1 /*
2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "incls/_precompiled.incl"
26 #include "incls/_stubGenerator_sparc.cpp.incl"
27
28 // Declaration and definition of StubGenerator (no .hpp file).
29 // For a more detailed description of the stub routine structure
30 // see the comment in stubRoutines.hpp.
31
32 #define __ _masm->
33
34 #ifdef PRODUCT
35 #define BLOCK_COMMENT(str) /* nothing */
36 #else
37 #define BLOCK_COMMENT(str) __ block_comment(str)
38 #endif
39
40 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
41
42 // Note: The register L7 is used as L7_thread_cache, and may not be used
43 // any other way within this module.
44
45
46 static const Register& Lstub_temp = L2;
47
48 // -------------------------------------------------------------------------------------------------------------------------
49 // Stub Code definitions
50
51 static address handle_unsafe_access() {
52 JavaThread* thread = JavaThread::current();
53 address pc = thread->saved_exception_pc();
54 address npc = thread->saved_exception_npc();
55 // pc is the instruction which we must emulate
56 // doing a no-op is fine: return garbage from the load
57
58 // request an async exception
59 thread->set_pending_unsafe_access_error();
60
61 // return address of next instruction to execute
62 return npc;
63 }
64
65 class StubGenerator: public StubCodeGenerator {
66 private:
67
68 #ifdef PRODUCT
69 #define inc_counter_np(a,b,c) (0)
70 #else
71 #define inc_counter_np(counter, t1, t2) \
72 BLOCK_COMMENT("inc_counter " #counter); \
73 __ inc_counter(&counter, t1, t2);
74 #endif
75
76 //----------------------------------------------------------------------------------------------------
77 // Call stubs are used to call Java from C
78
79 address generate_call_stub(address& return_pc) {
80 StubCodeMark mark(this, "StubRoutines", "call_stub");
81 address start = __ pc();
82
83 // Incoming arguments:
84 //
85 // o0 : call wrapper address
86 // o1 : result (address)
87 // o2 : result type
88 // o3 : method
89 // o4 : (interpreter) entry point
90 // o5 : parameters (address)
91 // [sp + 0x5c]: parameter size (in words)
92 // [sp + 0x60]: thread
93 //
94 // +---------------+ <--- sp + 0
95 // | |
96 // . reg save area .
97 // | |
98 // +---------------+ <--- sp + 0x40
99 // | |
100 // . extra 7 slots .
101 // | |
102 // +---------------+ <--- sp + 0x5c
103 // | param. size |
104 // +---------------+ <--- sp + 0x60
105 // | thread |
106 // +---------------+
107 // | |
108
109 // note: if the link argument position changes, adjust
110 // the code in frame::entry_frame_call_wrapper()
111
112 const Argument link = Argument(0, false); // used only for GC
113 const Argument result = Argument(1, false);
114 const Argument result_type = Argument(2, false);
115 const Argument method = Argument(3, false);
116 const Argument entry_point = Argument(4, false);
117 const Argument parameters = Argument(5, false);
118 const Argument parameter_size = Argument(6, false);
119 const Argument thread = Argument(7, false);
120
121 // setup thread register
122 __ ld_ptr(thread.as_address(), G2_thread);
123 __ reinit_heapbase();
124
125 #ifdef ASSERT
126 // make sure we have no pending exceptions
127 { const Register t = G3_scratch;
128 Label L;
129 __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), t);
130 __ br_null(t, false, Assembler::pt, L);
131 __ delayed()->nop();
132 __ stop("StubRoutines::call_stub: entered with pending exception");
133 __ bind(L);
134 }
135 #endif
136
137 // create activation frame & allocate space for parameters
138 { const Register t = G3_scratch;
139 __ ld_ptr(parameter_size.as_address(), t); // get parameter size (in words)
140 __ add(t, frame::memory_parameter_word_sp_offset, t); // add space for save area (in words)
141 __ round_to(t, WordsPerLong); // make sure it is multiple of 2 (in words)
142 __ sll(t, Interpreter::logStackElementSize, t); // compute number of bytes
143 __ neg(t); // negate so it can be used with save
144 __ save(SP, t, SP); // setup new frame
145 }
146
147 // +---------------+ <--- sp + 0
148 // | |
149 // . reg save area .
150 // | |
151 // +---------------+ <--- sp + 0x40
152 // | |
153 // . extra 7 slots .
154 // | |
155 // +---------------+ <--- sp + 0x5c
156 // | empty slot | (only if parameter size is even)
157 // +---------------+
158 // | |
159 // . parameters .
160 // | |
161 // +---------------+ <--- fp + 0
162 // | |
163 // . reg save area .
164 // | |
165 // +---------------+ <--- fp + 0x40
166 // | |
167 // . extra 7 slots .
168 // | |
169 // +---------------+ <--- fp + 0x5c
170 // | param. size |
171 // +---------------+ <--- fp + 0x60
172 // | thread |
173 // +---------------+
174 // | |
175
176 // pass parameters if any
177 BLOCK_COMMENT("pass parameters if any");
178 { const Register src = parameters.as_in().as_register();
179 const Register dst = Lentry_args;
180 const Register tmp = G3_scratch;
181 const Register cnt = G4_scratch;
182
183 // test if any parameters & setup of Lentry_args
184 Label exit;
185 __ ld_ptr(parameter_size.as_in().as_address(), cnt); // parameter counter
186 __ add( FP, STACK_BIAS, dst );
187 __ tst(cnt);
188 __ br(Assembler::zero, false, Assembler::pn, exit);
189 __ delayed()->sub(dst, BytesPerWord, dst); // setup Lentry_args
190
191 // copy parameters if any
192 Label loop;
193 __ BIND(loop);
194 // Store parameter value
195 __ ld_ptr(src, 0, tmp);
196 __ add(src, BytesPerWord, src);
197 __ st_ptr(tmp, dst, 0);
198 __ deccc(cnt);
199 __ br(Assembler::greater, false, Assembler::pt, loop);
200 __ delayed()->sub(dst, Interpreter::stackElementSize, dst);
201
202 // done
203 __ BIND(exit);
204 }
205
206 // setup parameters, method & call Java function
207 #ifdef ASSERT
208 // layout_activation_impl checks it's notion of saved SP against
209 // this register, so if this changes update it as well.
210 const Register saved_SP = Lscratch;
211 __ mov(SP, saved_SP); // keep track of SP before call
212 #endif
213
214 // setup parameters
215 const Register t = G3_scratch;
216 __ ld_ptr(parameter_size.as_in().as_address(), t); // get parameter size (in words)
217 __ sll(t, Interpreter::logStackElementSize, t); // compute number of bytes
218 __ sub(FP, t, Gargs); // setup parameter pointer
219 #ifdef _LP64
220 __ add( Gargs, STACK_BIAS, Gargs ); // Account for LP64 stack bias
221 #endif
222 __ mov(SP, O5_savedSP);
223
224
225 // do the call
226 //
227 // the following register must be setup:
228 //
229 // G2_thread
230 // G5_method
231 // Gargs
232 BLOCK_COMMENT("call Java function");
233 __ jmpl(entry_point.as_in().as_register(), G0, O7);
234 __ delayed()->mov(method.as_in().as_register(), G5_method); // setup method
235
236 BLOCK_COMMENT("call_stub_return_address:");
237 return_pc = __ pc();
238
239 // The callee, if it wasn't interpreted, can return with SP changed so
240 // we can no longer assert of change of SP.
241
242 // store result depending on type
243 // (everything that is not T_OBJECT, T_LONG, T_FLOAT, or T_DOUBLE
244 // is treated as T_INT)
245 { const Register addr = result .as_in().as_register();
246 const Register type = result_type.as_in().as_register();
247 Label is_long, is_float, is_double, is_object, exit;
248 __ cmp(type, T_OBJECT); __ br(Assembler::equal, false, Assembler::pn, is_object);
249 __ delayed()->cmp(type, T_FLOAT); __ br(Assembler::equal, false, Assembler::pn, is_float);
250 __ delayed()->cmp(type, T_DOUBLE); __ br(Assembler::equal, false, Assembler::pn, is_double);
251 __ delayed()->cmp(type, T_LONG); __ br(Assembler::equal, false, Assembler::pn, is_long);
252 __ delayed()->nop();
253
254 // store int result
255 __ st(O0, addr, G0);
256
257 __ BIND(exit);
258 __ ret();
259 __ delayed()->restore();
260
261 __ BIND(is_object);
262 __ ba(false, exit);
263 __ delayed()->st_ptr(O0, addr, G0);
264
265 __ BIND(is_float);
266 __ ba(false, exit);
267 __ delayed()->stf(FloatRegisterImpl::S, F0, addr, G0);
268
269 __ BIND(is_double);
270 __ ba(false, exit);
271 __ delayed()->stf(FloatRegisterImpl::D, F0, addr, G0);
272
273 __ BIND(is_long);
274 #ifdef _LP64
275 __ ba(false, exit);
276 __ delayed()->st_long(O0, addr, G0); // store entire long
277 #else
278 #if defined(COMPILER2)
279 // All return values are where we want them, except for Longs. C2 returns
280 // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1.
281 // Since the interpreter will return longs in G1 and O0/O1 in the 32bit
282 // build we simply always use G1.
283 // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to
284 // do this here. Unfortunately if we did a rethrow we'd see an machepilog node
285 // first which would move g1 -> O0/O1 and destroy the exception we were throwing.
286
287 __ ba(false, exit);
288 __ delayed()->stx(G1, addr, G0); // store entire long
289 #else
290 __ st(O1, addr, BytesPerInt);
291 __ ba(false, exit);
292 __ delayed()->st(O0, addr, G0);
293 #endif /* COMPILER2 */
294 #endif /* _LP64 */
295 }
296 return start;
297 }
298
299
300 //----------------------------------------------------------------------------------------------------
301 // Return point for a Java call if there's an exception thrown in Java code.
302 // The exception is caught and transformed into a pending exception stored in
303 // JavaThread that can be tested from within the VM.
304 //
305 // Oexception: exception oop
306
307 address generate_catch_exception() {
308 StubCodeMark mark(this, "StubRoutines", "catch_exception");
309
310 address start = __ pc();
311 // verify that thread corresponds
312 __ verify_thread();
313
314 const Register& temp_reg = Gtemp;
315 Address pending_exception_addr (G2_thread, Thread::pending_exception_offset());
316 Address exception_file_offset_addr(G2_thread, Thread::exception_file_offset ());
317 Address exception_line_offset_addr(G2_thread, Thread::exception_line_offset ());
318
319 // set pending exception
320 __ verify_oop(Oexception);
321 __ st_ptr(Oexception, pending_exception_addr);
322 __ set((intptr_t)__FILE__, temp_reg);
323 __ st_ptr(temp_reg, exception_file_offset_addr);
324 __ set((intptr_t)__LINE__, temp_reg);
325 __ st(temp_reg, exception_line_offset_addr);
326
327 // complete return to VM
328 assert(StubRoutines::_call_stub_return_address != NULL, "must have been generated before");
329
330 AddressLiteral stub_ret(StubRoutines::_call_stub_return_address);
331 __ jump_to(stub_ret, temp_reg);
332 __ delayed()->nop();
333
334 return start;
335 }
336
337
338 //----------------------------------------------------------------------------------------------------
339 // Continuation point for runtime calls returning with a pending exception
340 // The pending exception check happened in the runtime or native call stub
341 // The pending exception in Thread is converted into a Java-level exception
342 //
343 // Contract with Java-level exception handler: O0 = exception
344 // O1 = throwing pc
345
346 address generate_forward_exception() {
347 StubCodeMark mark(this, "StubRoutines", "forward_exception");
348 address start = __ pc();
349
350 // Upon entry, O7 has the return address returning into Java
351 // (interpreted or compiled) code; i.e. the return address
352 // becomes the throwing pc.
353
354 const Register& handler_reg = Gtemp;
355
356 Address exception_addr(G2_thread, Thread::pending_exception_offset());
357
358 #ifdef ASSERT
359 // make sure that this code is only executed if there is a pending exception
360 { Label L;
361 __ ld_ptr(exception_addr, Gtemp);
362 __ br_notnull(Gtemp, false, Assembler::pt, L);
363 __ delayed()->nop();
364 __ stop("StubRoutines::forward exception: no pending exception (1)");
365 __ bind(L);
366 }
367 #endif
368
369 // compute exception handler into handler_reg
370 __ get_thread();
371 __ ld_ptr(exception_addr, Oexception);
372 __ verify_oop(Oexception);
373 __ save_frame(0); // compensates for compiler weakness
374 __ add(O7->after_save(), frame::pc_return_offset, Lscratch); // save the issuing PC
375 BLOCK_COMMENT("call exception_handler_for_return_address");
376 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), G2_thread, Lscratch);
377 __ mov(O0, handler_reg);
378 __ restore(); // compensates for compiler weakness
379
380 __ ld_ptr(exception_addr, Oexception);
381 __ add(O7, frame::pc_return_offset, Oissuing_pc); // save the issuing PC
382
383 #ifdef ASSERT
384 // make sure exception is set
385 { Label L;
386 __ br_notnull(Oexception, false, Assembler::pt, L);
387 __ delayed()->nop();
388 __ stop("StubRoutines::forward exception: no pending exception (2)");
389 __ bind(L);
390 }
391 #endif
392 // jump to exception handler
393 __ jmp(handler_reg, 0);
394 // clear pending exception
395 __ delayed()->st_ptr(G0, exception_addr);
396
397 return start;
398 }
399
400
401 //------------------------------------------------------------------------------------------------------------------------
402 // Continuation point for throwing of implicit exceptions that are not handled in
403 // the current activation. Fabricates an exception oop and initiates normal
404 // exception dispatching in this frame. Only callee-saved registers are preserved
405 // (through the normal register window / RegisterMap handling).
406 // If the compiler needs all registers to be preserved between the fault
407 // point and the exception handler then it must assume responsibility for that in
408 // AbstractCompiler::continuation_for_implicit_null_exception or
409 // continuation_for_implicit_division_by_zero_exception. All other implicit
410 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
411 // either at call sites or otherwise assume that stack unwinding will be initiated,
412 // so caller saved registers were assumed volatile in the compiler.
413
414 // Note that we generate only this stub into a RuntimeStub, because it needs to be
415 // properly traversed and ignored during GC, so we change the meaning of the "__"
416 // macro within this method.
417 #undef __
418 #define __ masm->
419
420 address generate_throw_exception(const char* name, address runtime_entry, bool restore_saved_exception_pc) {
421 #ifdef ASSERT
422 int insts_size = VerifyThread ? 1 * K : 600;
423 #else
424 int insts_size = VerifyThread ? 1 * K : 256;
425 #endif /* ASSERT */
426 int locs_size = 32;
427
428 CodeBuffer code(name, insts_size, locs_size);
429 MacroAssembler* masm = new MacroAssembler(&code);
430
431 __ verify_thread();
432
433 // This is an inlined and slightly modified version of call_VM
434 // which has the ability to fetch the return PC out of thread-local storage
435 __ assert_not_delayed();
436
437 // Note that we always push a frame because on the SPARC
438 // architecture, for all of our implicit exception kinds at call
439 // sites, the implicit exception is taken before the callee frame
440 // is pushed.
441 __ save_frame(0);
442
443 int frame_complete = __ offset();
444
445 if (restore_saved_exception_pc) {
446 __ ld_ptr(G2_thread, JavaThread::saved_exception_pc_offset(), I7);
447 __ sub(I7, frame::pc_return_offset, I7);
448 }
449
450 // Note that we always have a runtime stub frame on the top of stack by this point
451 Register last_java_sp = SP;
452 // 64-bit last_java_sp is biased!
