Mercurial > hg > graal-compiler
comparison src/cpu/sparc/vm/stubGenerator_sparc.cpp @ 0:a61af66fc99e jdk7-b24
Initial load
author | duke |
---|---|
date | Sat, 01 Dec 2007 00:00:00 +0000 |
parents | |
children | f8236e79048a |
comparison
equal
deleted
inserted
replaced
-1:000000000000 | 0:a61af66fc99e |
---|---|
1 /* | |
2 * Copyright 1997-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
21 * have any 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 void inc_counter_np_(int& counter, Register t1, Register t2) { | |
72 Address counter_addr(t2, (address) &counter); | |
73 __ sethi(counter_addr); | |
74 __ ld(counter_addr, t1); | |
75 __ inc(t1); | |
76 __ st(t1, counter_addr); | |
77 } | |
78 #define inc_counter_np(counter, t1, t2) \ | |
79 BLOCK_COMMENT("inc_counter " #counter); \ | |
80 inc_counter_np_(counter, t1, t2); | |
81 #endif | |
82 | |
83 //---------------------------------------------------------------------------------------------------- | |
84 // Call stubs are used to call Java from C | |
85 | |
86 address generate_call_stub(address& return_pc) { | |
87 StubCodeMark mark(this, "StubRoutines", "call_stub"); | |
88 address start = __ pc(); | |
89 | |
90 // Incoming arguments: | |
91 // | |
92 // o0 : call wrapper address | |
93 // o1 : result (address) | |
94 // o2 : result type | |
95 // o3 : method | |
96 // o4 : (interpreter) entry point | |
97 // o5 : parameters (address) | |
98 // [sp + 0x5c]: parameter size (in words) | |
99 // [sp + 0x60]: thread | |
100 // | |
101 // +---------------+ <--- sp + 0 | |
102 // | | | |
103 // . reg save area . | |
104 // | | | |
105 // +---------------+ <--- sp + 0x40 | |
106 // | | | |
107 // . extra 7 slots . | |
108 // | | | |
109 // +---------------+ <--- sp + 0x5c | |
110 // | param. size | | |
111 // +---------------+ <--- sp + 0x60 | |
112 // | thread | | |
113 // +---------------+ | |
114 // | | | |
115 | |
116 // note: if the link argument position changes, adjust | |
117 // the code in frame::entry_frame_call_wrapper() | |
118 | |
119 const Argument link = Argument(0, false); // used only for GC | |
120 const Argument result = Argument(1, false); | |
121 const Argument result_type = Argument(2, false); | |
122 const Argument method = Argument(3, false); | |
123 const Argument entry_point = Argument(4, false); | |
124 const Argument parameters = Argument(5, false); | |
125 const Argument parameter_size = Argument(6, false); | |
126 const Argument thread = Argument(7, false); | |
127 | |
128 // setup thread register | |
129 __ ld_ptr(thread.as_address(), G2_thread); | |
130 | |
131 #ifdef ASSERT | |
132 // make sure we have no pending exceptions | |
133 { const Register t = G3_scratch; | |
134 Label L; | |
135 __ ld_ptr(G2_thread, in_bytes(Thread::pending_exception_offset()), t); | |
136 __ br_null(t, false, Assembler::pt, L); | |
137 __ delayed()->nop(); | |
138 __ stop("StubRoutines::call_stub: entered with pending exception"); | |
139 __ bind(L); | |
140 } | |
141 #endif | |
142 | |
143 // create activation frame & allocate space for parameters | |
144 { const Register t = G3_scratch; | |
145 __ ld_ptr(parameter_size.as_address(), t); // get parameter size (in words) | |
146 __ add(t, frame::memory_parameter_word_sp_offset, t); // add space for save area (in words) | |
147 __ round_to(t, WordsPerLong); // make sure it is multiple of 2 (in words) | |
148 __ sll(t, Interpreter::logStackElementSize(), t); // compute number of bytes | |
149 __ neg(t); // negate so it can be used with save | |
150 __ save(SP, t, SP); // setup new frame | |
151 } | |
152 | |
153 // +---------------+ <--- sp + 0 | |
154 // | | | |
155 // . reg save area . | |
156 // | | | |
157 // +---------------+ <--- sp + 0x40 | |
158 // | | | |
159 // . extra 7 slots . | |
160 // | | | |
161 // +---------------+ <--- sp + 0x5c | |
162 // | empty slot | (only if parameter size is even) | |
163 // +---------------+ | |
164 // | | | |
165 // . parameters . | |
166 // | | | |
167 // +---------------+ <--- fp + 0 | |
168 // | | | |
169 // . reg save area . | |
170 // | | | |
171 // +---------------+ <--- fp + 0x40 | |
172 // | | | |
173 // . extra 7 slots . | |
174 // | | | |
175 // +---------------+ <--- fp + 0x5c | |
176 // | param. size | | |
177 // +---------------+ <--- fp + 0x60 | |
178 // | thread | | |
179 // +---------------+ | |
180 // | | | |
181 | |
182 // pass parameters if any | |
183 BLOCK_COMMENT("pass parameters if any"); | |
184 { const Register src = parameters.as_in().as_register(); | |
185 const Register dst = Lentry_args; | |
186 const Register tmp = G3_scratch; | |
187 const Register cnt = G4_scratch; | |
188 | |
189 // test if any parameters & setup of Lentry_args | |
190 Label exit; | |
191 __ ld_ptr(parameter_size.as_in().as_address(), cnt); // parameter counter | |
192 __ add( FP, STACK_BIAS, dst ); | |
193 __ tst(cnt); | |
194 __ br(Assembler::zero, false, Assembler::pn, exit); | |
195 __ delayed()->sub(dst, BytesPerWord, dst); // setup Lentry_args | |
196 | |
197 // copy parameters if any | |
198 Label loop; | |
199 __ BIND(loop); | |
200 // Store tag first. | |
201 if (TaggedStackInterpreter) { | |
202 __ ld_ptr(src, 0, tmp); | |
203 __ add(src, BytesPerWord, src); // get next | |
204 __ st_ptr(tmp, dst, Interpreter::tag_offset_in_bytes()); | |
205 } | |
206 // Store parameter value | |
207 __ ld_ptr(src, 0, tmp); | |
208 __ add(src, BytesPerWord, src); | |
209 __ st_ptr(tmp, dst, Interpreter::value_offset_in_bytes()); | |
210 __ deccc(cnt); | |
211 __ br(Assembler::greater, false, Assembler::pt, loop); | |
212 __ delayed()->sub(dst, Interpreter::stackElementSize(), dst); | |
213 | |
214 // done | |
215 __ BIND(exit); | |
216 } | |
217 | |
218 // setup parameters, method & call Java function | |
219 #ifdef ASSERT | |
220 // layout_activation_impl checks it's notion of saved SP against | |
221 // this register, so if this changes update it as well. | |
222 const Register saved_SP = Lscratch; | |
223 __ mov(SP, saved_SP); // keep track of SP before call | |
224 #endif | |
225 | |
226 // setup parameters | |
227 const Register t = G3_scratch; | |
228 __ ld_ptr(parameter_size.as_in().as_address(), t); // get parameter size (in words) | |
229 __ sll(t, Interpreter::logStackElementSize(), t); // compute number of bytes | |
230 __ sub(FP, t, Gargs); // setup parameter pointer | |
231 #ifdef _LP64 | |
232 __ add( Gargs, STACK_BIAS, Gargs ); // Account for LP64 stack bias | |
233 #endif | |
234 __ mov(SP, O5_savedSP); | |
235 | |
236 | |
237 // do the call | |
238 // | |
239 // the following register must be setup: | |
240 // | |
241 // G2_thread | |
242 // G5_method | |
243 // Gargs | |
244 BLOCK_COMMENT("call Java function"); | |
245 __ jmpl(entry_point.as_in().as_register(), G0, O7); | |
246 __ delayed()->mov(method.as_in().as_register(), G5_method); // setup method | |
247 | |
248 BLOCK_COMMENT("call_stub_return_address:"); | |
249 return_pc = __ pc(); | |
250 | |
251 // The callee, if it wasn't interpreted, can return with SP changed so | |
252 // we can no longer assert of change of SP. | |
253 | |
254 // store result depending on type | |
255 // (everything that is not T_OBJECT, T_LONG, T_FLOAT, or T_DOUBLE | |
256 // is treated as T_INT) | |
257 { const Register addr = result .as_in().as_register(); | |
258 const Register type = result_type.as_in().as_register(); | |
259 Label is_long, is_float, is_double, is_object, exit; | |
260 __ cmp(type, T_OBJECT); __ br(Assembler::equal, false, Assembler::pn, is_object); | |
261 __ delayed()->cmp(type, T_FLOAT); __ br(Assembler::equal, false, Assembler::pn, is_float); | |
262 __ delayed()->cmp(type, T_DOUBLE); __ br(Assembler::equal, false, Assembler::pn, is_double); | |
263 __ delayed()->cmp(type, T_LONG); __ br(Assembler::equal, false, Assembler::pn, is_long); | |
264 __ delayed()->nop(); | |
265 | |
266 // store int result | |
267 __ st(O0, addr, G0); | |
268 | |
269 __ BIND(exit); | |
270 __ ret(); | |
271 __ delayed()->restore(); | |
272 | |
273 __ BIND(is_object); | |
274 __ ba(false, exit); | |
275 __ delayed()->st_ptr(O0, addr, G0); | |
276 | |
277 __ BIND(is_float); | |
278 __ ba(false, exit); | |
279 __ delayed()->stf(FloatRegisterImpl::S, F0, addr, G0); | |
280 | |
281 __ BIND(is_double); | |
282 __ ba(false, exit); | |
283 __ delayed()->stf(FloatRegisterImpl::D, F0, addr, G0); | |
284 | |
285 __ BIND(is_long); | |
286 #ifdef _LP64 | |
287 __ ba(false, exit); | |
288 __ delayed()->st_long(O0, addr, G0); // store entire long | |
289 #else | |
290 #if defined(COMPILER2) | |
291 // All return values are where we want them, except for Longs. C2 returns | |
292 // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1. | |
293 // Since the interpreter will return longs in G1 and O0/O1 in the 32bit | |
294 // build we simply always use G1. | |
295 // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to | |
296 // do this here. Unfortunately if we did a rethrow we'd see an machepilog node | |
297 // first which would move g1 -> O0/O1 and destroy the exception we were throwing. | |
298 | |
299 __ ba(false, exit); | |
300 __ delayed()->stx(G1, addr, G0); // store entire long | |
301 #else | |
302 __ st(O1, addr, BytesPerInt); | |
303 __ ba(false, exit); | |
304 __ delayed()->st(O0, addr, G0); | |
305 #endif /* COMPILER2 */ | |
306 #endif /* _LP64 */ | |
307 } | |
308 return start; | |
309 } | |
310 | |
311 | |
312 //---------------------------------------------------------------------------------------------------- | |
313 // Return point for a Java call if there's an exception thrown in Java code. | |
314 // The exception is caught and transformed into a pending exception stored in | |
315 // JavaThread that can be tested from within the VM. | |
316 // | |
317 // Oexception: exception oop | |
318 | |
319 address generate_catch_exception() { | |
320 StubCodeMark mark(this, "StubRoutines", "catch_exception"); | |
321 | |
322 address start = __ pc(); | |
323 // verify that thread corresponds | |
324 __ verify_thread(); | |
325 | |
326 const Register& temp_reg = Gtemp; | |
327 Address pending_exception_addr (G2_thread, 0, in_bytes(Thread::pending_exception_offset())); | |
328 Address exception_file_offset_addr(G2_thread, 0, in_bytes(Thread::exception_file_offset ())); | |
329 Address exception_line_offset_addr(G2_thread, 0, in_bytes(Thread::exception_line_offset ())); | |
330 | |
331 // set pending exception | |
332 __ verify_oop(Oexception); | |
333 __ st_ptr(Oexception, pending_exception_addr); | |
334 __ set((intptr_t)__FILE__, temp_reg); | |
335 __ st_ptr(temp_reg, exception_file_offset_addr); | |
336 __ set((intptr_t)__LINE__, temp_reg); | |
337 __ st(temp_reg, exception_line_offset_addr); | |
338 | |
339 // complete return to VM | |
340 assert(StubRoutines::_call_stub_return_address != NULL, "must have been generated before"); | |
341 | |
342 Address stub_ret(temp_reg, StubRoutines::_call_stub_return_address); | |
343 __ jump_to(stub_ret); | |
344 __ delayed()->nop(); | |
345 | |
346 return start; | |
347 } | |
348 | |
349 | |
350 //---------------------------------------------------------------------------------------------------- | |
351 // Continuation point for runtime calls returning with a pending exception | |
352 // The pending exception check happened in the runtime or native call stub | |
353 // The pending exception in Thread is converted into a Java-level exception | |
354 // | |
355 // Contract with Java-level exception handler: O0 = exception | |
356 // O1 = throwing pc | |
357 | |
358 address generate_forward_exception() { | |
359 StubCodeMark mark(this, "StubRoutines", "forward_exception"); | |
360 address start = __ pc(); | |
361 | |
362 // Upon entry, O7 has the return address returning into Java | |
363 // (interpreted or compiled) code; i.e. the return address | |
364 // becomes the throwing pc. | |
365 | |
366 const Register& handler_reg = Gtemp; | |
367 | |
368 Address exception_addr (G2_thread, 0, in_bytes(Thread::pending_exception_offset())); | |
369 | |
370 #ifdef ASSERT | |
371 // make sure that this code is only executed if there is a pending exception | |
372 { Label L; | |
373 __ ld_ptr(exception_addr, Gtemp); | |
374 __ br_notnull(Gtemp, false, Assembler::pt, L); | |
375 __ delayed()->nop(); | |
376 __ stop("StubRoutines::forward exception: no pending exception (1)"); | |
377 __ bind(L); | |
378 } | |
379 #endif | |
380 | |
381 // compute exception handler into handler_reg | |
382 __ get_thread(); | |
383 __ ld_ptr(exception_addr, Oexception); | |
384 __ verify_oop(Oexception); | |
385 __ save_frame(0); // compensates for compiler weakness | |
386 __ add(O7->after_save(), frame::pc_return_offset, Lscratch); // save the issuing PC | |
387 BLOCK_COMMENT("call exception_handler_for_return_address"); | |
388 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), Lscratch); | |
389 __ mov(O0, handler_reg); | |
390 __ restore(); // compensates for compiler weakness | |
391 | |
392 __ ld_ptr(exception_addr, Oexception); | |
393 __ add(O7, frame::pc_return_offset, Oissuing_pc); // save the issuing PC | |
394 | |
395 #ifdef ASSERT | |
396 // make sure exception is set | |
397 { Label L; | |
398 __ br_notnull(Oexception, false, Assembler::pt, L); | |
399 __ delayed()->nop(); | |
400 __ stop("StubRoutines::forward exception: no pending exception (2)"); | |
401 __ bind(L); | |
402 } | |
403 #endif | |
404 // jump to exception handler | |
405 __ jmp(handler_reg, 0); | |
406 // clear pending exception | |
407 __ delayed()->st_ptr(G0, exception_addr); | |
408 | |
409 return start; | |
410 } | |
411 | |
412 | |
413 //------------------------------------------------------------------------------------------------------------------------ | |
414 // Continuation point for throwing of implicit exceptions that are not handled in | |
415 // the current activation. Fabricates an exception oop and initiates normal | |
416 // exception dispatching in this frame. Only callee-saved registers are preserved | |
417 // (through the normal register window / RegisterMap handling). | |
418 // If the compiler needs all registers to be preserved between the fault | |
419 // point and the exception handler then it must assume responsibility for that in | |
420 // AbstractCompiler::continuation_for_implicit_null_exception or | |
421 // continuation_for_implicit_division_by_zero_exception. All other implicit | |
422 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are | |
