add(AT, T3, AT);
1403 __ lea(T3 , Address(AT, -8));
1404 //__ std();
1405 //__ leal(edi, Address(edi, ecx, Address::times_4, -4)); // to + count*4 - 4
1406 __ sll(AT, T1, Address::times_8);
1407 __ add(AT, T0, AT);
1408 __ lea(T0 , Address(AT, -8));
1410 // __ cmpl(ecx, 32);
1411 // __ jcc(Assembler::above, l_3); // > 32 dwords
1412 // __ testl(ecx, ecx);
1413 //__ jcc(Assembler::zero, l_4);
1414 __ beq(T1, R0, l_4);
1415 __ delayed()->nop();
1416 // __ subl(edi, esi);
1417 __ align(16);
1418 __ bind(l_2);
1419 // __ movl(edx, Address(esi));
1420 __ ld(AT, T3, 0);
1421 // __ movl(Address(esi, edi, Address::times_1), edx);
1422 __ sd(AT, T0, 0);
1423 // __ subl(esi, 4);
1424 __ addi(T3, T3, -8);
1425 __ addi(T0, T0, -8);
1426 // __ decl(ecx);
1427 __ addi(T1, T1, -1);
1428 //__ jcc(Assembler::notEqual, l_2);
1429 __ bne(T1, R0, l_2);
1430 __ delayed()->nop();
1431 if (is_oop) {
1432 // __ jmp(l_stchk);
1433 __ b( l_stchk);
1434 __ delayed()->nop();
1435 }
1436 __ bind(l_4);
1437 // __ cld();
1438 // __ popl(edi);
1439 // __ popl(esi);
1440 // __ ret(0);
1441 __ pop(T8);
1442 __ pop(T1);
1443 __ pop(T0);
1444 __ pop(T3);
1445 __ jr(RA);
1446 __ delayed()->nop();
1447 __ bind(l_3);
1448 // __ rep_movl();
1449 if (is_oop) {
1450 __ bind(l_stchk);
1451 // __ movl(edi, Address(esp, 8+ 8));
1452 __ move(T0, A1);
1453 // __ movl(ecx, Address(esp, 8+ 12));
1454 __ move(T1, A2);
1455 array_store_check();
1456 }
1457 // __ cld();
1458 // __ popl(edi);
1459 // __ popl(esi);
1460 // __ ret(0);
1461 __ pop(T8);
1462 __ pop(T1);
1463 __ pop(T0);
1464 __ pop(T3);
1465 __ jr(RA);
1466 __ delayed()->nop();
1467 return start;
1468 }
1469 #if 0
1470 // Arguments:
1471 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1472 // ignored
1473 // is_oop - true => oop array, so generate store check code
1474 // name - stub name string
1475 //
1476 // Inputs:
1477 // c_rarg0 - source array address
1478 // c_rarg1 - destination array address
1479 // c_rarg2 - element count, treated as ssize_t, can be zero
1480 //
1481 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop, const char *name) {
1482 __ align(CodeEntryAlignment);
1483 StubCodeMark mark(this, "StubRoutines", name);
1484 address start = __ pc();
1486 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1487 const Register from = rdi; // source array address
1488 const Register to = rsi; // destination array address
1489 const Register qword_count = rdx; // elements count
1490 const Register saved_count = rcx;
1492 __ enter(); // required for proper stackwalking of RuntimeStub frame
1493 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1495 address disjoint_copy_entry = NULL;
1496 if (is_oop) {
1497 assert(!UseCompressedOops, "shouldn't be called for compressed oops");
1498 disjoint_copy_entry = disjoint_oop_copy_entry;
1499 oop_copy_entry = __ pc();
1500 array_overlap_test(disjoint_oop_copy_entry, Address::times_8);
1501 } else {
1502 disjoint_copy_entry = disjoint_long_copy_entry;
1503 long_copy_entry = __ pc();
1504 array_overlap_test(disjoint_long_copy_entry, Address::times_8);
1505 }
1506 BLOCK_COMMENT("Entry:");
1507 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1509 array_overlap_test(disjoint_copy_entry, Address::times_8);
1510 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1511 // r9 and r10 may be used to save non-volatile registers
1513 // 'from', 'to' and 'qword_count' are now valid
1515 if (is_oop) {
1516 // Save to and count for store barrier
1517 __ movptr(saved_count, qword_count);
1518 // No registers are destroyed by this call
1519 gen_write_ref_array_pre_barrier(to, saved_count);
1520 }
1522 __ jmp(L_copy_32_bytes);
1524 // Copy trailing qwords
1525 __ BIND(L_copy_8_bytes);
1526 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1527 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1528 __ decrement(qword_count);
1529 __ jcc(Assembler::notZero, L_copy_8_bytes);
1531 if (is_oop) {
1532 __ jmp(L_exit);
1533 } else {
1534 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1535 restore_arg_regs();
1536 __ xorptr(rax, rax); // return 0
1537 __ leave(); // required for proper stackwalking of RuntimeStub frame
1538 __ ret(0);
1539 }
1541 // Copy in 32-bytes chunks
1542 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1544 if (is_oop) {
1545 __ BIND(L_exit);
1546 __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
1547 gen_write_ref_array_post_barrier(to, rcx, rax);
1548 inc_counter_np(SharedRuntime::_oop_array_copy_ctr);
1549 } else {
1550 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr);
1551 }
1552 restore_arg_regs();
1553 __ xorptr(rax, rax); // return 0
1554 __ leave(); // required for proper stackwalking of RuntimeStub frame
1555 __ ret(0);
1557 return start;
1558 }
1561 // Helper for generating a dynamic type check.
1562 // Smashes no registers.
1563 void generate_type_check(Register sub_klass,
1564 Register super_check_offset,
1565 Register super_klass,
1566 Label& L_success) {
1567 assert_different_registers(sub_klass, super_check_offset, super_klass);
1569 BLOCK_COMMENT("type_check:");
1571 Label L_miss;
1573 // a couple of useful fields in sub_klass:
1574 int ss_offset = (klassOopDesc::header_size() * HeapWordSize +
1575 Klass::secondary_supers_offset_in_bytes());
1576 int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1577 Klass::secondary_super_cache_offset_in_bytes());
1578 Address secondary_supers_addr(sub_klass, ss_offset);
1579 Address super_cache_addr( sub_klass, sc_offset);
1581 // if the pointers are equal, we are done (e.g., String[] elements)
1582 __ cmpptr(super_klass, sub_klass);
1583 __ jcc(Assembler::equal, L_success);
1585 // check the supertype display:
1586 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
1587 __ cmpptr(super_klass, super_check_addr); // test the super type
1588 __ jcc(Assembler::equal, L_success);
1590 // if it was a primary super, we can just fail immediately
1591 __ cmpl(super_check_offset, sc_offset);
1592 __ jcc(Assembler::notEqual, L_miss);
1594 // Now do a linear scan of the secondary super-klass chain.