453 __ set_last_Java_frame(last_java_sp, G0);
454 if (VerifyThread) __ mov(G2_thread, O0); // about to be smashed; pass early
455 __ save_thread(noreg);
456 // do the call
457 BLOCK_COMMENT("call runtime_entry");
458 __ call(runtime_entry, relocInfo::runtime_call_type);
459 if (!VerifyThread)
460 __ delayed()->mov(G2_thread, O0); // pass thread as first argument
461 else
462 __ delayed()->nop(); // (thread already passed)
463 __ restore_thread(noreg);
464 __ reset_last_Java_frame();
465
466 // check for pending exceptions. use Gtemp as scratch register.
467 #ifdef ASSERT
468 Label L;
469
470 Address exception_addr(G2_thread, Thread::pending_exception_offset());
471 Register scratch_reg = Gtemp;
472 __ ld_ptr(exception_addr, scratch_reg);
473 __ br_notnull(scratch_reg, false, Assembler::pt, L);
474 __ delayed()->nop();
475 __ should_not_reach_here();
476 __ bind(L);
477 #endif // ASSERT
478 BLOCK_COMMENT("call forward_exception_entry");
479 __ call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
480 // we use O7 linkage so that forward_exception_entry has the issuing PC
481 __ delayed()->restore();
482
483 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, masm->total_frame_size_in_bytes(0), NULL, false);
484 return stub->entry_point();
485 }
486
487 #undef __
488 #define __ _masm->
489
490
491 // Generate a routine that sets all the registers so we
492 // can tell if the stop routine prints them correctly.
493 address generate_test_stop() {
494 StubCodeMark mark(this, "StubRoutines", "test_stop");
495 address start = __ pc();
496
497 int i;
498
499 __ save_frame(0);
500
501 static jfloat zero = 0.0, one = 1.0;
502
503 // put addr in L0, then load through L0 to F0
504 __ set((intptr_t)&zero, L0); __ ldf( FloatRegisterImpl::S, L0, 0, F0);
505 __ set((intptr_t)&one, L0); __ ldf( FloatRegisterImpl::S, L0, 0, F1); // 1.0 to F1
506
507 // use add to put 2..18 in F2..F18
508 for ( i = 2; i <= 18; ++i ) {
509 __ fadd( FloatRegisterImpl::S, F1, as_FloatRegister(i-1), as_FloatRegister(i));
510 }
511
512 // Now put double 2 in F16, double 18 in F18
513 __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, F2, F16 );
514 __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, F18, F18 );
515
516 // use add to put 20..32 in F20..F32
517 for (i = 20; i < 32; i += 2) {
518 __ fadd( FloatRegisterImpl::D, F16, as_FloatRegister(i-2), as_FloatRegister(i));
519 }
520
521 // put 0..7 in i's, 8..15 in l's, 16..23 in o's, 24..31 in g's
522 for ( i = 0; i < 8; ++i ) {
523 if (i < 6) {
524 __ set( i, as_iRegister(i));
525 __ set(16 + i, as_oRegister(i));
526 __ set(24 + i, as_gRegister(i));
527 }
528 __ set( 8 + i, as_lRegister(i));
529 }
530
531 __ stop("testing stop");
532
533
534 __ ret();
535 __ delayed()->restore();
536
537 return start;
538 }
539
540
541 address generate_stop_subroutine() {
542 StubCodeMark mark(this, "StubRoutines", "stop_subroutine");
543 address start = __ pc();
544
545 __ stop_subroutine();
546
547 return start;
548 }
549
550 address generate_flush_callers_register_windows() {
551 StubCodeMark mark(this, "StubRoutines", "flush_callers_register_windows");
552 address start = __ pc();
553
554 __ flush_windows();
555 __ retl(false);
556 __ delayed()->add( FP, STACK_BIAS, O0 );
557 // The returned value must be a stack pointer whose register save area
558 // is flushed, and will stay flushed while the caller executes.
559
560 return start;
561 }
562
563 // Helper functions for v8 atomic operations.
564 //
565 void get_v8_oop_lock_ptr(Register lock_ptr_reg, Register mark_oop_reg, Register scratch_reg) {
566 if (mark_oop_reg == noreg) {
567 address lock_ptr = (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr();
568 __ set((intptr_t)lock_ptr, lock_ptr_reg);
569 } else {
570 assert(scratch_reg != noreg, "just checking");
571 address lock_ptr = (address)StubRoutines::Sparc::_v8_oop_lock_cache;
572 __ set((intptr_t)lock_ptr, lock_ptr_reg);
573 __ and3(mark_oop_reg, StubRoutines::Sparc::v8_oop_lock_mask_in_place, scratch_reg);
574 __ add(lock_ptr_reg, scratch_reg, lock_ptr_reg);
575 }
576 }
577
578 void generate_v8_lock_prologue(Register lock_reg, Register lock_ptr_reg, Register yield_reg, Label& retry, Label& dontyield, Register mark_oop_reg = noreg, Register scratch_reg = noreg) {
579
580 get_v8_oop_lock_ptr(lock_ptr_reg, mark_oop_reg, scratch_reg);
581 __ set(StubRoutines::Sparc::locked, lock_reg);
582 // Initialize yield counter
583 __ mov(G0,yield_reg);
584
585 __ BIND(retry);
586 __ cmp(yield_reg, V8AtomicOperationUnderLockSpinCount);
587 __ br(Assembler::less, false, Assembler::pt, dontyield);
588 __ delayed()->nop();
589
590 // This code can only be called from inside the VM, this
591 // stub is only invoked from Atomic::add(). We do not
592 // want to use call_VM, because _last_java_sp and such
593 // must already be set.
594 //
595 // Save the regs and make space for a C call
596 __ save(SP, -96, SP);
597 __ save_all_globals_into_locals();
598 BLOCK_COMMENT("call os::naked_sleep");
599 __ call(CAST_FROM_FN_PTR(address, os::naked_sleep));
600 __ delayed()->nop();
601 __ restore_globals_from_locals();
602 __ restore();
603 // reset the counter
604 __ mov(G0,yield_reg);
605
606 __ BIND(dontyield);
607
608 // try to get lock
609 __ swap(lock_ptr_reg, 0, lock_reg);
610
611 // did we get the lock?
612 __ cmp(lock_reg, StubRoutines::Sparc::unlocked);
613 __ br(Assembler::notEqual, true, Assembler::pn, retry);
614 __ delayed()->add(yield_reg,1,yield_reg);
615
616 // yes, got lock. do the operation here.
617 }
618
619 void generate_v8_lock_epilogue(Register lock_reg, Register lock_ptr_reg, Register yield_reg, Label& retry, Label& dontyield, Register mark_oop_reg = noreg, Register scratch_reg = noreg) {
620 __ st(lock_reg, lock_ptr_reg, 0); // unlock
621 }
622
623 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest).
624 //
625 // Arguments :
626 //
627 // exchange_value: O0
628 // dest: O1
629 //
630 // Results:
631 //
632 // O0: the value previously stored in dest
633 //
634 address generate_atomic_xchg() {
635 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
636 address start = __ pc();
637
638 if (UseCASForSwap) {
639 // Use CAS instead of swap, just in case the MP hardware
640 // prefers to work with just one kind of synch. instruction.
641 Label retry;
642 __ BIND(retry);
643 __ mov(O0, O3); // scratch copy of exchange value
644 __ ld(O1, 0, O2); // observe the previous value
645 // try to replace O2 with O3
646 __ cas_under_lock(O1, O2, O3,
647 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(),false);
648 __ cmp(O2, O3);
649 __ br(Assembler::notEqual, false, Assembler::pn, retry);
650 __ delayed()->nop();
651
652 __ retl(false);
653 __ delayed()->mov(O2, O0); // report previous value to caller
654
655 } else {
656 if (VM_Version::v9_instructions_work()) {
657 __ retl(false);
658 __ delayed()->swap(O1, 0, O0);
659 } else {
660 const Register& lock_reg = O2;
661 const Register& lock_ptr_reg = O3;
662 const Register& yield_reg = O4;
663
664 Label retry;
665 Label dontyield;
666
667 generate_v8_lock_prologue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield);
668 // got the lock, do the swap
669 __ swap(O1, 0, O0);
670
671 generate_v8_lock_epilogue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield);
672 __ retl(false);
673 __ delayed()->nop();
674 }
675 }
676
677 return start;
678 }
679
680
681 // Support for jint Atomic::cmpxchg(jint exchange_value, volatile jint* dest, jint compare_value)
682 //
683 // Arguments :
684 //
685 // exchange_value: O0
686 // dest: O1
687 // compare_value: O2
688 //
689 // Results:
690 //
691 // O0: the value previously stored in dest
692 //
693 // Overwrites (v8): O3,O4,O5
694 //
695 address generate_atomic_cmpxchg() {
696 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
697 address start = __ pc();
698
699 // cmpxchg(dest, compare_value, exchange_value)
700 __ cas_under_lock(O1, O2, O0,
701 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(),false);
702 __ retl(false);
703 __ delayed()->nop();
704
705 return start;
706 }
707
708 // Support for jlong Atomic::cmpxchg(jlong exchange_value, volatile jlong *dest, jlong compare_value)
709 //
710 // Arguments :
711 //
712 // exchange_value: O1:O0
713 // dest: O2
714 // compare_value: O4:O3
715 //
716 // Results:
717 //
718 // O1:O0: the value previously stored in dest
719 //
720 // This only works on V9, on V8 we don't generate any
721 // code and just return NULL.
722 //
723 // Overwrites: G1,G2,G3
724 //
725 address generate_atomic_cmpxchg_long() {
726 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
727 address start = __ pc();
728
729 if (!VM_Version::supports_cx8())
730 return NULL;;
731 __ sllx(O0, 32, O0);
732 __ srl(O1, 0, O1);
733 __ or3(O0,O1,O0); // O0 holds 64-bit value from compare_value
734 __ sllx(O3, 32, O3);
735 __ srl(O4, 0, O4);
736 __ or3(O3,O4,O3); // O3 holds 64-bit value from exchange_value
737 __ casx(O2, O3, O0);
738 __ srl(O0, 0, O1); // unpacked return value in O1:O0
739 __ retl(false);
740 __ delayed()->srlx(O0, 32, O0);
741
742 return start;
743 }
744
745
746 // Support for jint Atomic::add(jint add_value, volatile jint* dest).
747 //
748 // Arguments :
749 //
750 // add_value: O0 (e.g., +1 or -1)
751 // dest: O1
752 //
753 // Results:
754 //
755 // O0: the new value stored in dest
756 //
757 // Overwrites (v9): O3
758 // Overwrites (v8): O3,O4,O5
759 //
760 address generate_atomic_add() {
761 StubCodeMark mark(this, "StubRoutines", "atomic_add");
762 address start = __ pc();
763 __ BIND(_atomic_add_stub);
764
765 if (VM_Version::v9_instructions_work()) {
766 Label(retry);
767 __ BIND(retry);
768
769 __ lduw(O1, 0, O2);
770 __ add(O0, O2, O3);
771 __ cas(O1, O2, O3);
772 __ cmp( O2, O3);
773 __ br(Assembler::notEqual, false, Assembler::pn, retry);
774 __ delayed()->nop();
775 __ retl(false);
776 __ delayed()->add(O0, O2, O0); // note that cas made O2==O3
777 } else {
778 const Register& lock_reg = O2;
779 const Register& lock_ptr_reg = O3;
780 const Register& value_reg = O4;
781 const Register& yield_reg = O5;
782
783 Label(retry);
784 Label(dontyield);
785
786 generate_v8_lock_prologue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield);
787 // got lock, do the increment
788 __ ld(O1, 0, value_reg);
789 __ add(O0, value_reg, value_reg);
790 __ st(value_reg, O1, 0);
791
792 // %%% only for RMO and PSO
793 __ membar(Assembler::StoreStore);
794
795 generate_v8_lock_epilogue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield);
796
797 __ retl(false);
798 __ delayed()->mov(value_reg, O0);
799 }
800
801 return start;
802 }
803 Label _atomic_add_stub; // called from other stubs
804
805
806 //------------------------------------------------------------------------------------------------------------------------
807 // The following routine generates a subroutine to throw an asynchronous
808 // UnknownError when an unsafe access gets a fault that could not be
809 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
810 //
811 // Arguments :
812 //
813 // trapping PC: O7
814 //
815 // Results:
816 // posts an asynchronous exception, skips the trapping instruction
817 //
818
819 address generate_handler_for_unsafe_access() {
820 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
821 address start = __ pc();
822
823 const int preserve_register_words = (64 * 2);
824 Address preserve_addr(FP, (-preserve_register_words * wordSize) + STACK_BIAS);
825
826 Register Lthread = L7_thread_cache;
827 int i;
828
829 __ save_frame(0);
830 __ mov(G1, L1);
831 __ mov(G2, L2);
832 __ mov(G3, L3);
833 __ mov(G4, L4);
834 __ mov(G5, L5);
835 for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) {
836 __ stf(FloatRegisterImpl::D, as_FloatRegister(i), preserve_addr, i * wordSize);
837 }
838
839 address entry_point = CAST_FROM_FN_PTR(address, handle_unsafe_access);
840 BLOCK_COMMENT("call handle_unsafe_access");
841 __ call(entry_point, relocInfo::runtime_call_type);
842 __ delayed()->nop();
843
844 __ mov(L1, G1);
845 __ mov(L2, G2);
846 __ mov(L3, G3);
847 __ mov(L4, G4);
848 __ mov(L5, G5);
849 for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) {
850 __ ldf(FloatRegisterImpl::D, preserve_addr, as_FloatRegister(i), i * wordSize);
851 }
852
853 __ verify_thread();
854
855 __ jmp(O0, 0);
856 __ delayed()->restore();
857
858 return start;
859 }
860
861
862 // Support for uint StubRoutine::Sparc::partial_subtype_check( Klass sub, Klass super );
863 // Arguments :
864 //
865 // ret : O0, returned
866 // icc/xcc: set as O0 (depending on wordSize)
867 // sub : O1, argument, not changed
868 // super: O2, argument, not changed
869 // raddr: O7, blown by call
870 address generate_partial_subtype_check() {
871 __ align(CodeEntryAlignment);
872 StubCodeMark mark(this, "StubRoutines", "partial_subtype_check");
873 address start = __ pc();
874 Label miss;
875
876 #if defined(COMPILER2) && !defined(_LP64)
877 // Do not use a 'save' because it blows the 64-bit O registers.
878 __ add(SP,-4*wordSize,SP); // Make space for 4 temps (stack must be 2 words aligned)
879 __ st_ptr(L0,SP,(frame::register_save_words+0)*wordSize);
880 __ st_ptr(L1,SP,(frame::register_save_words+1)*wordSize);
881 __ st_ptr(L2,SP,(frame::register_save_words+2)*wordSize);
882 __ st_ptr(L3,SP,(frame::register_save_words+3)*wordSize);
883 Register Rret = O0;
884 Register Rsub = O1;
885 Register Rsuper = O2;
886 #else
887 __ save_frame(0);
888 Register Rret = I0;
889 Register Rsub = I1;
890 Register Rsuper = I2;
891 #endif
892
893 Register L0_ary_len = L0;
894 Register L1_ary_ptr = L1;
895 Register L2_super = L2;
896 Register L3_index = L3;
897
898 __ check_klass_subtype_slow_path(Rsub, Rsuper,
899 L0, L1, L2, L3,
900 NULL, &miss);
901
902 // Match falls through here.