423 // either at call sites or otherwise assume that stack unwinding will be initiated, | |
424 // so caller saved registers were assumed volatile in the compiler. | |
425 | |
426 // Note that we generate only this stub into a RuntimeStub, because it needs to be | |
427 // properly traversed and ignored during GC, so we change the meaning of the "__" | |
428 // macro within this method. | |
429 #undef __ | |
430 #define __ masm-> | |
431 | |
432 address generate_throw_exception(const char* name, address runtime_entry, bool restore_saved_exception_pc) { | |
433 #ifdef ASSERT | |
434 int insts_size = VerifyThread ? 1 * K : 600; | |
435 #else | |
436 int insts_size = VerifyThread ? 1 * K : 256; | |
437 #endif /* ASSERT */ | |
438 int locs_size = 32; | |
439 | |
440 CodeBuffer code(name, insts_size, locs_size); | |
441 MacroAssembler* masm = new MacroAssembler(&code); | |
442 | |
443 __ verify_thread(); | |
444 | |
445 // This is an inlined and slightly modified version of call_VM | |
446 // which has the ability to fetch the return PC out of thread-local storage | |
447 __ assert_not_delayed(); | |
448 | |
449 // Note that we always push a frame because on the SPARC | |
450 // architecture, for all of our implicit exception kinds at call | |
451 // sites, the implicit exception is taken before the callee frame | |
452 // is pushed. | |
453 __ save_frame(0); | |
454 | |
455 int frame_complete = __ offset(); | |
456 | |
457 if (restore_saved_exception_pc) { | |
458 Address saved_exception_pc(G2_thread, 0, in_bytes(JavaThread::saved_exception_pc_offset())); | |
459 __ ld_ptr(saved_exception_pc, I7); | |
460 __ sub(I7, frame::pc_return_offset, I7); | |
461 } | |
462 | |
463 // Note that we always have a runtime stub frame on the top of stack by this point | |
464 Register last_java_sp = SP; | |
465 // 64-bit last_java_sp is biased! | |
466 __ set_last_Java_frame(last_java_sp, G0); | |
467 if (VerifyThread) __ mov(G2_thread, O0); // about to be smashed; pass early | |
468 __ save_thread(noreg); | |
469 // do the call | |
470 BLOCK_COMMENT("call runtime_entry"); | |
471 __ call(runtime_entry, relocInfo::runtime_call_type); | |
472 if (!VerifyThread) | |
473 __ delayed()->mov(G2_thread, O0); // pass thread as first argument | |
474 else | |
475 __ delayed()->nop(); // (thread already passed) | |
476 __ restore_thread(noreg); | |
477 __ reset_last_Java_frame(); | |
478 | |
479 // check for pending exceptions. use Gtemp as scratch register. | |
480 #ifdef ASSERT | |
481 Label L; | |
482 | |
483 Address exception_addr(G2_thread, 0, in_bytes(Thread::pending_exception_offset())); | |
484 Register scratch_reg = Gtemp; | |
485 __ ld_ptr(exception_addr, scratch_reg); | |
486 __ br_notnull(scratch_reg, false, Assembler::pt, L); | |
487 __ delayed()->nop(); | |
488 __ should_not_reach_here(); | |
489 __ bind(L); | |
490 #endif // ASSERT | |
491 BLOCK_COMMENT("call forward_exception_entry"); | |
492 __ call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type); | |
493 // we use O7 linkage so that forward_exception_entry has the issuing PC | |
494 __ delayed()->restore(); | |
495 | |
496 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, masm->total_frame_size_in_bytes(0), NULL, false); | |
497 return stub->entry_point(); | |
498 } | |
499 | |
500 #undef __ | |
501 #define __ _masm-> | |
502 | |
503 | |
504 // Generate a routine that sets all the registers so we | |
505 // can tell if the stop routine prints them correctly. | |
506 address generate_test_stop() { | |
507 StubCodeMark mark(this, "StubRoutines", "test_stop"); | |
508 address start = __ pc(); | |
509 | |
510 int i; | |
511 | |
512 __ save_frame(0); | |
513 | |
514 static jfloat zero = 0.0, one = 1.0; | |
515 | |
516 // put addr in L0, then load through L0 to F0 | |
517 __ set((intptr_t)&zero, L0); __ ldf( FloatRegisterImpl::S, L0, 0, F0); | |
518 __ set((intptr_t)&one, L0); __ ldf( FloatRegisterImpl::S, L0, 0, F1); // 1.0 to F1 | |
519 | |
520 // use add to put 2..18 in F2..F18 | |
521 for ( i = 2; i <= 18; ++i ) { | |
522 __ fadd( FloatRegisterImpl::S, F1, as_FloatRegister(i-1), as_FloatRegister(i)); | |
523 } | |
524 | |
525 // Now put double 2 in F16, double 18 in F18 | |
526 __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, F2, F16 ); | |
527 __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, F18, F18 ); | |
528 | |
529 // use add to put 20..32 in F20..F32 | |
530 for (i = 20; i < 32; i += 2) { | |
531 __ fadd( FloatRegisterImpl::D, F16, as_FloatRegister(i-2), as_FloatRegister(i)); | |
532 } | |
533 | |
534 // put 0..7 in i's, 8..15 in l's, 16..23 in o's, 24..31 in g's | |
535 for ( i = 0; i < 8; ++i ) { | |
536 if (i < 6) { | |
537 __ set( i, as_iRegister(i)); | |
538 __ set(16 + i, as_oRegister(i)); | |
539 __ set(24 + i, as_gRegister(i)); | |
540 } | |
541 __ set( 8 + i, as_lRegister(i)); | |
542 } | |
543 | |
544 __ stop("testing stop"); | |
545 | |
546 | |
547 __ ret(); | |
548 __ delayed()->restore(); | |
549 | |
550 return start; | |
551 } | |
552 | |
553 | |
554 address generate_stop_subroutine() { | |
555 StubCodeMark mark(this, "StubRoutines", "stop_subroutine"); | |
556 address start = __ pc(); | |
557 | |
558 __ stop_subroutine(); | |
559 | |
560 return start; | |
561 } | |
562 | |
563 address generate_flush_callers_register_windows() { | |
564 StubCodeMark mark(this, "StubRoutines", "flush_callers_register_windows"); | |
565 address start = __ pc(); | |
566 | |
567 __ flush_windows(); | |
568 __ retl(false); | |
569 __ delayed()->add( FP, STACK_BIAS, O0 ); | |
570 // The returned value must be a stack pointer whose register save area | |
571 // is flushed, and will stay flushed while the caller executes. | |
572 | |
573 return start; | |
574 } | |
575 | |
576 // Helper functions for v8 atomic operations. | |
577 // | |
578 void get_v8_oop_lock_ptr(Register lock_ptr_reg, Register mark_oop_reg, Register scratch_reg) { | |
579 if (mark_oop_reg == noreg) { | |
580 address lock_ptr = (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(); | |
581 __ set((intptr_t)lock_ptr, lock_ptr_reg); | |
582 } else { | |
583 assert(scratch_reg != noreg, "just checking"); | |
584 address lock_ptr = (address)StubRoutines::Sparc::_v8_oop_lock_cache; | |
585 __ set((intptr_t)lock_ptr, lock_ptr_reg); | |
586 __ and3(mark_oop_reg, StubRoutines::Sparc::v8_oop_lock_mask_in_place, scratch_reg); | |
587 __ add(lock_ptr_reg, scratch_reg, lock_ptr_reg); | |
588 } | |
589 } | |
590 | |
591 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) { | |
592 | |
593 get_v8_oop_lock_ptr(lock_ptr_reg, mark_oop_reg, scratch_reg); | |
594 __ set(StubRoutines::Sparc::locked, lock_reg); | |
595 // Initialize yield counter | |
596 __ mov(G0,yield_reg); | |
597 | |
598 __ BIND(retry); | |
599 __ cmp(yield_reg, V8AtomicOperationUnderLockSpinCount); | |
600 __ br(Assembler::less, false, Assembler::pt, dontyield); | |
601 __ delayed()->nop(); | |
602 | |
603 // This code can only be called from inside the VM, this | |
604 // stub is only invoked from Atomic::add(). We do not | |
605 // want to use call_VM, because _last_java_sp and such | |
606 // must already be set. | |
607 // | |
608 // Save the regs and make space for a C call | |
609 __ save(SP, -96, SP); | |
610 __ save_all_globals_into_locals(); | |
611 BLOCK_COMMENT("call os::naked_sleep"); | |
612 __ call(CAST_FROM_FN_PTR(address, os::naked_sleep)); | |
613 __ delayed()->nop(); | |
614 __ restore_globals_from_locals(); | |
615 __ restore(); | |
616 // reset the counter | |
617 __ mov(G0,yield_reg); | |
618 | |
619 __ BIND(dontyield); | |
620 | |
621 // try to get lock | |
622 __ swap(lock_ptr_reg, 0, lock_reg); | |
623 | |
624 // did we get the lock? | |
625 __ cmp(lock_reg, StubRoutines::Sparc::unlocked); | |
626 __ br(Assembler::notEqual, true, Assembler::pn, retry); | |
627 __ delayed()->add(yield_reg,1,yield_reg); | |
628 | |
629 // yes, got lock. do the operation here. | |
630 } | |
631 | |
632 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) { | |
633 __ st(lock_reg, lock_ptr_reg, 0); // unlock | |
634 } | |
635 | |
636 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest). | |
637 // | |
638 // Arguments : | |
639 // | |
640 // exchange_value: O0 | |
641 // dest: O1 | |
642 // | |
643 // Results: | |
644 // | |
645 // O0: the value previously stored in dest | |
646 // | |
647 address generate_atomic_xchg() { | |
648 StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); | |
649 address start = __ pc(); | |
650 | |
651 if (UseCASForSwap) { | |
652 // Use CAS instead of swap, just in case the MP hardware | |
653 // prefers to work with just one kind of synch. instruction. | |
654 Label retry; | |
655 __ BIND(retry); | |
656 __ mov(O0, O3); // scratch copy of exchange value | |
657 __ ld(O1, 0, O2); // observe the previous value | |
658 // try to replace O2 with O3 | |
659 __ cas_under_lock(O1, O2, O3, | |
660 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(),false); | |
661 __ cmp(O2, O3); | |
662 __ br(Assembler::notEqual, false, Assembler::pn, retry); | |
663 __ delayed()->nop(); | |
664 | |
665 __ retl(false); | |
666 __ delayed()->mov(O2, O0); // report previous value to caller | |
667 | |
668 } else { | |
669 if (VM_Version::v9_instructions_work()) { | |
670 __ retl(false); | |
671 __ delayed()->swap(O1, 0, O0); | |
672 } else { | |
673 const Register& lock_reg = O2; | |
674 const Register& lock_ptr_reg = O3; | |
675 const Register& yield_reg = O4; | |
676 | |
677 Label retry; | |
678 Label dontyield; | |
679 | |
680 generate_v8_lock_prologue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield); | |
681 // got the lock, do the swap | |
682 __ swap(O1, 0, O0); | |
683 | |
684 generate_v8_lock_epilogue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield); | |
685 __ retl(false); | |
686 __ delayed()->nop(); | |
687 } | |
688 } | |
689 | |
690 return start; | |
691 } | |
692 | |
693 | |
694 // Support for jint Atomic::cmpxchg(jint exchange_value, volatile jint* dest, jint compare_value) | |
695 // | |
696 // Arguments : | |
697 // | |
698 // exchange_value: O0 | |
699 // dest: O1 | |
700 // compare_value: O2 | |
701 // | |
702 // Results: | |
703 // | |
704 // O0: the value previously stored in dest | |
705 // | |
706 // Overwrites (v8): O3,O4,O5 | |
707 // | |
708 address generate_atomic_cmpxchg() { | |
709 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg"); | |
710 address start = __ pc(); | |
711 | |
712 // cmpxchg(dest, compare_value, exchange_value) | |
713 __ cas_under_lock(O1, O2, O0, | |
714 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr(),false); | |
715 __ retl(false); | |
716 __ delayed()->nop(); | |
717 | |
718 return start; | |
719 } | |
720 | |
721 // Support for jlong Atomic::cmpxchg(jlong exchange_value, volatile jlong *dest, jlong compare_value) | |
722 // | |
723 // Arguments : | |
724 // | |
725 // exchange_value: O1:O0 | |
726 // dest: O2 | |
727 // compare_value: O4:O3 | |
728 // | |
729 // Results: | |
730 // | |
731 // O1:O0: the value previously stored in dest | |
732 // | |
733 // This only works on V9, on V8 we don't generate any | |
734 // code and just return NULL. | |
735 // | |
736 // Overwrites: G1,G2,G3 | |
737 // | |
738 address generate_atomic_cmpxchg_long() { | |
739 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long"); | |
740 address start = __ pc(); | |
741 | |
742 if (!VM_Version::supports_cx8()) | |
743 return NULL;; | |
744 __ sllx(O0, 32, O0); | |
745 __ srl(O1, 0, O1); | |
746 __ or3(O0,O1,O0); // O0 holds 64-bit value from compare_value | |
747 __ sllx(O3, 32, O3); | |
748 __ srl(O4, 0, O4); | |
749 __ or3(O3,O4,O3); // O3 holds 64-bit value from exchange_value | |
750 __ casx(O2, O3, O0); | |
751 __ srl(O0, 0, O1); // unpacked return value in O1:O0 | |
752 __ retl(false); | |
753 __ delayed()->srlx(O0, 32, O0); | |
754 | |
755 return start; | |
756 } | |
757 | |
758 | |
759 // Support for jint Atomic::add(jint add_value, volatile jint* dest). | |
760 // | |
761 // Arguments : | |
762 // | |
763 // add_value: O0 (e.g., +1 or -1) | |
764 // dest: O1 | |
765 // | |
766 // Results: | |
767 // | |
768 // O0: the new value stored in dest | |
769 // | |
770 // Overwrites (v9): O3 | |
771 // Overwrites (v8): O3,O4,O5 | |
772 // | |
773 address generate_atomic_add() { | |
774 StubCodeMark mark(this, "StubRoutines", "atomic_add"); | |
775 address start = __ pc(); | |
776 __ BIND(_atomic_add_stub); | |
777 | |
778 if (VM_Version::v9_instructions_work()) { | |
779 Label(retry); | |
780 __ BIND(retry); | |
781 | |
782 __ lduw(O1, 0, O2); | |
783 __ add(O0, O2, O3); | |
784 __ cas(O1, O2, O3); | |
785 __ cmp( O2, O3); | |
786 __ br(Assembler::notEqual, false, Assembler::pn, retry); | |
787 __ delayed()->nop(); | |
788 __ retl(false); | |
789 __ delayed()->add(O0, O2, O0); // note that cas made O2==O3 | |
790 } else { | |
791 const Register& lock_reg = O2; | |
792 const Register& lock_ptr_reg = O3; | |
793 const Register& value_reg = O4; | |
794 const Register& yield_reg = O5; | |
795 | |
796 Label(retry); | |
797 Label(dontyield); | |
798 | |
799 generate_v8_lock_prologue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield); | |
800 // got lock, do the increment | |
801 __ ld(O1, 0, value_reg); | |
802 __ add(O0, value_reg, value_reg); | |
803 __ st(value_reg, O1, 0); | |
804 | |
805 // %%% only for RMO and PSO | |
806 __ membar(Assembler::StoreStore); | |
807 | |
808 generate_v8_lock_epilogue(lock_reg, lock_ptr_reg, yield_reg, retry, dontyield); | |
809 | |
810 __ retl(false); | |
811 __ delayed()->mov(value_reg, O0); | |
812 } | |
813 | |
814 return start; | |
815 } | |
816 Label _atomic_add_stub; // called from other stubs | |
817 | |
818 | |
819 // Support for void OrderAccess::fence(). | |
820 // | |
821 address generate_fence() { | |
822 StubCodeMark mark(this, "StubRoutines", "fence"); | |
823 address start = __ pc(); | |
824 | |
825 __ membar(Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore | | |
826 Assembler::StoreLoad | Assembler::StoreStore)); | |
827 __ retl(false); | |
828 __ delayed()->nop(); | |
829 | |
830 return start; | |
831 } | |
832 | |
833 | |
834 //------------------------------------------------------------------------------------------------------------------------ | |
835 // The following routine generates a subroutine to throw an asynchronous | |
836 // UnknownError when an unsafe access gets a fault that could not be | |
837 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.) | |
838 // | |
