Mercurial > hg > graal-compiler
annotate src/cpu/sparc/vm/interp_masm_sparc.cpp @ 144:e3729351c946
6697534: Premature GC and invalid lgrp selection with NUMA-aware allocator.
Summary: Don't move tops of the chunks in ensure_parsibility(). Handle the situation with Solaris when a machine has a locality group with no memory.
Reviewed-by: apetrusenko, jcoomes, ysr
author | iveresov |
---|---|
date | Fri, 09 May 2008 16:34:08 +0400 |
parents | ba764ed4b6f2 |
children | d1605aabd0a1 |
rev | line source |
---|---|
0 | 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/_interp_masm_sparc.cpp.incl" | |
27 | |
28 #ifndef CC_INTERP | |
29 #ifndef FAST_DISPATCH | |
30 #define FAST_DISPATCH 1 | |
31 #endif | |
32 #undef FAST_DISPATCH | |
33 | |
34 // Implementation of InterpreterMacroAssembler | |
35 | |
36 // This file specializes the assember with interpreter-specific macros | |
37 | |
38 const Address InterpreterMacroAssembler::l_tmp( FP, 0, (frame::interpreter_frame_l_scratch_fp_offset * wordSize ) + STACK_BIAS); | |
39 const Address InterpreterMacroAssembler::d_tmp( FP, 0, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); | |
40 | |
41 #else // CC_INTERP | |
42 #ifndef STATE | |
43 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name)) | |
44 #endif // STATE | |
45 | |
46 #endif // CC_INTERP | |
47 | |
48 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { | |
49 // Note: this algorithm is also used by C1's OSR entry sequence. | |
50 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). | |
51 assert_different_registers(args_size, locals_size); | |
52 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. | |
53 if (TaggedStackInterpreter) sll(locals_size, 1, locals_size); | |
54 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words | |
55 // Use br/mov combination because it works on both V8 and V9 and is | |
56 // faster. | |
57 Label skip_move; | |
58 br(Assembler::negative, true, Assembler::pt, skip_move); | |
59 delayed()->mov(G0, delta); | |
60 bind(skip_move); | |
61 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) | |
62 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes | |
63 } | |
64 | |
65 #ifndef CC_INTERP | |
66 | |
67 // Dispatch code executed in the prolog of a bytecode which does not do it's | |
68 // own dispatch. The dispatch address is computed and placed in IdispatchAddress | |
69 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { | |
70 assert_not_delayed(); | |
71 #ifdef FAST_DISPATCH | |
72 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since | |
73 // they both use I2. | |
74 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); | |
75 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
76 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); | |
77 // add offset to correct dispatch table | |
78 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
79 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr | |
80 #else | |
81 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
82 // dispatch table to use | |
83 Address tbl(G3_scratch, (address)Interpreter::dispatch_table(state)); | |
84 | |
85 sethi(tbl); | |
86 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
87 add(tbl, tbl.base(), 0); | |
88 ld_ptr( G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr | |
89 #endif | |
90 } | |
91 | |
92 | |
93 // Dispatch code executed in the epilog of a bytecode which does not do it's | |
94 // own dispatch. The dispatch address in IdispatchAddress is used for the | |
95 // dispatch. | |
96 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { | |
97 assert_not_delayed(); | |
98 verify_FPU(1, state); | |
99 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
100 jmp( IdispatchAddress, 0 ); | |
101 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); | |
102 else delayed()->nop(); | |
103 } | |
104 | |
105 | |
106 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { | |
107 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) | |
108 assert_not_delayed(); | |
109 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
110 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr); | |
111 } | |
112 | |
113 | |
114 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { | |
115 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) | |
116 assert_not_delayed(); | |
117 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
118 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); | |
119 } | |
120 | |
121 | |
122 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { | |
123 // load current bytecode | |
124 assert_not_delayed(); | |
125 ldub( Lbcp, 0, Lbyte_code); // load next bytecode | |
126 dispatch_base(state, table); | |
127 } | |
128 | |
129 | |
130 void InterpreterMacroAssembler::call_VM_leaf_base( | |
131 Register java_thread, | |
132 address entry_point, | |
133 int number_of_arguments | |
134 ) { | |
135 if (!java_thread->is_valid()) | |
136 java_thread = L7_thread_cache; | |
137 // super call | |