1595 // The repne_scan instruction uses fixed registers, which we must spill.
1596 // (We need a couple more temps in any case.)
1597 // This code is rarely used, so simplicity is a virtue here.
1598 inc_counter_np(SharedRuntime::_partial_subtype_ctr);
1599 {
1600 __ push(rax);
1601 __ push(rcx);
1602 __ push(rdi);
1603 assert_different_registers(sub_klass, super_klass, rax, rcx, rdi);
1605 __ movptr(rdi, secondary_supers_addr);
1606 // Load the array length.
1607 __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
1608 // Skip to start of data.
1609 __ addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1610 // Scan rcx words at [rdi] for occurance of rax
1611 // Set NZ/Z based on last compare
1612 __ movptr(rax, super_klass);
1613 if (UseCompressedOops) {
1614 // Compare against compressed form. Don't need to uncompress because
1615 // looks like orig rax is restored in popq below.
1616 __ encode_heap_oop(rax);
1617 __ repne_scanl();
1618 } else {
1619 __ repne_scan();
1620 }
1622 // Unspill the temp. registers:
1623 __ pop(rdi);
1624 __ pop(rcx);
1625 __ pop(rax);
1627 __ jcc(Assembler::notEqual, L_miss);
1628 }
1630 // Success. Cache the super we found and proceed in triumph.
1631 __ movptr(super_cache_addr, super_klass); // note: rax is dead
1632 __ jmp(L_success);
1634 // Fall through on failure!
1635 __ BIND(L_miss);
1636 }
1638 //
1639 // Generate checkcasting array copy stub
1640 //
1641 // Input:
1642 // c_rarg0 - source array address
1643 // c_rarg1 - destination array address
1644 // c_rarg2 - element count, treated as ssize_t, can be zero
1645 // c_rarg3 - size_t ckoff (super_check_offset)
1646 // not Win64
1647 // c_rarg4 - oop ckval (super_klass)
1648 // Win64
1649 // rsp+40 - oop ckval (super_klass)
1650 //
1651 // Output:
1652 // rax == 0 - success
1653 // rax == -1^K - failure, where K is partial transfer count
1654 //
1655 address generate_checkcast_copy(const char *name) {
1657 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1659 // Input registers (after setup_arg_regs)
1660 const Register from = rdi; // source array address
1661 const Register to = rsi; // destination array address
1662 const Register length = rdx; // elements count
1663 const Register ckoff = rcx; // super_check_offset
1664 const Register ckval = r8; // super_klass
1666 // Registers used as temps (r13, r14 are save-on-entry)
1667 const Register end_from = from; // source array end address
1668 const Register end_to = r13; // destination array end address
1669 const Register count = rdx; // -(count_remaining)
1670 const Register r14_length = r14; // saved copy of length
1671 // End pointers are inclusive, and if length is not zero they point
1672 // to the last unit copied: end_to[0] := end_from[0]
1674 const Register rax_oop = rax; // actual oop copied
1675 const Register r11_klass = r11; // oop._klass
1677 //---------------------------------------------------------------
1678 // Assembler stub will be used for this call to arraycopy
1679 // if the two arrays are subtypes of Object[] but the
1680 // destination array type is not equal to or a supertype
1681 // of the source type. Each element must be separately
1682 // checked.
1684 __ align(CodeEntryAlignment);
1685 StubCodeMark mark(this, "StubRoutines", name);
1686 address start = __ pc();
1688 __ enter(); // required for proper stackwalking of RuntimeStub frame
1690 checkcast_copy_entry = __ pc();
1691 BLOCK_COMMENT("Entry:");
1693 #ifdef ASSERT
1694 // caller guarantees that the arrays really are different
1695 // otherwise, we would have to make conjoint checks
1696 { Label L;
1697 array_overlap_test(L, TIMES_OOP);
1698 __ stop("checkcast_copy within a single array");
1699 __ bind(L);
1700 }
1701 #endif //ASSERT
1703 // allocate spill slots for r13, r14
1704 enum {
1705 saved_r13_offset,
1706 saved_r14_offset,
1707 saved_rbp_offset,
1708 saved_rip_offset,
1709 saved_rarg0_offset
1710 };
1711 __ subptr(rsp, saved_rbp_offset * wordSize);
1712 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
1713 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
1714 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
1715 // ckoff => rcx, ckval => r8
1716 // r9 and r10 may be used to save non-volatile registers
1717 #ifdef _WIN64
1718 // last argument (#4) is on stack on Win64
1719 const int ckval_offset = saved_rarg0_offset + 4;
1720 __ movptr(ckval, Address(rsp, ckval_offset * wordSize));
1721 #endif
1723 // check that int operands are properly extended to size_t
1724 assert_clean_int(length, rax);
1725 assert_clean_int(ckoff, rax);
1727 #ifdef ASSERT
1728 BLOCK_COMMENT("assert consistent ckoff/ckval");
1729 // The ckoff and ckval must be mutually consistent,
1730 // even though caller generates both.
1731 { Label L;
1732 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
1733 Klass::super_check_offset_offset_in_bytes());
1734 __ cmpl(ckoff, Address(ckval, sco_offset));
1735 __ jcc(Assembler::equal, L);
1736 __ stop("super_check_offset inconsistent");
1737 __ bind(L);
1738 }
1739 #endif //ASSERT
1741 // Loop-invariant addresses. They are exclusive end pointers.
1742 Address end_from_addr(from, length, TIMES_OOP, 0);
1743 Address end_to_addr(to, length, TIMES_OOP, 0);
1744 // Loop-variant addresses. They assume post-incremented count < 0.
1745 Address from_element_addr(end_from, count, TIMES_OOP, 0);
1746 Address to_element_addr(end_to, count, TIMES_OOP, 0);
1748 gen_write_ref_array_pre_barrier(to, count);
1750 // Copy from low to high addresses, indexed from the end of each array.
1751 __ lea(end_from, end_from_addr);
1752 __ lea(end_to, end_to_addr);
1753 __ movptr(r14_length, length); // save a copy of the length
1754 assert(length == count, ""); // else fix next line:
1755 __ negptr(count); // negate and test the length
1756 __ jcc(Assembler::notZero, L_load_element);
1758 // Empty array: Nothing to do.
1759 __ xorptr(rax, rax); // return 0 on (trivial) success
1760 __ jmp(L_done);
1762 // ======== begin loop ========
1763 // (Loop is rotated; its entry is L_load_element.)
1764 // Loop control:
1765 // for (count = -count; count != 0; count++)
1766 // Base pointers src, dst are biased by 8*(count-1),to last element.