839 // Arguments : | |
840 // | |
841 // trapping PC: O7 | |
842 // | |
843 // Results: | |
844 // posts an asynchronous exception, skips the trapping instruction | |
845 // | |
846 | |
847 address generate_handler_for_unsafe_access() { | |
848 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); | |
849 address start = __ pc(); | |
850 | |
851 const int preserve_register_words = (64 * 2); | |
852 Address preserve_addr(FP, 0, (-preserve_register_words * wordSize) + STACK_BIAS); | |
853 | |
854 Register Lthread = L7_thread_cache; | |
855 int i; | |
856 | |
857 __ save_frame(0); | |
858 __ mov(G1, L1); | |
859 __ mov(G2, L2); | |
860 __ mov(G3, L3); | |
861 __ mov(G4, L4); | |
862 __ mov(G5, L5); | |
863 for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) { | |
864 __ stf(FloatRegisterImpl::D, as_FloatRegister(i), preserve_addr, i * wordSize); | |
865 } | |
866 | |
867 address entry_point = CAST_FROM_FN_PTR(address, handle_unsafe_access); | |
868 BLOCK_COMMENT("call handle_unsafe_access"); | |
869 __ call(entry_point, relocInfo::runtime_call_type); | |
870 __ delayed()->nop(); | |
871 | |
872 __ mov(L1, G1); | |
873 __ mov(L2, G2); | |
874 __ mov(L3, G3); | |
875 __ mov(L4, G4); | |
876 __ mov(L5, G5); | |
877 for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) { | |
878 __ ldf(FloatRegisterImpl::D, preserve_addr, as_FloatRegister(i), i * wordSize); | |
879 } | |
880 | |
881 __ verify_thread(); | |
882 | |
883 __ jmp(O0, 0); | |
884 __ delayed()->restore(); | |
885 | |
886 return start; | |
887 } | |
888 | |
889 | |
890 // Support for uint StubRoutine::Sparc::partial_subtype_check( Klass sub, Klass super ); | |
891 // Arguments : | |
892 // | |
893 // ret : O0, returned | |
894 // icc/xcc: set as O0 (depending on wordSize) | |
895 // sub : O1, argument, not changed | |
896 // super: O2, argument, not changed | |
897 // raddr: O7, blown by call | |
898 address generate_partial_subtype_check() { | |
899 StubCodeMark mark(this, "StubRoutines", "partial_subtype_check"); | |
900 address start = __ pc(); | |
901 Label loop, miss; | |
902 | |
903 // Compare super with sub directly, since super is not in its own SSA. | |
904 // The compiler used to emit this test, but we fold it in here, | |
905 // to increase overall code density, with no real loss of speed. | |
906 { Label L; | |
907 __ cmp(O1, O2); | |
908 __ brx(Assembler::notEqual, false, Assembler::pt, L); | |
909 __ delayed()->nop(); | |
910 __ retl(); | |
911 __ delayed()->addcc(G0,0,O0); // set Z flags, zero result | |
912 __ bind(L); | |
913 } | |
914 | |
915 #if defined(COMPILER2) && !defined(_LP64) | |
916 // Do not use a 'save' because it blows the 64-bit O registers. | |
917 __ add(SP,-4*wordSize,SP); // Make space for 4 temps | |
918 __ st_ptr(L0,SP,(frame::register_save_words+0)*wordSize); | |
919 __ st_ptr(L1,SP,(frame::register_save_words+1)*wordSize); | |
920 __ st_ptr(L2,SP,(frame::register_save_words+2)*wordSize); | |
921 __ st_ptr(L3,SP,(frame::register_save_words+3)*wordSize); | |
922 Register Rret = O0; | |
923 Register Rsub = O1; | |
924 Register Rsuper = O2; | |
925 #else | |
926 __ save_frame(0); | |
927 Register Rret = I0; | |
928 Register Rsub = I1; | |
929 Register Rsuper = I2; | |
930 #endif | |
931 | |
932 Register L0_ary_len = L0; | |
933 Register L1_ary_ptr = L1; | |
934 Register L2_super = L2; | |
935 Register L3_index = L3; | |
936 | |
937 inc_counter_np(SharedRuntime::_partial_subtype_ctr, L0, L1); | |
938 | |
939 __ ld_ptr( Rsub, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes(), L3 ); | |
940 __ lduw(L3,arrayOopDesc::length_offset_in_bytes(),L0_ary_len); | |
941 __ add(L3,arrayOopDesc::base_offset_in_bytes(T_OBJECT),L1_ary_ptr); | |
942 __ clr(L3_index); // zero index | |
943 // Load a little early; will load 1 off the end of the array. | |
944 // Ok for now; revisit if we have other uses of this routine. | |
945 __ ld_ptr(L1_ary_ptr,0,L2_super);// Will load a little early | |
946 __ align(CodeEntryAlignment); | |
947 | |
948 // The scan loop | |
949 __ BIND(loop); | |
950 __ add(L1_ary_ptr,wordSize,L1_ary_ptr); // Bump by OOP size | |
951 __ cmp(L3_index,L0_ary_len); | |
952 __ br(Assembler::equal,false,Assembler::pn,miss); | |
953 __ delayed()->inc(L3_index); // Bump index | |
954 __ subcc(L2_super,Rsuper,Rret); // Check for match; zero in Rret for a hit | |
955 __ brx( Assembler::notEqual, false, Assembler::pt, loop ); | |
956 __ delayed()->ld_ptr(L1_ary_ptr,0,L2_super); // Will load a little early | |
957 | |
958 // Got a hit; report success; set cache. Cache load doesn't | |
959 // happen here; for speed it is directly emitted by the compiler. | |
960 __ st_ptr( Rsuper, Rsub, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() ); | |
961 | |
962 #if defined(COMPILER2) && !defined(_LP64) | |
963 __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0); | |
964 __ ld_ptr(SP,(frame::register_save_words+1)*wordSize,L1); | |
965 __ ld_ptr(SP,(frame::register_save_words+2)*wordSize,L2); | |
966 __ ld_ptr(SP,(frame::register_save_words+3)*wordSize,L3); | |
967 __ retl(); // Result in Rret is zero; flags set to Z | |
968 __ delayed()->add(SP,4*wordSize,SP); | |
969 #else | |
970 __ ret(); // Result in Rret is zero; flags set to Z | |
971 __ delayed()->restore(); | |
972 #endif | |
973 | |
974 // Hit or miss falls through here | |
975 __ BIND(miss); | |
976 __ addcc(G0,1,Rret); // set NZ flags, NZ result | |
977 | |
978 #if defined(COMPILER2) && !defined(_LP64) | |
979 __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0); | |
980 __ ld_ptr(SP,(frame::register_save_words+1)*wordSize,L1); | |
981 __ ld_ptr(SP,(frame::register_save_words+2)*wordSize,L2); | |
982 __ ld_ptr(SP,(frame::register_save_words+3)*wordSize,L3); | |
983 __ retl(); // Result in Rret is != 0; flags set to NZ | |
984 __ delayed()->add(SP,4*wordSize,SP); | |
985 #else | |
986 __ ret(); // Result in Rret is != 0; flags set to NZ | |
987 __ delayed()->restore(); | |
988 #endif | |
989 | |
990 return start; | |
991 } | |
992 | |
993 | |
994 // Called from MacroAssembler::verify_oop | |
995 // | |
996 address generate_verify_oop_subroutine() { | |
997 StubCodeMark mark(this, "StubRoutines", "verify_oop_stub"); | |
998 | |
999 address start = __ pc(); | |
1000 | |
1001 __ verify_oop_subroutine(); | |
1002 | |
1003 return start; | |
1004 } | |
1005 | |
1006 static address disjoint_byte_copy_entry; | |
1007 static address disjoint_short_copy_entry; | |
1008 static address disjoint_int_copy_entry; | |
1009 static address disjoint_long_copy_entry; | |
1010 static address disjoint_oop_copy_entry; | |
1011 | |
1012 static address byte_copy_entry; | |
1013 static address short_copy_entry; | |
1014 static address int_copy_entry; | |
1015 static address long_copy_entry; | |
1016 static address oop_copy_entry; | |
1017 | |
1018 static address checkcast_copy_entry; | |
1019 | |
1020 // | |
1021 // Verify that a register contains clean 32-bits positive value | |
1022 // (high 32-bits are 0) so it could be used in 64-bits shifts (sllx, srax). | |
1023 // | |
1024 // Input: | |
1025 // Rint - 32-bits value | |
1026 // Rtmp - scratch | |
1027 // | |
1028 void assert_clean_int(Register Rint, Register Rtmp) { | |
1029 #if defined(ASSERT) && defined(_LP64) | |
1030 __ signx(Rint, Rtmp); | |
1031 __ cmp(Rint, Rtmp); | |
1032 __ breakpoint_trap(Assembler::notEqual, Assembler::xcc); | |
1033 #endif | |
1034 } | |
1035 | |
1036 // | |
1037 // Generate overlap test for array copy stubs | |
1038 // | |
1039 // Input: | |
1040 // O0 - array1 | |
1041 // O1 - array2 | |
1042 // O2 - element count | |
1043 // | |
1044 // Kills temps: O3, O4 | |
1045 // | |
1046 void array_overlap_test(address no_overlap_target, int log2_elem_size) { | |
1047 assert(no_overlap_target != NULL, "must be generated"); | |
1048 array_overlap_test(no_overlap_target, NULL, log2_elem_size); | |
1049 } | |
1050 void array_overlap_test(Label& L_no_overlap, int log2_elem_size) { | |
1051 array_overlap_test(NULL, &L_no_overlap, log2_elem_size); | |
1052 } | |
1053 void array_overlap_test(address no_overlap_target, Label* NOLp, int log2_elem_size) { | |
1054 const Register from = O0; | |
1055 const Register to = O1; | |
1056 const Register count = O2; | |
1057 const Register to_from = O3; // to - from | |
1058 const Register byte_count = O4; // count << log2_elem_size | |
1059 | |
1060 __ subcc(to, from, to_from); | |
1061 __ sll_ptr(count, log2_elem_size, byte_count); | |
1062 if (NOLp == NULL) | |
1063 __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, no_overlap_target); | |
1064 else | |
1065 __ brx(Assembler::lessEqualUnsigned, false, Assembler::pt, (*NOLp)); | |
1066 __ delayed()->cmp(to_from, byte_count); | |
1067 if (NOLp == NULL) | |
1068 __ brx(Assembler::greaterEqual, false, Assembler::pt, no_overlap_target); | |
1069 else | |
1070 __ brx(Assembler::greaterEqual, false, Assembler::pt, (*NOLp)); | |
1071 __ delayed()->nop(); | |
1072 } | |
1073 | |
1074 // | |
1075 // Generate pre-write barrier for array. | |
1076 // | |
1077 // Input: | |
1078 // addr - register containing starting address | |
1079 // count - register containing element count | |
1080 // tmp - scratch register | |
1081 // | |
1082 // The input registers are overwritten. | |
1083 // | |
1084 void gen_write_ref_array_pre_barrier(Register addr, Register count) { | |
1085 #if 0 // G1 only | |
1086 BarrierSet* bs = Universe::heap()->barrier_set(); | |
1087 if (bs->has_write_ref_pre_barrier()) { | |
1088 assert(bs->has_write_ref_array_pre_opt(), | |
1089 "Else unsupported barrier set."); | |
1090 | |
1091 assert(addr->is_global() && count->is_global(), | |
1092 "If not, then we have to fix this code to handle more " | |
1093 "general cases."); | |
1094 // Get some new fresh output registers. | |
1095 __ save_frame(0); | |
1096 // Save the necessary global regs... will be used after. | |
1097 __ mov(addr, L0); | |
1098 __ mov(count, L1); | |
1099 | |
1100 __ mov(addr, O0); | |
1101 // Get the count into O1 | |
1102 __ call(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre)); | |
1103 __ delayed()->mov(count, O1); | |
1104 __ mov(L0, addr); | |
1105 __ mov(L1, count); | |
1106 __ restore(); | |
1107 } | |
1108 #endif // 0 | |
1109 } | |
1110 | |
1111 // | |
1112 // Generate post-write barrier for array. | |
1113 // | |
1114 // Input: | |
1115 // addr - register containing starting address | |
1116 // count - register containing element count | |
1117 // tmp - scratch register | |
1118 // | |
1119 // The input registers are overwritten. | |
1120 // | |
1121 void gen_write_ref_array_post_barrier(Register addr, Register count, | |
1122 Register tmp) { | |
1123 BarrierSet* bs = Universe::heap()->barrier_set(); | |
1124 | |
1125 switch (bs->kind()) { | |
1126 #if 0 // G1 - only | |
1127 case BarrierSet::G1SATBCT: | |
1128 case BarrierSet::G1SATBCTLogging: | |
1129 { | |
1130 assert(addr->is_global() && count->is_global(), | |
1131 "If not, then we have to fix this code to handle more " | |
1132 "general cases."); | |
1133 // Get some new fresh output registers. | |
1134 __ save_frame(0); | |
1135 __ mov(addr, O0); | |
1136 __ call(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)); | |
1137 __ delayed()->mov(count, O1); | |
1138 __ restore(); | |
1139 } | |
1140 break; | |
1141 #endif // 0 G1 - only | |
1142 case BarrierSet::CardTableModRef: | |
1143 case BarrierSet::CardTableExtension: | |
1144 { | |
1145 CardTableModRefBS* ct = (CardTableModRefBS*)bs; | |
1146 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); | |
1147 assert_different_registers(addr, count, tmp); | |
1148 | |
1149 Label L_loop; | |
1150 | |
1151 __ sll_ptr(count, LogBytesPerOop, count); | |
1152 __ sub(count, BytesPerOop, count); | |
1153 __ add(count, addr, count); | |
1154 // Use two shifts to clear out those low order two bits! (Cannot opt. into 1.) | |
1155 __ srl_ptr(addr, CardTableModRefBS::card_shift, addr); | |
1156 __ srl_ptr(count, CardTableModRefBS::card_shift, count); | |
1157 __ sub(count, addr, count); | |
1158 Address rs(tmp, (address)ct->byte_map_base); | |
1159 __ load_address(rs); | |
1160 __ BIND(L_loop); | |
1161 __ stb(G0, rs.base(), addr); | |
1162 __ subcc(count, 1, count); | |
1163 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop); | |
1164 __ delayed()->add(addr, 1, addr); | |
1165 | |
1166 } | |
1167 break; | |
1168 case BarrierSet::ModRef: | |
1169 break; | |
1170 default : | |
1171 ShouldNotReachHere(); | |
1172 | |
1173 } | |
1174 | |
1175 } | |
1176 | |
1177 | |
1178 // Copy big chunks forward with shift | |
1179 // | |
1180 // Inputs: | |
1181 // from - source arrays | |
1182 // to - destination array aligned to 8-bytes | |
1183 // count - elements count to copy >= the count equivalent to 16 bytes | |
1184 // count_dec - elements count's decrement equivalent to 16 bytes | |
1185 // L_copy_bytes - copy exit label | |
1186 // | |
1187 void copy_16_bytes_forward_with_shift(Register from, Register to, | |
1188 Register count, int count_dec, Label& L_copy_bytes) { | |
1189 Label L_loop, L_aligned_copy, L_copy_last_bytes; | |
1190 | |
1191 // if both arrays have the same alignment mod 8, do 8 bytes aligned copy | |
1192 __ andcc(from, 7, G1); // misaligned bytes | |
1193 __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy); | |
1194 __ delayed()->nop(); | |
1195 | |
1196 const Register left_shift = G1; // left shift bit counter | |
1197 const Register right_shift = G5; // right shift bit counter | |
1198 | |
1199 __ sll(G1, LogBitsPerByte, left_shift); | |
1200 __ mov(64, right_shift); | |
1201 __ sub(right_shift, left_shift, right_shift); | |
1202 | |
1203 // | |
1204 // Load 2 aligned 8-bytes chunks and use one from previous iteration | |
1205 // to form 2 aligned 8-bytes chunks to store. | |
1206 // | |
1207 __ deccc(count, count_dec); // Pre-decrement 'count' | |
1208 __ andn(from, 7, from); // Align address | |
1209 __ ldx(from, 0, O3); | |
1210 __ inc(from, 8); | |
1211 __ align(16); | |
1212 __ BIND(L_loop); | |
1213 __ ldx(from, 0, O4); | |
1214 __ deccc(count, count_dec); // Can we do next iteration after this one? | |
1215 __ ldx(from, 8, G4); | |
1216 __ inc(to, 16); | |
1217 __ inc(from, 16); | |
1218 __ sllx(O3, left_shift, O3); | |
1219 __ srlx(O4, right_shift, G3); | |
1220 __ bset(G3, O3); | |
1221 __ stx(O3, to, -16); | |
1222 __ sllx(O4, left_shift, O4); | |
1223 __ srlx(G4, right_shift, G3); | |
1224 __ bset(G3, O4); | |
1225 __ stx(O4, to, -8); | |
1226 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop); | |