138 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); | |
139 } | |
140 | |
141 | |
142 void InterpreterMacroAssembler::call_VM_base( | |
143 Register oop_result, | |
144 Register java_thread, | |
145 Register last_java_sp, | |
146 address entry_point, | |
147 int number_of_arguments, | |
148 bool check_exception | |
149 ) { | |
150 if (!java_thread->is_valid()) | |
151 java_thread = L7_thread_cache; | |
152 // See class ThreadInVMfromInterpreter, which assumes that the interpreter | |
153 // takes responsibility for setting its own thread-state on call-out. | |
154 // However, ThreadInVMfromInterpreter resets the state to "in_Java". | |
155 | |
156 //save_bcp(); // save bcp | |
157 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); | |
158 //restore_bcp(); // restore bcp | |
159 //restore_locals(); // restore locals pointer | |
160 } | |
161 | |
162 | |
163 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { | |
164 if (JvmtiExport::can_pop_frame()) { | |
165 Label L; | |
166 | |
167 // Check the "pending popframe condition" flag in the current thread | |
168 Address popframe_condition_addr(G2_thread, 0, in_bytes(JavaThread::popframe_condition_offset())); | |
169 ld(popframe_condition_addr, scratch_reg); | |
170 | |
171 // Initiate popframe handling only if it is not already being processed. If the flag | |
172 // has the popframe_processing bit set, it means that this code is called *during* popframe | |
173 // handling - we don't want to reenter. | |
174 btst(JavaThread::popframe_pending_bit, scratch_reg); | |
175 br(zero, false, pt, L); | |
176 delayed()->nop(); | |
177 btst(JavaThread::popframe_processing_bit, scratch_reg); | |
178 br(notZero, false, pt, L); | |
179 delayed()->nop(); | |
180 | |
181 // Call Interpreter::remove_activation_preserving_args_entry() to get the | |
182 // address of the same-named entrypoint in the generated interpreter code. | |
183 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); | |
184 | |
185 // Jump to Interpreter::_remove_activation_preserving_args_entry | |
186 jmpl(O0, G0, G0); | |
187 delayed()->nop(); | |
188 bind(L); | |
189 } | |
190 } | |
191 | |
192 | |
193 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { | |
194 Register thr_state = G4_scratch; | |
195 ld_ptr(Address(G2_thread, 0, in_bytes(JavaThread::jvmti_thread_state_offset())), | |
196 thr_state); | |
197 const Address tos_addr(thr_state, 0, in_bytes(JvmtiThreadState::earlyret_tos_offset())); | |
198 const Address oop_addr(thr_state, 0, in_bytes(JvmtiThreadState::earlyret_oop_offset())); | |
199 const Address val_addr(thr_state, 0, in_bytes(JvmtiThreadState::earlyret_value_offset())); | |
200 switch (state) { | |
201 case ltos: ld_long(val_addr, Otos_l); break; | |
202 case atos: ld_ptr(oop_addr, Otos_l); | |
203 st_ptr(G0, oop_addr); break; | |
204 case btos: // fall through | |
205 case ctos: // fall through | |
206 case stos: // fall through | |
207 case itos: ld(val_addr, Otos_l1); break; | |
208 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; | |
209 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; | |
210 case vtos: /* nothing to do */ break; | |
211 default : ShouldNotReachHere(); | |
212 } | |
213 // Clean up tos value in the jvmti thread state | |
214 or3(G0, ilgl, G3_scratch); | |
215 stw(G3_scratch, tos_addr); | |
216 st_long(G0, val_addr); | |
217 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
218 } | |
219 | |
220 | |
221 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { | |
222 if (JvmtiExport::can_force_early_return()) { | |
223 Label L; | |
224 Register thr_state = G3_scratch; | |
225 ld_ptr(Address(G2_thread, 0, in_bytes(JavaThread::jvmti_thread_state_offset())), | |
226 thr_state); | |
227 tst(thr_state); | |
228 br(zero, false, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; | |
229 delayed()->nop(); | |
230 | |
231 // Initiate earlyret handling only if it is not already being processed. | |
232 // If the flag has the earlyret_processing bit set, it means that this code | |
233 // is called *during* earlyret handling - we don't want to reenter. | |
234 ld(Address(thr_state, 0, in_bytes(JvmtiThreadState::earlyret_state_offset())), | |
235 G4_scratch); | |
236 cmp(G4_scratch, JvmtiThreadState::earlyret_pending); | |
237 br(Assembler::notEqual, false, pt, L); | |
238 delayed()->nop(); | |
239 | |
240 // Call Interpreter::remove_activation_early_entry() to get the address of the | |
241 // same-named entrypoint in the generated interpreter code | |
242 Address tos_addr(thr_state, 0, in_bytes(JvmtiThreadState::earlyret_tos_offset())); | |
243 ld(tos_addr, Otos_l1); | |
244 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); | |
245 | |
246 // Jump to Interpreter::_remove_activation_early_entry | |
247 jmpl(O0, G0, G0); | |
248 delayed()->nop(); | |
249 bind(L); | |
250 } | |
251 } | |
252 | |
253 | |
254 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1) { | |
255 mov(arg_1, O0); | |