903 __ addcc(G0,0,Rret); // set Z flags, Z result
904
905 #if defined(COMPILER2) && !defined(_LP64)
906 __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0);
907 __ ld_ptr(SP,(frame::register_save_words+1)*wordSize,L1);
908 __ ld_ptr(SP,(frame::register_save_words+2)*wordSize,L2);
909 __ ld_ptr(SP,(frame::register_save_words+3)*wordSize,L3);
910 __ retl(); // Result in Rret is zero; flags set to Z
911 __ delayed()->add(SP,4*wordSize,SP);
912 #else
913 __ ret(); // Result in Rret is zero; flags set to Z
914 __ delayed()->restore();
915 #endif
916
917 __ BIND(miss);
918 __ addcc(G0,1,Rret); // set NZ flags, NZ result
919
920 #if defined(COMPILER2) && !defined(_LP64)
921 __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0);
922 __ ld_ptr(SP,(frame::register_save_words+1)*wordSize,L1);
923 __ ld_ptr(SP,(frame::register_save_words+2)*wordSize,L2);
924 __ ld_ptr(SP,(frame::register_save_words+3)*wordSize,L3);
925 __ retl(); // Result in Rret is != 0; flags set to NZ
926 __ delayed()->add(SP,4*wordSize,SP);
927 #else
928 __ ret(); // Result in Rret is != 0; flags set to NZ
929 __ delayed()->restore();
930 #endif
931
932 return start;
933 }
934
935
936 // Called from MacroAssembler::verify_oop
937 //
938 address generate_verify_oop_subroutine() {
939 StubCodeMark mark(this, "StubRoutines", "verify_oop_stub");
940
941 address start = __ pc();
942
943 __ verify_oop_subroutine();
944
945 return start;
946 }
947
948 static address disjoint_byte_copy_entry;
949 static address disjoint_short_copy_entry;
950 static address disjoint_int_copy_entry;
951 static address disjoint_long_copy_entry;
952 static address disjoint_oop_copy_entry;
953
954 static address byte_copy_entry;
955 static address short_copy_entry;
956 static address int_copy_entry;
957 static address long_copy_entry;
958 static address oop_copy_entry;
959
960 static address checkcast_copy_entry;
961
962 //
963 // Verify that a register contains clean 32-bits positive value
964 // (high 32-bits are 0) so it could be used in 64-bits shifts (sllx, srax).
965 //
966 // Input:
967 // Rint - 32-bits value
968 // Rtmp - scratch
969 //
970 void assert_clean_int(Register Rint, Register Rtmp) {
971 #if defined(ASSERT) && defined(_LP64)
972 __ signx(Rint, Rtmp);
973 __ cmp(Rint, Rtmp);
974 __ breakpoint_trap(Assembler::notEqual, Assembler::xcc);
975 #endif
976 }
977
978 //
979 // Generate overlap test for array copy stubs
980 //
981 // Input:
982 // O0 - array1
983 // O1 - array2
984 // O2 - element count
985 //
986 // Kills temps: O3, O4
987 //
988 void array_overlap_test(address no_overlap_target, int log2_elem_size) {
989 assert(no_overlap_target != NULL, "must be generated");
990 array_overlap_test(no_overlap_target, NULL, log2_elem_size);
991 }
992 void array_overlap_test(Label& L_no_overlap, int log2_elem_size) {
993 array_overlap_test(NULL, &L_no_overlap, log2_elem_size);
994 }
995 void array_overlap_test(address no_overlap_target, Label* NOLp, int log2_elem_size) {
996 const Register from = O0;
997 const Register to = O1;
998 const Register count = O2;
999 const Register to_from = O3; // to - from
1000 const Register byte_count = O4; // count << log2_elem_size
1001
1002 __ subcc(to, from, to_from);
1003 __ sll_ptr(count, log2_elem_size, byte_count);
1004 if (NOLp == NULL)
1005 __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, no_overlap_target);
1006 else
1007 __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, (*NOLp));
1008 __ delayed()->cmp(to_from, byte_count);
1009 if (NOLp == NULL)
1010 __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, no_overlap_target);
1011 else
1012 __ brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, (*NOLp));
1013 __ delayed()->nop();
1014 }
1015
1016 //
1017 // Generate pre-write barrier for array.
1018 //
1019 // Input:
1020 // addr - register containing starting address
1021 // count - register containing element count
1022 // tmp - scratch register
1023 //
1024 // The input registers are overwritten.
1025 //
1026 void gen_write_ref_array_pre_barrier(Register addr, Register count) {
1027 BarrierSet* bs = Universe::heap()->barrier_set();
1028 if (bs->has_write_ref_pre_barrier()) {
1029 assert(bs->has_write_ref_array_pre_opt(),
1030 "Else unsupported barrier set.");
1031
1032 __ save_frame(0);
1033 // Save the necessary global regs... will be used after.
1034 if (addr->is_global()) {
1035 __ mov(addr, L0);
1036 }
1037 if (count->is_global()) {
1038 __ mov(count, L1);
1039 }
1040 __ mov(addr->after_save(), O0);
1041 // Get the count into O1
1042 __ call(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre));
1043 __ delayed()->mov(count->after_save(), O1);
1044 if (addr->is_global()) {
1045 __ mov(L0, addr);
1046 }
1047 if (count->is_global()) {
1048 __ mov(L1, count);
1049 }
1050 __ restore();
1051 }
1052 }
1053 //
1054 // Generate post-write barrier for array.
1055 //
1056 // Input:
1057 // addr - register containing starting address
1058 // count - register containing element count
1059 // tmp - scratch register
1060 //
1061 // The input registers are overwritten.
1062 //
1063 void gen_write_ref_array_post_barrier(Register addr, Register count,
1064 Register tmp) {
1065 BarrierSet* bs = Universe::heap()->barrier_set();
1066
1067 switch (bs->kind()) {
1068 case BarrierSet::G1SATBCT:
1069 case BarrierSet::G1SATBCTLogging:
1070 {
1071 // Get some new fresh output registers.
1072 __ save_frame(0);
1073 __ mov(addr->after_save(), O0);
1074 __ call(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post));
1075 __ delayed()->mov(count->after_save(), O1);
1076 __ restore();
1077 }
1078 break;
1079 case BarrierSet::CardTableModRef:
1080 case BarrierSet::CardTableExtension:
1081 {
1082 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1083 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1084 assert_different_registers(addr, count, tmp);
1085
1086 Label L_loop;
1087
1088 __ sll_ptr(count, LogBytesPerHeapOop, count);
1089 __ sub(count, BytesPerHeapOop, count);
1090 __ add(count, addr, count);
1091 // Use two shifts to clear out those low order two bits! (Cannot opt. into 1.)
1092 __ srl_ptr(addr, CardTableModRefBS::card_shift, addr);
1093 __ srl_ptr(count, CardTableModRefBS::card_shift, count);
1094 __ sub(count, addr, count);
1095 AddressLiteral rs(ct->byte_map_base);
1096 __ set(rs, tmp);
1097 __ BIND(L_loop);
1098 __ stb(G0, tmp, addr);
1099 __ subcc(count, 1, count);
1100 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop);
1101 __ delayed()->add(addr, 1, addr);
1102 }
1103 break;
1104 case BarrierSet::ModRef:
1105 break;
1106 default:
1107 ShouldNotReachHere();
1108 }
1109 }
1110
1111
1112 // Copy big chunks forward with shift
1113 //
1114 // Inputs:
1115 // from - source arrays
1116 // to - destination array aligned to 8-bytes
1117 // count - elements count to copy >= the count equivalent to 16 bytes
1118 // count_dec - elements count's decrement equivalent to 16 bytes
1119 // L_copy_bytes - copy exit label
1120 //
1121 void copy_16_bytes_forward_with_shift(Register from, Register to,
1122 Register count, int count_dec, Label& L_copy_bytes) {
1123 Label L_loop, L_aligned_copy, L_copy_last_bytes;
1124
1125 // if both arrays have the same alignment mod 8, do 8 bytes aligned copy
1126 __ andcc(from, 7, G1); // misaligned bytes
1127 __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy);
1128 __ delayed()->nop();
1129
1130 const Register left_shift = G1; // left shift bit counter
1131 const Register right_shift = G5; // right shift bit counter
1132
1133 __ sll(G1, LogBitsPerByte, left_shift);
1134 __ mov(64, right_shift);
1135 __ sub(right_shift, left_shift, right_shift);
1136
1137 //
1138 // Load 2 aligned 8-bytes chunks and use one from previous iteration
1139 // to form 2 aligned 8-bytes chunks to store.
1140 //
1141 __ deccc(count, count_dec); // Pre-decrement 'count'
1142 __ andn(from, 7, from); // Align address
1143 __ ldx(from, 0, O3);
1144 __ inc(from, 8);
1145 __ align(OptoLoopAlignment);
1146 __ BIND(L_loop);
1147 __ ldx(from, 0, O4);
1148 __ deccc(count, count_dec); // Can we do next iteration after this one?
1149 __ ldx(from, 8, G4);
1150 __ inc(to, 16);
1151 __ inc(from, 16);
1152 __ sllx(O3, left_shift, O3);
1153 __ srlx(O4, right_shift, G3);
1154 __ bset(G3, O3);
1155 __ stx(O3, to, -16);
1156 __ sllx(O4, left_shift, O4);
1157 __ srlx(G4, right_shift, G3);
1158 __ bset(G3, O4);
1159 __ stx(O4, to, -8);
1160 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop);
1161 __ delayed()->mov(G4, O3);
1162
1163 __ inccc(count, count_dec>>1 ); // + 8 bytes
1164 __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes);
1165 __ delayed()->inc(count, count_dec>>1); // restore 'count'
1166
1167 // copy 8 bytes, part of them already loaded in O3
1168 __ ldx(from, 0, O4);
1169 __ inc(to, 8);
1170 __ inc(from, 8);
1171 __ sllx(O3, left_shift, O3);
1172 __ srlx(O4, right_shift, G3);
1173 __ bset(O3, G3);
1174 __ stx(G3, to, -8);
1175
1176 __ BIND(L_copy_last_bytes);
1177 __ srl(right_shift, LogBitsPerByte, right_shift); // misaligned bytes
1178 __ br(Assembler::always, false, Assembler::pt, L_copy_bytes);
1179 __ delayed()->sub(from, right_shift, from); // restore address
1180
1181 __ BIND(L_aligned_copy);
1182 }
1183
1184 // Copy big chunks backward with shift
1185 //
1186 // Inputs:
1187 // end_from - source arrays end address
1188 // end_to - destination array end address aligned to 8-bytes
1189 // count - elements count to copy >= the count equivalent to 16 bytes
1190 // count_dec - elements count's decrement equivalent to 16 bytes
1191 // L_aligned_copy - aligned copy exit label
1192 // L_copy_bytes - copy exit label
1193 //
1194 void copy_16_bytes_backward_with_shift(Register end_from, Register end_to,
1195 Register count, int count_dec,
1196 Label& L_aligned_copy, Label& L_copy_bytes) {
1197 Label L_loop, L_copy_last_bytes;
1198
1199 // if both arrays have the same alignment mod 8, do 8 bytes aligned copy
1200 __ andcc(end_from, 7, G1); // misaligned bytes
1201 __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy);
1202 __ delayed()->deccc(count, count_dec); // Pre-decrement 'count'
1203
1204 const Register left_shift = G1; // left shift bit counter
1205 const Register right_shift = G5; // right shift bit counter
1206
1207 __ sll(G1, LogBitsPerByte, left_shift);
1208 __ mov(64, right_shift);
1209 __ sub(right_shift, left_shift, right_shift);
1210
1211 //
1212 // Load 2 aligned 8-bytes chunks and use one from previous iteration
1213 // to form 2 aligned 8-bytes chunks to store.
1214 //
1215 __ andn(end_from, 7, end_from); // Align address
1216 __ ldx(end_from, 0, O3);
1217 __ align(OptoLoopAlignment);
1218 __ BIND(L_loop);
1219 __ ldx(end_from, -8, O4);
1220 __ deccc(count, count_dec); // Can we do next iteration after this one?
1221 __ ldx(end_from, -16, G4);
1222 __ dec(end_to, 16);
1223 __ dec(end_from, 16);
1224 __ srlx(O3, right_shift, O3);
1225 __ sllx(O4, left_shift, G3);
1226 __ bset(G3, O3);
1227 __ stx(O3, end_to, 8);
1228 __ srlx(O4, right_shift, O4);
1229 __ sllx(G4, left_shift, G3);
1230 __ bset(G3, O4);
1231 __ stx(O4, end_to, 0);
1232 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop);
1233 __ delayed()->mov(G4, O3);
1234
1235 __ inccc(count, count_dec>>1 ); // + 8 bytes
1236 __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes);
1237 __ delayed()->inc(count, count_dec>>1); // restore 'count'
1238
1239 // copy 8 bytes, part of them already loaded in O3
1240 __ ldx(end_from, -8, O4);
1241 __ dec(end_to, 8);
1242 __ dec(end_from, 8);
1243 __ srlx(O3, right_shift, O3);
1244 __ sllx(O4, left_shift, G3);
1245 __ bset(O3, G3);
1246 __ stx(G3, end_to, 0);
1247
1248 __ BIND(L_copy_last_bytes);
1249 __ srl(left_shift, LogBitsPerByte, left_shift); // misaligned bytes
1250 __ br(Assembler::always, false, Assembler::pt, L_copy_bytes);
1251 __ delayed()->add(end_from, left_shift, end_from); // restore address
1252 }
1253
1254 //
1255 // Generate stub for disjoint byte copy. If "aligned" is true, the
1256 // "from" and "to" addresses are assumed to be heapword aligned.
1257 //
1258 // Arguments for generated stub:
1259 // from: O0
1260 // to: O1
1261 // count: O2 treated as signed
1262 //
1263 address generate_disjoint_byte_copy(bool aligned, const char * name) {
1264 __ align(CodeEntryAlignment);
1265 StubCodeMark mark(this, "StubRoutines", name);
1266 address start = __ pc();
1267
1268 Label L_skip_alignment, L_align;
1269 Label L_copy_byte, L_copy_byte_loop, L_exit;
1270
1271 const Register from = O0; // source array address
1272 const Register to = O1; // destination array address
1273 const Register count = O2; // elements count
1274 const Register offset = O5; // offset from start of arrays
1275 // O3, O4, G3, G4 are used as temp registers
1276
1277 assert_clean_int(count, O3); // Make sure 'count' is clean int.
1278
1279 if (!aligned) disjoint_byte_copy_entry = __ pc();
1280 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1281 if (!aligned) BLOCK_COMMENT("Entry:");
1282
1283 // for short arrays, just do single element copy
1284 __ cmp(count, 23); // 16 + 7
1285 __ brx(Assembler::less, false, Assembler::pn, L_copy_byte);
1286 __ delayed()->mov(G0, offset);
1287
1288 if (aligned) {
1289 // 'aligned' == true when it is known statically during compilation
1290 // of this arraycopy call site that both 'from' and 'to' addresses
1291 // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()).