1227 __ delayed()->mov(G4, O3); | |
1228 | |
1229 __ inccc(count, count_dec>>1 ); // + 8 bytes | |
1230 __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes); | |
1231 __ delayed()->inc(count, count_dec>>1); // restore 'count' | |
1232 | |
1233 // copy 8 bytes, part of them already loaded in O3 | |
1234 __ ldx(from, 0, O4); | |
1235 __ inc(to, 8); | |
1236 __ inc(from, 8); | |
1237 __ sllx(O3, left_shift, O3); | |
1238 __ srlx(O4, right_shift, G3); | |
1239 __ bset(O3, G3); | |
1240 __ stx(G3, to, -8); | |
1241 | |
1242 __ BIND(L_copy_last_bytes); | |
1243 __ srl(right_shift, LogBitsPerByte, right_shift); // misaligned bytes | |
1244 __ br(Assembler::always, false, Assembler::pt, L_copy_bytes); | |
1245 __ delayed()->sub(from, right_shift, from); // restore address | |
1246 | |
1247 __ BIND(L_aligned_copy); | |
1248 } | |
1249 | |
1250 // Copy big chunks backward with shift | |
1251 // | |
1252 // Inputs: | |
1253 // end_from - source arrays end address | |
1254 // end_to - destination array end address aligned to 8-bytes | |
1255 // count - elements count to copy >= the count equivalent to 16 bytes | |
1256 // count_dec - elements count's decrement equivalent to 16 bytes | |
1257 // L_aligned_copy - aligned copy exit label | |
1258 // L_copy_bytes - copy exit label | |
1259 // | |
1260 void copy_16_bytes_backward_with_shift(Register end_from, Register end_to, | |
1261 Register count, int count_dec, | |
1262 Label& L_aligned_copy, Label& L_copy_bytes) { | |
1263 Label L_loop, L_copy_last_bytes; | |
1264 | |
1265 // if both arrays have the same alignment mod 8, do 8 bytes aligned copy | |
1266 __ andcc(end_from, 7, G1); // misaligned bytes | |
1267 __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy); | |
1268 __ delayed()->deccc(count, count_dec); // Pre-decrement 'count' | |
1269 | |
1270 const Register left_shift = G1; // left shift bit counter | |
1271 const Register right_shift = G5; // right shift bit counter | |
1272 | |
1273 __ sll(G1, LogBitsPerByte, left_shift); | |
1274 __ mov(64, right_shift); | |
1275 __ sub(right_shift, left_shift, right_shift); | |
1276 | |
1277 // | |
1278 // Load 2 aligned 8-bytes chunks and use one from previous iteration | |
1279 // to form 2 aligned 8-bytes chunks to store. | |
1280 // | |
1281 __ andn(end_from, 7, end_from); // Align address | |
1282 __ ldx(end_from, 0, O3); | |
1283 __ align(16); | |
1284 __ BIND(L_loop); | |
1285 __ ldx(end_from, -8, O4); | |
1286 __ deccc(count, count_dec); // Can we do next iteration after this one? | |
1287 __ ldx(end_from, -16, G4); | |
1288 __ dec(end_to, 16); | |
1289 __ dec(end_from, 16); | |
1290 __ srlx(O3, right_shift, O3); | |
1291 __ sllx(O4, left_shift, G3); | |
1292 __ bset(G3, O3); | |
1293 __ stx(O3, end_to, 8); | |
1294 __ srlx(O4, right_shift, O4); | |
1295 __ sllx(G4, left_shift, G3); | |
1296 __ bset(G3, O4); | |
1297 __ stx(O4, end_to, 0); | |
1298 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_loop); | |
1299 __ delayed()->mov(G4, O3); | |
1300 | |
1301 __ inccc(count, count_dec>>1 ); // + 8 bytes | |
1302 __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes); | |
1303 __ delayed()->inc(count, count_dec>>1); // restore 'count' | |
1304 | |
1305 // copy 8 bytes, part of them already loaded in O3 | |
1306 __ ldx(end_from, -8, O4); | |
1307 __ dec(end_to, 8); | |
1308 __ dec(end_from, 8); | |
1309 __ srlx(O3, right_shift, O3); | |
1310 __ sllx(O4, left_shift, G3); | |
1311 __ bset(O3, G3); | |
1312 __ stx(G3, end_to, 0); | |
1313 | |
1314 __ BIND(L_copy_last_bytes); | |
1315 __ srl(left_shift, LogBitsPerByte, left_shift); // misaligned bytes | |
1316 __ br(Assembler::always, false, Assembler::pt, L_copy_bytes); | |
1317 __ delayed()->add(end_from, left_shift, end_from); // restore address | |
1318 } | |
1319 | |
1320 // | |
1321 // Generate stub for disjoint byte copy. If "aligned" is true, the | |
1322 // "from" and "to" addresses are assumed to be heapword aligned. | |
1323 // | |
1324 // Arguments for generated stub: | |
1325 // from: O0 | |
1326 // to: O1 | |
1327 // count: O2 treated as signed | |
1328 // | |
1329 address generate_disjoint_byte_copy(bool aligned, const char * name) { | |
1330 __ align(CodeEntryAlignment); | |
1331 StubCodeMark mark(this, "StubRoutines", name); | |
1332 address start = __ pc(); | |
1333 | |
1334 Label L_skip_alignment, L_align; | |
1335 Label L_copy_byte, L_copy_byte_loop, L_exit; | |
1336 | |
1337 const Register from = O0; // source array address | |
1338 const Register to = O1; // destination array address | |
1339 const Register count = O2; // elements count | |
1340 const Register offset = O5; // offset from start of arrays | |
1341 // O3, O4, G3, G4 are used as temp registers | |
1342 | |
1343 assert_clean_int(count, O3); // Make sure 'count' is clean int. | |
1344 | |
1345 if (!aligned) disjoint_byte_copy_entry = __ pc(); | |
1346 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
1347 if (!aligned) BLOCK_COMMENT("Entry:"); | |
1348 | |
1349 // for short arrays, just do single element copy | |
1350 __ cmp(count, 23); // 16 + 7 | |
1351 __ brx(Assembler::less, false, Assembler::pn, L_copy_byte); | |
1352 __ delayed()->mov(G0, offset); | |
1353 | |
1354 if (aligned) { | |
1355 // 'aligned' == true when it is known statically during compilation | |
1356 // of this arraycopy call site that both 'from' and 'to' addresses | |
1357 // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()). | |
1358 // | |
1359 // Aligned arrays have 4 bytes alignment in 32-bits VM | |
1360 // and 8 bytes - in 64-bits VM. So we do it only for 32-bits VM | |
1361 // | |
1362 #ifndef _LP64 | |
1363 // copy a 4-bytes word if necessary to align 'to' to 8 bytes | |
1364 __ andcc(to, 7, G0); | |
1365 __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment); | |
1366 __ delayed()->ld(from, 0, O3); | |
1367 __ inc(from, 4); | |
1368 __ inc(to, 4); | |
1369 __ dec(count, 4); | |
1370 __ st(O3, to, -4); | |
1371 __ BIND(L_skip_alignment); | |
1372 #endif | |
1373 } else { | |
1374 // copy bytes to align 'to' on 8 byte boundary | |
1375 __ andcc(to, 7, G1); // misaligned bytes | |
1376 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); | |
1377 __ delayed()->neg(G1); | |
1378 __ inc(G1, 8); // bytes need to copy to next 8-bytes alignment | |
1379 __ sub(count, G1, count); | |
1380 __ BIND(L_align); | |
1381 __ ldub(from, 0, O3); | |
1382 __ deccc(G1); | |
1383 __ inc(from); | |
1384 __ stb(O3, to, 0); | |
1385 __ br(Assembler::notZero, false, Assembler::pt, L_align); | |
1386 __ delayed()->inc(to); | |
1387 __ BIND(L_skip_alignment); | |
1388 } | |
1389 #ifdef _LP64 | |
1390 if (!aligned) | |
1391 #endif | |
1392 { | |
1393 // Copy with shift 16 bytes per iteration if arrays do not have | |
1394 // the same alignment mod 8, otherwise fall through to the next | |
1395 // code for aligned copy. | |
1396 // The compare above (count >= 23) guarantes 'count' >= 16 bytes. | |
1397 // Also jump over aligned copy after the copy with shift completed. | |
1398 | |
1399 copy_16_bytes_forward_with_shift(from, to, count, 16, L_copy_byte); | |
1400 } | |
1401 | |
1402 // Both array are 8 bytes aligned, copy 16 bytes at a time | |
1403 __ and3(count, 7, G4); // Save count | |
1404 __ srl(count, 3, count); | |
1405 generate_disjoint_long_copy_core(aligned); | |
1406 __ mov(G4, count); // Restore count | |
1407 | |
1408 // copy tailing bytes | |
1409 __ BIND(L_copy_byte); | |
1410 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit); | |
1411 __ delayed()->nop(); | |
1412 __ align(16); | |
1413 __ BIND(L_copy_byte_loop); | |
1414 __ ldub(from, offset, O3); | |
1415 __ deccc(count); | |
1416 __ stb(O3, to, offset); | |
1417 __ brx(Assembler::notZero, false, Assembler::pt, L_copy_byte_loop); | |
1418 __ delayed()->inc(offset); | |
1419 | |
1420 __ BIND(L_exit); | |
1421 // O3, O4 are used as temp registers | |
1422 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr, O3, O4); | |
1423 __ retl(); | |
1424 __ delayed()->mov(G0, O0); // return 0 | |
1425 return start; | |
1426 } | |
1427 | |
1428 // | |
1429 // Generate stub for conjoint byte copy. If "aligned" is true, the | |
1430 // "from" and "to" addresses are assumed to be heapword aligned. | |
1431 // | |
1432 // Arguments for generated stub: | |
1433 // from: O0 | |
1434 // to: O1 | |
1435 // count: O2 treated as signed | |
1436 // | |
1437 address generate_conjoint_byte_copy(bool aligned, const char * name) { | |
1438 // Do reverse copy. | |
1439 | |
1440 __ align(CodeEntryAlignment); | |
1441 StubCodeMark mark(this, "StubRoutines", name); | |
1442 address start = __ pc(); | |
1443 address nooverlap_target = aligned ? | |
1444 StubRoutines::arrayof_jbyte_disjoint_arraycopy() : | |
1445 disjoint_byte_copy_entry; | |
1446 | |
1447 Label L_skip_alignment, L_align, L_aligned_copy; | |
1448 Label L_copy_byte, L_copy_byte_loop, L_exit; | |
1449 | |
1450 const Register from = O0; // source array address | |
1451 const Register to = O1; // destination array address | |
1452 const Register count = O2; // elements count | |
1453 const Register end_from = from; // source array end address | |
1454 const Register end_to = to; // destination array end address | |
1455 | |
1456 assert_clean_int(count, O3); // Make sure 'count' is clean int. | |
1457 | |
1458 if (!aligned) byte_copy_entry = __ pc(); | |
1459 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
1460 if (!aligned) BLOCK_COMMENT("Entry:"); | |
1461 | |
1462 array_overlap_test(nooverlap_target, 0); | |
1463 | |
1464 __ add(to, count, end_to); // offset after last copied element | |
1465 | |
1466 // for short arrays, just do single element copy | |
1467 __ cmp(count, 23); // 16 + 7 | |
1468 __ brx(Assembler::less, false, Assembler::pn, L_copy_byte); | |
1469 __ delayed()->add(from, count, end_from); | |
1470 | |
1471 { | |
1472 // Align end of arrays since they could be not aligned even | |
1473 // when arrays itself are aligned. | |
1474 | |
1475 // copy bytes to align 'end_to' on 8 byte boundary | |
1476 __ andcc(end_to, 7, G1); // misaligned bytes | |
1477 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); | |
1478 __ delayed()->nop(); | |
1479 __ sub(count, G1, count); | |
1480 __ BIND(L_align); | |
1481 __ dec(end_from); | |
1482 __ dec(end_to); | |
1483 __ ldub(end_from, 0, O3); | |
1484 __ deccc(G1); | |
1485 __ brx(Assembler::notZero, false, Assembler::pt, L_align); | |
1486 __ delayed()->stb(O3, end_to, 0); | |
1487 __ BIND(L_skip_alignment); | |
1488 } | |
1489 #ifdef _LP64 | |
1490 if (aligned) { | |
1491 // Both arrays are aligned to 8-bytes in 64-bits VM. | |
1492 // The 'count' is decremented in copy_16_bytes_backward_with_shift() | |
1493 // in unaligned case. | |
1494 __ dec(count, 16); | |
1495 } else | |
1496 #endif | |
1497 { | |
1498 // Copy with shift 16 bytes per iteration if arrays do not have | |
1499 // the same alignment mod 8, otherwise jump to the next | |
1500 // code for aligned copy (and substracting 16 from 'count' before jump). | |
1501 // The compare above (count >= 11) guarantes 'count' >= 16 bytes. | |
1502 // Also jump over aligned copy after the copy with shift completed. | |
1503 | |
1504 copy_16_bytes_backward_with_shift(end_from, end_to, count, 16, | |
1505 L_aligned_copy, L_copy_byte); | |
1506 } | |
1507 // copy 4 elements (16 bytes) at a time | |
1508 __ align(16); | |
1509 __ BIND(L_aligned_copy); | |
1510 __ dec(end_from, 16); | |
1511 __ ldx(end_from, 8, O3); | |
1512 __ ldx(end_from, 0, O4); | |
1513 __ dec(end_to, 16); | |
1514 __ deccc(count, 16); | |
1515 __ stx(O3, end_to, 8); | |
1516 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy); | |
1517 __ delayed()->stx(O4, end_to, 0); | |
1518 __ inc(count, 16); | |
1519 | |
1520 // copy 1 element (2 bytes) at a time | |
1521 __ BIND(L_copy_byte); | |
1522 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit); | |
1523 __ delayed()->nop(); | |
1524 __ align(16); | |
1525 __ BIND(L_copy_byte_loop); | |
1526 __ dec(end_from); | |
1527 __ dec(end_to); | |
1528 __ ldub(end_from, 0, O4); | |
1529 __ deccc(count); | |
1530 __ brx(Assembler::greater, false, Assembler::pt, L_copy_byte_loop); | |
1531 __ delayed()->stb(O4, end_to, 0); | |
1532 | |
1533 __ BIND(L_exit); | |
1534 // O3, O4 are used as temp registers | |
1535 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr, O3, O4); | |
1536 __ retl(); | |
1537 __ delayed()->mov(G0, O0); // return 0 | |
1538 return start; | |
1539 } | |
1540 | |
1541 // | |
1542 // Generate stub for disjoint short copy. If "aligned" is true, the | |
1543 // "from" and "to" addresses are assumed to be heapword aligned. | |
1544 // | |
1545 // Arguments for generated stub: | |
1546 // from: O0 | |
1547 // to: O1 | |
1548 // count: O2 treated as signed | |
1549 // | |
1550 address generate_disjoint_short_copy(bool aligned, const char * name) { | |
1551 __ align(CodeEntryAlignment); | |
1552 StubCodeMark mark(this, "StubRoutines", name); | |
1553 address start = __ pc(); | |
1554 | |
1555 Label L_skip_alignment, L_skip_alignment2; | |
1556 Label L_copy_2_bytes, L_copy_2_bytes_loop, L_exit; | |
1557 | |
1558 const Register from = O0; // source array address | |
1559 const Register to = O1; // destination array address | |
1560 const Register count = O2; // elements count | |
1561 const Register offset = O5; // offset from start of arrays | |
1562 // O3, O4, G3, G4 are used as temp registers | |
1563 | |
1564 assert_clean_int(count, O3); // Make sure 'count' is clean int. | |
1565 | |
1566 if (!aligned) disjoint_short_copy_entry = __ pc(); | |
1567 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
1568 if (!aligned) BLOCK_COMMENT("Entry:"); | |
1569 | |
1570 // for short arrays, just do single element copy | |
1571 __ cmp(count, 11); // 8 + 3 (22 bytes) | |
1572 __ brx(Assembler::less, false, Assembler::pn, L_copy_2_bytes); | |
1573 __ delayed()->mov(G0, offset); | |
1574 | |
1575 if (aligned) { | |
1576 // 'aligned' == true when it is known statically during compilation | |
1577 // of this arraycopy call site that both 'from' and 'to' addresses | |
1578 // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()). | |
1579 // | |
1580 // Aligned arrays have 4 bytes alignment in 32-bits VM | |
1581 // and 8 bytes - in 64-bits VM. | |
1582 // | |
1583 #ifndef _LP64 | |
1584 // copy a 2-elements word if necessary to align 'to' to 8 bytes | |
1585 __ andcc(to, 7, G0); | |
1586 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); | |