256 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 1); | |
257 } | |
258 #endif /* CC_INTERP */ | |
259 | |
260 | |
261 #ifndef CC_INTERP | |
262 | |
263 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { | |
264 assert_not_delayed(); | |
265 dispatch_Lbyte_code(state, table); | |
266 } | |
267 | |
268 | |
269 void InterpreterMacroAssembler::dispatch_normal(TosState state) { | |
270 dispatch_base(state, Interpreter::normal_table(state)); | |
271 } | |
272 | |
273 | |
274 void InterpreterMacroAssembler::dispatch_only(TosState state) { | |
275 dispatch_base(state, Interpreter::dispatch_table(state)); | |
276 } | |
277 | |
278 | |
279 // common code to dispatch and dispatch_only | |
280 // dispatch value in Lbyte_code and increment Lbcp | |
281 | |
282 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) { | |
283 verify_FPU(1, state); | |
284 // %%%%% maybe implement +VerifyActivationFrameSize here | |
285 //verify_thread(); //too slow; we will just verify on method entry & exit | |
286 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
287 #ifdef FAST_DISPATCH | |
288 if (table == Interpreter::dispatch_table(state)) { | |
289 // use IdispatchTables | |
290 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); | |
291 // add offset to correct dispatch table | |
292 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
293 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr | |
294 } else { | |
295 #endif | |
296 // dispatch table to use | |
297 Address tbl(G3_scratch, (address)table); | |
298 | |
299 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
300 load_address(tbl); // compute addr of table | |
301 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr | |
302 #ifdef FAST_DISPATCH | |
303 } | |
304 #endif | |
305 jmp( G3_scratch, 0 ); | |
306 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); | |
307 else delayed()->nop(); | |
308 } | |
309 | |
310 | |
311 // Helpers for expression stack | |
312 | |
313 // Longs and doubles are Category 2 computational types in the | |
314 // JVM specification (section 3.11.1) and take 2 expression stack or | |
315 // local slots. | |
316 // Aligning them on 32 bit with tagged stacks is hard because the code generated | |
317 // for the dup* bytecodes depends on what types are already on the stack. | |
318 // If the types are split into the two stack/local slots, that is much easier | |
319 // (and we can use 0 for non-reference tags). | |
320 | |
321 // Known good alignment in _LP64 but unknown otherwise | |
322 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { | |
323 assert_not_delayed(); | |
324 | |
325 #ifdef _LP64 | |
326 ldf(FloatRegisterImpl::D, r1, offset, d); | |
327 #else | |
328 ldf(FloatRegisterImpl::S, r1, offset, d); | |
329 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize(), d->successor()); | |
330 #endif | |
331 } | |
332 | |
333 // Known good alignment in _LP64 but unknown otherwise | |
334 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { | |
335 assert_not_delayed(); | |
336 | |
337 #ifdef _LP64 | |
338 stf(FloatRegisterImpl::D, d, r1, offset); | |
339 // store something more useful here | |
340 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize());) | |
341 #else | |
342 stf(FloatRegisterImpl::S, d, r1, offset); | |
343 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize()); | |
344 #endif | |
345 } | |
346 | |
347 | |
348 // Known good alignment in _LP64 but unknown otherwise | |
349 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { | |
350 assert_not_delayed(); | |
351 #ifdef _LP64 | |
352 ldx(r1, offset, rd); | |
353 #else | |
354 ld(r1, offset, rd); | |
355 ld(r1, offset + Interpreter::stackElementSize(), rd->successor()); | |
356 #endif | |
357 } | |
358 | |
359 // Known good alignment in _LP64 but unknown otherwise | |
360 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { | |
361 assert_not_delayed(); | |
362 | |
363 #ifdef _LP64 | |
364 stx(l, r1, offset); | |
365 // store something more useful here | |
366 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize());) | |
367 #else | |
368 st(l, r1, offset); | |
369 st(l->successor(), r1, offset + Interpreter::stackElementSize()); | |
370 #endif | |
371 } | |
372 | |
373 #ifdef ASSERT | |
374 void InterpreterMacroAssembler::verify_stack_tag(frame::Tag t, | |
375 Register r, | |
376 Register scratch) { | |
377 if (TaggedStackInterpreter) { | |
378 Label ok, long_ok; | |
379 ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(0), r); | |
380 if (t == frame::TagCategory2) { | |
381 cmp(r, G0); | |
382 brx(Assembler::equal, false, Assembler::pt, long_ok); | |
383 delayed()->ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(1), r); | |
384 stop("stack long/double tag value bad"); | |
385 bind(long_ok); | |
386 cmp(r, G0); | |
387 } else if (t == frame::TagValue) { | |
388 cmp(r, G0); | |
389 } else { | |
390 assert_different_registers(r, scratch); | |
391 mov(t, scratch); | |
392 cmp(r, scratch); | |
393 } | |