1292 //
1293 // Aligned arrays have 4 bytes alignment in 32-bits VM
1294 // and 8 bytes - in 64-bits VM. So we do it only for 32-bits VM
1295 //
1296 #ifndef _LP64
1297 // copy a 4-bytes word if necessary to align 'to' to 8 bytes
1298 __ andcc(to, 7, G0);
1299 __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment);
1300 __ delayed()->ld(from, 0, O3);
1301 __ inc(from, 4);
1302 __ inc(to, 4);
1303 __ dec(count, 4);
1304 __ st(O3, to, -4);
1305 __ BIND(L_skip_alignment);
1306 #endif
1307 } else {
1308 // copy bytes to align 'to' on 8 byte boundary
1309 __ andcc(to, 7, G1); // misaligned bytes
1310 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1311 __ delayed()->neg(G1);
1312 __ inc(G1, 8); // bytes need to copy to next 8-bytes alignment
1313 __ sub(count, G1, count);
1314 __ BIND(L_align);
1315 __ ldub(from, 0, O3);
1316 __ deccc(G1);
1317 __ inc(from);
1318 __ stb(O3, to, 0);
1319 __ br(Assembler::notZero, false, Assembler::pt, L_align);
1320 __ delayed()->inc(to);
1321 __ BIND(L_skip_alignment);
1322 }
1323 #ifdef _LP64
1324 if (!aligned)
1325 #endif
1326 {
1327 // Copy with shift 16 bytes per iteration if arrays do not have
1328 // the same alignment mod 8, otherwise fall through to the next
1329 // code for aligned copy.
1330 // The compare above (count >= 23) guarantes 'count' >= 16 bytes.
1331 // Also jump over aligned copy after the copy with shift completed.
1332
1333 copy_16_bytes_forward_with_shift(from, to, count, 16, L_copy_byte);
1334 }
1335
1336 // Both array are 8 bytes aligned, copy 16 bytes at a time
1337 __ and3(count, 7, G4); // Save count
1338 __ srl(count, 3, count);
1339 generate_disjoint_long_copy_core(aligned);
1340 __ mov(G4, count); // Restore count
1341
1342 // copy tailing bytes
1343 __ BIND(L_copy_byte);
1344 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1345 __ delayed()->nop();
1346 __ align(OptoLoopAlignment);
1347 __ BIND(L_copy_byte_loop);
1348 __ ldub(from, offset, O3);
1349 __ deccc(count);
1350 __ stb(O3, to, offset);
1351 __ brx(Assembler::notZero, false, Assembler::pt, L_copy_byte_loop);
1352 __ delayed()->inc(offset);
1353
1354 __ BIND(L_exit);
1355 // O3, O4 are used as temp registers
1356 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr, O3, O4);
1357 __ retl();
1358 __ delayed()->mov(G0, O0); // return 0
1359 return start;
1360 }
1361
1362 //
1363 // Generate stub for conjoint byte copy. If "aligned" is true, the
1364 // "from" and "to" addresses are assumed to be heapword aligned.
1365 //
1366 // Arguments for generated stub:
1367 // from: O0
1368 // to: O1
1369 // count: O2 treated as signed
1370 //
1371 address generate_conjoint_byte_copy(bool aligned, const char * name) {
1372 // Do reverse copy.
1373
1374 __ align(CodeEntryAlignment);
1375 StubCodeMark mark(this, "StubRoutines", name);
1376 address start = __ pc();
1377 address nooverlap_target = aligned ?
1378 StubRoutines::arrayof_jbyte_disjoint_arraycopy() :
1379 disjoint_byte_copy_entry;
1380
1381 Label L_skip_alignment, L_align, L_aligned_copy;
1382 Label L_copy_byte, L_copy_byte_loop, L_exit;
1383
1384 const Register from = O0; // source array address
1385 const Register to = O1; // destination array address
1386 const Register count = O2; // elements count
1387 const Register end_from = from; // source array end address
1388 const Register end_to = to; // destination array end address
1389
1390 assert_clean_int(count, O3); // Make sure 'count' is clean int.
1391
1392 if (!aligned) byte_copy_entry = __ pc();
1393 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1394 if (!aligned) BLOCK_COMMENT("Entry:");
1395
1396 array_overlap_test(nooverlap_target, 0);
1397
1398 __ add(to, count, end_to); // offset after last copied element
1399
1400 // for short arrays, just do single element copy
1401 __ cmp(count, 23); // 16 + 7
1402 __ brx(Assembler::less, false, Assembler::pn, L_copy_byte);
1403 __ delayed()->add(from, count, end_from);
1404
1405 {
1406 // Align end of arrays since they could be not aligned even
1407 // when arrays itself are aligned.
1408
1409 // copy bytes to align 'end_to' on 8 byte boundary
1410 __ andcc(end_to, 7, G1); // misaligned bytes
1411 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1412 __ delayed()->nop();
1413 __ sub(count, G1, count);
1414 __ BIND(L_align);
1415 __ dec(end_from);
1416 __ dec(end_to);
1417 __ ldub(end_from, 0, O3);
1418 __ deccc(G1);
1419 __ brx(Assembler::notZero, false, Assembler::pt, L_align);
1420 __ delayed()->stb(O3, end_to, 0);
1421 __ BIND(L_skip_alignment);
1422 }
1423 #ifdef _LP64
1424 if (aligned) {
1425 // Both arrays are aligned to 8-bytes in 64-bits VM.
1426 // The 'count' is decremented in copy_16_bytes_backward_with_shift()
1427 // in unaligned case.
1428 __ dec(count, 16);
1429 } else
1430 #endif
1431 {
1432 // Copy with shift 16 bytes per iteration if arrays do not have
1433 // the same alignment mod 8, otherwise jump to the next
1434 // code for aligned copy (and substracting 16 from 'count' before jump).
1435 // The compare above (count >= 11) guarantes 'count' >= 16 bytes.
1436 // Also jump over aligned copy after the copy with shift completed.
1437
1438 copy_16_bytes_backward_with_shift(end_from, end_to, count, 16,
1439 L_aligned_copy, L_copy_byte);
1440 }
1441 // copy 4 elements (16 bytes) at a time
1442 __ align(OptoLoopAlignment);
1443 __ BIND(L_aligned_copy);
1444 __ dec(end_from, 16);
1445 __ ldx(end_from, 8, O3);
1446 __ ldx(end_from, 0, O4);
1447 __ dec(end_to, 16);
1448 __ deccc(count, 16);
1449 __ stx(O3, end_to, 8);
1450 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy);
1451 __ delayed()->stx(O4, end_to, 0);
1452 __ inc(count, 16);
1453
1454 // copy 1 element (2 bytes) at a time
1455 __ BIND(L_copy_byte);
1456 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1457 __ delayed()->nop();
1458 __ align(OptoLoopAlignment);
1459 __ BIND(L_copy_byte_loop);
1460 __ dec(end_from);
1461 __ dec(end_to);
1462 __ ldub(end_from, 0, O4);
1463 __ deccc(count);
1464 __ brx(Assembler::greater, false, Assembler::pt, L_copy_byte_loop);
1465 __ delayed()->stb(O4, end_to, 0);
1466
1467 __ BIND(L_exit);
1468 // O3, O4 are used as temp registers
1469 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr, O3, O4);
1470 __ retl();
1471 __ delayed()->mov(G0, O0); // return 0
1472 return start;
1473 }
1474
1475 //
1476 // Generate stub for disjoint short copy. If "aligned" is true, the
1477 // "from" and "to" addresses are assumed to be heapword aligned.
1478 //
1479 // Arguments for generated stub:
1480 // from: O0
1481 // to: O1
1482 // count: O2 treated as signed
1483 //
1484 address generate_disjoint_short_copy(bool aligned, const char * name) {
1485 __ align(CodeEntryAlignment);
1486 StubCodeMark mark(this, "StubRoutines", name);
1487 address start = __ pc();
1488
1489 Label L_skip_alignment, L_skip_alignment2;
1490 Label L_copy_2_bytes, L_copy_2_bytes_loop, L_exit;
1491
1492 const Register from = O0; // source array address
1493 const Register to = O1; // destination array address
1494 const Register count = O2; // elements count
1495 const Register offset = O5; // offset from start of arrays
1496 // O3, O4, G3, G4 are used as temp registers
1497
1498 assert_clean_int(count, O3); // Make sure 'count' is clean int.
1499
1500 if (!aligned) disjoint_short_copy_entry = __ pc();
1501 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1502 if (!aligned) BLOCK_COMMENT("Entry:");
1503
1504 // for short arrays, just do single element copy
1505 __ cmp(count, 11); // 8 + 3 (22 bytes)
1506 __ brx(Assembler::less, false, Assembler::pn, L_copy_2_bytes);
1507 __ delayed()->mov(G0, offset);
1508
1509 if (aligned) {
1510 // 'aligned' == true when it is known statically during compilation
1511 // of this arraycopy call site that both 'from' and 'to' addresses
1512 // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()).
1513 //
1514 // Aligned arrays have 4 bytes alignment in 32-bits VM
1515 // and 8 bytes - in 64-bits VM.
1516 //
1517 #ifndef _LP64
1518 // copy a 2-elements word if necessary to align 'to' to 8 bytes
1519 __ andcc(to, 7, G0);
1520 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1521 __ delayed()->ld(from, 0, O3);
1522 __ inc(from, 4);
1523 __ inc(to, 4);
1524 __ dec(count, 2);
1525 __ st(O3, to, -4);
1526 __ BIND(L_skip_alignment);
1527 #endif
1528 } else {
1529 // copy 1 element if necessary to align 'to' on an 4 bytes
1530 __ andcc(to, 3, G0);
1531 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1532 __ delayed()->lduh(from, 0, O3);
1533 __ inc(from, 2);
1534 __ inc(to, 2);
1535 __ dec(count);
1536 __ sth(O3, to, -2);
1537 __ BIND(L_skip_alignment);
1538
1539 // copy 2 elements to align 'to' on an 8 byte boundary
1540 __ andcc(to, 7, G0);
1541 __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment2);
1542 __ delayed()->lduh(from, 0, O3);
1543 __ dec(count, 2);
1544 __ lduh(from, 2, O4);
1545 __ inc(from, 4);
1546 __ inc(to, 4);
1547 __ sth(O3, to, -4);
1548 __ sth(O4, to, -2);
1549 __ BIND(L_skip_alignment2);
1550 }
1551 #ifdef _LP64
1552 if (!aligned)
1553 #endif
1554 {
1555 // Copy with shift 16 bytes per iteration if arrays do not have
1556 // the same alignment mod 8, otherwise fall through to the next
1557 // code for aligned copy.
1558 // The compare above (count >= 11) guarantes 'count' >= 16 bytes.
1559 // Also jump over aligned copy after the copy with shift completed.
1560
1561 copy_16_bytes_forward_with_shift(from, to, count, 8, L_copy_2_bytes);
1562 }
1563
1564 // Both array are 8 bytes aligned, copy 16 bytes at a time
1565 __ and3(count, 3, G4); // Save
1566 __ srl(count, 2, count);
1567 generate_disjoint_long_copy_core(aligned);
1568 __ mov(G4, count); // restore
1569
1570 // copy 1 element at a time
1571 __ BIND(L_copy_2_bytes);
1572 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1573 __ delayed()->nop();
1574 __ align(OptoLoopAlignment);
1575 __ BIND(L_copy_2_bytes_loop);
1576 __ lduh(from, offset, O3);
1577 __ deccc(count);
1578 __ sth(O3, to, offset);
1579 __ brx(Assembler::notZero, false, Assembler::pt, L_copy_2_bytes_loop);
1580 __ delayed()->inc(offset, 2);
1581
1582 __ BIND(L_exit);
1583 // O3, O4 are used as temp registers
1584 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr, O3, O4);
1585 __ retl();
1586 __ delayed()->mov(G0, O0); // return 0
1587 return start;
1588 }
1589
1590 //
1591 // Generate stub for disjoint short fill. If "aligned" is true, the
1592 // "to" address is assumed to be heapword aligned.
1593 //
1594 // Arguments for generated stub:
1595 // to: O0
1596 // value: O1
1597 // count: O2 treated as signed
1598 //
1599 address generate_fill(BasicType t, bool aligned, const char* name) {
1600 __ align(CodeEntryAlignment);
1601 StubCodeMark mark(this, "StubRoutines", name);
1602 address start = __ pc();
1603
1604 const Register to = O0; // source array address
1605 const Register value = O1; // fill value
1606 const Register count = O2; // elements count
1607 // O3 is used as a temp register
1608
1609 assert_clean_int(count, O3); // Make sure 'count' is clean int.
1610
1611 Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte;
1612 Label L_fill_2_bytes, L_fill_elements, L_fill_32_bytes;
1613
1614 int shift = -1;
1615 switch (t) {
1616 case T_BYTE:
1617 shift = 2;
1618 break;
1619 case T_SHORT:
1620 shift = 1;
1621 break;
1622 case T_INT:
1623 shift = 0;
1624 break;
1625 default: ShouldNotReachHere();
1626 }
1627
1628 BLOCK_COMMENT("Entry:");
1629
1630 if (t == T_BYTE) {
1631 // Zero extend value
1632 __ and3(value, 0xff, value);
1633 __ sllx(value, 8, O3);
1634 __ or3(value, O3, value);
1635 }
1636 if (t == T_SHORT) {
1637 // Zero extend value
1638 __ sllx(value, 48, value);
1639 __ srlx(value, 48, value);
1640 }
1641 if (t == T_BYTE || t == T_SHORT) {
1642 __ sllx(value, 16, O3);
1643 __ or3(value, O3, value);
1644 }
1645
1646 __ cmp(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
1647 __ brx(Assembler::lessUnsigned, false, Assembler::pn, L_fill_elements); // use unsigned cmp
1648 __ delayed()->andcc(count, 1, G0);
1649
1650 if (!aligned && (t == T_BYTE || t == T_SHORT)) {
1651 // align source address at 4 bytes address boundary
1652 if (t == T_BYTE) {
1653 // One byte misalignment happens only for byte arrays
1654 __ andcc(to, 1, G0);
1655 __ br(Assembler::zero, false, Assembler::pt, L_skip_align1);
1656 __ delayed()->nop();
1657 __ stb(value, to, 0);
1658 __ inc(to, 1);
1659 __ dec(count, 1);
1660 __ BIND(L_skip_align1);
1661 }
1662 // Two bytes misalignment happens only for byte and short (char) arrays
1663 __ andcc(to, 2, G0);
1664 __ br(Assembler::zero, false, Assembler::pt, L_skip_align2);
1665 __ delayed()->nop();
1666 __ sth(value, to, 0);
1667 __ inc(to, 2);
1668 __ dec(count, 1 << (shift - 1));
1669 __ BIND(L_skip_align2);
1670 }
1671 #ifdef _LP64
1672 if (!aligned) {
1673 #endif
1674 // align to 8 bytes, we know we are 4 byte aligned to start
1675 __ andcc(to, 7, G0);
1676 __ br(Assembler::zero, false, Assembler::pt, L_fill_32_bytes);
1677 __ delayed()->nop();
1678 __ stw(value, to, 0);
1679 __ inc(to, 4);
1680 __ dec(count, 1 << shift);
1681 __ BIND(L_fill_32_bytes);
1682 #ifdef _LP64
1683 }
1684 #endif
1685
1686 if (t == T_INT) {
1687 // Zero extend value
1688 __ srl(value, 0, value);
1689 }
1690 if (t == T_BYTE || t == T_SHORT || t == T_INT) {
1691 __ sllx(value, 32, O3);
1692 __ or3(value, O3, value);
1693 }
1694
1695 Label L_check_fill_8_bytes;
1696 // Fill 32-byte chunks
1697 __ subcc(count, 8 << shift, count);
1698 __ brx(Assembler::less, false, Assembler::pt, L_check_fill_8_bytes);
1699 __ delayed()->nop();
1700
1701 Label L_fill_32_bytes_loop, L_fill_4_bytes;
1702 __ align(16);
1703 __ BIND(L_fill_32_bytes_loop);
1704
1705 __ stx(value, to, 0);
1706 __ stx(value, to, 8);
1707 __ stx(value, to, 16);
1708 __ stx(value, to, 24);
1709
1710 __ subcc(count, 8 << shift, count);
1711 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_fill_32_bytes_loop);
1712 __ delayed()->add(to, 32, to);
1713
1714 __ BIND(L_check_fill_8_bytes);
1715 __ addcc(count, 8 << shift, count);
1716 __ brx(Assembler::zero, false, Assembler::pn, L_exit);
1717 __ delayed()->subcc(count, 1 << (shift + 1), count);
1718 __ brx(Assembler::less, false, Assembler::pn, L_fill_4_bytes);
1719 __ delayed()->andcc(count, 1<<shift, G0);
1720
1721 //
1722 // length is too short, just fill 8 bytes at a time
1723 //
1724 Label L_fill_8_bytes_loop;
1725 __ BIND(L_fill_8_bytes_loop);
1726 __ stx(value, to, 0);
1727 __ subcc(count, 1 << (shift + 1), count);
1728 __ brx(Assembler::greaterEqual, false, Assembler::pn, L_fill_8_bytes_loop);
1729 __ delayed()->add(to, 8, to);
1730
1731 // fill trailing 4 bytes
1732 __ andcc(count, 1<<shift, G0); // in delay slot of branches
1733 if (t == T_INT) {
1734 __ BIND(L_fill_elements);
1735 }
1736 __ BIND(L_fill_4_bytes);
1737 __ brx(Assembler::zero, false, Assembler::pt, L_fill_2_bytes);
1738 if (t == T_BYTE || t == T_SHORT) {
1739 __ delayed()->andcc(count, 1<<(shift-1), G0);
1740 } else {
1741 __ delayed()->nop();
1742 }
1743 __ stw(value, to, 0);
1744 if (t == T_BYTE || t == T_SHORT) {
1745 __ inc(to, 4);
1746 // fill trailing 2 bytes
1747 __ andcc(count, 1<<(shift-1), G0); // in delay slot of branches
1748 __ BIND(L_fill_2_bytes);
1749 __ brx(Assembler::zero, false, Assembler::pt, L_fill_byte);
1750 __ delayed()->andcc(count, 1, count);
1751 __ sth(value, to, 0);
1752 if (t == T_BYTE) {
1753 __ inc(to, 2);
1754 // fill trailing byte
1755 __ andcc(count, 1, count); // in delay slot of branches
1756 __ BIND(L_fill_byte);
1757 __ brx(Assembler::zero, false, Assembler::pt, L_exit);
1758 __ delayed()->nop();
1759 __ stb(value, to, 0);
1760 } else {
1761 __ BIND(L_fill_byte);
1762 }
1763 } else {
1764 __ BIND(L_fill_2_bytes);
1765 }
1766 __ BIND(L_exit);
1767 __ retl();
1768 __ delayed()->nop();
1769
1770 // Handle copies less than 8 bytes. Int is handled elsewhere.