1767 __ align(16);
1769 __ BIND(L_store_element);
1770 __ store_heap_oop(rax_oop, to_element_addr); // store the oop
1771 __ increment(count); // increment the count toward zero
1772 __ jcc(Assembler::zero, L_do_card_marks);
1774 // ======== loop entry is here ========
1775 __ BIND(L_load_element);
1776 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
1777 __ testptr(rax_oop, rax_oop);
1778 __ jcc(Assembler::zero, L_store_element);
1780 __ load_klass(r11_klass, rax_oop);// query the object klass
1781 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
1782 // ======== end loop ========
1784 // It was a real error; we must depend on the caller to finish the job.
1785 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
1786 // Emit GC store barriers for the oops we have copied (r14 + rdx),
1787 // and report their number to the caller.
1788 assert_different_registers(rax, r14_length, count, to, end_to, rcx);
1789 __ lea(end_to, to_element_addr);
1790 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
1791 __ movptr(rax, r14_length); // original oops
1792 __ addptr(rax, count); // K = (original - remaining) oops
1793 __ notptr(rax); // report (-1^K) to caller
1794 __ jmp(L_done);
1796 // Come here on success only.
1797 __ BIND(L_do_card_marks);
1798 __ addptr(end_to, -wordSize); // make an inclusive end pointer
1799 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
1800 __ xorptr(rax, rax); // return 0 on success
1802 // Common exit point (success or failure).
1803 __ BIND(L_done);
1804 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
1805 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
1806 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1807 restore_arg_regs();
1808 __ leave(); // required for proper stackwalking of RuntimeStub frame
1809 __ ret(0);
1811 return start;
1812 }
1814 //
1815 // Generate 'unsafe' array copy stub
1816 // Though just as safe as the other stubs, it takes an unscaled
1817 // size_t argument instead of an element count.
1818 //
1819 // Input:
1820 // c_rarg0 - source array address
1821 // c_rarg1 - destination array address
1822 // c_rarg2 - byte count, treated as ssize_t, can be zero
1823 //
1824 // Examines the alignment of the operands and dispatches
1825 // to a long, int, short, or byte copy loop.
1826 //
1827 address generate_unsafe_copy(const char *name) {
1829 Label L_long_aligned, L_int_aligned, L_short_aligned;
1831 // Input registers (before setup_arg_regs)
1832 const Register from = c_rarg0; // source array address
1833 const Register to = c_rarg1; // destination array address
1834 const Register size = c_rarg2; // byte count (size_t)
1836 // Register used as a temp
1837 const Register bits = rax; // test copy of low bits
1839 __ align(CodeEntryAlignment);
1840 StubCodeMark mark(this, "StubRoutines", name);
1841 address start = __ pc();
1843 __ enter(); // required for proper stackwalking of RuntimeStub frame
1845 // bump this on entry, not on exit:
1846 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1848 __ mov(bits, from);
1849 __ orptr(bits, to);
1850 __ orptr(bits, size);
1852 __ testb(bits, BytesPerLong-1);
1853 __ jccb(Assembler::zero, L_long_aligned);
1855 __ testb(bits, BytesPerInt-1);
1856 __ jccb(Assembler::zero, L_int_aligned);
1858 __ testb(bits, BytesPerShort-1);
1859 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1861 __ BIND(L_short_aligned);
1862 __ shrptr(size, LogBytesPerShort); // size => short_count
1863 __ jump(RuntimeAddress(short_copy_entry));
1865 __ BIND(L_int_aligned);
1866 __ shrptr(size, LogBytesPerInt); // size => int_count
1867 __ jump(RuntimeAddress(int_copy_entry));
1869 __ BIND(L_long_aligned);
1870 __ shrptr(size, LogBytesPerLong); // size => qword_count
1871 __ jump(RuntimeAddress(long_copy_entry));
1873 return start;
1874 }
1876 // Perform range checks on the proposed arraycopy.
1877 // Kills temp, but nothing else.
1878 // Also, clean the sign bits of src_pos and dst_pos.
1879 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
1880 Register src_pos, // source position (c_rarg1)
1881 Register dst, // destination array oo (c_rarg2)
1882 Register dst_pos, // destination position (c_rarg3)
1883 Register length,
1884 Register temp,
1885 Label& L_failed) {
1886 BLOCK_COMMENT("arraycopy_range_checks:");
1888 // if (src_pos + length > arrayOop(src)->length()) FAIL;
1889 __ movl(temp, length);
1890 __ addl(temp, src_pos); // src_pos + length
1891 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
1892 __ jcc(Assembler::above, L_failed);
1894 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
1895 __ movl(temp, length);
1896 __ addl(temp, dst_pos); // dst_pos + length
1897 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1898 __ jcc(Assembler::above, L_failed);
1900 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
1901 // Move with sign extension can be used since they are positive.
1902 __ movslq(src_pos, src_pos);
1903 __ movslq(dst_pos, dst_pos);
1905 BLOCK_COMMENT("arraycopy_range_checks done");
1906 }
1908 //
1909 // Generate generic array copy stubs
1910 //
1911 // Input:
1912 // c_rarg0 - src oop
1913 // c_rarg1 - src_pos (32-bits)
1914 // c_rarg2 - dst oop
1915 // c_rarg3 - dst_pos (32-bits)
1916 // not Win64
1917 // c_rarg4 - element count (32-bits)
1918 // Win64
1919 // rsp+40 - element count (32-bits)
1920 //
1921 // Output:
1922 // rax == 0 - success
1923 // rax == -1^K - failure, where K is partial transfer count
1924 //
1925 address generate_generic_copy(const char *name) {
1927 Label L_failed, L_failed_0, L_objArray;
1928 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
1930 // Input registers
1931 const Register src = c_rarg0; // source array oop
1932 const Register src_pos = c_rarg1; // source position
1933 const Register dst = c_rarg2; // destination array oop
1934 const Register dst_pos = c_rarg3; // destination position
1935 // elements count is on stack on Win64
1936 #ifdef _WIN64
1937 #define C_RARG4 Address(rsp, 6 * wordSize)
1938 #else
1939 #define C_RARG4 c_rarg4
1940 #endif
1942 { int modulus = CodeEntryAlignment;
1943 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1944 int advance = target - (__ offset() % modulus);
1945 if (advance < 0) advance += modulus;
1946 if (advance > 0) __ nop(advance);
1947 }
1948 StubCodeMark mark(this, "StubRoutines", name);
1950 // Short-hop target to L_failed. Makes for denser prologue code.
1951 __ BIND(L_failed_0);
1952 __ jmp(L_failed);
1953 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1955 __ align(CodeEntryAlignment);
1956 address start = __ pc();
1958 __ enter(); // required for proper stackwalking of RuntimeStub frame
1960 // bump this on entry, not on exit:
1961 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1963 //-----------------------------------------------------------------------
1964 // Assembler stub will be used for this call to arraycopy
1965 // if the following conditions are met:
1966 //
1967 // (1) src and dst must not be null.