1587 __ delayed()->ld(from, 0, O3); | |
1588 __ inc(from, 4); | |
1589 __ inc(to, 4); | |
1590 __ dec(count, 2); | |
1591 __ st(O3, to, -4); | |
1592 __ BIND(L_skip_alignment); | |
1593 #endif | |
1594 } else { | |
1595 // copy 1 element if necessary to align 'to' on an 4 bytes | |
1596 __ andcc(to, 3, G0); | |
1597 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); | |
1598 __ delayed()->lduh(from, 0, O3); | |
1599 __ inc(from, 2); | |
1600 __ inc(to, 2); | |
1601 __ dec(count); | |
1602 __ sth(O3, to, -2); | |
1603 __ BIND(L_skip_alignment); | |
1604 | |
1605 // copy 2 elements to align 'to' on an 8 byte boundary | |
1606 __ andcc(to, 7, G0); | |
1607 __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment2); | |
1608 __ delayed()->lduh(from, 0, O3); | |
1609 __ dec(count, 2); | |
1610 __ lduh(from, 2, O4); | |
1611 __ inc(from, 4); | |
1612 __ inc(to, 4); | |
1613 __ sth(O3, to, -4); | |
1614 __ sth(O4, to, -2); | |
1615 __ BIND(L_skip_alignment2); | |
1616 } | |
1617 #ifdef _LP64 | |
1618 if (!aligned) | |
1619 #endif | |
1620 { | |
1621 // Copy with shift 16 bytes per iteration if arrays do not have | |
1622 // the same alignment mod 8, otherwise fall through to the next | |
1623 // code for aligned copy. | |
1624 // The compare above (count >= 11) guarantes 'count' >= 16 bytes. | |
1625 // Also jump over aligned copy after the copy with shift completed. | |
1626 | |
1627 copy_16_bytes_forward_with_shift(from, to, count, 8, L_copy_2_bytes); | |
1628 } | |
1629 | |
1630 // Both array are 8 bytes aligned, copy 16 bytes at a time | |
1631 __ and3(count, 3, G4); // Save | |
1632 __ srl(count, 2, count); | |
1633 generate_disjoint_long_copy_core(aligned); | |
1634 __ mov(G4, count); // restore | |
1635 | |
1636 // copy 1 element at a time | |
1637 __ BIND(L_copy_2_bytes); | |
1638 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit); | |
1639 __ delayed()->nop(); | |
1640 __ align(16); | |
1641 __ BIND(L_copy_2_bytes_loop); | |
1642 __ lduh(from, offset, O3); | |
1643 __ deccc(count); | |
1644 __ sth(O3, to, offset); | |
1645 __ brx(Assembler::notZero, false, Assembler::pt, L_copy_2_bytes_loop); | |
1646 __ delayed()->inc(offset, 2); | |
1647 | |
1648 __ BIND(L_exit); | |
1649 // O3, O4 are used as temp registers | |
1650 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr, O3, O4); | |
1651 __ retl(); | |
1652 __ delayed()->mov(G0, O0); // return 0 | |
1653 return start; | |
1654 } | |
1655 | |
1656 // | |
1657 // Generate stub for conjoint short copy. If "aligned" is true, the | |
1658 // "from" and "to" addresses are assumed to be heapword aligned. | |
1659 // | |
1660 // Arguments for generated stub: | |
1661 // from: O0 | |
1662 // to: O1 | |
1663 // count: O2 treated as signed | |
1664 // | |
1665 address generate_conjoint_short_copy(bool aligned, const char * name) { | |
1666 // Do reverse copy. | |
1667 | |
1668 __ align(CodeEntryAlignment); | |
1669 StubCodeMark mark(this, "StubRoutines", name); | |
1670 address start = __ pc(); | |
1671 address nooverlap_target = aligned ? | |
1672 StubRoutines::arrayof_jshort_disjoint_arraycopy() : | |
1673 disjoint_short_copy_entry; | |
1674 | |
1675 Label L_skip_alignment, L_skip_alignment2, L_aligned_copy; | |
1676 Label L_copy_2_bytes, L_copy_2_bytes_loop, L_exit; | |
1677 | |
1678 const Register from = O0; // source array address | |
1679 const Register to = O1; // destination array address | |
1680 const Register count = O2; // elements count | |
1681 const Register end_from = from; // source array end address | |
1682 const Register end_to = to; // destination array end address | |
1683 | |
1684 const Register byte_count = O3; // bytes count to copy | |
1685 | |
1686 assert_clean_int(count, O3); // Make sure 'count' is clean int. | |
1687 | |
1688 if (!aligned) short_copy_entry = __ pc(); | |
1689 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
1690 if (!aligned) BLOCK_COMMENT("Entry:"); | |
1691 | |
1692 array_overlap_test(nooverlap_target, 1); | |
1693 | |
1694 __ sllx(count, LogBytesPerShort, byte_count); | |
1695 __ add(to, byte_count, end_to); // offset after last copied element | |
1696 | |
1697 // for short arrays, just do single element copy | |
1698 __ cmp(count, 11); // 8 + 3 (22 bytes) | |
1699 __ brx(Assembler::less, false, Assembler::pn, L_copy_2_bytes); | |
1700 __ delayed()->add(from, byte_count, end_from); | |
1701 | |
1702 { | |
1703 // Align end of arrays since they could be not aligned even | |
1704 // when arrays itself are aligned. | |
1705 | |
1706 // copy 1 element if necessary to align 'end_to' on an 4 bytes | |
1707 __ andcc(end_to, 3, G0); | |
1708 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); | |
1709 __ delayed()->lduh(end_from, -2, O3); | |
1710 __ dec(end_from, 2); | |
1711 __ dec(end_to, 2); | |
1712 __ dec(count); | |
1713 __ sth(O3, end_to, 0); | |
1714 __ BIND(L_skip_alignment); | |
1715 | |
1716 // copy 2 elements to align 'end_to' on an 8 byte boundary | |
1717 __ andcc(end_to, 7, G0); | |
1718 __ br(Assembler::zero, false, Assembler::pn, L_skip_alignment2); | |
1719 __ delayed()->lduh(end_from, -2, O3); | |
1720 __ dec(count, 2); | |
1721 __ lduh(end_from, -4, O4); | |
1722 __ dec(end_from, 4); | |
1723 __ dec(end_to, 4); | |
1724 __ sth(O3, end_to, 2); | |
1725 __ sth(O4, end_to, 0); | |
1726 __ BIND(L_skip_alignment2); | |
1727 } | |
1728 #ifdef _LP64 | |
1729 if (aligned) { | |
1730 // Both arrays are aligned to 8-bytes in 64-bits VM. | |
1731 // The 'count' is decremented in copy_16_bytes_backward_with_shift() | |
1732 // in unaligned case. | |
1733 __ dec(count, 8); | |
1734 } else | |
1735 #endif | |
1736 { | |
1737 // Copy with shift 16 bytes per iteration if arrays do not have | |
1738 // the same alignment mod 8, otherwise jump to the next | |
1739 // code for aligned copy (and substracting 8 from 'count' before jump). | |
1740 // The compare above (count >= 11) guarantes 'count' >= 16 bytes. | |
1741 // Also jump over aligned copy after the copy with shift completed. | |
1742 | |
1743 copy_16_bytes_backward_with_shift(end_from, end_to, count, 8, | |
1744 L_aligned_copy, L_copy_2_bytes); | |
1745 } | |
1746 // copy 4 elements (16 bytes) at a time | |
1747 __ align(16); | |
1748 __ BIND(L_aligned_copy); | |
1749 __ dec(end_from, 16); | |
1750 __ ldx(end_from, 8, O3); | |
1751 __ ldx(end_from, 0, O4); | |
1752 __ dec(end_to, 16); | |
1753 __ deccc(count, 8); | |
1754 __ stx(O3, end_to, 8); | |
1755 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy); | |
1756 __ delayed()->stx(O4, end_to, 0); | |
1757 __ inc(count, 8); | |
1758 | |
1759 // copy 1 element (2 bytes) at a time | |
1760 __ BIND(L_copy_2_bytes); | |
1761 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit); | |
1762 __ delayed()->nop(); | |
1763 __ BIND(L_copy_2_bytes_loop); | |
1764 __ dec(end_from, 2); | |
1765 __ dec(end_to, 2); | |
1766 __ lduh(end_from, 0, O4); | |
1767 __ deccc(count); | |
1768 __ brx(Assembler::greater, false, Assembler::pt, L_copy_2_bytes_loop); | |
1769 __ delayed()->sth(O4, end_to, 0); | |
1770 | |
1771 __ BIND(L_exit); | |
1772 // O3, O4 are used as temp registers | |
1773 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr, O3, O4); | |
1774 __ retl(); | |
1775 __ delayed()->mov(G0, O0); // return 0 | |
1776 return start; | |
1777 } | |
1778 | |
1779 // | |
1780 // Generate core code for disjoint int copy (and oop copy on 32-bit). | |
1781 // If "aligned" is true, the "from" and "to" addresses are assumed | |
1782 // to be heapword aligned. | |
1783 // | |
1784 // Arguments: | |
1785 // from: O0 | |
1786 // to: O1 | |
1787 // count: O2 treated as signed | |
1788 // | |
1789 void generate_disjoint_int_copy_core(bool aligned) { | |
1790 | |
1791 Label L_skip_alignment, L_aligned_copy; | |
1792 Label L_copy_16_bytes, L_copy_4_bytes, L_copy_4_bytes_loop, L_exit; | |
1793 | |
1794 const Register from = O0; // source array address | |
1795 const Register to = O1; // destination array address | |
1796 const Register count = O2; // elements count | |
1797 const Register offset = O5; // offset from start of arrays | |
1798 // O3, O4, G3, G4 are used as temp registers | |
1799 | |
1800 // 'aligned' == true when it is known statically during compilation | |
1801 // of this arraycopy call site that both 'from' and 'to' addresses | |
1802 // are HeapWordSize aligned (see LibraryCallKit::basictype2arraycopy()). | |
1803 // | |
1804 // Aligned arrays have 4 bytes alignment in 32-bits VM | |
1805 // and 8 bytes - in 64-bits VM. | |
1806 // | |
1807 #ifdef _LP64 | |
1808 if (!aligned) | |
1809 #endif | |
1810 { | |
1811 // The next check could be put under 'ifndef' since the code in | |
1812 // generate_disjoint_long_copy_core() has own checks and set 'offset'. | |
1813 | |
1814 // for short arrays, just do single element copy | |
1815 __ cmp(count, 5); // 4 + 1 (20 bytes) | |
1816 __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_4_bytes); | |
1817 __ delayed()->mov(G0, offset); | |
1818 | |
1819 // copy 1 element to align 'to' on an 8 byte boundary | |
1820 __ andcc(to, 7, G0); | |
1821 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); | |
1822 __ delayed()->ld(from, 0, O3); | |
1823 __ inc(from, 4); | |
1824 __ inc(to, 4); | |
1825 __ dec(count); | |
1826 __ st(O3, to, -4); | |
1827 __ BIND(L_skip_alignment); | |
1828 | |
1829 // if arrays have same alignment mod 8, do 4 elements copy | |
1830 __ andcc(from, 7, G0); | |
1831 __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy); | |
1832 __ delayed()->ld(from, 0, O3); | |
1833 | |
1834 // | |
1835 // Load 2 aligned 8-bytes chunks and use one from previous iteration | |
1836 // to form 2 aligned 8-bytes chunks to store. | |
1837 // | |
1838 // copy_16_bytes_forward_with_shift() is not used here since this | |
1839 // code is more optimal. | |
1840 | |
1841 // copy with shift 4 elements (16 bytes) at a time | |
1842 __ dec(count, 4); // The cmp at the beginning guaranty count >= 4 | |
1843 | |
1844 __ align(16); | |
1845 __ BIND(L_copy_16_bytes); | |
1846 __ ldx(from, 4, O4); | |
1847 __ deccc(count, 4); // Can we do next iteration after this one? | |
1848 __ ldx(from, 12, G4); | |
1849 __ inc(to, 16); | |
1850 __ inc(from, 16); | |
1851 __ sllx(O3, 32, O3); | |
1852 __ srlx(O4, 32, G3); | |
1853 __ bset(G3, O3); | |
1854 __ stx(O3, to, -16); | |
1855 __ sllx(O4, 32, O4); | |
1856 __ srlx(G4, 32, G3); | |
1857 __ bset(G3, O4); | |
1858 __ stx(O4, to, -8); | |
1859 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes); | |
1860 __ delayed()->mov(G4, O3); | |
1861 | |
1862 __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes); | |
1863 __ delayed()->inc(count, 4); // restore 'count' | |
1864 | |
1865 __ BIND(L_aligned_copy); | |
1866 } | |
1867 // copy 4 elements (16 bytes) at a time | |
1868 __ and3(count, 1, G4); // Save | |
1869 __ srl(count, 1, count); | |
1870 generate_disjoint_long_copy_core(aligned); | |
1871 __ mov(G4, count); // Restore | |
1872 | |
1873 // copy 1 element at a time | |
1874 __ BIND(L_copy_4_bytes); | |
1875 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit); | |
1876 __ delayed()->nop(); | |
1877 __ BIND(L_copy_4_bytes_loop); | |
1878 __ ld(from, offset, O3); | |
1879 __ deccc(count); | |
1880 __ st(O3, to, offset); | |
1881 __ brx(Assembler::notZero, false, Assembler::pt, L_copy_4_bytes_loop); | |
1882 __ delayed()->inc(offset, 4); | |
1883 __ BIND(L_exit); | |
1884 } | |
1885 | |
1886 // | |
1887 // Generate stub for disjoint int copy. If "aligned" is true, the | |
1888 // "from" and "to" addresses are assumed to be heapword aligned. | |
1889 // | |
1890 // Arguments for generated stub: | |
1891 // from: O0 | |
1892 // to: O1 | |
1893 // count: O2 treated as signed | |
1894 // | |
1895 address generate_disjoint_int_copy(bool aligned, const char * name) { | |
1896 __ align(CodeEntryAlignment); | |
1897 StubCodeMark mark(this, "StubRoutines", name); | |
1898 address start = __ pc(); | |
1899 | |
1900 const Register count = O2; | |
1901 assert_clean_int(count, O3); // Make sure 'count' is clean int. | |
1902 | |
1903 if (!aligned) disjoint_int_copy_entry = __ pc(); | |
1904 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
1905 if (!aligned) BLOCK_COMMENT("Entry:"); | |
1906 | |
1907 generate_disjoint_int_copy_core(aligned); | |
1908 | |
1909 // O3, O4 are used as temp registers | |
1910 inc_counter_np(SharedRuntime::_jint_array_copy_ctr, O3, O4); | |
1911 __ retl(); | |
1912 __ delayed()->mov(G0, O0); // return 0 | |
1913 return start; | |
1914 } | |
1915 | |
1916 // | |
1917 // Generate core code for conjoint int copy (and oop copy on 32-bit). | |
1918 // If "aligned" is true, the "from" and "to" addresses are assumed | |
1919 // to be heapword aligned. | |
1920 // | |
1921 // Arguments: | |
1922 // from: O0 | |
1923 // to: O1 | |
1924 // count: O2 treated as signed | |
1925 // | |
1926 void generate_conjoint_int_copy_core(bool aligned) { | |
1927 // Do reverse copy. | |
1928 | |
1929 Label L_skip_alignment, L_aligned_copy; | |
1930 Label L_copy_16_bytes, L_copy_4_bytes, L_copy_4_bytes_loop, L_exit; | |
1931 | |
1932 const Register from = O0; // source array address | |
1933 const Register to = O1; // destination array address | |
1934 const Register count = O2; // elements count | |
1935 const Register end_from = from; // source array end address | |
1936 const Register end_to = to; // destination array end address | |
1937 // O3, O4, O5, G3 are used as temp registers | |
1938 | |
1939 const Register byte_count = O3; // bytes count to copy | |
1940 | |
1941 __ sllx(count, LogBytesPerInt, byte_count); | |
1942 __ add(to, byte_count, end_to); // offset after last copied element | |
1943 | |
1944 __ cmp(count, 5); // for short arrays, just do single element copy | |
1945 __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_4_bytes); | |
1946 __ delayed()->add(from, byte_count, end_from); | |
1947 | |
1948 // copy 1 element to align 'to' on an 8 byte boundary | |
1949 __ andcc(end_to, 7, G0); | |
1950 __ br(Assembler::zero, false, Assembler::pt, L_skip_alignment); | |
1951 __ delayed()->nop(); | |
1952 __ dec(count); | |
1953 __ dec(end_from, 4); | |
1954 __ dec(end_to, 4); | |
1955 __ ld(end_from, 0, O4); | |
1956 __ st(O4, end_to, 0); | |
1957 __ BIND(L_skip_alignment); | |
1958 | |
1959 // Check if 'end_from' and 'end_to' has the same alignment. | |
1960 __ andcc(end_from, 7, G0); | |
1961 __ br(Assembler::zero, false, Assembler::pt, L_aligned_copy); | |