1771 if (t == T_BYTE) {
1772 __ BIND(L_fill_elements);
1773 Label L_fill_2, L_fill_4;
1774 // in delay slot __ andcc(count, 1, G0);
1775 __ brx(Assembler::zero, false, Assembler::pt, L_fill_2);
1776 __ delayed()->andcc(count, 2, G0);
1777 __ stb(value, to, 0);
1778 __ inc(to, 1);
1779 __ BIND(L_fill_2);
1780 __ brx(Assembler::zero, false, Assembler::pt, L_fill_4);
1781 __ delayed()->andcc(count, 4, G0);
1782 __ stb(value, to, 0);
1783 __ stb(value, to, 1);
1784 __ inc(to, 2);
1785 __ BIND(L_fill_4);
1786 __ brx(Assembler::zero, false, Assembler::pt, L_exit);
1787 __ delayed()->nop();
1788 __ stb(value, to, 0);
1789 __ stb(value, to, 1);
1790 __ stb(value, to, 2);
1791 __ retl();
1792 __ delayed()->stb(value, to, 3);
1793 }
1794
1795 if (t == T_SHORT) {
1796 Label L_fill_2;
1797 __ BIND(L_fill_elements);
1798 // in delay slot __ andcc(count, 1, G0);
1799 __ brx(Assembler::zero, false, Assembler::pt, L_fill_2);
1800 __ delayed()->andcc(count, 2, G0);
1801 __ sth(value, to, 0);
1802 __ inc(to, 2);
1803 __ BIND(L_fill_2);
1804 __ brx(Assembler::zero, false, Assembler::pt, L_exit);
1805 __ delayed()->nop();
1806 __ sth(value, to, 0);
1807 __ retl();
1808 __ delayed()->sth(value, to, 2);
1809 }
1810 return start;
1811 }
1812
1813 //
1814 // Generate stub for conjoint short copy. If "aligned" is true, the
1815 // "from" and "to" addresses are assumed to be heapword aligned.
1816 //
1817 // Arguments for generated stub:
1818 // from: O0
1819 // to: O1
1820 // count: O2 treated as signed
1821 //
1822 address generate_conjoint_short_copy(bool aligned, const char * name) {
1823 // Do reverse copy.
1824
1825 __ align(CodeEntryAlignment);
1826 StubCodeMark mark(this, "StubRoutines", name);
1827 address start = __ pc();
1828 address nooverlap_target = aligned ?
1829 StubRoutines::arrayof_jshort_disjoint_arraycopy() :
1830 disjoint_short_copy_entry;
1831
1832 Label L_skip_alignment, L_skip_alignment2, L_aligned_copy;
1833 Label L_copy_2_bytes, L_copy_2_bytes_loop, L_exit;
1834
1835 const Register from = O0; // source array address
1836 const Register to = O1; // destination array address
1837 const Register count = O2; // elements count
1838 const Register end_from = from; // source array end address
1839 const Register end_to = to; // destination array end address
1840
1841 const Register byte_count = O3; // bytes count to copy
1842
1843 assert_clean_int(count, O3); // Make sure 'count' is clean int.
1844
1845 if (!aligned) short_copy_entry = __ pc();
1846 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1847 if (!aligned) BLOCK_COMMENT("Entry:");
1848
1849 array_overlap_test(nooverlap_target, 1);
1850
1851 __ sllx(count, LogBytesPerShort, byte_count);
1852 __ add(to, byte_count, end_to); // offset after last copied element
1853
1854 // for short arrays, just do single element copy
1855 __ cmp(count, 11); // 8 + 3 (22 bytes)
1856 __ brx(Assembler::less, false, Assembler::pn, L_copy_2_bytes);
1857 __ delayed()->add(from, byte_count, end_from);
1858
1859 {
1860 // Align end of arrays since they could be not aligned even
1861 // when arrays itself are aligned.
1862
1863 // copy 1 element if necessary to align 'end_to' on an 4 bytes
1864 __ andcc(end_to, 3, G0);
1865 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1866 __ delayed()->lduh(end_from, -2, O3);
1867 __ dec(end_from, 2);
1868 __ dec(end_to, 2);
1869 __ dec(count);
1870 __ sth(O3, end_to, 0);
1871 __ BIND(L_skip_alignment);
1872
1873 // copy 2 elements to align 'end_to' on an 8 byte boundary
1874 __ andcc(end_to, 7, G0);
1875 __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment2);
1876 __ delayed()->lduh(end_from, -2, O3);
1877 __ dec(count, 2);
1878 __ lduh(end_from, -4, O4);
1879 __ dec(end_from, 4);
1880 __ dec(end_to, 4);
1881 __ sth(O3, end_to, 2);
1882 __ sth(O4, end_to, 0);
1883 __ BIND(L_skip_alignment2);
1884 }
1885 #ifdef _LP64
1886 if (aligned) {
1887 // Both arrays are aligned to 8-bytes in 64-bits VM.
1888 // The 'count' is decremented in copy_16_bytes_backward_with_shift()
1889 // in unaligned case.
1890 __ dec(count, 8);
1891 } else
1892 #endif
1893 {
1894 // Copy with shift 16 bytes per iteration if arrays do not have
1895 // the same alignment mod 8, otherwise jump to the next
1896 // code for aligned copy (and substracting 8 from 'count' before jump).
1897 // The compare above (count >= 11) guarantes 'count' >= 16 bytes.
1898 // Also jump over aligned copy after the copy with shift completed.
1899
1900 copy_16_bytes_backward_with_shift(end_from, end_to, count, 8,
1901 L_aligned_copy, L_copy_2_bytes);
1902 }
1903 // copy 4 elements (16 bytes) at a time
1904 __ align(OptoLoopAlignment);
1905 __ BIND(L_aligned_copy);
1906 __ dec(end_from, 16);
1907 __ ldx(end_from, 8, O3);
1908 __ ldx(end_from, 0, O4);
1909 __ dec(end_to, 16);
1910 __ deccc(count, 8);
1911 __ stx(O3, end_to, 8);
1912 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy);
1913 __ delayed()->stx(O4, end_to, 0);
1914 __ inc(count, 8);
1915
1916 // copy 1 element (2 bytes) at a time
1917 __ BIND(L_copy_2_bytes);
1918 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
1919 __ delayed()->nop();
1920 __ BIND(L_copy_2_bytes_loop);
1921 __ dec(end_from, 2);
1922 __ dec(end_to, 2);
1923 __ lduh(end_from, 0, O4);
1924 __ deccc(count);
1925 __ brx(Assembler::greater, false, Assembler::pt, L_copy_2_bytes_loop);
1926 __ delayed()->sth(O4, end_to, 0);
1927
1928 __ BIND(L_exit);
1929 // O3, O4 are used as temp registers
1930 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr, O3, O4);
1931 __ retl();
1932 __ delayed()->mov(G0, O0); // return 0
1933 return start;
1934 }
1935
1936 //
1937 // Generate core code for disjoint int copy (and oop copy on 32-bit).
1938 // If "aligned" is true, the "from" and "to" addresses are assumed
1939 // to be heapword aligned.
1940 //
1941 // Arguments:
1942 // from: O0
1943 // to: O1
1944 // count: O2 treated as signed
1945 //
1946 void generate_disjoint_int_copy_core(bool aligned) {
1947
1948 Label L_skip_alignment, L_aligned_copy;
1949 Label L_copy_16_bytes, L_copy_4_bytes, L_copy_4_bytes_loop, L_exit;
1950
1951 const Register from = O0; // source array address
1952 const Register to = O1; // destination array address
1953 const Register count = O2; // elements count
1954 const Register offset = O5; // offset from start of arrays
1955 // O3, O4, G3, G4 are used as temp registers
1956
1957 // 'aligned' == true when it is known statically during compilation
1958 // of this arraycopy call site that both 'from' and 'to' addresses
1959 // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()).
1960 //
1961 // Aligned arrays have 4 bytes alignment in 32-bits VM
1962 // and 8 bytes - in 64-bits VM.
1963 //
1964 #ifdef _LP64
1965 if (!aligned)
1966 #endif
1967 {
1968 // The next check could be put under 'ifndef' since the code in
1969 // generate_disjoint_long_copy_core() has own checks and set 'offset'.
1970
1971 // for short arrays, just do single element copy
1972 __ cmp(count, 5); // 4 + 1 (20 bytes)
1973 __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_4_bytes);
1974 __ delayed()->mov(G0, offset);
1975
1976 // copy 1 element to align 'to' on an 8 byte boundary
1977 __ andcc(to, 7, G0);
1978 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
1979 __ delayed()->ld(from, 0, O3);
1980 __ inc(from, 4);
1981 __ inc(to, 4);
1982 __ dec(count);
1983 __ st(O3, to, -4);
1984 __ BIND(L_skip_alignment);
1985
1986 // if arrays have same alignment mod 8, do 4 elements copy
1987 __ andcc(from, 7, G0);
1988 __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy);
1989 __ delayed()->ld(from, 0, O3);
1990
1991 //
1992 // Load 2 aligned 8-bytes chunks and use one from previous iteration
1993 // to form 2 aligned 8-bytes chunks to store.
1994 //
1995 // copy_16_bytes_forward_with_shift() is not used here since this
1996 // code is more optimal.
1997
1998 // copy with shift 4 elements (16 bytes) at a time
1999 __ dec(count, 4); // The cmp at the beginning guaranty count >= 4
2000
2001 __ align(OptoLoopAlignment);
2002 __ BIND(L_copy_16_bytes);
2003 __ ldx(from, 4, O4);
2004 __ deccc(count, 4); // Can we do next iteration after this one?
2005 __ ldx(from, 12, G4);
2006 __ inc(to, 16);
2007 __ inc(from, 16);
2008 __ sllx(O3, 32, O3);
2009 __ srlx(O4, 32, G3);
2010 __ bset(G3, O3);
2011 __ stx(O3, to, -16);
2012 __ sllx(O4, 32, O4);
2013 __ srlx(G4, 32, G3);
2014 __ bset(G3, O4);
2015 __ stx(O4, to, -8);
2016 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes);
2017 __ delayed()->mov(G4, O3);
2018
2019 __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes);
2020 __ delayed()->inc(count, 4); // restore 'count'
2021
2022 __ BIND(L_aligned_copy);
2023 }
2024 // copy 4 elements (16 bytes) at a time
2025 __ and3(count, 1, G4); // Save
2026 __ srl(count, 1, count);
2027 generate_disjoint_long_copy_core(aligned);
2028 __ mov(G4, count); // Restore
2029
2030 // copy 1 element at a time
2031 __ BIND(L_copy_4_bytes);
2032 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
2033 __ delayed()->nop();
2034 __ BIND(L_copy_4_bytes_loop);
2035 __ ld(from, offset, O3);
2036 __ deccc(count);
2037 __ st(O3, to, offset);
2038 __ brx(Assembler::notZero, false, Assembler::pt, L_copy_4_bytes_loop);
2039 __ delayed()->inc(offset, 4);
2040 __ BIND(L_exit);
2041 }
2042
2043 //
2044 // Generate stub for disjoint int copy. If "aligned" is true, the
2045 // "from" and "to" addresses are assumed to be heapword aligned.
2046 //
2047 // Arguments for generated stub:
2048 // from: O0
2049 // to: O1
2050 // count: O2 treated as signed
2051 //
2052 address generate_disjoint_int_copy(bool aligned, const char * name) {
2053 __ align(CodeEntryAlignment);
2054 StubCodeMark mark(this, "StubRoutines", name);
2055 address start = __ pc();
2056
2057 const Register count = O2;
2058 assert_clean_int(count, O3); // Make sure 'count' is clean int.
2059
2060 if (!aligned) disjoint_int_copy_entry = __ pc();
2061 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2062 if (!aligned) BLOCK_COMMENT("Entry:");
2063
2064 generate_disjoint_int_copy_core(aligned);
2065
2066 // O3, O4 are used as temp registers
2067 inc_counter_np(SharedRuntime::_jint_array_copy_ctr, O3, O4);
2068 __ retl();
2069 __ delayed()->mov(G0, O0); // return 0
2070 return start;
2071 }
2072
2073 //
2074 // Generate core code for conjoint int copy (and oop copy on 32-bit).
2075 // If "aligned" is true, the "from" and "to" addresses are assumed
2076 // to be heapword aligned.
2077 //
2078 // Arguments:
2079 // from: O0
2080 // to: O1
2081 // count: O2 treated as signed
2082 //
2083 void generate_conjoint_int_copy_core(bool aligned) {
2084 // Do reverse copy.