394 brx(Assembler::equal, false, Assembler::pt, ok); | |
395 delayed()->nop(); | |
396 // Also compare if the stack value is zero, then the tag might | |
397 // not have been set coming from deopt. | |
398 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
399 cmp(r, G0); | |
400 brx(Assembler::equal, false, Assembler::pt, ok); | |
401 delayed()->nop(); | |
402 stop("Stack tag value is bad"); | |
403 bind(ok); | |
404 } | |
405 } | |
406 #endif // ASSERT | |
407 | |
408 void InterpreterMacroAssembler::pop_i(Register r) { | |
409 assert_not_delayed(); | |
410 // Uses destination register r for scratch | |
411 debug_only(verify_stack_tag(frame::TagValue, r)); | |
412 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
413 inc(Lesp, Interpreter::stackElementSize()); | |
414 debug_only(verify_esp(Lesp)); | |
415 } | |
416 | |
417 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { | |
418 assert_not_delayed(); | |
419 // Uses destination register r for scratch | |
420 debug_only(verify_stack_tag(frame::TagReference, r, scratch)); | |
421 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
422 inc(Lesp, Interpreter::stackElementSize()); | |
423 debug_only(verify_esp(Lesp)); | |
424 } | |
425 | |
426 void InterpreterMacroAssembler::pop_l(Register r) { | |
427 assert_not_delayed(); | |
428 // Uses destination register r for scratch | |
429 debug_only(verify_stack_tag(frame::TagCategory2, r)); | |
430 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
431 inc(Lesp, 2*Interpreter::stackElementSize()); | |
432 debug_only(verify_esp(Lesp)); | |
433 } | |
434 | |
435 | |
436 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { | |
437 assert_not_delayed(); | |
438 debug_only(verify_stack_tag(frame::TagValue, scratch)); | |
439 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); | |
440 inc(Lesp, Interpreter::stackElementSize()); | |
441 debug_only(verify_esp(Lesp)); | |
442 } | |
443 | |
444 | |
445 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { | |
446 assert_not_delayed(); | |
447 debug_only(verify_stack_tag(frame::TagCategory2, scratch)); | |
448 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); | |
449 inc(Lesp, 2*Interpreter::stackElementSize()); | |
450 debug_only(verify_esp(Lesp)); | |
451 } | |
452 | |
453 | |
454 // (Note use register first, then decrement so dec can be done during store stall) | |
455 void InterpreterMacroAssembler::tag_stack(Register r) { | |
456 if (TaggedStackInterpreter) { | |
457 st_ptr(r, Lesp, Interpreter::tag_offset_in_bytes()); | |
458 } | |
459 } | |
460 | |
461 void InterpreterMacroAssembler::tag_stack(frame::Tag t, Register r) { | |
462 if (TaggedStackInterpreter) { | |
463 assert (frame::TagValue == 0, "TagValue must be zero"); | |
464 if (t == frame::TagValue) { | |
465 st_ptr(G0, Lesp, Interpreter::tag_offset_in_bytes()); | |
466 } else if (t == frame::TagCategory2) { | |
467 st_ptr(G0, Lesp, Interpreter::tag_offset_in_bytes()); | |
468 // Tag next slot down too | |
469 st_ptr(G0, Lesp, -Interpreter::stackElementSize() + Interpreter::tag_offset_in_bytes()); | |
470 } else { | |
471 assert_different_registers(r, O3); | |
472 mov(t, O3); | |
473 st_ptr(O3, Lesp, Interpreter::tag_offset_in_bytes()); | |
474 } | |
475 } | |
476 } | |
477 | |
478 void InterpreterMacroAssembler::push_i(Register r) { | |
479 assert_not_delayed(); | |
480 debug_only(verify_esp(Lesp)); | |
481 tag_stack(frame::TagValue, r); | |
482 st( r, Lesp, Interpreter::value_offset_in_bytes()); | |
483 dec( Lesp, Interpreter::stackElementSize()); | |
484 } | |
485 | |
486 void InterpreterMacroAssembler::push_ptr(Register r) { | |
487 assert_not_delayed(); | |
488 tag_stack(frame::TagReference, r); | |
489 st_ptr( r, Lesp, Interpreter::value_offset_in_bytes()); | |
490 dec( Lesp, Interpreter::stackElementSize()); | |
491 } | |
492 | |
493 void InterpreterMacroAssembler::push_ptr(Register r, Register tag) { | |
494 assert_not_delayed(); | |
495 tag_stack(tag); | |
496 st_ptr(r, Lesp, Interpreter::value_offset_in_bytes()); | |
497 dec( Lesp, Interpreter::stackElementSize()); | |
498 } | |
499 | |
500 // remember: our convention for longs in SPARC is: | |
501 // O0 (Otos_l1) has high-order part in first word, | |
502 // O1 (Otos_l2) has low-order part in second word | |
503 | |
504 void InterpreterMacroAssembler::push_l(Register r) { | |
505 assert_not_delayed(); | |
506 debug_only(verify_esp(Lesp)); | |
507 tag_stack(frame::TagCategory2, r); | |
508 // Longs are in stored in memory-correct order, even if unaligned. | |
509 // and may be separated by stack tags. | |
510 int offset = -Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); | |
511 store_unaligned_long(r, Lesp, offset); | |
512 dec(Lesp, 2 * Interpreter::stackElementSize()); | |
513 } | |
514 | |
515 | |
516 void InterpreterMacroAssembler::push_f(FloatRegister f) { | |
517 assert_not_delayed(); | |
518 debug_only(verify_esp(Lesp)); | |
519 tag_stack(frame::TagValue, Otos_i); | |