1968 // (2) src_pos must not be negative.
1969 // (3) dst_pos must not be negative.
1970 // (4) length must not be negative.
1971 // (5) src klass and dst klass should be the same and not NULL.
1972 // (6) src and dst should be arrays.
1973 // (7) src_pos + length must not exceed length of src.
1974 // (8) dst_pos + length must not exceed length of dst.
1975 //
1977 // if (src == NULL) return -1;
1978 __ testptr(src, src); // src oop
1979 size_t j1off = __ offset();
1980 __ jccb(Assembler::zero, L_failed_0);
1982 // if (src_pos < 0) return -1;
1983 __ testl(src_pos, src_pos); // src_pos (32-bits)
1984 __ jccb(Assembler::negative, L_failed_0);
1986 // if (dst == NULL) return -1;
1987 __ testptr(dst, dst); // dst oop
1988 __ jccb(Assembler::zero, L_failed_0);
1990 // if (dst_pos < 0) return -1;
1991 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
1992 size_t j4off = __ offset();
1993 __ jccb(Assembler::negative, L_failed_0);
1995 // The first four tests are very dense code,
1996 // but not quite dense enough to put four
1997 // jumps in a 16-byte instruction fetch buffer.
1998 // That's good, because some branch predicters
1999 // do not like jumps so close together.
2000 // Make sure of this.
2001 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2003 // registers used as temp
2004 const Register r11_length = r11; // elements count to copy
2005 const Register r10_src_klass = r10; // array klass
2006 const Register r9_dst_klass = r9; // dest array klass
2008 // if (length < 0) return -1;
2009 __ movl(r11_length, C_RARG4); // length (elements count, 32-bits value)
2010 __ testl(r11_length, r11_length);
2011 __ jccb(Assembler::negative, L_failed_0);
2013 __ load_klass(r10_src_klass, src);
2014 #ifdef ASSERT
2015 // assert(src->klass() != NULL);
2016 BLOCK_COMMENT("assert klasses not null");
2017 { Label L1, L2;
2018 __ testptr(r10_src_klass, r10_src_klass);
2019 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2020 __ bind(L1);
2021 __ stop("broken null klass");
2022 __ bind(L2);
2023 __ load_klass(r9_dst_klass, dst);
2024 __ cmpq(r9_dst_klass, 0);
2025 __ jcc(Assembler::equal, L1); // this would be broken also
2026 BLOCK_COMMENT("assert done");
2027 }
2028 #endif
2030 // Load layout helper (32-bits)
2031 //
2032 // |array_tag| | header_size | element_type | |log2_element_size|
2033 // 32 30 24 16 8 2 0
2034 //
2035 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2036 //
2038 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2039 Klass::layout_helper_offset_in_bytes();
2041 const Register rax_lh = rax; // layout helper
2043 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2045 // Handle objArrays completely differently...
2046 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2047 __ cmpl(rax_lh, objArray_lh);
2048 __ jcc(Assembler::equal, L_objArray);
2050 // if (src->klass() != dst->klass()) return -1;
2051 __ load_klass(r9_dst_klass, dst);
2052 __ cmpq(r10_src_klass, r9_dst_klass);
2053 __ jcc(Assembler::notEqual, L_failed);
2055 // if (!src->is_Array()) return -1;
2056 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2057 __ jcc(Assembler::greaterEqual, L_failed);
2059 // At this point, it is known to be a typeArray (array_tag 0x3).
2060 #ifdef ASSERT
2061 { Label L;
2062 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2063 __ jcc(Assembler::greaterEqual, L);
2064 __ stop("must be a primitive array");
2065 __ bind(L);
2066 }
2067 #endif
2069 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2070 r10, L_failed);
2072 // typeArrayKlass
2073 //
2074 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2075 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2076 //
2078 const Register r10_offset = r10; // array offset
2079 const Register rax_elsize = rax_lh; // element size
2081 __ movl(r10_offset, rax_lh);
2082 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2083 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2084 __ addptr(src, r10_offset); // src array offset
2085 __ addptr(dst, r10_offset); // dst array offset
2086 BLOCK_COMMENT("choose copy loop based on element size");
2087 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2089 // next registers should be set before the jump to corresponding stub
2090 const Register from = c_rarg0; // source array address
2091 const Register to = c_rarg1; // destination array address
2092 const Register count = c_rarg2; // elements count
2094 // 'from', 'to', 'count' registers should be set in such order
2095 // since they are the same as 'src', 'src_pos', 'dst'.
2097 __ BIND(L_copy_bytes);
2098 __ cmpl(rax_elsize, 0);
2099 __ jccb(Assembler::notEqual, L_copy_shorts);
2100 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2101 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2102 __ movl2ptr(count, r11_length); // length
2103 __ jump(RuntimeAddress(byte_copy_entry));
2105 __ BIND(L_copy_shorts);
2106 __ cmpl(rax_elsize, LogBytesPerShort);
2107 __ jccb(Assembler::notEqual, L_copy_ints);
2108 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2109 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2110 __ movl2ptr(count, r11_length); // length
2111 __ jump(RuntimeAddress(short_copy_entry));
2113 __ BIND(L_copy_ints);
2114 __ cmpl(rax_elsize, LogBytesPerInt);
2115 __ jccb(Assembler::notEqual, L_copy_longs);
2116 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2117 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2118 __ movl2ptr(count, r11_length); // length
2119 __ jump(RuntimeAddress(int_copy_entry));
2121 __ BIND(L_copy_longs);
2122 #ifdef ASSERT
2123 { Label L;
2124 __ cmpl(rax_elsize, LogBytesPerLong);
2125 __ jcc(Assembler::equal, L);
2126 __ stop("must be long copy, but elsize is wrong");
2127 __ bind(L);
2128 }
2129 #endif
2130 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2131 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2132 __ movl2ptr(count, r11_length); // length
2133 __ jump(RuntimeAddress(long_copy_entry));
2135 // objArrayKlass
2136 __ BIND(L_objArray);
2137 // live at this point: r10_src_klass, src[_pos], dst[_pos]
2139 Label L_plain_copy, L_checkcast_copy;
2140 // test array classes for subtyping
2141 __ load_klass(r9_dst_klass, dst);
2142 __ cmpq(r10_src_klass, r9_dst_klass); // usual case is exact equality
2143 __ jcc(Assembler::notEqual, L_checkcast_copy);
2145 // Identically typed arrays can be copied without element-wise checks.