1962 __ delayed()->dec(count, 4); // The cmp at the start guaranty cnt >= 4 | |
1963 | |
1964 // copy with shift 4 elements (16 bytes) at a time | |
1965 // | |
1966 // Load 2 aligned 8-bytes chunks and use one from previous iteration | |
1967 // to form 2 aligned 8-bytes chunks to store. | |
1968 // | |
1969 __ ldx(end_from, -4, O3); | |
1970 __ align(16); | |
1971 __ BIND(L_copy_16_bytes); | |
1972 __ ldx(end_from, -12, O4); | |
1973 __ deccc(count, 4); | |
1974 __ ldx(end_from, -20, O5); | |
1975 __ dec(end_to, 16); | |
1976 __ dec(end_from, 16); | |
1977 __ srlx(O3, 32, O3); | |
1978 __ sllx(O4, 32, G3); | |
1979 __ bset(G3, O3); | |
1980 __ stx(O3, end_to, 8); | |
1981 __ srlx(O4, 32, O4); | |
1982 __ sllx(O5, 32, G3); | |
1983 __ bset(O4, G3); | |
1984 __ stx(G3, end_to, 0); | |
1985 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes); | |
1986 __ delayed()->mov(O5, O3); | |
1987 | |
1988 __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes); | |
1989 __ delayed()->inc(count, 4); | |
1990 | |
1991 // copy 4 elements (16 bytes) at a time | |
1992 __ align(16); | |
1993 __ BIND(L_aligned_copy); | |
1994 __ dec(end_from, 16); | |
1995 __ ldx(end_from, 8, O3); | |
1996 __ ldx(end_from, 0, O4); | |
1997 __ dec(end_to, 16); | |
1998 __ deccc(count, 4); | |
1999 __ stx(O3, end_to, 8); | |
2000 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_aligned_copy); | |
2001 __ delayed()->stx(O4, end_to, 0); | |
2002 __ inc(count, 4); | |
2003 | |
2004 // copy 1 element (4 bytes) at a time | |
2005 __ BIND(L_copy_4_bytes); | |
2006 __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit); | |
2007 __ delayed()->nop(); | |
2008 __ BIND(L_copy_4_bytes_loop); | |
2009 __ dec(end_from, 4); | |
2010 __ dec(end_to, 4); | |
2011 __ ld(end_from, 0, O4); | |
2012 __ deccc(count); | |
2013 __ brx(Assembler::greater, false, Assembler::pt, L_copy_4_bytes_loop); | |
2014 __ delayed()->st(O4, end_to, 0); | |
2015 __ BIND(L_exit); | |
2016 } | |
2017 | |
2018 // | |
2019 // Generate stub for conjoint int copy. If "aligned" is true, the | |
2020 // "from" and "to" addresses are assumed to be heapword aligned. | |
2021 // | |
2022 // Arguments for generated stub: | |
2023 // from: O0 | |
2024 // to: O1 | |
2025 // count: O2 treated as signed | |
2026 // | |
2027 address generate_conjoint_int_copy(bool aligned, const char * name) { | |
2028 __ align(CodeEntryAlignment); | |
2029 StubCodeMark mark(this, "StubRoutines", name); | |
2030 address start = __ pc(); | |
2031 | |
2032 address nooverlap_target = aligned ? | |
2033 StubRoutines::arrayof_jint_disjoint_arraycopy() : | |
2034 disjoint_int_copy_entry; | |
2035 | |
2036 assert_clean_int(O2, O3); // Make sure 'count' is clean int. | |
2037 | |
2038 if (!aligned) int_copy_entry = __ pc(); | |
2039 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
2040 if (!aligned) BLOCK_COMMENT("Entry:"); | |
2041 | |
2042 array_overlap_test(nooverlap_target, 2); | |
2043 | |
2044 generate_conjoint_int_copy_core(aligned); | |
2045 | |
2046 // O3, O4 are used as temp registers | |
2047 inc_counter_np(SharedRuntime::_jint_array_copy_ctr, O3, O4); | |
2048 __ retl(); | |
2049 __ delayed()->mov(G0, O0); // return 0 | |
2050 return start; | |
2051 } | |
2052 | |
2053 // | |
2054 // Generate core code for disjoint long copy (and oop copy on 64-bit). | |
2055 // "aligned" is ignored, because we must make the stronger | |
2056 // assumption that both addresses are always 64-bit aligned. | |
2057 // | |
2058 // Arguments: | |
2059 // from: O0 | |
2060 // to: O1 | |
2061 // count: O2 treated as signed | |
2062 // | |
2063 void generate_disjoint_long_copy_core(bool aligned) { | |
2064 Label L_copy_8_bytes, L_copy_16_bytes, L_exit; | |
2065 const Register from = O0; // source array address | |
2066 const Register to = O1; // destination array address | |
2067 const Register count = O2; // elements count | |
2068 const Register offset0 = O4; // element offset | |
2069 const Register offset8 = O5; // next element offset | |
2070 | |
2071 __ deccc(count, 2); | |
2072 __ mov(G0, offset0); // offset from start of arrays (0) | |
2073 __ brx(Assembler::negative, false, Assembler::pn, L_copy_8_bytes ); | |
2074 __ delayed()->add(offset0, 8, offset8); | |
2075 __ align(16); | |
2076 __ BIND(L_copy_16_bytes); | |
2077 __ ldx(from, offset0, O3); | |
2078 __ ldx(from, offset8, G3); | |
2079 __ deccc(count, 2); | |
2080 __ stx(O3, to, offset0); | |
2081 __ inc(offset0, 16); | |
2082 __ stx(G3, to, offset8); | |
2083 __ brx(Assembler::greaterEqual, false, Assembler::pt, L_copy_16_bytes); | |
2084 __ delayed()->inc(offset8, 16); | |
2085 | |
2086 __ BIND(L_copy_8_bytes); | |
2087 __ inccc(count, 2); | |
2088 __ brx(Assembler::zero, true, Assembler::pn, L_exit ); | |
2089 __ delayed()->mov(offset0, offset8); // Set O5 used by other stubs | |
2090 __ ldx(from, offset0, O3); | |
2091 __ stx(O3, to, offset0); | |
2092 __ BIND(L_exit); | |
2093 } | |
2094 | |
2095 // | |
2096 // Generate stub for disjoint long copy. | |
2097 // "aligned" is ignored, because we must make the stronger | |
2098 // assumption that both addresses are always 64-bit aligned. | |
2099 // | |
2100 // Arguments for generated stub: | |
2101 // from: O0 | |
2102 // to: O1 | |
2103 // count: O2 treated as signed | |
2104 // | |
2105 address generate_disjoint_long_copy(bool aligned, const char * name) { | |
2106 __ align(CodeEntryAlignment); | |
2107 StubCodeMark mark(this, "StubRoutines", name); | |
2108 address start = __ pc(); | |
2109 | |
2110 assert_clean_int(O2, O3); // Make sure 'count' is clean int. | |
2111 | |
2112 if (!aligned) disjoint_long_copy_entry = __ pc(); | |
2113 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
2114 if (!aligned) BLOCK_COMMENT("Entry:"); | |
2115 | |
2116 generate_disjoint_long_copy_core(aligned); | |
2117 | |
2118 // O3, O4 are used as temp registers | |
2119 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr, O3, O4); | |
2120 __ retl(); | |
2121 __ delayed()->mov(G0, O0); // return 0 | |
2122 return start; | |
2123 } | |
2124 | |
2125 // | |
2126 // Generate core code for conjoint long copy (and oop copy on 64-bit). | |
2127 // "aligned" is ignored, because we must make the stronger | |
2128 // assumption that both addresses are always 64-bit aligned. | |
2129 // | |
2130 // Arguments: | |
2131 // from: O0 | |
2132 // to: O1 | |
2133 // count: O2 treated as signed | |
2134 // | |
2135 void generate_conjoint_long_copy_core(bool aligned) { | |
2136 // Do reverse copy. | |
2137 Label L_copy_8_bytes, L_copy_16_bytes, L_exit; | |
2138 const Register from = O0; // source array address | |
2139 const Register to = O1; // destination array address | |
2140 const Register count = O2; // elements count | |
2141 const Register offset8 = O4; // element offset | |
2142 const Register offset0 = O5; // previous element offset | |
2143 | |
2144 __ subcc(count, 1, count); | |
2145 __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_8_bytes ); | |
2146 __ delayed()->sllx(count, LogBytesPerLong, offset8); | |
2147 __ sub(offset8, 8, offset0); | |
2148 __ align(16); | |
2149 __ BIND(L_copy_16_bytes); | |
2150 __ ldx(from, offset8, O2); | |
2151 __ ldx(from, offset0, O3); | |
2152 __ stx(O2, to, offset8); | |
2153 __ deccc(offset8, 16); // use offset8 as counter | |
2154 __ stx(O3, to, offset0); | |
2155 __ brx(Assembler::greater, false, Assembler::pt, L_copy_16_bytes); | |
2156 __ delayed()->dec(offset0, 16); | |
2157 | |
2158 __ BIND(L_copy_8_bytes); | |
2159 __ brx(Assembler::negative, false, Assembler::pn, L_exit ); | |
2160 __ delayed()->nop(); | |
2161 __ ldx(from, 0, O3); | |
2162 __ stx(O3, to, 0); | |
2163 __ BIND(L_exit); | |
2164 } | |
2165 | |
2166 // Generate stub for conjoint long copy. | |
2167 // "aligned" is ignored, because we must make the stronger | |
2168 // assumption that both addresses are always 64-bit aligned. | |
2169 // | |
2170 // Arguments for generated stub: | |
2171 // from: O0 | |
2172 // to: O1 | |
2173 // count: O2 treated as signed | |
2174 // | |
2175 address generate_conjoint_long_copy(bool aligned, const char * name) { | |
2176 __ align(CodeEntryAlignment); | |
2177 StubCodeMark mark(this, "StubRoutines", name); | |
2178 address start = __ pc(); | |
2179 | |
2180 assert(!aligned, "usage"); | |
2181 address nooverlap_target = disjoint_long_copy_entry; | |
2182 | |
2183 assert_clean_int(O2, O3); // Make sure 'count' is clean int. | |
2184 | |
2185 if (!aligned) long_copy_entry = __ pc(); | |
2186 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory) | |
2187 if (!aligned) BLOCK_COMMENT("Entry:"); | |
2188 | |
2189 array_overlap_test(nooverlap_target, 3); | |
2190 | |
2191 generate_conjoint_long_copy_core(aligned); | |
2192 | |
2193 // O3, O4 are used as temp registers | |
2194 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr, O3, O4); | |
2195 __ retl(); | |
2196 __ delayed()->mov(G0, O0); // return 0 | |
2197 return start; | |
2198 } | |
2199 | |
2200 // Generate stub for disjoint oop copy. If "aligned" is true, the | |
2201 // "from" and "to" addresses are assumed to be heapword aligned. | |
2202 // | |
2203 // Arguments for generated stub: | |
2204 // from: O0 | |
2205 // to: O1 | |
2206 // count: O2 treated as signed | |
2207 // | |
2208 address generate_disjoint_oop_copy(bool aligned, const char * name) { | |
2209 | |
2210 const Register from = O0; // source array address | |
2211 const Register to = O1; // destination array address | |
2212 const Register count = O2; // elements count | |
2213 | |
2214 __ align(CodeEntryAlignment); | |
2215 StubCodeMark mark(this, "StubRoutines", name); | |
2216 address start = __ pc(); | |
2217 | |
2218 assert_clean_int(count, O3); // Make sure 'count' is clean int. | |
2219 | |
2220 if (!aligned) disjoint_oop_copy_entry = __ pc(); | |
2221 // caller can pass a 64-bit byte count here | |
2222 if (!aligned) BLOCK_COMMENT("Entry:"); | |
2223 | |
2224 // save arguments for barrier generation | |
2225 __ mov(to, G1); | |
2226 __ mov(count, G5); | |
2227 gen_write_ref_array_pre_barrier(G1, G5); | |
2228 #ifdef _LP64 | |
2229 generate_disjoint_long_copy_core(aligned); | |
2230 #else | |
2231 generate_disjoint_int_copy_core(aligned); | |
2232 #endif | |
2233 // O0 is used as temp register | |
2234 gen_write_ref_array_post_barrier(G1, G5, O0); | |
2235 | |
2236 // O3, O4 are used as temp registers | |
2237 inc_counter_np(SharedRuntime::_oop_array_copy_ctr, O3, O4); | |
2238 __ retl(); | |
2239 __ delayed()->mov(G0, O0); // return 0 | |
2240 return start; | |
2241 } | |
2242 | |
2243 // Generate stub for conjoint oop copy. If "aligned" is true, the | |
2244 // "from" and "to" addresses are assumed to be heapword aligned. | |
2245 // | |
2246 // Arguments for generated stub: | |
2247 // from: O0 | |
2248 // to: O1 | |
2249 // count: O2 treated as signed | |
2250 // | |
2251 address generate_conjoint_oop_copy(bool aligned, const char * name) { | |
2252 | |
2253 const Register from = O0; // source array address | |
2254 const Register to = O1; // destination array address | |
2255 const Register count = O2; // elements count | |
2256 | |
2257 __ align(CodeEntryAlignment); | |
2258 StubCodeMark mark(this, "StubRoutines", name); | |
2259 address start = __ pc(); | |
2260 | |
2261 assert_clean_int(count, O3); // Make sure 'count' is clean int. | |
2262 | |
2263 if (!aligned) oop_copy_entry = __ pc(); | |
2264 // caller can pass a 64-bit byte count here | |
2265 if (!aligned) BLOCK_COMMENT("Entry:"); | |
2266 | |
2267 // save arguments for barrier generation | |
2268 __ mov(to, G1); | |
2269 __ mov(count, G5); | |
2270 | |
2271 gen_write_ref_array_pre_barrier(G1, G5); | |
2272 | |
2273 address nooverlap_target = aligned ? | |
2274 StubRoutines::arrayof_oop_disjoint_arraycopy() : | |
2275 disjoint_oop_copy_entry; | |
2276 | |
2277 array_overlap_test(nooverlap_target, LogBytesPerWord); | |
2278 | |
2279 #ifdef _LP64 | |
2280 generate_conjoint_long_copy_core(aligned); | |
2281 #else | |
2282 generate_conjoint_int_copy_core(aligned); | |
2283 #endif | |
2284 | |
2285 // O0 is used as temp register | |
2286 gen_write_ref_array_post_barrier(G1, G5, O0); | |
2287 | |
2288 // O3, O4 are used as temp registers | |
2289 inc_counter_np(SharedRuntime::_oop_array_copy_ctr, O3, O4); | |
2290 __ retl(); | |
2291 __ delayed()->mov(G0, O0); // return 0 | |
2292 return start; | |
2293 } | |
2294 | |
2295 | |
2296 // Helper for generating a dynamic type check. | |
2297 // Smashes only the given temp registers. | |
2298 void generate_type_check(Register sub_klass, | |
2299 Register super_check_offset, | |
2300 Register super_klass, | |
2301 Register temp, | |
2302 Label& L_success, | |
2303 Register deccc_hack = noreg) { | |
2304 assert_different_registers(sub_klass, super_check_offset, super_klass, temp); | |
2305 | |
2306 BLOCK_COMMENT("type_check:"); | |
2307 | |
2308 Label L_miss; | |
2309 | |
2310 assert_clean_int(super_check_offset, temp); | |
2311 | |
2312 // maybe decrement caller's trip count: | |
2313 #define DELAY_SLOT delayed(); \ | |
2314 { if (deccc_hack == noreg) __ nop(); else __ deccc(deccc_hack); } | |
2315 | |
2316 // if the pointers are equal, we are done (e.g., String[] elements) | |
2317 __ cmp(sub_klass, super_klass); | |
2318 __ brx(Assembler::equal, true, Assembler::pt, L_success); | |
2319 __ DELAY_SLOT; | |
2320 | |
2321 // check the supertype display: | |
2322 __ ld_ptr(sub_klass, super_check_offset, temp); // query the super type | |
2323 __ cmp(super_klass, temp); // test the super type | |
2324 __ brx(Assembler::equal, true, Assembler::pt, L_success); | |
2325 __ DELAY_SLOT; | |
2326 | |
2327 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + | |
2328 Klass::secondary_super_cache_offset_in_bytes()); | |
2329 __ cmp(super_klass, sc_offset); | |
2330 __ brx(Assembler::notEqual, true, Assembler::pt, L_miss); | |
2331 __ delayed()->nop(); | |
2332 | |
2333 __ save_frame(0); | |
2334 __ mov(sub_klass->after_save(), O1); | |
2335 // mov(super_klass->after_save(), O2); //fill delay slot | |
2336 assert(StubRoutines::Sparc::_partial_subtype_check != NULL, "order of generation"); | |
2337 __ call(StubRoutines::Sparc::_partial_subtype_check); | |
2338 __ delayed()->mov(super_klass->after_save(), O2); | |
2339 __ restore(); | |
2340 | |
2341 // Upon return, the condition codes are already set. | |
2342 __ brx(Assembler::equal, true, Assembler::pt, L_success); | |
2343 __ DELAY_SLOT; | |
2344 | |
2345 #undef DELAY_SLOT | |