520 stf(FloatRegisterImpl::S, f, Lesp, Interpreter::value_offset_in_bytes()); | |
521 dec(Lesp, Interpreter::stackElementSize()); | |
522 } | |
523 | |
524 | |
525 void InterpreterMacroAssembler::push_d(FloatRegister d) { | |
526 assert_not_delayed(); | |
527 debug_only(verify_esp(Lesp)); | |
528 tag_stack(frame::TagCategory2, Otos_i); | |
529 // Longs are in stored in memory-correct order, even if unaligned. | |
530 // and may be separated by stack tags. | |
531 int offset = -Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); | |
532 store_unaligned_double(d, Lesp, offset); | |
533 dec(Lesp, 2 * Interpreter::stackElementSize()); | |
534 } | |
535 | |
536 | |
537 void InterpreterMacroAssembler::push(TosState state) { | |
538 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
539 switch (state) { | |
540 case atos: push_ptr(); break; | |
541 case btos: push_i(); break; | |
542 case ctos: | |
543 case stos: push_i(); break; | |
544 case itos: push_i(); break; | |
545 case ltos: push_l(); break; | |
546 case ftos: push_f(); break; | |
547 case dtos: push_d(); break; | |
548 case vtos: /* nothing to do */ break; | |
549 default : ShouldNotReachHere(); | |
550 } | |
551 } | |
552 | |
553 | |
554 void InterpreterMacroAssembler::pop(TosState state) { | |
555 switch (state) { | |
556 case atos: pop_ptr(); break; | |
557 case btos: pop_i(); break; | |
558 case ctos: | |
559 case stos: pop_i(); break; | |
560 case itos: pop_i(); break; | |
561 case ltos: pop_l(); break; | |
562 case ftos: pop_f(); break; | |
563 case dtos: pop_d(); break; | |
564 case vtos: /* nothing to do */ break; | |
565 default : ShouldNotReachHere(); | |
566 } | |
567 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
568 } | |
569 | |
570 | |
571 // Tagged stack helpers for swap and dup | |
572 void InterpreterMacroAssembler::load_ptr_and_tag(int n, Register val, | |
573 Register tag) { | |
574 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); | |
575 if (TaggedStackInterpreter) { | |
576 ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(n), tag); | |
577 } | |
578 } | |
579 void InterpreterMacroAssembler::store_ptr_and_tag(int n, Register val, | |
580 Register tag) { | |
581 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); | |
582 if (TaggedStackInterpreter) { | |
583 st_ptr(tag, Lesp, Interpreter::expr_tag_offset_in_bytes(n)); | |
584 } | |
585 } | |
586 | |
587 | |
588 void InterpreterMacroAssembler::load_receiver(Register param_count, | |
589 Register recv) { | |
590 | |
591 sll(param_count, Interpreter::logStackElementSize(), param_count); | |
592 if (TaggedStackInterpreter) { | |
593 add(param_count, Interpreter::value_offset_in_bytes(), param_count); // get obj address | |
594 } | |
595 ld_ptr(Lesp, param_count, recv); // gets receiver Oop | |
596 } | |
597 | |
598 void InterpreterMacroAssembler::empty_expression_stack() { | |
599 // Reset Lesp. | |
600 sub( Lmonitors, wordSize, Lesp ); | |
601 | |
602 // Reset SP by subtracting more space from Lesp. | |
603 Label done; | |
604 | |
605 const Address max_stack (Lmethod, 0, in_bytes(methodOopDesc::max_stack_offset())); | |
606 const Address access_flags(Lmethod, 0, in_bytes(methodOopDesc::access_flags_offset())); | |
607 | |
608 verify_oop(Lmethod); | |
609 | |
610 | |
611 assert( G4_scratch != Gframe_size, | |
612 "Only you can prevent register aliasing!"); | |
613 | |
614 // A native does not need to do this, since its callee does not change SP. | |
615 ld(access_flags, Gframe_size); | |
616 btst(JVM_ACC_NATIVE, Gframe_size); | |
617 br(Assembler::notZero, false, Assembler::pt, done); | |
618 delayed()->nop(); | |
619 | |
620 // | |
621 // Compute max expression stack+register save area | |
622 // | |
623 lduh( max_stack, Gframe_size ); | |
624 if (TaggedStackInterpreter) sll ( Gframe_size, 1, Gframe_size); // max_stack * 2 for TAGS | |
625 add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size ); | |
626 | |
627 // | |
628 // now set up a stack frame with the size computed above | |
629 // | |
630 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below | |
631 sll( Gframe_size, LogBytesPerWord, Gframe_size ); | |
632 sub( Lesp, Gframe_size, Gframe_size ); | |
633 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary | |
634 debug_only(verify_sp(Gframe_size, G4_scratch)); | |
635 #ifdef _LP64 | |
636 sub(Gframe_size, STACK_BIAS, Gframe_size ); | |
637 #endif | |
638 mov(Gframe_size, SP); | |
639 | |
640 bind(done); | |
641 } | |
642 | |
643 | |
644 #ifdef ASSERT | |
645 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { | |
646 Label Bad, OK; | |
647 | |
648 // Saved SP must be aligned. | |
649 #ifdef _LP64 | |
650 btst(2*BytesPerWord-1, Rsp); | |
651 #else | |
652 btst(LongAlignmentMask, Rsp); | |
653 #endif | |
654 br(Assembler::notZero, false, Assembler::pn, Bad); | |
655 delayed()->nop(); | |
656 | |
657 // Saved SP, plus register window size, must not be above FP. | |
658 add(Rsp, frame::register_save_words * wordSize, Rtemp); | |
659 #ifdef _LP64 | |