2146 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2147 r10, L_failed);
2149 __ lea(from, Address(src, src_pos, TIMES_OOP,
2150 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2151 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2152 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2153 __ movl2ptr(count, r11_length); // length
2154 __ BIND(L_plain_copy);
2155 __ jump(RuntimeAddress(oop_copy_entry));
2157 __ BIND(L_checkcast_copy);
2158 // live at this point: r10_src_klass, !r11_length
2159 {
2160 // assert(r11_length == C_RARG4); // will reload from here
2161 Register r11_dst_klass = r11;
2162 __ load_klass(r11_dst_klass, dst);
2164 // Before looking at dst.length, make sure dst is also an objArray.
2165 __ cmpl(Address(r11_dst_klass, lh_offset), objArray_lh);
2166 __ jcc(Assembler::notEqual, L_failed);
2168 // It is safe to examine both src.length and dst.length.
2169 #ifndef _WIN64
2170 arraycopy_range_checks(src, src_pos, dst, dst_pos, C_RARG4,
2171 rax, L_failed);
2172 #else
2173 __ movl(r11_length, C_RARG4); // reload
2174 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2175 rax, L_failed);
2176 __ load_klass(r11_dst_klass, dst); // reload
2177 #endif
2179 // Marshal the base address arguments now, freeing registers.
2180 __ lea(from, Address(src, src_pos, TIMES_OOP,
2181 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2182 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2183 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2184 __ movl(count, C_RARG4); // length (reloaded)
2185 Register sco_temp = c_rarg3; // this register is free now
2186 assert_different_registers(from, to, count, sco_temp,
2187 r11_dst_klass, r10_src_klass);
2188 assert_clean_int(count, sco_temp);
2190 // Generate the type check.
2191 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2192 Klass::super_check_offset_offset_in_bytes());
2193 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2194 assert_clean_int(sco_temp, rax);
2195 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2197 // Fetch destination element klass from the objArrayKlass header.
2198 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2199 objArrayKlass::element_klass_offset_in_bytes());
2200 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2201 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2202 assert_clean_int(sco_temp, rax);
2204 // the checkcast_copy loop needs two extra arguments:
2205 assert(c_rarg3 == sco_temp, "#3 already in place");
2206 __ movptr(C_RARG4, r11_dst_klass); // dst.klass.element_klass
2207 __ jump(RuntimeAddress(checkcast_copy_entry));
2208 }
2210 __ BIND(L_failed);
2211 __ xorptr(rax, rax);
2212 __ notptr(rax); // return -1
2213 __ leave(); // required for proper stackwalking of RuntimeStub frame
2214 __ ret(0);
2216 return start;
2217 }
2219 #undef length_arg
2220 #endif
2222 //FIXME
2223 address generate_disjoint_long_copy(bool aligned, const char *name) {
2224 Label l_1, l_2;
2225 StubCodeMark mark(this, "StubRoutines", name);
2226 __ align(CodeEntryAlignment);
2227 address start = __ pc();
2229 // __ movl(ecx, Address(esp, 4+8)); // count
2230 // __ movl(eax, Address(esp, 4+0)); // from
2231 // __ movl(edx, Address(esp, 4+4)); // to
2232 __ move(T1, A2);
2233 __ move(T3, A0);
2234 __ move(T0, A1);
2235 __ push(T3);
2236 __ push(T0);
2237 __ push(T1);
2238 //__ subl(edx, eax);
2239 //__ jmp(l_2);
2240 __ b(l_2);
2241 __ delayed()->nop();
2242 __ align(16);
2243 __ bind(l_1);
2244 // if (VM_Version::supports_mmx()) {
2245 // __ movq(mmx0, Address(eax));
2246 // __ movq(Address(eax, edx, Address::times_1), mmx0);
2247 // } else {
2248 // __ fild_d(Address(eax));
2249 __ ld(AT, T3, 0);
2250 // __ fistp_d(Address(eax, edx, Address::times_1));
2251 __ sd (AT, T0, 0);
2252 // }
2253 // __ addl(eax, 8);
2254 __ addi(T3, T3, 8);
2255 __ addi(T0, T0, 8);
2256 __ bind(l_2);
2257 // __ decl(ecx);
2258 __ addi(T1, T1, -1);
2259 // __ jcc(Assembler::greaterEqual, l_1);
2260 __ bgez(T1, l_1);
2261 __ delayed()->nop();
2262 // if (VM_Version::supports_mmx()) {
2263 // __ emms();
2264 // }
2265 // __ ret(0);
2266 __ pop(T1);
2267 __ pop(T0);
2268 __ pop(T3);
2269 __ jr(RA);
2270 __ delayed()->nop();
2271 return start;
2272 }
2275 address generate_conjoint_long_copy(bool aligned, const char *name) {
2276 Label l_1, l_2;
2277 StubCodeMark mark(this, "StubRoutines", name);
2278 __ align(CodeEntryAlignment);
2279 address start = __ pc();
2280 address nooverlap_target = aligned ?