2346 | |
2347 // Fall through on failure! | |
2348 __ BIND(L_miss); | |
2349 } | |
2350 | |
2351 | |
2352 // Generate stub for checked oop copy. | |
2353 // | |
2354 // Arguments for generated stub: | |
2355 // from: O0 | |
2356 // to: O1 | |
2357 // count: O2 treated as signed | |
2358 // ckoff: O3 (super_check_offset) | |
2359 // ckval: O4 (super_klass) | |
2360 // ret: O0 zero for success; (-1^K) where K is partial transfer count | |
2361 // | |
2362 address generate_checkcast_copy(const char* name) { | |
2363 | |
2364 const Register O0_from = O0; // source array address | |
2365 const Register O1_to = O1; // destination array address | |
2366 const Register O2_count = O2; // elements count | |
2367 const Register O3_ckoff = O3; // super_check_offset | |
2368 const Register O4_ckval = O4; // super_klass | |
2369 | |
2370 const Register O5_offset = O5; // loop var, with stride wordSize | |
2371 const Register G1_remain = G1; // loop var, with stride -1 | |
2372 const Register G3_oop = G3; // actual oop copied | |
2373 const Register G4_klass = G4; // oop._klass | |
2374 const Register G5_super = G5; // oop._klass._primary_supers[ckval] | |
2375 | |
2376 __ align(CodeEntryAlignment); | |
2377 StubCodeMark mark(this, "StubRoutines", name); | |
2378 address start = __ pc(); | |
2379 | |
2380 int klass_off = oopDesc::klass_offset_in_bytes(); | |
2381 | |
2382 gen_write_ref_array_pre_barrier(G1, G5); | |
2383 | |
2384 | |
2385 #ifdef ASSERT | |
2386 // We sometimes save a frame (see partial_subtype_check below). | |
2387 // If this will cause trouble, let's fail now instead of later. | |
2388 __ save_frame(0); | |
2389 __ restore(); | |
2390 #endif | |
2391 | |
2392 #ifdef ASSERT | |
2393 // caller guarantees that the arrays really are different | |
2394 // otherwise, we would have to make conjoint checks | |
2395 { Label L; | |
2396 __ mov(O3, G1); // spill: overlap test smashes O3 | |
2397 __ mov(O4, G4); // spill: overlap test smashes O4 | |
2398 array_overlap_test(L, LogBytesPerWord); | |
2399 __ stop("checkcast_copy within a single array"); | |
2400 __ bind(L); | |
2401 __ mov(G1, O3); | |
2402 __ mov(G4, O4); | |
2403 } | |
2404 #endif //ASSERT | |
2405 | |
2406 assert_clean_int(O2_count, G1); // Make sure 'count' is clean int. | |
2407 | |
2408 checkcast_copy_entry = __ pc(); | |
2409 // caller can pass a 64-bit byte count here (from generic stub) | |
2410 BLOCK_COMMENT("Entry:"); | |
2411 | |
2412 Label load_element, store_element, do_card_marks, fail, done; | |
2413 __ addcc(O2_count, 0, G1_remain); // initialize loop index, and test it | |
2414 __ brx(Assembler::notZero, false, Assembler::pt, load_element); | |
2415 __ delayed()->mov(G0, O5_offset); // offset from start of arrays | |
2416 | |
2417 // Empty array: Nothing to do. | |
2418 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4); | |
2419 __ retl(); | |
2420 __ delayed()->set(0, O0); // return 0 on (trivial) success | |
2421 | |
2422 // ======== begin loop ======== | |
2423 // (Loop is rotated; its entry is load_element.) | |
2424 // Loop variables: | |
2425 // (O5 = 0; ; O5 += wordSize) --- offset from src, dest arrays | |
2426 // (O2 = len; O2 != 0; O2--) --- number of oops *remaining* | |
2427 // G3, G4, G5 --- current oop, oop.klass, oop.klass.super | |
2428 __ align(16); | |
2429 | |
2430 __ bind(store_element); | |
2431 // deccc(G1_remain); // decrement the count (hoisted) | |
2432 __ st_ptr(G3_oop, O1_to, O5_offset); // store the oop | |
2433 __ inc(O5_offset, wordSize); // step to next offset | |
2434 __ brx(Assembler::zero, true, Assembler::pt, do_card_marks); | |
2435 __ delayed()->set(0, O0); // return -1 on success | |
2436 | |
2437 // ======== loop entry is here ======== | |
2438 __ bind(load_element); | |
2439 __ ld_ptr(O0_from, O5_offset, G3_oop); // load the oop | |
2440 __ br_null(G3_oop, true, Assembler::pt, store_element); | |
2441 __ delayed()->deccc(G1_remain); // decrement the count | |
2442 | |
2443 __ ld_ptr(G3_oop, klass_off, G4_klass); // query the object klass | |
2444 | |
2445 generate_type_check(G4_klass, O3_ckoff, O4_ckval, G5_super, | |
2446 // branch to this on success: | |
2447 store_element, | |
2448 // decrement this on success: | |
2449 G1_remain); | |
2450 // ======== end loop ======== | |
2451 | |
2452 // It was a real error; we must depend on the caller to finish the job. | |
2453 // Register G1 has number of *remaining* oops, O2 number of *total* oops. | |
2454 // Emit GC store barriers for the oops we have copied (O2 minus G1), | |
2455 // and report their number to the caller. | |
2456 __ bind(fail); | |
2457 __ subcc(O2_count, G1_remain, O2_count); | |
2458 __ brx(Assembler::zero, false, Assembler::pt, done); | |
2459 __ delayed()->not1(O2_count, O0); // report (-1^K) to caller | |
2460 | |
2461 __ bind(do_card_marks); | |
2462 gen_write_ref_array_post_barrier(O1_to, O2_count, O3); // store check on O1[0..O2] | |
2463 | |
2464 __ bind(done); | |
2465 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4); | |
2466 __ retl(); | |
2467 __ delayed()->nop(); // return value in 00 | |
2468 | |
2469 return start; | |
2470 } | |
2471 | |
2472 | |
2473 // Generate 'unsafe' array copy stub | |
2474 // Though just as safe as the other stubs, it takes an unscaled | |
2475 // size_t argument instead of an element count. | |
2476 // | |
2477 // Arguments for generated stub: | |
2478 // from: O0 | |
2479 // to: O1 | |
2480 // count: O2 byte count, treated as ssize_t, can be zero | |
2481 // | |
2482 // Examines the alignment of the operands and dispatches | |
2483 // to a long, int, short, or byte copy loop. | |
2484 // | |
2485 address generate_unsafe_copy(const char* name) { | |
2486 | |
2487 const Register O0_from = O0; // source array address | |
2488 const Register O1_to = O1; // destination array address | |
2489 const Register O2_count = O2; // elements count | |
2490 | |
2491 const Register G1_bits = G1; // test copy of low bits | |
2492 | |
2493 __ align(CodeEntryAlignment); | |
2494 StubCodeMark mark(this, "StubRoutines", name); | |
2495 address start = __ pc(); | |
2496 | |
2497 // bump this on entry, not on exit: | |
2498 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr, G1, G3); | |
2499 | |
2500 __ or3(O0_from, O1_to, G1_bits); | |
2501 __ or3(O2_count, G1_bits, G1_bits); | |
2502 | |
2503 __ btst(BytesPerLong-1, G1_bits); | |
2504 __ br(Assembler::zero, true, Assembler::pt, | |
2505 long_copy_entry, relocInfo::runtime_call_type); | |
2506 // scale the count on the way out: | |
2507 __ delayed()->srax(O2_count, LogBytesPerLong, O2_count); | |
2508 | |
2509 __ btst(BytesPerInt-1, G1_bits); | |
2510 __ br(Assembler::zero, true, Assembler::pt, | |
2511 int_copy_entry, relocInfo::runtime_call_type); | |
2512 // scale the count on the way out: | |
2513 __ delayed()->srax(O2_count, LogBytesPerInt, O2_count); | |
2514 | |
2515 __ btst(BytesPerShort-1, G1_bits); | |
2516 __ br(Assembler::zero, true, Assembler::pt, | |
2517 short_copy_entry, relocInfo::runtime_call_type); | |
2518 // scale the count on the way out: | |
2519 __ delayed()->srax(O2_count, LogBytesPerShort, O2_count); | |
2520 | |
2521 __ br(Assembler::always, false, Assembler::pt, | |
2522 byte_copy_entry, relocInfo::runtime_call_type); | |
2523 __ delayed()->nop(); | |
2524 | |
2525 return start; | |
2526 } | |
2527 | |
2528 | |
2529 // Perform range checks on the proposed arraycopy. | |
2530 // Kills the two temps, but nothing else. | |
2531 // Also, clean the sign bits of src_pos and dst_pos. | |
2532 void arraycopy_range_checks(Register src, // source array oop (O0) | |
2533 Register src_pos, // source position (O1) | |
2534 Register dst, // destination array oo (O2) | |
2535 Register dst_pos, // destination position (O3) | |
2536 Register length, // length of copy (O4) | |
2537 Register temp1, Register temp2, | |
2538 Label& L_failed) { | |
2539 BLOCK_COMMENT("arraycopy_range_checks:"); | |
2540 | |
2541 // if (src_pos + length > arrayOop(src)->length() ) FAIL; | |
2542 | |
2543 const Register array_length = temp1; // scratch | |
2544 const Register end_pos = temp2; // scratch | |
2545 | |
2546 // Note: This next instruction may be in the delay slot of a branch: | |
2547 __ add(length, src_pos, end_pos); // src_pos + length | |
2548 __ lduw(src, arrayOopDesc::length_offset_in_bytes(), array_length); | |
2549 __ cmp(end_pos, array_length); | |
2550 __ br(Assembler::greater, false, Assembler::pn, L_failed); | |
2551 | |
2552 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL; | |
2553 __ delayed()->add(length, dst_pos, end_pos); // dst_pos + length | |
2554 __ lduw(dst, arrayOopDesc::length_offset_in_bytes(), array_length); | |
2555 __ cmp(end_pos, array_length); | |
2556 __ br(Assembler::greater, false, Assembler::pn, L_failed); | |
2557 | |
2558 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'. | |
2559 // Move with sign extension can be used since they are positive. | |
2560 __ delayed()->signx(src_pos, src_pos); | |
2561 __ signx(dst_pos, dst_pos); | |
2562 | |
2563 BLOCK_COMMENT("arraycopy_range_checks done"); | |
2564 } | |
2565 | |
2566 | |
2567 // | |
2568 // Generate generic array copy stubs | |
2569 // | |
2570 // Input: | |
2571 // O0 - src oop | |
2572 // O1 - src_pos | |
2573 // O2 - dst oop | |
2574 // O3 - dst_pos | |
2575 // O4 - element count | |
2576 // | |
2577 // Output: | |
2578 // O0 == 0 - success | |
2579 // O0 == -1 - need to call System.arraycopy | |
2580 // | |
2581 address generate_generic_copy(const char *name) { | |
2582 | |
2583 Label L_failed, L_objArray; | |
2584 | |
2585 // Input registers | |
2586 const Register src = O0; // source array oop | |
2587 const Register src_pos = O1; // source position | |
2588 const Register dst = O2; // destination array oop | |
2589 const Register dst_pos = O3; // destination position | |
2590 const Register length = O4; // elements count | |
2591 | |
2592 // registers used as temp | |
2593 const Register G3_src_klass = G3; // source array klass | |
2594 const Register G4_dst_klass = G4; // destination array klass | |
2595 const Register G5_lh = G5; // layout handler | |
2596 const Register O5_temp = O5; | |
2597 | |
2598 __ align(CodeEntryAlignment); | |
2599 StubCodeMark mark(this, "StubRoutines", name); | |
2600 address start = __ pc(); | |
2601 | |
2602 // bump this on entry, not on exit: | |
2603 inc_counter_np(SharedRuntime::_generic_array_copy_ctr, G1, G3); | |
2604 | |
2605 // In principle, the int arguments could be dirty. | |
2606 //assert_clean_int(src_pos, G1); | |
2607 //assert_clean_int(dst_pos, G1); | |
2608 //assert_clean_int(length, G1); | |
2609 | |
2610 //----------------------------------------------------------------------- | |
2611 // Assembler stubs will be used for this call to arraycopy | |
2612 // if the following conditions are met: | |
2613 // | |
2614 // (1) src and dst must not be null. | |
2615 // (2) src_pos must not be negative. | |
2616 // (3) dst_pos must not be negative. | |
2617 // (4) length must not be negative. | |
2618 // (5) src klass and dst klass should be the same and not NULL. | |
2619 // (6) src and dst should be arrays. | |
2620 // (7) src_pos + length must not exceed length of src. | |
2621 // (8) dst_pos + length must not exceed length of dst. | |
2622 BLOCK_COMMENT("arraycopy initial argument checks"); | |
2623 | |
2624 // if (src == NULL) return -1; | |
2625 __ br_null(src, false, Assembler::pn, L_failed); | |
2626 | |
2627 // if (src_pos < 0) return -1; | |
2628 __ delayed()->tst(src_pos); | |
2629 __ br(Assembler::negative, false, Assembler::pn, L_failed); | |
2630 __ delayed()->nop(); | |
2631 | |
2632 // if (dst == NULL) return -1; | |
2633 __ br_null(dst, false, Assembler::pn, L_failed); | |
2634 | |
2635 // if (dst_pos < 0) return -1; | |
2636 __ delayed()->tst(dst_pos); | |
2637 __ br(Assembler::negative, false, Assembler::pn, L_failed); | |
2638 | |
2639 // if (length < 0) return -1; | |
2640 __ delayed()->tst(length); | |
2641 __ br(Assembler::negative, false, Assembler::pn, L_failed); | |
2642 | |
2643 BLOCK_COMMENT("arraycopy argument klass checks"); | |
2644 // get src->klass() | |
2645 __ delayed()->ld_ptr(src, oopDesc::klass_offset_in_bytes(), G3_src_klass); | |
2646 | |
2647 #ifdef ASSERT | |
2648 // assert(src->klass() != NULL); | |
2649 BLOCK_COMMENT("assert klasses not null"); | |
2650 { Label L_a, L_b; | |
2651 __ br_notnull(G3_src_klass, false, Assembler::pt, L_b); // it is broken if klass is NULL | |
2652 __ delayed()->ld_ptr(dst, oopDesc::klass_offset_in_bytes(), G4_dst_klass); | |
2653 __ bind(L_a); | |
2654 __ stop("broken null klass"); | |
2655 __ bind(L_b); | |
2656 __ br_null(G4_dst_klass, false, Assembler::pn, L_a); // this would be broken also | |
2657 __ delayed()->mov(G0, G4_dst_klass); // scribble the temp | |
2658 BLOCK_COMMENT("assert done"); | |
2659 } | |
2660 #endif | |
2661 | |
2662 // Load layout helper | |
2663 // | |
2664 // |array_tag| | header_size | element_type | |log2_element_size| | |
2665 // 32 30 24 16 8 2 0 | |
2666 // | |
2667 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 | |
2668 // | |
2669 | |
2670 int lh_offset = klassOopDesc::header_size() * HeapWordSize + | |
2671 Klass::layout_helper_offset_in_bytes(); | |
2672 | |
2673 // Load 32-bits signed value. Use br() instruction with it to check icc. | |
2674 __ lduw(G3_src_klass, lh_offset, G5_lh); | |
2675 | |
2676 // Handle objArrays completely differently... | |
2677 juint objArray_lh = Klass::array_layout_helper(T_OBJECT); | |
2678 __ set(objArray_lh, O5_temp); | |
2679 __ cmp(G5_lh, O5_temp); | |
2680 __ br(Assembler::equal, false, Assembler::pt, L_objArray); | |
2681 __ delayed()->ld_ptr(dst, oopDesc::klass_offset_in_bytes(), G4_dst_klass); | |
2682 | |
2683 // if (src->klass() != dst->klass()) return -1; | |
2684 __ cmp(G3_src_klass, G4_dst_klass); | |
2685 __ brx(Assembler::notEqual, false, Assembler::pn, L_failed); | |
2686 __ delayed()->nop(); | |
2687 | |
2688 // if (!src->is_Array()) return -1; | |
2689 __ cmp(G5_lh, Klass::_lh_neutral_value); // < 0 | |
2690 __ br(Assembler::greaterEqual, false, Assembler::pn, L_failed); | |
2691 | |
2692 // At this point, it is known to be a typeArray (array_tag 0x3). | |
2693 #ifdef ASSERT | |
2694 __ delayed()->nop(); | |
2695 { Label L; | |
2696 jint lh_prim_tag_in_place = (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift); | |
2697 __ set(lh_prim_tag_in_place, O5_temp); | |
2698 __ cmp(G5_lh, O5_temp); | |