660 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP | |
661 #endif | |
662 cmp(Rtemp, FP); | |
663 brx(Assembler::greaterUnsigned, false, Assembler::pn, Bad); | |
664 delayed()->nop(); | |
665 | |
666 // Saved SP must not be ridiculously below current SP. | |
667 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); | |
668 set(maxstack, Rtemp); | |
669 sub(SP, Rtemp, Rtemp); | |
670 #ifdef _LP64 | |
671 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp | |
672 #endif | |
673 cmp(Rsp, Rtemp); | |
674 brx(Assembler::lessUnsigned, false, Assembler::pn, Bad); | |
675 delayed()->nop(); | |
676 | |
677 br(Assembler::always, false, Assembler::pn, OK); | |
678 delayed()->nop(); | |
679 | |
680 bind(Bad); | |
681 stop("on return to interpreted call, restored SP is corrupted"); | |
682 | |
683 bind(OK); | |
684 } | |
685 | |
686 | |
687 void InterpreterMacroAssembler::verify_esp(Register Resp) { | |
688 // about to read or write Resp[0] | |
689 // make sure it is not in the monitors or the register save area | |
690 Label OK1, OK2; | |
691 | |
692 cmp(Resp, Lmonitors); | |
693 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); | |
694 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); | |
695 stop("too many pops: Lesp points into monitor area"); | |
696 bind(OK1); | |
697 #ifdef _LP64 | |
698 sub(Resp, STACK_BIAS, Resp); | |
699 #endif | |
700 cmp(Resp, SP); | |
701 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); | |
702 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); | |
703 stop("too many pushes: Lesp points into register window"); | |
704 bind(OK2); | |
705 } | |
706 #endif // ASSERT | |
707 | |
708 // Load compiled (i2c) or interpreter entry when calling from interpreted and | |
709 // do the call. Centralized so that all interpreter calls will do the same actions. | |
710 // If jvmti single stepping is on for a thread we must not call compiled code. | |
711 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { | |
712 | |
713 // Assume we want to go compiled if available | |
714 | |
715 ld_ptr(G5_method, in_bytes(methodOopDesc::from_interpreted_offset()), target); | |
716 | |
717 if (JvmtiExport::can_post_interpreter_events()) { | |
718 // JVMTI events, such as single-stepping, are implemented partly by avoiding running | |
719 // compiled code in threads for which the event is enabled. Check here for | |
720 // interp_only_mode if these events CAN be enabled. | |
721 verify_thread(); | |
722 Label skip_compiled_code; | |
723 | |
724 const Address interp_only (G2_thread, 0, in_bytes(JavaThread::interp_only_mode_offset())); | |
725 | |
726 ld(interp_only, scratch); | |
727 tst(scratch); | |
728 br(Assembler::notZero, true, Assembler::pn, skip_compiled_code); | |
729 delayed()->ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), target); | |
730 bind(skip_compiled_code); | |
731 } | |
732 | |
733 // the i2c_adapters need methodOop in G5_method (right? %%%) | |
734 // do the call | |
735 #ifdef ASSERT | |
736 { | |
737 Label ok; | |
738 br_notnull(target, false, Assembler::pt, ok); | |
739 delayed()->nop(); | |
740 stop("null entry point"); | |
741 bind(ok); | |
742 } | |
743 #endif // ASSERT | |
744 | |
745 // Adjust Rret first so Llast_SP can be same as Rret | |
746 add(Rret, -frame::pc_return_offset, O7); | |
747 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer | |
748 // Record SP so we can remove any stack space allocated by adapter transition | |
749 jmp(target, 0); | |
750 delayed()->mov(SP, Llast_SP); | |
751 } | |
752 | |
753 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { | |
754 assert_not_delayed(); | |
755 | |
756 Label not_taken; | |
757 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); | |
758 else br (cc, false, Assembler::pn, not_taken); | |
759 delayed()->nop(); | |
760 | |
761 TemplateTable::branch(false,false); | |
762 | |
763 bind(not_taken); | |
764 | |
765 profile_not_taken_branch(G3_scratch); | |
766 } | |
767 | |
768 | |
769 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( | |
770 int bcp_offset, | |
771 Register Rtmp, | |
772 Register Rdst, | |
773 signedOrNot is_signed, | |
774 setCCOrNot should_set_CC ) { | |
775 assert(Rtmp != Rdst, "need separate temp register"); | |
776 assert_not_delayed(); | |
777 switch (is_signed) { | |
778 default: ShouldNotReachHere(); | |
779 | |
780 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte | |
781 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte | |
782 } | |
783 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte | |
784 sll( Rdst, BitsPerByte, Rdst); | |
785 switch (should_set_CC ) { | |
786 default: ShouldNotReachHere(); | |
787 | |
788 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; | |
789 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; | |
790 } | |
791 } | |
792 | |
793 | |
794 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( | |
795 int bcp_offset, | |
796 Register Rtmp, | |
797 Register Rdst, | |