2281 StubRoutines::arrayof_jlong_disjoint_arraycopy() :
2282 StubRoutines::jlong_disjoint_arraycopy();
2283 array_overlap_test(nooverlap_target, 3);
2285 __ push(T3);
2286 __ push(T0);
2287 __ push(T1);
2289 /* __ movl(ecx, Address(esp, 4+8)); // count
2290 __ movl(eax, Address(esp, 4+0)); // from
2291 __ movl(edx, Address(esp, 4+4)); // to
2292 __ jmp(l_2);
2294 */
2295 __ move(T1, A2);
2296 __ move(T3, A0);
2297 __ move(T0, A1);
2298 __ sll(AT, T1, Address::times_8);
2299 __ add(AT, T3, AT);
2300 __ lea(T3 , Address(AT, -8));
2301 __ sll(AT, T1, Address::times_8);
2302 __ add(AT, T0, AT);
2303 __ lea(T0 , Address(AT, -8));
2307 __ b(l_2);
2308 __ delayed()->nop();
2309 __ align(16);
2310 __ bind(l_1);
2311 /* if (VM_Version::supports_mmx()) {
2312 __ movq(mmx0, Address(eax, ecx, Address::times_8));
2313 __ movq(Address(edx, ecx,Address::times_8), mmx0);
2314 } else {
2315 __ fild_d(Address(eax, ecx, Address::times_8));
2316 __ fistp_d(Address(edx, ecx,Address::times_8));
2317 }
2318 */
2319 __ ld(AT, T3, 0);
2320 __ sd (AT, T0, 0);
2321 __ addi(T3, T3, -8);
2322 __ addi(T0, T0,-8);
2323 __ bind(l_2);
2324 // __ decl(ecx);
2325 __ addi(T1, T1, -1);
2326 //__ jcc(Assembler::greaterEqual, l_1);
2327 __ bgez(T1, l_1);
2328 __ delayed()->nop();
2329 // if (VM_Version::supports_mmx()) {
2330 // __ emms();
2331 // }
2332 // __ ret(0);
2333 __ pop(T1);
2334 __ pop(T0);
2335 __ pop(T3);
2336 __ jr(RA);
2337 __ delayed()->nop();
2338 return start;
2339 }
2341 void generate_arraycopy_stubs() {
2342 if (UseCompressedOops) {
2343 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, "oop_disjoint_arraycopy");
2344 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, "oop_arraycopy");
2345 } else {
2346 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, "oop_disjoint_arraycopy");
2347 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, "oop_arraycopy");
2348 }
2350 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
2351 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
2352 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, "jint_disjoint_arraycopy");
2353 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_copy(false, "jlong_disjoint_arraycopy");
2354 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(true, "arrayof_jbyte_disjoint_arraycopy");
2356 // if (VM_Version::supports_mmx())
2357 //if (false)
2358 // StubRoutines::_arrayof_jshort_disjoint_arraycopy = generate_disjoint_short_mmx_copy_aligned("arrayof_jshort_disjoint_arraycopy");
2359 // else
2360 StubRoutines::_arrayof_jshort_disjoint_arraycopy = generate_disjoint_short_copy(true, "arrayof_jshort_disjoint_arraycopy");
2361 StubRoutines::_arrayof_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(true, false, "arrayof_jint_disjoint_arraycopy");
2362 //StubRoutines::_arrayof_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(true, true, "arrayof_oop_disjoint_arraycopy");
2363 StubRoutines::_arrayof_jlong_disjoint_arraycopy = generate_disjoint_long_copy(true, "arrayof_jlong_disjoint_arraycopy");
2365 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
2366 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
2367 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, "jint_arraycopy");
2368 StubRoutines::_jlong_arraycopy = generate_conjoint_long_copy(false, "jlong_arraycopy");
2370 StubRoutines::_arrayof_jbyte_arraycopy = generate_conjoint_byte_copy(true, "arrayof_jbyte_arraycopy");
2371 StubRoutines::_arrayof_jshort_arraycopy = generate_conjoint_short_copy(true, "arrayof_jshort_arraycopy");
2372 StubRoutines::_arrayof_jint_arraycopy = generate_conjoint_int_oop_copy(true, false, "arrayof_jint_arraycopy");
2373 //StubRoutines::_arrayof_oop_arraycopy = generate_conjoint_int_oop_copy(true, true, "arrayof_oop_arraycopy");
2374 StubRoutines::_arrayof_jlong_arraycopy = generate_conjoint_long_copy(true, "arrayof_jlong_arraycopy");
2376 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2377 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2378 }
2380 //Wang: add a function to implement SafeFetch32 and SafeFetchN
2381 void generate_safefetch(const char* name, int size, address* entry,
2382 address* fault_pc, address* continuation_pc) {
2383 // safefetch signatures:
2384 // int SafeFetch32(int* adr, int errValue);
2385 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
2386 //
2387 // arguments:
2388 // A0 = adr
2389 // A1 = errValue
2390 //
2391 // result:
2392 // PPC_RET = *adr or errValue
2394 StubCodeMark mark(this, "StubRoutines", name);
2396 // Entry point, pc or function descriptor.
2397 *entry = __ pc();
2399 // Load *adr into A1, may fault.
2400 *fault_pc = __ pc();
2401 switch (size) {
2402 case 4:
2403 // int32_t
2404 __ lw(A1, A0, 0);
2405 break;
2406 case 8:
2407 // int64_t
2408 __ ld(A1, A0, 0);
2409 break;
2410 default:
2411 ShouldNotReachHere();
2412 }
2414 // return errValue or *adr
2415 *continuation_pc = __ pc();
2416 __ addu(V0,A1,R0);
2417 __ jr(RA);
2418 __ delayed()->nop();
2419 }
2422 #undef __
2423 #define __ masm->
2425 // Continuation point for throwing of implicit exceptions that are
2426 // not handled in the current activation. Fabricates an exception
2427 // oop and initiates normal exception dispatching in this
2428 // frame. Since we need to preserve callee-saved values (currently
2429 // only for C2, but done for C1 as well) we need a callee-saved oop
2430 // map and therefore have to make these stubs into RuntimeStubs
2431 // rather than BufferBlobs. If the compiler needs all registers to
2432 // be preserved between the fault point and the exception handler
2433 // then it must assume responsibility for that in
2434 // AbstractCompiler::continuation_for_implicit_null_exception or
2435 // continuation_for_implicit_division_by_zero_exception. All other
2436 // implicit exceptions (e.g., NullPointerException or
2437 // AbstractMethodError on entry) are either at call sites or
2438 // otherwise assume that stack unwinding will be initiated, so
2439 // caller saved registers were assumed volatile in the compiler.
2440 address generate_throw_exception(const char* name,
2441 address runtime_entry,
2442 bool restore_saved_exception_pc) {
2443 // Information about frame layout at time of blocking runtime call.
2444 // Note that we only have to preserve callee-saved registers since
2445 // the compilers are responsible for supplying a continuation point
2446 // if they expect all registers to be preserved.