2699 __ br(Assembler::greaterEqual, false, Assembler::pt, L); | |
2700 __ delayed()->nop(); | |
2701 __ stop("must be a primitive array"); | |
2702 __ bind(L); | |
2703 } | |
2704 #else | |
2705 __ delayed(); // match next insn to prev branch | |
2706 #endif | |
2707 | |
2708 arraycopy_range_checks(src, src_pos, dst, dst_pos, length, | |
2709 O5_temp, G4_dst_klass, L_failed); | |
2710 | |
2711 // typeArrayKlass | |
2712 // | |
2713 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); | |
2714 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); | |
2715 // | |
2716 | |
2717 const Register G4_offset = G4_dst_klass; // array offset | |
2718 const Register G3_elsize = G3_src_klass; // log2 element size | |
2719 | |
2720 __ srl(G5_lh, Klass::_lh_header_size_shift, G4_offset); | |
2721 __ and3(G4_offset, Klass::_lh_header_size_mask, G4_offset); // array_offset | |
2722 __ add(src, G4_offset, src); // src array offset | |
2723 __ add(dst, G4_offset, dst); // dst array offset | |
2724 __ and3(G5_lh, Klass::_lh_log2_element_size_mask, G3_elsize); // log2 element size | |
2725 | |
2726 // next registers should be set before the jump to corresponding stub | |
2727 const Register from = O0; // source array address | |
2728 const Register to = O1; // destination array address | |
2729 const Register count = O2; // elements count | |
2730 | |
2731 // 'from', 'to', 'count' registers should be set in this order | |
2732 // since they are the same as 'src', 'src_pos', 'dst'. | |
2733 | |
2734 BLOCK_COMMENT("scale indexes to element size"); | |
2735 __ sll_ptr(src_pos, G3_elsize, src_pos); | |
2736 __ sll_ptr(dst_pos, G3_elsize, dst_pos); | |
2737 __ add(src, src_pos, from); // src_addr | |
2738 __ add(dst, dst_pos, to); // dst_addr | |
2739 | |
2740 BLOCK_COMMENT("choose copy loop based on element size"); | |
2741 __ cmp(G3_elsize, 0); | |
2742 __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jbyte_arraycopy); | |
2743 __ delayed()->signx(length, count); // length | |
2744 | |
2745 __ cmp(G3_elsize, LogBytesPerShort); | |
2746 __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jshort_arraycopy); | |
2747 __ delayed()->signx(length, count); // length | |
2748 | |
2749 __ cmp(G3_elsize, LogBytesPerInt); | |
2750 __ br(Assembler::equal,true,Assembler::pt,StubRoutines::_jint_arraycopy); | |
2751 __ delayed()->signx(length, count); // length | |
2752 #ifdef ASSERT | |
2753 { Label L; | |
2754 __ cmp(G3_elsize, LogBytesPerLong); | |
2755 __ br(Assembler::equal, false, Assembler::pt, L); | |
2756 __ delayed()->nop(); | |
2757 __ stop("must be long copy, but elsize is wrong"); | |
2758 __ bind(L); | |
2759 } | |
2760 #endif | |
2761 __ br(Assembler::always,false,Assembler::pt,StubRoutines::_jlong_arraycopy); | |
2762 __ delayed()->signx(length, count); // length | |
2763 | |
2764 // objArrayKlass | |
2765 __ BIND(L_objArray); | |
2766 // live at this point: G3_src_klass, G4_dst_klass, src[_pos], dst[_pos], length | |
2767 | |
2768 Label L_plain_copy, L_checkcast_copy; | |
2769 // test array classes for subtyping | |
2770 __ cmp(G3_src_klass, G4_dst_klass); // usual case is exact equality | |
2771 __ brx(Assembler::notEqual, true, Assembler::pn, L_checkcast_copy); | |
2772 __ delayed()->lduw(G4_dst_klass, lh_offset, O5_temp); // hoisted from below | |
2773 | |
2774 // Identically typed arrays can be copied without element-wise checks. | |
2775 arraycopy_range_checks(src, src_pos, dst, dst_pos, length, | |
2776 O5_temp, G5_lh, L_failed); | |
2777 | |
2778 __ add(src, arrayOopDesc::base_offset_in_bytes(T_OBJECT), src); //src offset | |
2779 __ add(dst, arrayOopDesc::base_offset_in_bytes(T_OBJECT), dst); //dst offset | |
2780 __ sll_ptr(src_pos, LogBytesPerOop, src_pos); | |
2781 __ sll_ptr(dst_pos, LogBytesPerOop, dst_pos); | |
2782 __ add(src, src_pos, from); // src_addr | |
2783 __ add(dst, dst_pos, to); // dst_addr | |
2784 __ BIND(L_plain_copy); | |
2785 __ br(Assembler::always, false, Assembler::pt,StubRoutines::_oop_arraycopy); | |
2786 __ delayed()->signx(length, count); // length | |
2787 | |
2788 __ BIND(L_checkcast_copy); | |
2789 // live at this point: G3_src_klass, G4_dst_klass | |
2790 { | |
2791 // Before looking at dst.length, make sure dst is also an objArray. | |
2792 // lduw(G4_dst_klass, lh_offset, O5_temp); // hoisted to delay slot | |
2793 __ cmp(G5_lh, O5_temp); | |
2794 __ br(Assembler::notEqual, false, Assembler::pn, L_failed); | |
2795 | |
2796 // It is safe to examine both src.length and dst.length. | |
2797 __ delayed(); // match next insn to prev branch | |
2798 arraycopy_range_checks(src, src_pos, dst, dst_pos, length, | |
2799 O5_temp, G5_lh, L_failed); | |
2800 | |
2801 // Marshal the base address arguments now, freeing registers. | |
2802 __ add(src, arrayOopDesc::base_offset_in_bytes(T_OBJECT), src); //src offset | |
2803 __ add(dst, arrayOopDesc::base_offset_in_bytes(T_OBJECT), dst); //dst offset | |
2804 __ sll_ptr(src_pos, LogBytesPerOop, src_pos); | |
2805 __ sll_ptr(dst_pos, LogBytesPerOop, dst_pos); | |
2806 __ add(src, src_pos, from); // src_addr | |
2807 __ add(dst, dst_pos, to); // dst_addr | |
2808 __ signx(length, count); // length (reloaded) | |
2809 | |
2810 Register sco_temp = O3; // this register is free now | |
2811 assert_different_registers(from, to, count, sco_temp, | |
2812 G4_dst_klass, G3_src_klass); | |
2813 | |
2814 // Generate the type check. | |
2815 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + | |
2816 Klass::super_check_offset_offset_in_bytes()); | |
2817 __ lduw(G4_dst_klass, sco_offset, sco_temp); | |
2818 generate_type_check(G3_src_klass, sco_temp, G4_dst_klass, | |
2819 O5_temp, L_plain_copy); | |
2820 | |
2821 // Fetch destination element klass from the objArrayKlass header. | |
2822 int ek_offset = (klassOopDesc::header_size() * HeapWordSize + | |
2823 objArrayKlass::element_klass_offset_in_bytes()); | |
2824 | |
2825 // the checkcast_copy loop needs two extra arguments: | |
2826 __ ld_ptr(G4_dst_klass, ek_offset, O4); // dest elem klass | |
2827 // lduw(O4, sco_offset, O3); // sco of elem klass | |
2828 | |
2829 __ br(Assembler::always, false, Assembler::pt, checkcast_copy_entry); | |
2830 __ delayed()->lduw(O4, sco_offset, O3); | |
2831 } | |
2832 | |
2833 __ BIND(L_failed); | |
2834 __ retl(); | |
2835 __ delayed()->sub(G0, 1, O0); // return -1 | |
2836 return start; | |
2837 } | |
2838 | |
2839 void generate_arraycopy_stubs() { | |
2840 | |
2841 // Note: the disjoint stubs must be generated first, some of | |
2842 // the conjoint stubs use them. | |
2843 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, "jbyte_disjoint_arraycopy"); | |
2844 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, "jshort_disjoint_arraycopy"); | |
2845 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_copy(false, "jint_disjoint_arraycopy"); | |
2846 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_copy(false, "jlong_disjoint_arraycopy"); | |
2847 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_oop_copy(false, "oop_disjoint_arraycopy"); | |
2848 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(true, "arrayof_jbyte_disjoint_arraycopy"); | |
2849 StubRoutines::_arrayof_jshort_disjoint_arraycopy = generate_disjoint_short_copy(true, "arrayof_jshort_disjoint_arraycopy"); | |
2850 StubRoutines::_arrayof_jint_disjoint_arraycopy = generate_disjoint_int_copy(true, "arrayof_jint_disjoint_arraycopy"); | |
2851 StubRoutines::_arrayof_jlong_disjoint_arraycopy = generate_disjoint_long_copy(true, "arrayof_jlong_disjoint_arraycopy"); | |
2852 StubRoutines::_arrayof_oop_disjoint_arraycopy = generate_disjoint_oop_copy(true, "arrayof_oop_disjoint_arraycopy"); | |
2853 | |
2854 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, "jbyte_arraycopy"); | |
2855 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, "jshort_arraycopy"); | |
2856 StubRoutines::_jint_arraycopy = generate_conjoint_int_copy(false, "jint_arraycopy"); | |
2857 StubRoutines::_jlong_arraycopy = generate_conjoint_long_copy(false, "jlong_arraycopy"); | |
2858 StubRoutines::_oop_arraycopy = generate_conjoint_oop_copy(false, "oop_arraycopy"); | |
2859 StubRoutines::_arrayof_jbyte_arraycopy = generate_conjoint_byte_copy(true, "arrayof_jbyte_arraycopy"); | |
2860 StubRoutines::_arrayof_jshort_arraycopy = generate_conjoint_short_copy(true, "arrayof_jshort_arraycopy"); | |
2861 #ifdef _LP64 | |
2862 // since sizeof(jint) < sizeof(HeapWord), there's a different flavor: | |
2863 StubRoutines::_arrayof_jint_arraycopy = generate_conjoint_int_copy(true, "arrayof_jint_arraycopy"); | |
2864 #else | |
2865 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; | |
2866 #endif | |
2867 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; | |
2868 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; | |
2869 | |
2870 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy"); | |
2871 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy"); | |
2872 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy"); | |
2873 } | |
2874 | |
2875 void generate_initial() { | |
2876 // Generates all stubs and initializes the entry points | |
2877 | |
2878 //------------------------------------------------------------------------------------------------------------------------ | |
2879 // entry points that exist in all platforms | |
2880 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than | |
2881 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp. | |
2882 StubRoutines::_forward_exception_entry = generate_forward_exception(); | |
2883 | |
2884 StubRoutines::_call_stub_entry = generate_call_stub(StubRoutines::_call_stub_return_address); | |
2885 StubRoutines::_catch_exception_entry = generate_catch_exception(); | |
2886 | |
2887 //------------------------------------------------------------------------------------------------------------------------ | |
2888 // entry points that are platform specific | |
2889 StubRoutines::Sparc::_test_stop_entry = generate_test_stop(); | |
2890 | |
2891 StubRoutines::Sparc::_stop_subroutine_entry = generate_stop_subroutine(); | |
2892 StubRoutines::Sparc::_flush_callers_register_windows_entry = generate_flush_callers_register_windows(); | |
2893 | |
2894 #if !defined(COMPILER2) && !defined(_LP64) | |
2895 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); | |
2896 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg(); | |
2897 StubRoutines::_atomic_add_entry = generate_atomic_add(); | |
2898 StubRoutines::_atomic_xchg_ptr_entry = StubRoutines::_atomic_xchg_entry; | |
2899 StubRoutines::_atomic_cmpxchg_ptr_entry = StubRoutines::_atomic_cmpxchg_entry; | |
2900 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long(); | |
2901 StubRoutines::_atomic_add_ptr_entry = StubRoutines::_atomic_add_entry; | |
2902 StubRoutines::_fence_entry = generate_fence(); | |
2903 #endif // COMPILER2 !=> _LP64 | |
2904 | |
2905 StubRoutines::Sparc::_partial_subtype_check = generate_partial_subtype_check(); | |
2906 } | |
2907 | |
2908 | |
2909 void generate_all() { | |
2910 // Generates all stubs and initializes the entry points | |
2911 | |
2912 // These entry points require SharedInfo::stack0 to be set up in non-core builds | |
2913 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false); | |
2914 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true); | |
2915 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true); | |
2916 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); | |
2917 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false); | |
2918 | |
2919 StubRoutines::_handler_for_unsafe_access_entry = | |
2920 generate_handler_for_unsafe_access(); | |
2921 | |
2922 // support for verify_oop (must happen after universe_init) | |
2923 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop_subroutine(); | |
2924 | |
2925 // arraycopy stubs used by compilers | |
2926 generate_arraycopy_stubs(); | |
2927 } | |
2928 | |
2929 | |
2930 public: | |
2931 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { | |
2932 // replace the standard masm with a special one: | |
2933 _masm = new MacroAssembler(code); | |
2934 | |
2935 _stub_count = !all ? 0x100 : 0x200; | |
2936 if (all) { | |
2937 generate_all(); | |
2938 } else { | |
2939 generate_initial(); | |
2940 } | |
2941 | |
2942 // make sure this stub is available for all local calls | |
2943 if (_atomic_add_stub.is_unbound()) { | |
2944 // generate a second time, if necessary | |
2945 (void) generate_atomic_add(); | |
2946 } | |
2947 } | |
2948 | |
2949 | |
2950 private: | |
2951 int _stub_count; | |
2952 void stub_prolog(StubCodeDesc* cdesc) { | |
2953 # ifdef ASSERT | |
2954 // put extra information in the stub code, to make it more readable | |
2955 #ifdef _LP64 | |
2956 // Write the high part of the address | |
2957 // [RGV] Check if there is a dependency on the size of this prolog | |
2958 __ emit_data((intptr_t)cdesc >> 32, relocInfo::none); | |
2959 #endif | |
2960 __ emit_data((intptr_t)cdesc, relocInfo::none); | |
2961 __ emit_data(++_stub_count, relocInfo::none); | |
2962 # endif | |
2963 align(true); | |
2964 } | |
2965 | |
2966 void align(bool at_header = false) { | |
2967 // %%%%% move this constant somewhere else | |
2968 // UltraSPARC cache line size is 8 instructions: | |
2969 const unsigned int icache_line_size = 32; | |
2970 const unsigned int icache_half_line_size = 16; | |
2971 | |
2972 if (at_header) { | |
2973 while ((intptr_t)(__ pc()) % icache_line_size != 0) { | |
2974 __ emit_data(0, relocInfo::none); | |
2975 } | |
2976 } else { | |
2977 while ((intptr_t)(__ pc()) % icache_half_line_size != 0) { | |
2978 __ nop(); | |
2979 } | |
2980 } | |
2981 } | |
2982 | |
2983 }; // end class declaration | |
2984 | |
2985 | |
2986 address StubGenerator::disjoint_byte_copy_entry = NULL; | |
2987 address StubGenerator::disjoint_short_copy_entry = NULL; | |
2988 address StubGenerator::disjoint_int_copy_entry = NULL; | |
2989 address StubGenerator::disjoint_long_copy_entry = NULL; | |
2990 address StubGenerator::disjoint_oop_copy_entry = NULL; | |
2991 | |
2992 address StubGenerator::byte_copy_entry = NULL; | |
2993 address StubGenerator::short_copy_entry = NULL; | |
2994 address StubGenerator::int_copy_entry = NULL; | |
2995 address StubGenerator::long_copy_entry = NULL; | |
2996 address StubGenerator::oop_copy_entry = NULL; | |
2997 | |
2998 address StubGenerator::checkcast_copy_entry = NULL; | |
2999 | |
3000 void StubGenerator_generate(CodeBuffer* code, bool all) { | |
3001 StubGenerator g(code, all); | |
3002 } |