798 setCCOrNot should_set_CC ) { | |
799 assert(Rtmp != Rdst, "need separate temp register"); | |
800 assert_not_delayed(); | |
801 add( Lbcp, bcp_offset, Rtmp); | |
802 andcc( Rtmp, 3, G0); | |
803 Label aligned; | |
804 switch (should_set_CC ) { | |
805 default: ShouldNotReachHere(); | |
806 | |
807 case set_CC: break; | |
808 case dont_set_CC: break; | |
809 } | |
810 | |
811 br(Assembler::zero, true, Assembler::pn, aligned); | |
812 #ifdef _LP64 | |
813 delayed()->ldsw(Rtmp, 0, Rdst); | |
814 #else | |
815 delayed()->ld(Rtmp, 0, Rdst); | |
816 #endif | |
817 | |
818 ldub(Lbcp, bcp_offset + 3, Rdst); | |
819 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); | |
820 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); | |
821 #ifdef _LP64 | |
822 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); | |
823 #else | |
824 // Unsigned load is faster than signed on some implementations | |
825 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); | |
826 #endif | |
827 or3(Rtmp, Rdst, Rdst ); | |
828 | |
829 bind(aligned); | |
830 if (should_set_CC == set_CC) tst(Rdst); | |
831 } | |
832 | |
833 | |
834 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp, int bcp_offset) { | |
835 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); | |
836 assert_different_registers(cache, tmp); | |
837 assert_not_delayed(); | |
838 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); | |
839 // convert from field index to ConstantPoolCacheEntry index | |
840 // and from word index to byte offset | |
841 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); | |
842 add(LcpoolCache, tmp, cache); | |
843 } | |
844 | |
845 | |
846 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, int bcp_offset) { | |
847 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); | |
848 assert_different_registers(cache, tmp); | |
849 assert_not_delayed(); | |
850 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); | |
851 // convert from field index to ConstantPoolCacheEntry index | |
852 // and from word index to byte offset | |
853 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); | |
854 // skip past the header | |
855 add(tmp, in_bytes(constantPoolCacheOopDesc::base_offset()), tmp); | |
856 // construct pointer to cache entry | |
857 add(LcpoolCache, tmp, cache); | |
858 } | |
859 | |
860 | |
861 // Generate a subtype check: branch to ok_is_subtype if sub_klass is | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
862 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. |
0 | 863 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, |
864 Register Rsuper_klass, | |
865 Register Rtmp1, | |
866 Register Rtmp2, | |
867 Register Rtmp3, | |
868 Label &ok_is_subtype ) { | |
869 Label not_subtype, loop; | |
870 | |
871 // Profile the not-null value's klass. | |
872 profile_typecheck(Rsub_klass, Rtmp1); | |
873 | |
874 // Load the super-klass's check offset into Rtmp1 | |
875 ld( Rsuper_klass, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes(), Rtmp1 ); | |
876 // Load from the sub-klass's super-class display list, or a 1-word cache of | |
877 // the secondary superclass list, or a failing value with a sentinel offset | |
878 // if the super-klass is an interface or exceptionally deep in the Java | |
879 // hierarchy and we have to scan the secondary superclass list the hard way. | |
880 ld_ptr( Rsub_klass, Rtmp1, Rtmp2 ); | |
881 // See if we get an immediate positive hit | |
882 cmp( Rtmp2, Rsuper_klass ); | |
883 brx( Assembler::equal, false, Assembler::pt, ok_is_subtype ); | |
884 // In the delay slot, check for immediate negative hit | |
885 delayed()->cmp( Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() ); | |
886 br( Assembler::notEqual, false, Assembler::pt, not_subtype ); | |
887 // In the delay slot, check for self | |
888 delayed()->cmp( Rsub_klass, Rsuper_klass ); | |
889 brx( Assembler::equal, false, Assembler::pt, ok_is_subtype ); | |
890 | |
891 // Now do a linear scan of the secondary super-klass chain. | |
892 delayed()->ld_ptr( Rsub_klass, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes(), Rtmp2 ); | |
893 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
894 // compress superclass |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
895 if (UseCompressedOops) encode_heap_oop(Rsuper_klass); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
896 |
0 | 897 // Rtmp2 holds the objArrayOop of secondary supers. |
898 ld( Rtmp2, arrayOopDesc::length_offset_in_bytes(), Rtmp1 );// Load the array length | |
899 // Check for empty secondary super list | |
900 tst(Rtmp1); | |
901 | |
902 // Top of search loop | |
903 bind( loop ); | |
904 br( Assembler::equal, false, Assembler::pn, not_subtype ); | |
905 delayed()->nop(); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
906 |
0 | 907 // load next super to check |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
908 if (UseCompressedOops) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