2447 //#define aoqi_test
2448 #ifdef aoqi_test
2449 tty->print_cr("%s:%d name:%s", __func__, __LINE__, name);
2450 #endif
2451 enum layout {
2452 thread_off, // last_java_sp
2453 S7_off, // callee saved register sp + 1
2454 S6_off, // callee saved register sp + 2
2455 S5_off, // callee saved register sp + 3
2456 S4_off, // callee saved register sp + 4
2457 S3_off, // callee saved register sp + 5
2458 S2_off, // callee saved register sp + 6
2459 S1_off, // callee saved register sp + 7
2460 S0_off, // callee saved register sp + 8
2461 FP_off,
2462 ret_address,
2463 framesize
2464 };
2466 int insts_size = 2048;
2467 int locs_size = 32;
2469 // CodeBuffer* code = new CodeBuffer(insts_size, locs_size, 0, 0, 0, false,
2470 // NULL, NULL, NULL, false, NULL, name, false);
2471 CodeBuffer code (name , insts_size, locs_size);
2472 #ifdef aoqi_test
2473 tty->print_cr("%s:%d name:%s", __func__, __LINE__, name);
2474 #endif
2475 OopMapSet* oop_maps = new OopMapSet();
2476 #ifdef aoqi_test
2477 tty->print_cr("%s:%d name:%s", __func__, __LINE__, name);
2478 #endif
2479 MacroAssembler* masm = new MacroAssembler(&code);
2480 #ifdef aoqi_test
2481 tty->print_cr("%s:%d name:%s", __func__, __LINE__, name);
2482 #endif
2484 address start = __ pc();
2485 //__ stop("generate_throw_exception");
2486 /*
2487 __ move(AT, (int)&jerome1 );
2488 __ sw(SP, AT, 0);
2489 __ move(AT, (int)&jerome2 );
2490 __ sw(FP, AT, 0);
2491 __ move(AT, (int)&jerome3 );
2492 __ sw(RA, AT, 0);
2493 __ move(AT, (int)&jerome4 );
2494 __ sw(R0, AT, 0);
2495 __ move(AT, (int)&jerome5 );
2496 __ sw(R0, AT, 0);
2497 __ move(AT, (int)&jerome6 );
2498 __ sw(R0, AT, 0);
2499 __ move(AT, (int)&jerome7 );
2500 __ sw(R0, AT, 0);
2501 __ move(AT, (int)&jerome10 );
2502 __ sw(R0, AT, 0);
2504 __ pushad();
2506 //__ enter();
2507 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::print_call_statistics),
2508 relocInfo::runtime_call_type);
2509 __ delayed()->nop();
2511 //__ leave();
2512 __ popad();
2514 */
2516 // This is an inlined and slightly modified version of call_VM
2517 // which has the ability to fetch the return PC out of
2518 // thread-local storage and also sets up last_Java_sp slightly
2519 // differently than the real call_VM
2520 #ifndef OPT_THREAD
2521 Register java_thread = TREG;
2522 __ get_thread(java_thread);
2523 #else
2524 Register java_thread = TREG;
2525 #endif
2526 #ifdef aoqi_test
2527 tty->print_cr("%s:%d name:%s", __func__, __LINE__, name);
2528 #endif
2529 if (restore_saved_exception_pc) {
2530 __ ld(RA, java_thread, in_bytes(JavaThread::saved_exception_pc_offset())); // eax
2531 }
2533 __ enter(); // required for proper stackwalking of RuntimeStub frame
2535 __ addi(SP, SP, (-1) * (framesize-2) * wordSize); // prolog
2536 __ sd(S0, SP, S0_off * wordSize);
2537 __ sd(S1, SP, S1_off * wordSize);
2538 __ sd(S2, SP, S2_off * wordSize);
2539 __ sd(S3, SP, S3_off * wordSize);
2540 __ sd(S4, SP, S4_off * wordSize);
2541 __ sd(S5, SP, S5_off * wordSize);
2542 __ sd(S6, SP, S6_off * wordSize);
2543 __ sd(S7, SP, S7_off * wordSize);
2545 int frame_complete = __ pc() - start;
2546 // push java thread (becomes first argument of C function)
2547 __ sd(java_thread, SP, thread_off * wordSize);
2548 if (java_thread!=A0)
2549 __ move(A0, java_thread);
2551 // Set up last_Java_sp and last_Java_fp
2552 __ set_last_Java_frame(java_thread, SP, FP, NULL);
2553 __ relocate(relocInfo::internal_pc_type);
2554 {
2555 intptr_t save_pc = (intptr_t)__ pc() + NativeMovConstReg::instruction_size + NativeCall::return_address_offset + 4;
2556 __ li48(AT, save_pc);
2557 }
2558 __ sd(AT, java_thread, in_bytes(JavaThread::last_Java_pc_offset()));
2560 // Call runtime
2561 __ call(runtime_entry);
2562 __ delayed()->nop();
2563 // Generate oop map
2564 OopMap* map = new OopMap(framesize, 0);
2565 oop_maps->add_gc_map(__ offset(), map);
2567 // restore the thread (cannot use the pushed argument since arguments
2568 // may be overwritten by C code generated by an optimizing compiler);
2569 // however can use the register value directly if it is callee saved.
2570 #ifndef OPT_THREAD
2571 __ get_thread(java_thread);
2572 #endif
2574 __ ld(SP, java_thread, in_bytes(JavaThread::last_Java_sp_offset()));
2575 // __ reset_last_Java_frame(java_thread, true);
2576 __ reset_last_Java_frame(java_thread, true, true);
2578 // Restore callee save registers. This must be done after resetting the Java frame
2579 __ ld(S0, SP, S0_off * wordSize);
2580 __ ld(S1, SP, S1_off * wordSize);
2581 __ ld(S2, SP, S2_off * wordSize);
2582 __ ld(S3, SP, S3_off * wordSize);
2583 __ ld(S4, SP, S4_off * wordSize);
2584 __ ld(S5, SP, S5_off * wordSize);
2585 __ ld(S6, SP, S6_off * wordSize);
2586 __ ld(S7, SP, S7_off * wordSize);
2588 // discard arguments
2589 __ addi(SP, SP, (framesize-2) * wordSize); // epilog
2590 // __ leave(); // required for proper stackwalking of RuntimeStub frame
2591 __ addi(SP, FP, wordSize);
2592 __ ld(FP, SP, -1*wordSize);
2593 // check for pending exceptions
2594 #ifdef ASSERT
2595 Label L;
2596 __ lw(AT, java_thread, in_bytes(Thread::pending_exception_offset()));
2597 __ bne(AT, R0, L);
2598 __ delayed()->nop();
2599 __ should_not_reach_here();
2600 __ bind(L);
2601 #endif //ASSERT
2602 __ jmp(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
2603 __ delayed()->nop();
2604 #ifdef aoqi_test
2605 tty->print_cr("%s:%d name:%s", __func__, __LINE__, name);
2606 #endif
2607 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code,frame_complete,
2608 framesize, oop_maps, false);
2609 #ifdef aoqi_test
2610 tty->print_cr("%s:%d name:%s", __func__, __LINE__, name);
2611 #endif
2612 return stub->entry_point();
2613 }
2615 // Initialization
2616 void generate_initial() {
2617 /*
2618 // Generates all stubs and initializes the entry points
2620 // This platform-specific stub is needed by generate_call_stub()
2621 StubRoutines::mips::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
2623 // entry points that exist in all platforms Note: This is code
2624 // that could be shared among different platforms - however the
2625 // benefit seems to be smaller than the disadvantage of having a
2626 // much more complicated generator structure. See also comment in
2627 // stubRoutines.hpp.
2629 StubRoutines::_forward_exception_entry = generate_forward_exception();
2631 StubRoutines::_call_stub_entry =
2632 generate_call_stub(StubRoutines::_call_stub_return_address);
2634 // is referenced by megamorphic call
2635 StubRoutines::_catch_exception_entry = generate_catch_exception();
2637 // atomic calls
2638 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2639 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
2640 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
2641 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
2642 StubRoutines::_atomic_add_entry = generate_atomic_add();
2643 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
2644 StubRoutines::_fence_entry = generate_orderaccess_fence();
2646 StubRoutines::_handler_for_unsafe_access_entry =
2647 generate_handler_for_unsafe_access();
2649 // platform dependent
2650 StubRoutines::mips::_get_previous_fp_entry = generate_get_previous_fp();
2652 StubRoutines::mips::_verify_mxcsr_entry = generate_verify_mxcsr();
2653 */
2654 // Generates all stubs and initializes the entry points
2656 //-------------------------------------------------------------
2657 //-----------------------------------------------------------
2658 // entry points that exist in all platforms
2659 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller
2660 // than the disadvantage of having a much more complicated generator structure.