2085
2086 Label L_skip_alignment, L_aligned_copy;
2087 Label L_copy_16_bytes, L_copy_4_bytes, L_copy_4_bytes_loop, L_exit;
2088
2089 const Register from = O0; // source array address
2090 const Register to = O1; // destination array address
2091 const Register count = O2; // elements count
2092 const Register end_from = from; // source array end address
2093 const Register end_to = to; // destination array end address
2094 // O3, O4, O5, G3 are used as temp registers
2095
2096 const Register byte_count = O3; // bytes count to copy
2097
2098 __ sllx(count, LogBytesPerInt, byte_count);
2099 __ add(to, byte_count, end_to); // offset after last copied element
2100
2101 __ cmp(count, 5); // for short arrays, just do single element copy
2102 __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_4_bytes);
2103 __ delayed()->add(from, byte_count, end_from);
2104
2105 // copy 1 element to align 'to' on an 8 byte boundary
2106 __ andcc(end_to, 7, G0);
2107 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment);
2108 __ delayed()->nop();
2109 __ dec(count);
2110 __ dec(end_from, 4);
2111 __ dec(end_to, 4);
2112 __ ld(end_from, 0, O4);
2113 __ st(O4, end_to, 0);
2114 __ BIND(L_skip_alignment);
2115
2116 // Check if 'end_from' and 'end_to' has the same alignment.
2117 __ andcc(end_from, 7, G0);
2118 __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy);
2119 __ delayed()->dec(count, 4); // The cmp at the start guaranty cnt >= 4
2120
2121 // copy with shift 4 elements (16 bytes) at a time
2122 //
2123 // Load 2 aligned 8-bytes chunks and use one from previous iteration
2124 // to form 2 aligned 8-bytes chunks to store.
2125 //
2126 __ ldx(end_from, -4, O3);
2127 __ align(OptoLoopAlignment);
2128 __ BIND(L_copy_16_bytes);
2129 __ ldx(end_from, -12, O4);
2130 __ deccc(count, 4);
2131 __ ldx(end_from, -20, O5);
2132 __ dec(end_to, 16);
2133 __ dec(end_from, 16);
2134 __ srlx(O3, 32, O3);
2135 __ sllx(O4, 32, G3);
2136 __ bset(G3, O3);
2137 __ stx(O3, end_to, 8);
2138 __ srlx(O4, 32, O4);
2139 __ sllx(O5, 32, G3);
2140 __ bset(O4, G3);
2141 __ stx(G3, end_to, 0);
2142 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes);
2143 __ delayed()->mov(O5, O3);
2144
2145 __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes);
2146 __ delayed()->inc(count, 4);
2147
2148 // copy 4 elements (16 bytes) at a time
2149 __ align(OptoLoopAlignment);
2150 __ BIND(L_aligned_copy);
2151 __ dec(end_from, 16);
2152 __ ldx(end_from, 8, O3);
2153 __ ldx(end_from, 0, O4);
2154 __ dec(end_to, 16);
2155 __ deccc(count, 4);
2156 __ stx(O3, end_to, 8);
2157 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy);
2158 __ delayed()->stx(O4, end_to, 0);
2159 __ inc(count, 4);
2160
2161 // copy 1 element (4 bytes) at a time
2162 __ BIND(L_copy_4_bytes);
2163 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
2164 __ delayed()->nop();
2165 __ BIND(L_copy_4_bytes_loop);
2166 __ dec(end_from, 4);
2167 __ dec(end_to, 4);
2168 __ ld(end_from, 0, O4);
2169 __ deccc(count);
2170 __ brx(Assembler::greater, false, Assembler::pt, L_copy_4_bytes_loop);
2171 __ delayed()->st(O4, end_to, 0);
2172 __ BIND(L_exit);
2173 }
2174
2175 //
2176 // Generate stub for conjoint int copy. If "aligned" is true, the
2177 // "from" and "to" addresses are assumed to be heapword aligned.
2178 //
2179 // Arguments for generated stub:
2180 // from: O0
2181 // to: O1
2182 // count: O2 treated as signed
2183 //
2184 address generate_conjoint_int_copy(bool aligned, const char * name) {
2185 __ align(CodeEntryAlignment);
2186 StubCodeMark mark(this, "StubRoutines", name);
2187 address start = __ pc();
2188
2189 address nooverlap_target = aligned ?
2190 StubRoutines::arrayof_jint_disjoint_arraycopy() :
2191 disjoint_int_copy_entry;
2192
2193 assert_clean_int(O2, O3); // Make sure 'count' is clean int.
2194
2195 if (!aligned) int_copy_entry = __ pc();
2196 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2197 if (!aligned) BLOCK_COMMENT("Entry:");
2198
2199 array_overlap_test(nooverlap_target, 2);
2200
2201 generate_conjoint_int_copy_core(aligned);
2202
2203 // O3, O4 are used as temp registers
2204 inc_counter_np(SharedRuntime::_jint_array_copy_ctr, O3, O4);
2205 __ retl();
2206 __ delayed()->mov(G0, O0); // return 0
2207 return start;
2208 }
2209
2210 //
2211 // Generate core code for disjoint long copy (and oop copy on 64-bit).
2212 // "aligned" is ignored, because we must make the stronger
2213 // assumption that both addresses are always 64-bit aligned.
2214 //
2215 // Arguments:
2216 // from: O0
2217 // to: O1
2218 // count: O2 treated as signed
2219 //
2220 // count -= 2;
2221 // if ( count >= 0 ) { // >= 2 elements
2222 // if ( count > 6) { // >= 8 elements
2223 // count -= 6; // original count - 8
2224 // do {
2225 // copy_8_elements;
2226 // count -= 8;
2227 // } while ( count >= 0 );
2228 // count += 6;
2229 // }
2230 // if ( count >= 0 ) { // >= 2 elements
2231 // do {
2232 // copy_2_elements;
2233 // } while ( (count=count-2) >= 0 );
2234 // }
2235 // }
2236 // count += 2;
2237 // if ( count != 0 ) { // 1 element left
2238 // copy_1_element;
2239 // }
2240 //
2241 void generate_disjoint_long_copy_core(bool aligned) {
2242 Label L_copy_8_bytes, L_copy_16_bytes, L_exit;
2243 const Register from = O0; // source array address
2244 const Register to = O1; // destination array address
2245 const Register count = O2; // elements count
2246 const Register offset0 = O4; // element offset
2247 const Register offset8 = O5; // next element offset
2248
2249 __ deccc(count, 2);
2250 __ mov(G0, offset0); // offset from start of arrays (0)
2251 __ brx(Assembler::negative, false, Assembler::pn, L_copy_8_bytes );
2252 __ delayed()->add(offset0, 8, offset8);
2253
2254 // Copy by 64 bytes chunks
2255 Label L_copy_64_bytes;
2256 const Register from64 = O3; // source address
2257 const Register to64 = G3; // destination address
2258 __ subcc(count, 6, O3);
2259 __ brx(Assembler::negative, false, Assembler::pt, L_copy_16_bytes );
2260 __ delayed()->mov(to, to64);
2261 // Now we can use O4(offset0), O5(offset8) as temps
2262 __ mov(O3, count);
2263 __ mov(from, from64);
2264
2265 __ align(OptoLoopAlignment);
2266 __ BIND(L_copy_64_bytes);
2267 for( int off = 0; off < 64; off += 16 ) {
2268 __ ldx(from64, off+0, O4);
2269 __ ldx(from64, off+8, O5);
2270 __ stx(O4, to64, off+0);
2271 __ stx(O5, to64, off+8);
2272 }
2273 __ deccc(count, 8);
2274 __ inc(from64, 64);
2275 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_64_bytes);
2276 __ delayed()->inc(to64, 64);
2277
2278 // Restore O4(offset0), O5(offset8)
2279 __ sub(from64, from, offset0);
2280 __ inccc(count, 6);
2281 __ brx(Assembler::negative, false, Assembler::pn, L_copy_8_bytes );
2282 __ delayed()->add(offset0, 8, offset8);
2283
2284 // Copy by 16 bytes chunks
2285 __ align(OptoLoopAlignment);
2286 __ BIND(L_copy_16_bytes);
2287 __ ldx(from, offset0, O3);
2288 __ ldx(from, offset8, G3);
2289 __ deccc(count, 2);
2290 __ stx(O3, to, offset0);
2291 __ inc(offset0, 16);
2292 __ stx(G3, to, offset8);
2293 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes);
2294 __ delayed()->inc(offset8, 16);
2295
2296 // Copy last 8 bytes
2297 __ BIND(L_copy_8_bytes);
2298 __ inccc(count, 2);
2299 __ brx(Assembler::zero, true, Assembler::pn, L_exit );
2300 __ delayed()->mov(offset0, offset8); // Set O5 used by other stubs
2301 __ ldx(from, offset0, O3);
2302 __ stx(O3, to, offset0);
2303 __ BIND(L_exit);
2304 }
2305
2306 //
2307 // Generate stub for disjoint long copy.
2308 // "aligned" is ignored, because we must make the stronger
2309 // assumption that both addresses are always 64-bit aligned.
2310 //
2311 // Arguments for generated stub:
2312 // from: O0
2313 // to: O1
2314 // count: O2 treated as signed
2315 //
2316 address generate_disjoint_long_copy(bool aligned, const char * name) {
2317 __ align(CodeEntryAlignment);
2318 StubCodeMark mark(this, "StubRoutines", name);
2319 address start = __ pc();
2320
2321 assert_clean_int(O2, O3); // Make sure 'count' is clean int.
2322
2323 if (!aligned) disjoint_long_copy_entry = __ pc();
2324 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2325 if (!aligned) BLOCK_COMMENT("Entry:");
2326
2327 generate_disjoint_long_copy_core(aligned);
2328
2329 // O3, O4 are used as temp registers
2330 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr, O3, O4);
2331 __ retl();
2332 __ delayed()->mov(G0, O0); // return 0
2333 return start;
2334 }
2335
2336 //
2337 // Generate core code for conjoint long copy (and oop copy on 64-bit).
2338 // "aligned" is ignored, because we must make the stronger
2339 // assumption that both addresses are always 64-bit aligned.
2340 //
2341 // Arguments:
2342 // from: O0
2343 // to: O1
2344 // count: O2 treated as signed
2345 //
2346 void generate_conjoint_long_copy_core(bool aligned) {
2347 // Do reverse copy.
2348 Label L_copy_8_bytes, L_copy_16_bytes, L_exit;
2349 const Register from = O0; // source array address
2350 const Register to = O1; // destination array address
2351 const Register count = O2; // elements count
2352 const Register offset8 = O4; // element offset
2353 const Register offset0 = O5; // previous element offset
2354
2355 __ subcc(count, 1, count);
2356 __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_8_bytes );
2357 __ delayed()->sllx(count, LogBytesPerLong, offset8);
2358 __ sub(offset8, 8, offset0);
2359 __ align(OptoLoopAlignment);
2360 __ BIND(L_copy_16_bytes);
2361 __ ldx(from, offset8, O2);
2362 __ ldx(from, offset0, O3);
2363 __ stx(O2, to, offset8);
2364 __ deccc(offset8, 16); // use offset8 as counter
2365 __ stx(O3, to, offset0);
2366 __ brx(Assembler::greater, false, Assembler::pt, L_copy_16_bytes);
2367 __ delayed()->dec(offset0, 16);
2368
2369 __ BIND(L_copy_8_bytes);
2370 __ brx(Assembler::negative, false, Assembler::pn, L_exit );
2371 __ delayed()->nop();
2372 __ ldx(from, 0, O3);
2373 __ stx(O3, to, 0);
2374 __ BIND(L_exit);
2375 }
2376
2377 // Generate stub for conjoint long copy.
2378 // "aligned" is ignored, because we must make the stronger
2379 // assumption that both addresses are always 64-bit aligned.
2380 //
2381 // Arguments for generated stub:
2382 // from: O0
2383 // to: O1
2384 // count: O2 treated as signed
2385 //
2386 address generate_conjoint_long_copy(bool aligned, const char * name) {
2387 __ align(CodeEntryAlignment);
2388 StubCodeMark mark(this, "StubRoutines", name);
2389 address start = __ pc();
2390
2391 assert(!aligned, "usage");
2392 address nooverlap_target = disjoint_long_copy_entry;
2393
2394 assert_clean_int(O2, O3); // Make sure 'count' is clean int.
2395
2396 if (!aligned) long_copy_entry = __ pc();
2397 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2398 if (!aligned) BLOCK_COMMENT("Entry:");
2399
2400 array_overlap_test(nooverlap_target, 3);
2401
2402 generate_conjoint_long_copy_core(aligned);
2403
2404 // O3, O4 are used as temp registers
2405 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr, O3, O4);
2406 __ retl();
2407 __ delayed()->mov(G0, O0); // return 0
2408 return start;
2409 }
2410
2411 // Generate stub for disjoint oop copy. If "aligned" is true, the
2412 // "from" and "to" addresses are assumed to be heapword aligned.
2413 //
2414 // Arguments for generated stub:
2415 // from: O0
2416 // to: O1
2417 // count: O2 treated as signed
2418 //
2419 address generate_disjoint_oop_copy(bool aligned, const char * name) {
2420
2421 const Register from = O0; // source array address
2422 const Register to = O1; // destination array address
2423 const Register count = O2; // elements count
2424
2425 __ align(CodeEntryAlignment);
2426 StubCodeMark mark(this, "StubRoutines", name);
2427 address start = __ pc();
2428
2429 assert_clean_int(count, O3); // Make sure 'count' is clean int.
2430
2431 if (!aligned) disjoint_oop_copy_entry = __ pc();
2432 // caller can pass a 64-bit byte count here
2433 if (!aligned) BLOCK_COMMENT("Entry:");
2434
2435 // save arguments for barrier generation
2436 __ mov(to, G1);
2437 __ mov(count, G5);
2438 gen_write_ref_array_pre_barrier(G1, G5);
2439 #ifdef _LP64
2440 assert_clean_int(count, O3); // Make sure 'count' is clean int.
2441 if (UseCompressedOops) {
2442 generate_disjoint_int_copy_core(aligned);
2443 } else {
2444 generate_disjoint_long_copy_core(aligned);
2445 }
2446 #else
2447 generate_disjoint_int_copy_core(aligned);
2448 #endif
2449 // O0 is used as temp register
2450 gen_write_ref_array_post_barrier(G1, G5, O0);
2451
2452 // O3, O4 are used as temp registers
2453 inc_counter_np(SharedRuntime::_oop_array_copy_ctr, O3, O4);
2454 __ retl();
2455 __ delayed()->mov(G0, O0); // return 0
2456 return start;
2457 }
2458
2459 // Generate stub for conjoint oop copy. If "aligned" is true, the
2460 // "from" and "to" addresses are assumed to be heapword aligned.
2461 //
2462 // Arguments for generated stub:
2463 // from: O0
2464 // to: O1
2465 // count: O2 treated as signed
2466 //
2467 address generate_conjoint_oop_copy(bool aligned, const char * name) {
2468
2469 const Register from = O0; // source array address
2470 const Register to = O1; // destination array address
2471 const Register count = O2; // elements count
2472
2473 __ align(CodeEntryAlignment);
2474 StubCodeMark mark(this, "StubRoutines", name);
2475 address start = __ pc();
2476
2477 assert_clean_int(count, O3); // Make sure 'count' is clean int.
2478
2479 if (!aligned) oop_copy_entry = __ pc();
2480 // caller can pass a 64-bit byte count here
2481 if (!aligned) BLOCK_COMMENT("Entry:");
2482
2483 // save arguments for barrier generation
2484 __ mov(to, G1);
2485 __ mov(count, G5);
2486
2487 gen_write_ref_array_pre_barrier(G1, G5);
2488
2489 address nooverlap_target = aligned ?
2490 StubRoutines::arrayof_oop_disjoint_arraycopy() :
2491 disjoint_oop_copy_entry;
2492
2493 array_overlap_test(nooverlap_target, LogBytesPerHeapOop);
2494
2495 #ifdef _LP64
2496 if (UseCompressedOops) {
2497 generate_conjoint_int_copy_core(aligned);
2498 } else {
2499 generate_conjoint_long_copy_core(aligned);
2500 }
2501 #else
2502 generate_conjoint_int_copy_core(aligned);
2503 #endif
2504
2505 // O0 is used as temp register
2506 gen_write_ref_array_post_barrier(G1, G5, O0);
2507
2508 // O3, O4 are used as temp registers
2509 inc_counter_np(SharedRuntime::_oop_array_copy_ctr, O3, O4);
2510 __ retl();
2511 __ delayed()->mov(G0, O0); // return 0
2512 return start;
2513 }
2514
2515
2516 // Helper for generating a dynamic type check.
2517 // Smashes only the given temp registers.
2518 void generate_type_check(Register sub_klass,
2519 Register super_check_offset,
2520 Register super_klass,
2521 Register temp,
2522 Label& L_success) {
2523 assert_different_registers(sub_klass, super_check_offset, super_klass, temp);
2524
2525 BLOCK_COMMENT("type_check:");
2526
2527 Label L_miss, L_pop_to_miss;
2528
2529 assert_clean_int(super_check_offset, temp);
2530
2531 __ check_klass_subtype_fast_path(sub_klass, super_klass, temp, noreg,
2532 &L_success, &L_miss, NULL,
2533 super_check_offset);
2534
2535 BLOCK_COMMENT("type_check_slow_path:");
2536 __ save_frame(0);
2537 __ check_klass_subtype_slow_path(sub_klass->after_save(),
2538 super_klass->after_save(),
2539 L0, L1, L2, L4,
2540 NULL, &L_pop_to_miss);
2541 __ ba(false, L_success);
2542 __ delayed()->restore();
2543
2544 __ bind(L_pop_to_miss);
2545 __ restore();
2546
2547 // Fall through on failure!
2548 __ BIND(L_miss);
2549 }
2550
2551
2552 // Generate stub for checked oop copy.
2553 //
2554 // Arguments for generated stub:
2555 // from: O0
2556 // to: O1
2557 // count: O2 treated as signed
2558 // ckoff: O3 (super_check_offset)
2559 // ckval: O4 (super_klass)
2560 // ret: O0 zero for success; (-1^K) where K is partial transfer count
2561 //
2562 address generate_checkcast_copy(const char* name) {
2563
2564 const Register O0_from = O0; // source array address
2565 const Register O1_to = O1; // destination array address
2566 const Register O2_count = O2; // elements count
2567 const Register O3_ckoff = O3; // super_check_offset
2568 const Register O4_ckval = O4; // super_klass
2569
2570 const Register O5_offset = O5; // loop var, with stride wordSize
2571 const Register G1_remain = G1; // loop var, with stride -1
2572 const Register G3_oop = G3; // actual oop copied
2573 const Register G4_klass = G4; // oop._klass
2574 const Register G5_super = G5; // oop._klass._primary_supers[ckval]
2575
2576 __ align(CodeEntryAlignment);
2577 StubCodeMark mark(this, "StubRoutines", name);
2578 address start = __ pc();
2579
2580 gen_write_ref_array_pre_barrier(O1, O2);
2581
2582 #ifdef ASSERT
2583 // We sometimes save a frame (see generate_type_check below).
2584 // If this will cause trouble, let's fail now instead of later.
2585 __ save_frame(0);
2586 __ restore();
2587 #endif
2588
2589 assert_clean_int(O2_count, G1); // Make sure 'count' is clean int.
2590
2591 #ifdef ASSERT
2592 // caller guarantees that the arrays really are different
2593 // otherwise, we would have to make conjoint checks
2594 { Label L;
2595 __ mov(O3, G1); // spill: overlap test smashes O3
2596 __ mov(O4, G4); // spill: overlap test smashes O4
2597 array_overlap_test(L, LogBytesPerHeapOop);
2598 __ stop("checkcast_copy within a single array");
2599 __ bind(L);
2600 __ mov(G1, O3);
2601 __ mov(G4, O4);
2602 }
2603 #endif //ASSERT
2604
2605 checkcast_copy_entry = __ pc();
2606 // caller can pass a 64-bit byte count here (from generic stub)
2607 BLOCK_COMMENT("Entry:");
2608
2609 Label load_element, store_element, do_card_marks, fail, done;
2610 __ addcc(O2_count, 0, G1_remain); // initialize loop index, and test it
2611 __ brx(Assembler::notZero, false, Assembler::pt, load_element);
2612 __ delayed()->mov(G0, O5_offset); // offset from start of arrays
2613
2614 // Empty array: Nothing to do.
2615 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4);
2616 __ retl();
2617 __ delayed()->set(0, O0); // return 0 on (trivial) success
2618
2619 // ======== begin loop ========
2620 // (Loop is rotated; its entry is load_element.)
2621 // Loop variables:
2622 // (O5 = 0; ; O5 += wordSize) --- offset from src, dest arrays
2623 // (O2 = len; O2 != 0; O2--) --- number of oops *remaining*
2624 // G3, G4, G5 --- current oop, oop.klass, oop.klass.super
2625 __ align(OptoLoopAlignment);
2626
2627 __ BIND(store_element);
2628 __ deccc(G1_remain); // decrement the count
2629 __ store_heap_oop(G3_oop, O1_to, O5_offset); // store the oop
2630 __ inc(O5_offset, heapOopSize); // step to next offset
2631 __ brx(Assembler::zero, true, Assembler::pt, do_card_marks);
2632 __ delayed()->set(0, O0); // return -1 on success
2633
2634 // ======== loop entry is here ========
2635 __ BIND(load_element);
2636 __ load_heap_oop(O0_from, O5_offset, G3_oop); // load the oop
2637 __ br_null(G3_oop, true, Assembler::pt, store_element);
2638 __ delayed()->nop();
2639
2640 __ load_klass(G3_oop, G4_klass); // query the object klass
2641
2642 generate_type_check(G4_klass, O3_ckoff, O4_ckval, G5_super,
2643 // branch to this on success:
2644 store_element);
2645 // ======== end loop ========
2646
2647 // It was a real error; we must depend on the caller to finish the job.
2648 // Register G1 has number of *remaining* oops, O2 number of *total* oops.
2649 // Emit GC store barriers for the oops we have copied (O2 minus G1),
2650 // and report their number to the caller.
2651 __ BIND(fail);
2652 __ subcc(O2_count, G1_remain, O2_count);
2653 __ brx(Assembler::zero, false, Assembler::pt, done);
2654 __ delayed()->not1(O2_count, O0); // report (-1^K) to caller
2655
2656 __ BIND(do_card_marks);
2657 gen_write_ref_array_post_barrier(O1_to, O2_count, O3); // store check on O1[0..O2]
2658
2659 __ BIND(done);
2660 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4);
2661 __ retl();
2662 __ delayed()->nop(); // return value in 00
2663
2664 return start;
2665 }
2666
2667
2668 // Generate 'unsafe' array copy stub
2669 // Though just as safe as the other stubs, it takes an unscaled
2670 // size_t argument instead of an element count.
2671 //
2672 // Arguments for generated stub:
2673 // from: O0
2674 // to: O1
2675 // count: O2 byte count, treated as ssize_t, can be zero
2676 //
2677 // Examines the alignment of the operands and dispatches
2678 // to a long, int, short, or byte copy loop.
2679 //
2680 address generate_unsafe_copy(const char* name) {
2681
2682 const Register O0_from = O0; // source array address
2683 const Register O1_to = O1; // destination array address
2684 const Register O2_count = O2; // elements count
2685
2686 const Register G1_bits = G1; // test copy of low bits
2687
2688 __ align(CodeEntryAlignment);
2689 StubCodeMark mark(this, "StubRoutines", name);
2690 address start = __ pc();
2691
2692 // bump this on entry, not on exit:
2693 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr, G1, G3);
2694
2695 __ or3(O0_from, O1_to, G1_bits);
2696 __ or3(O2_count, G1_bits, G1_bits);
2697
2698 __ btst(BytesPerLong-1, G1_bits);
2699 __ br(Assembler::zero, true, Assembler::pt,
2700 long_copy_entry, relocInfo::runtime_call_type);
2701 // scale the count on the way out:
2702 __ delayed()->srax(O2_count, LogBytesPerLong, O2_count);
2703
2704 __ btst(BytesPerInt-1, G1_bits);
2705 __ br(Assembler::zero, true, Assembler::pt,
2706 int_copy_entry, relocInfo::runtime_call_type);
2707 // scale the count on the way out:
2708 __ delayed()->srax(O2_count, LogBytesPerInt, O2_count);
2709
2710 __ btst(BytesPerShort-1, G1_bits);
2711 __ br(Assembler::zero, true, Assembler::pt,
2712 short_copy_entry, relocInfo::runtime_call_type);
2713 // scale the count on the way out:
2714 __ delayed()->srax(O2_count, LogBytesPerShort, O2_count);
2715
2716 __ br(Assembler::always, false, Assembler::pt,
2717 byte_copy_entry, relocInfo::runtime_call_type);
2718 __ delayed()->nop();
2719
2720 return start;
2721 }
2722
2723
2724 // Perform range checks on the proposed arraycopy.
2725 // Kills the two temps, but nothing else.
2726 // Also, clean the sign bits of src_pos and dst_pos.
2727 void arraycopy_range_checks(Register src, // source array oop (O0)
2728 Register src_pos, // source position (O1)
2729 Register dst, // destination array oo (O2)
2730 Register dst_pos, // destination position (O3)
2731 Register length, // length of copy (O4)
2732 Register temp1, Register temp2,
2733 Label& L_failed) {
2734 BLOCK_COMMENT("arraycopy_range_checks:");
2735
2736 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
2737
2738 const Register array_length = temp1; // scratch
2739 const Register end_pos = temp2; // scratch
2740
2741 // Note: This next instruction may be in the delay slot of a branch:
2742 __ add(length, src_pos, end_pos); // src_pos + length
2743 __ lduw(src, arrayOopDesc::length_offset_in_bytes(), array_length);
2744 __ cmp(end_pos, array_length);
2745 __ br(Assembler::greater, false, Assembler::pn, L_failed);
2746
2747 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
2748 __ delayed()->add(length, dst_pos, end_pos); // dst_pos + length
2749 __ lduw(dst, arrayOopDesc::length_offset_in_bytes(), array_length);
2750 __ cmp(end_pos, array_length);
2751 __ br(Assembler::greater, false, Assembler::pn, L_failed);
2752
2753 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2754 // Move with sign extension can be used since they are positive.
2755 __ delayed()->signx(src_pos, src_pos);
2756 __ signx(dst_pos, dst_pos);
2757
2758 BLOCK_COMMENT("arraycopy_range_checks done");
2759 }
2760
2761
2762 //
2763 // Generate generic array copy stubs
2764 //
2765 // Input:
2766 // O0 - src oop
2767 // O1 - src_pos
2768 // O2 - dst oop
2769 // O3 - dst_pos
2770 // O4 - element count
2771 //
2772 // Output:
2773 // O0 == 0 - success
2774 // O0 == -1 - need to call System.arraycopy
2775 //
2776 address generate_generic_copy(const char *name) {
2777
2778 Label L_failed, L_objArray;
2779
2780 // Input registers
2781 const Register src = O0; // source array oop
2782 const Register src_pos = O1; // source position
2783 const Register dst = O2; // destination array oop
2784 const Register dst_pos = O3; // destination position
2785 const Register length = O4; // elements count
2786
2787 // registers used as temp
2788 const Register G3_src_klass = G3; // source array klass
2789 const Register G4_dst_klass = G4; // destination array klass
2790 const Register G5_lh = G5; // layout handler
2791 const Register O5_temp = O5;
2792
2793 __ align(CodeEntryAlignment);
2794 StubCodeMark mark(this, "StubRoutines", name);
2795 address start = __ pc();
2796
2797 // bump this on entry, not on exit:
2798 inc_counter_np(SharedRuntime::_generic_array_copy_ctr, G1, G3);
2799
2800 // In principle, the int arguments could be dirty.
2801 //assert_clean_int(src_pos, G1);
2802 //assert_clean_int(dst_pos, G1);
2803 //assert_clean_int(length, G1);
2804
2805 //-----------------------------------------------------------------------
2806 // Assembler stubs will be used for this call to arraycopy
2807 // if the following conditions are met:
2808 //
2809 // (1) src and dst must not be null.
2810 // (2) src_pos must not be negative.
2811 // (3) dst_pos must not be negative.
2812 // (4) length must not be negative.
2813 // (5) src klass and dst klass should be the same and not NULL.
2814 // (6) src and dst should be arrays.
2815 // (7) src_pos + length must not exceed length of src.
2816 // (8) dst_pos + length must not exceed length of dst.
2817 BLOCK_COMMENT("arraycopy initial argument checks");
2818
2819 // if (src == NULL) return -1;
2820 __ br_null(src, false, Assembler::pn, L_failed);
2821
2822 // if (src_pos < 0) return -1;
2823 __ delayed()->tst(src_pos);
2824 __ br(Assembler::negative, false, Assembler::pn, L_failed);
2825 __ delayed()->nop();
2826
2827 // if (dst == NULL) return -1;
2828 __ br_null(dst, false, Assembler::pn, L_failed);
2829
2830 // if (dst_pos < 0) return -1;
2831 __ delayed()->tst(dst_pos);
2832 __ br(Assembler::negative, false, Assembler::pn, L_failed);
2833
2834 // if (length < 0) return -1;
2835 __ delayed()->tst(length);
2836 __ br(Assembler::negative, false, Assembler::pn, L_failed);
2837
2838 BLOCK_COMMENT("arraycopy argument klass checks");
2839 // get src->klass()
2840 if (UseCompressedOops) {
2841 __ delayed()->nop(); // ??? not good
2842 __ load_klass(src, G3_src_klass);
2843 } else {
2844 __ delayed()->ld_ptr(src, oopDesc::klass_offset_in_bytes(), G3_src_klass);
2845 }
2846
2847 #ifdef ASSERT
2848 // assert(src->klass() != NULL);
2849 BLOCK_COMMENT("assert klasses not null");
2850 { Label L_a, L_b;
2851 __ br_notnull(G3_src_klass, false, Assembler::pt, L_b); // it is broken if klass is NULL
2852 __ delayed()->nop();
2853 __ bind(L_a);
2854 __ stop("broken null klass");
2855 __ bind(L_b);
2856 __ load_klass(dst, G4_dst_klass);
2857 __ br_null(G4_dst_klass, false, Assembler::pn, L_a); // this would be broken also
2858 __ delayed()->mov(G0, G4_dst_klass); // scribble the temp
2859 BLOCK_COMMENT("assert done");
2860 }
2861 #endif
2862
2863 // Load layout helper
2864 //
2865 // |array_tag| | header_size | element_type | |log2_element_size|
2866 // 32 30 24 16 8 2 0
2867 //
2868 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2869 //
2870
2871 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2872 Klass::layout_helper_offset_in_bytes();
2873
2874 // Load 32-bits signed value. Use br() instruction with it to check icc.