2661 // See also comment in stubRoutines.hpp.
2662 StubRoutines::_forward_exception_entry = generate_forward_exception();
2663 StubRoutines::_call_stub_entry = generate_call_stub(StubRoutines::_call_stub_return_address);
2664 // is referenced by megamorphic call
2665 StubRoutines::_catch_exception_entry = generate_catch_exception();
2667 StubRoutines::_handler_for_unsafe_access_entry = generate_handler_for_unsafe_access();
2669 // platform dependent
2670 StubRoutines::gs2::_get_previous_fp_entry = generate_get_previous_fp();
2671 }
2673 void generate_all() {
2674 #ifdef aoqi_test
2675 tty->print_cr("%s:%d", __func__, __LINE__);
2676 #endif
2677 // Generates all stubs and initializes the entry points
2679 // These entry points require SharedInfo::stack0 to be set up in
2680 // non-core builds and need to be relocatable, so they each
2681 // fabricate a RuntimeStub internally.
2682 /*
2683 StubRoutines::_throw_AbstractMethodError_entry =
2684 generate_throw_exception("AbstractMethodError throw_exception",
2685 CAST_FROM_FN_PTR(address,
2686 SharedRuntime::
2687 throw_AbstractMethodError),
2688 false);
2690 StubRoutines::_throw_IncompatibleClassChangeError_entry =
2691 generate_throw_exception("IncompatibleClassChangeError throw_exception",
2692 CAST_FROM_FN_PTR(address,
2693 SharedRuntime::
2694 throw_IncompatibleClassChangeError),
2695 false);
2697 StubRoutines::_throw_ArithmeticException_entry =
2698 generate_throw_exception("ArithmeticException throw_exception",
2699 CAST_FROM_FN_PTR(address,
2700 SharedRuntime::
2701 throw_ArithmeticException),
2702 true);
2704 StubRoutines::_throw_NullPointerException_entry =
2705 generate_throw_exception("NullPointerException throw_exception",
2706 CAST_FROM_FN_PTR(address,
2707 SharedRuntime::
2708 throw_NullPointerException),
2709 true);
2711 StubRoutines::_throw_NullPointerException_at_call_entry =
2712 generate_throw_exception("NullPointerException at call throw_exception",
2713 CAST_FROM_FN_PTR(address,
2714 SharedRuntime::
2715 throw_NullPointerException_at_call),
2716 false);
2718 StubRoutines::_throw_StackOverflowError_entry =
2719 generate_throw_exception("StackOverflowError throw_exception",
2720 CAST_FROM_FN_PTR(address,
2721 SharedRuntime::
2722 throw_StackOverflowError),
2723 false);
2725 // entry points that are platform specific
2726 StubRoutines::mips::_f2i_fixup = generate_f2i_fixup();
2727 StubRoutines::mips::_f2l_fixup = generate_f2l_fixup();
2728 StubRoutines::mips::_d2i_fixup = generate_d2i_fixup();
2729 StubRoutines::mips::_d2l_fixup = generate_d2l_fixup();
2731 StubRoutines::mips::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
2732 StubRoutines::mips::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
2733 StubRoutines::mips::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
2734 StubRoutines::mips::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
2736 // support for verify_oop (must happen after universe_init)
2737 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2739 // arraycopy stubs used by compilers
2740 generate_arraycopy_stubs();
2741 */
2742 #ifdef aoqi_test
2743 tty->print_cr("%s:%d", __func__, __LINE__);
2744 #endif
2745 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false);
2746 #ifdef aoqi_test
2747 tty->print_cr("%s:%d", __func__, __LINE__);
2748 #endif
2749 // StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true);
2750 #ifdef aoqi_test
2751 tty->print_cr("%s:%d", __func__, __LINE__);
2752 #endif
2753 // StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2754 #ifdef aoqi_test
2755 tty->print_cr("%s:%d", __func__, __LINE__);
2756 #endif
2757 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call), false);
2758 #ifdef aoqi_test
2759 tty->print_cr("%s:%d", __func__, __LINE__);
2760 #endif
2761 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
2762 #ifdef aoqi_test
2763 tty->print_cr("%s:%d", __func__, __LINE__);
2764 #endif
2766 //------------------------------------------------------
2767 //------------------------------------------------------------------
2768 // entry points that are platform specific
2770 // support for verify_oop (must happen after universe_init)
2771 #ifdef aoqi_test
2772 tty->print_cr("%s:%d", __func__, __LINE__);
2773 #endif
2774 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2775 #ifdef aoqi_test
2776 tty->print_cr("%s:%d", __func__, __LINE__);
2777 #endif
2778 #ifndef CORE
2779 // arraycopy stubs used by compilers
2780 generate_arraycopy_stubs();
2781 #ifdef aoqi_test
2782 tty->print_cr("%s:%d", __func__, __LINE__);
2783 #endif
2784 #endif
2786 // Safefetch stubs.
2787 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
2788 &StubRoutines::_safefetch32_fault_pc,
2789 &StubRoutines::_safefetch32_continuation_pc);
2790 generate_safefetch("SafeFetchN", sizeof(intptr_t), &StubRoutines::_safefetchN_entry,
2791 &StubRoutines::_safefetchN_fault_pc,
2792 &StubRoutines::_safefetchN_continuation_pc);
2793 }
2795 public:
2796 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2797 if (all) {
2798 generate_all();
2799 } else {
2800 generate_initial();
2801 }
2802 }
2803 }; // end class declaration
2804 /*
2805 address StubGenerator::disjoint_byte_copy_entry = NULL;
2806 address StubGenerator::disjoint_short_copy_entry = NULL;
2807 address StubGenerator::disjoint_int_copy_entry = NULL;
2808 address StubGenerator::disjoint_long_copy_entry = NULL;
2809 address StubGenerator::disjoint_oop_copy_entry = NULL;
2811 address StubGenerator::byte_copy_entry = NULL;
2812 address StubGenerator::short_copy_entry = NULL;
2813 address StubGenerator::int_copy_entry = NULL;
2814 address StubGenerator::long_copy_entry = NULL;
2815 address StubGenerator::oop_copy_entry = NULL;
2817 address StubGenerator::checkcast_copy_entry = NULL;
2818 */
2819 void StubGenerator_generate(CodeBuffer* code, bool all) {
2820 StubGenerator g(code, all);
2821 }