2875 __ lduw(G3_src_klass, lh_offset, G5_lh);
2876
2877 if (UseCompressedOops) {
2878 __ load_klass(dst, G4_dst_klass);
2879 }
2880 // Handle objArrays completely differently...
2881 juint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2882 __ set(objArray_lh, O5_temp);
2883 __ cmp(G5_lh, O5_temp);
2884 __ br(Assembler::equal, false, Assembler::pt, L_objArray);
2885 if (UseCompressedOops) {
2886 __ delayed()->nop();
2887 } else {
2888 __ delayed()->ld_ptr(dst, oopDesc::klass_offset_in_bytes(), G4_dst_klass);
2889 }
2890
2891 // if (src->klass() != dst->klass()) return -1;
2892 __ cmp(G3_src_klass, G4_dst_klass);
2893 __ brx(Assembler::notEqual, false, Assembler::pn, L_failed);
2894 __ delayed()->nop();
2895
2896 // if (!src->is_Array()) return -1;
2897 __ cmp(G5_lh, Klass::_lh_neutral_value); // < 0
2898 __ br(Assembler::greaterEqual, false, Assembler::pn, L_failed);
2899
2900 // At this point, it is known to be a typeArray (array_tag 0x3).
2901 #ifdef ASSERT
2902 __ delayed()->nop();
2903 { Label L;
2904 jint lh_prim_tag_in_place = (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift);
2905 __ set(lh_prim_tag_in_place, O5_temp);
2906 __ cmp(G5_lh, O5_temp);
2907 __ br(Assembler::greaterEqual, false, Assembler::pt, L);
2908 __ delayed()->nop();
2909 __ stop("must be a primitive array");
2910 __ bind(L);
2911 }
2912 #else
2913 __ delayed(); // match next insn to prev branch
2914 #endif
2915
2916 arraycopy_range_checks(src, src_pos, dst, dst_pos, length,
2917 O5_temp, G4_dst_klass, L_failed);
2918
2919 // typeArrayKlass
2920 //
2921 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2922 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2923 //
2924
2925 const Register G4_offset = G4_dst_klass; // array offset
2926 const Register G3_elsize = G3_src_klass; // log2 element size
2927
2928 __ srl(G5_lh, Klass::_lh_header_size_shift, G4_offset);
2929 __ and3(G4_offset, Klass::_lh_header_size_mask, G4_offset); // array_offset
2930 __ add(src, G4_offset, src); // src array offset
2931 __ add(dst, G4_offset, dst); // dst array offset
2932 __ and3(G5_lh, Klass::_lh_log2_element_size_mask, G3_elsize); // log2 element size
2933
2934 // next registers should be set before the jump to corresponding stub
2935 const Register from = O0; // source array address
2936 const Register to = O1; // destination array address
2937 const Register count = O2; // elements count
2938
2939 // 'from', 'to', 'count' registers should be set in this order
2940 // since they are the same as 'src', 'src_pos', 'dst'.
2941
2942 BLOCK_COMMENT("scale indexes to element size");
2943 __ sll_ptr(src_pos, G3_elsize, src_pos);
2944 __ sll_ptr(dst_pos, G3_elsize, dst_pos);
2945 __ add(src, src_pos, from); // src_addr
2946 __ add(dst, dst_pos, to); // dst_addr
2947
2948 BLOCK_COMMENT("choose copy loop based on element size");
2949 __ cmp(G3_elsize, 0);
2950 __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jbyte_arraycopy);
2951 __ delayed()->signx(length, count); // length
2952
2953 __ cmp(G3_elsize, LogBytesPerShort);
2954 __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jshort_arraycopy);
2955 __ delayed()->signx(length, count); // length
2956
2957 __ cmp(G3_elsize, LogBytesPerInt);
2958 __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jint_arraycopy);
2959 __ delayed()->signx(length, count); // length
2960 #ifdef ASSERT
2961 { Label L;
2962 __ cmp(G3_elsize, LogBytesPerLong);
2963 __ br(Assembler::equal, false, Assembler::pt, L);
2964 __ delayed()->nop();
2965 __ stop("must be long copy, but elsize is wrong");
2966 __ bind(L);
2967 }
2968 #endif
2969 __ br(Assembler::always,false,Assembler::pt,StubRoutines::_jlong_arraycopy);
2970 __ delayed()->signx(length, count); // length
2971
2972 // objArrayKlass
2973 __ BIND(L_objArray);
2974 // live at this point: G3_src_klass, G4_dst_klass, src[_pos], dst[_pos], length
2975
2976 Label L_plain_copy, L_checkcast_copy;
2977 // test array classes for subtyping
2978 __ cmp(G3_src_klass, G4_dst_klass); // usual case is exact equality
2979 __ brx(Assembler::notEqual, true, Assembler::pn, L_checkcast_copy);
2980 __ delayed()->lduw(G4_dst_klass, lh_offset, O5_temp); // hoisted from below
2981
2982 // Identically typed arrays can be copied without element-wise checks.
2983 arraycopy_range_checks(src, src_pos, dst, dst_pos, length,
2984 O5_temp, G5_lh, L_failed);
2985
2986 __ add(src, arrayOopDesc::base_offset_in_bytes(T_OBJECT), src); //src offset
2987 __ add(dst, arrayOopDesc::base_offset_in_bytes(T_OBJECT), dst); //dst offset
2988 __ sll_ptr(src_pos, LogBytesPerHeapOop, src_pos);
2989 __ sll_ptr(dst_pos, LogBytesPerHeapOop, dst_pos);
2990 __ add(src, src_pos, from); // src_addr
2991 __ add(dst, dst_pos, to); // dst_addr
2992 __ BIND(L_plain_copy);
2993 __ br(Assembler::always, false, Assembler::pt,StubRoutines::_oop_arraycopy);
2994 __ delayed()->signx(length, count); // length
2995
2996 __ BIND(L_checkcast_copy);
2997 // live at this point: G3_src_klass, G4_dst_klass
2998 {
2999 // Before looking at dst.length, make sure dst is also an objArray.
3000 // lduw(G4_dst_klass, lh_offset, O5_temp); // hoisted to delay slot
3001 __ cmp(G5_lh, O5_temp);
3002 __ br(Assembler::notEqual, false, Assembler::pn, L_failed);
3003
3004 // It is safe to examine both src.length and dst.length.
3005 __ delayed(); // match next insn to prev branch
3006 arraycopy_range_checks(src, src_pos, dst, dst_pos, length,
3007 O5_temp, G5_lh, L_failed);
3008
3009 // Marshal the base address arguments now, freeing registers.
3010 __ add(src, arrayOopDesc::base_offset_in_bytes(T_OBJECT), src); //src offset
3011 __ add(dst, arrayOopDesc::base_offset_in_bytes(T_OBJECT), dst); //dst offset
3012 __ sll_ptr(src_pos, LogBytesPerHeapOop, src_pos);
3013 __ sll_ptr(dst_pos, LogBytesPerHeapOop, dst_pos);
3014 __ add(src, src_pos, from); // src_addr
3015 __ add(dst, dst_pos, to); // dst_addr
3016 __ signx(length, count); // length (reloaded)
3017
3018 Register sco_temp = O3; // this register is free now
3019 assert_different_registers(from, to, count, sco_temp,
3020 G4_dst_klass, G3_src_klass);
3021
3022 // Generate the type check.
3023 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
3024 Klass::super_check_offset_offset_in_bytes());
3025 __ lduw(G4_dst_klass, sco_offset, sco_temp);
3026 generate_type_check(G3_src_klass, sco_temp, G4_dst_klass,
3027 O5_temp, L_plain_copy);
3028
3029 // Fetch destination element klass from the objArrayKlass header.
3030 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
3031 objArrayKlass::element_klass_offset_in_bytes());
3032
3033 // the checkcast_copy loop needs two extra arguments:
3034 __ ld_ptr(G4_dst_klass, ek_offset, O4); // dest elem klass
3035 // lduw(O4, sco_offset, O3); // sco of elem klass
3036
3037 __ br(Assembler::always, false, Assembler::pt, checkcast_copy_entry);
3038 __ delayed()->lduw(O4, sco_offset, O3);
3039 }
3040
3041 __ BIND(L_failed);
3042 __ retl();
3043 __ delayed()->sub(G0, 1, O0); // return -1
3044 return start;
3045 }
3046
3047 void generate_arraycopy_stubs() {
3048
3049 // Note: the disjoint stubs must be generated first, some of
3050 // the conjoint stubs use them.
3051 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy");
3052 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy");
3053 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_copy(false, "jint_disjoint_arraycopy");
3054 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_copy(false, "jlong_disjoint_arraycopy");
3055 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_oop_copy(false, "oop_disjoint_arraycopy");
3056 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(true, "arrayof_jbyte_disjoint_arraycopy");
3057 StubRoutines::_arrayof_jshort_disjoint_arraycopy = generate_disjoint_short_copy(true, "arrayof_jshort_disjoint_arraycopy");
3058 StubRoutines::_arrayof_jint_disjoint_arraycopy = generate_disjoint_int_copy(true, "arrayof_jint_disjoint_arraycopy");
3059 StubRoutines::_arrayof_jlong_disjoint_arraycopy = generate_disjoint_long_copy(true, "arrayof_jlong_disjoint_arraycopy");
3060 StubRoutines::_arrayof_oop_disjoint_arraycopy = generate_disjoint_oop_copy(true, "arrayof_oop_disjoint_arraycopy");
3061
3062 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy");
3063 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy");
3064 StubRoutines::_jint_arraycopy = generate_conjoint_int_copy(false, "jint_arraycopy");
3065 StubRoutines::_jlong_arraycopy = generate_conjoint_long_copy(false, "jlong_arraycopy");
3066 StubRoutines::_oop_arraycopy = generate_conjoint_oop_copy(false, "oop_arraycopy");
3067 StubRoutines::_arrayof_jbyte_arraycopy = generate_conjoint_byte_copy(true, "arrayof_jbyte_arraycopy");
3068 StubRoutines::_arrayof_jshort_arraycopy = generate_conjoint_short_copy(true, "arrayof_jshort_arraycopy");
3069 #ifdef _LP64
3070 // since sizeof(jint) < sizeof(HeapWord), there's a different flavor:
3071 StubRoutines::_arrayof_jint_arraycopy = generate_conjoint_int_copy(true, "arrayof_jint_arraycopy");
3072 #else
3073 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
3074 #endif
3075 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
3076 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
3077
3078 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy");
3079 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy");
3080 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy");
3081
3082 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
3083 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
3084 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
3085 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
3086 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
3087 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
3088 }
3089
3090 void generate_initial() {
3091 // Generates all stubs and initializes the entry points
3092
3093 //------------------------------------------------------------------------------------------------------------------------
3094 // entry points that exist in all platforms
3095 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
3096 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
3097 StubRoutines::_forward_exception_entry = generate_forward_exception();
3098
3099 StubRoutines::_call_stub_entry = generate_call_stub(StubRoutines::_call_stub_return_address);
3100 StubRoutines::_catch_exception_entry = generate_catch_exception();
3101
3102 //------------------------------------------------------------------------------------------------------------------------
3103 // entry points that are platform specific
3104 StubRoutines::Sparc::_test_stop_entry = generate_test_stop();
3105
3106 StubRoutines::Sparc::_stop_subroutine_entry = generate_stop_subroutine();
3107 StubRoutines::Sparc::_flush_callers_register_windows_entry = generate_flush_callers_register_windows();
3108
3109 #if !defined(COMPILER2) && !defined(_LP64)
3110 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3111 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
3112 StubRoutines::_atomic_add_entry = generate_atomic_add();
3113 StubRoutines::_atomic_xchg_ptr_entry = StubRoutines::_atomic_xchg_entry;
3114 StubRoutines::_atomic_cmpxchg_ptr_entry = StubRoutines::_atomic_cmpxchg_entry;
3115 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
3116 StubRoutines::_atomic_add_ptr_entry = StubRoutines::_atomic_add_entry;
3117 #endif // COMPILER2 !=> _LP64
3118 }
3119
3120
3121 void generate_all() {
3122 // Generates all stubs and initializes the entry points
3123
3124 // Generate partial_subtype_check first here since its code depends on
3125 // UseZeroBaseCompressedOops which is defined after heap initialization.
3126 StubRoutines::Sparc::_partial_subtype_check = generate_partial_subtype_check();
3127 // These entry points require SharedInfo::stack0 to be set up in non-core builds
3128 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false);
3129 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false);
3130 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true);
3131 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
3132 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);
3133 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
3134
3135 StubRoutines::_handler_for_unsafe_access_entry =
3136 generate_handler_for_unsafe_access();
3137
3138 // support for verify_oop (must happen after universe_init)
3139 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop_subroutine();
3140
3141 // arraycopy stubs used by compilers
3142 generate_arraycopy_stubs();
3143
3144 // Don't initialize the platform math functions since sparc
3145 // doesn't have intrinsics for these operations.
3146 }
3147
3148
3149 public:
3150 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3151 // replace the standard masm with a special one:
3152 _masm = new MacroAssembler(code);
3153
3154 _stub_count = !all ? 0x100 : 0x200;
3155 if (all) {
3156 generate_all();
3157 } else {
3158 generate_initial();
3159 }
3160
3161 // make sure this stub is available for all local calls
3162 if (_atomic_add_stub.is_unbound()) {
3163 // generate a second time, if necessary
3164 (void) generate_atomic_add();
3165 }
3166 }
3167
3168
3169 private:
3170 int _stub_count;
3171 void stub_prolog(StubCodeDesc* cdesc) {
3172 # ifdef ASSERT
3173 // put extra information in the stub code, to make it more readable
3174 #ifdef _LP64
3175 // Write the high part of the address
3176 // [RGV] Check if there is a dependency on the size of this prolog
3177 __ emit_data((intptr_t)cdesc >> 32, relocInfo::none);
3178 #endif
3179 __ emit_data((intptr_t)cdesc, relocInfo::none);
3180 __ emit_data(++_stub_count, relocInfo::none);
3181 # endif
3182 align(true);
3183 }
3184
3185 void align(bool at_header = false) {
3186 // %%%%% move this constant somewhere else
3187 // UltraSPARC cache line size is 8 instructions:
3188 const unsigned int icache_line_size = 32;
3189 const unsigned int icache_half_line_size = 16;
3190
3191 if (at_header) {
3192 while ((intptr_t)(__ pc()) % icache_line_size != 0) {
3193 __ emit_data(0, relocInfo::none);
3194 }
3195 } else {
3196 while ((intptr_t)(__ pc()) % icache_half_line_size != 0) {
3197 __ nop();
3198 }
3199 }
3200 }
3201
3202 }; // end class declaration
3203
3204
3205 address StubGenerator::disjoint_byte_copy_entry = NULL;
3206 address StubGenerator::disjoint_short_copy_entry = NULL;
3207 address StubGenerator::disjoint_int_copy_entry = NULL;
3208 address StubGenerator::disjoint_long_copy_entry = NULL;
3209 address StubGenerator::disjoint_oop_copy_entry = NULL;
3210
3211 address StubGenerator::byte_copy_entry = NULL;
3212 address StubGenerator::short_copy_entry = NULL;
3213 address StubGenerator::int_copy_entry = NULL;
3214 address StubGenerator::long_copy_entry = NULL;
3215 address StubGenerator::oop_copy_entry = NULL;
3216
3217 address StubGenerator::checkcast_copy_entry = NULL;
3218
3219 void StubGenerator_generate(CodeBuffer* code, bool all) {
3220 StubGenerator g(code, all);
3221 }