909 ld( Rtmp2, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rtmp3); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
910 // Bump array pointer forward one oop |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
911 add( Rtmp2, 4, Rtmp2 ); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
912 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
913 ld_ptr( Rtmp2, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rtmp3); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
914 // Bump array pointer forward one oop |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
915 add( Rtmp2, wordSize, Rtmp2); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
916 } |
0 | 917 // Look for Rsuper_klass on Rsub_klass's secondary super-class-overflow list |
918 cmp( Rtmp3, Rsuper_klass ); | |
919 // A miss means we are NOT a subtype and need to keep looping | |
920 brx( Assembler::notEqual, false, Assembler::pt, loop ); | |
921 delayed()->deccc( Rtmp1 ); // dec trip counter in delay slot | |
922 // Falling out the bottom means we found a hit; we ARE a subtype | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
923 if (UseCompressedOops) decode_heap_oop(Rsuper_klass); |
0 | 924 br( Assembler::always, false, Assembler::pt, ok_is_subtype ); |
925 // Update the cache | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
926 delayed()->st_ptr( Rsuper_klass, Rsub_klass, |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
927 sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() ); |
0 | 928 |
929 bind(not_subtype); | |
930 profile_typecheck_failed(Rtmp1); | |
931 } | |
932 | |
933 // Separate these two to allow for delay slot in middle | |
934 // These are used to do a test and full jump to exception-throwing code. | |
935 | |
936 // %%%%% Could possibly reoptimize this by testing to see if could use | |
937 // a single conditional branch (i.e. if span is small enough. | |
938 // If you go that route, than get rid of the split and give up | |
939 // on the delay-slot hack. | |
940 | |
941 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, | |
942 Label& ok ) { | |
943 assert_not_delayed(); | |
944 br(ok_condition, true, pt, ok); | |
945 // DELAY SLOT | |
946 } | |
947 | |
948 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, | |
949 Label& ok ) { | |
950 assert_not_delayed(); | |
951 bp( ok_condition, true, Assembler::xcc, pt, ok); | |
952 // DELAY SLOT | |
953 } | |
954 | |
955 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, | |
956 Label& ok ) { | |
957 assert_not_delayed(); | |
958 brx(ok_condition, true, pt, ok); | |
959 // DELAY SLOT | |
960 } | |
961 | |
962 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, | |
963 Register Rscratch, | |
964 Label& ok ) { | |
965 assert(throw_entry_point != NULL, "entry point must be generated by now"); | |
966 Address dest(Rscratch, throw_entry_point); | |
967 jump_to(dest); | |
968 delayed()->nop(); | |
969 bind(ok); | |
970 } | |
971 | |
972 | |
973 // And if you cannot use the delay slot, here is a shorthand: | |
974 | |
975 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, | |
976 address throw_entry_point, | |
977 Register Rscratch ) { | |
978 Label ok; | |
979 if (ok_condition != never) { | |
980 throw_if_not_1_icc( ok_condition, ok); | |
981 delayed()->nop(); | |
982 } | |
983 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
984 } | |
985 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, | |
986 address throw_entry_point, | |
987 Register Rscratch ) { | |
988 Label ok; | |
989 if (ok_condition != never) { | |
990 throw_if_not_1_xcc( ok_condition, ok); | |
991 delayed()->nop(); | |
992 } | |
993 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
994 } | |
995 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, | |
996 address throw_entry_point, | |
997 Register Rscratch ) { | |
998 Label ok; | |
999 if (ok_condition != never) { | |
1000 throw_if_not_1_x( ok_condition, ok); | |
1001 delayed()->nop(); | |
1002 } | |
1003 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
1004 } | |
1005 | |
1006 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res | |
1007 // Note: res is still shy of address by array offset into object. | |
1008 | |
1009 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { | |
1010 assert_not_delayed(); | |
1011 | |
1012 verify_oop(array); | |
1013 #ifdef _LP64 | |
1014 // sign extend since tos (index) can be a 32bit value | |
1015 sra(index, G0, index); | |
1016 #endif // _LP64 | |
1017 | |
1018 // check array | |
1019 Label ptr_ok; | |
1020 tst(array); | |
1021 throw_if_not_1_x( notZero, ptr_ok ); | |
1022 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index | |
1023 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); | |
1024 | |
1025 Label index_ok; | |
1026 cmp(index, tmp); | |
1027 throw_if_not_1_icc( lessUnsigned, index_ok ); | |
1028 if (index_shift > 0) delayed()->sll(index, index_shift, index); | |
1029 else delayed()->add(array, index, res); // addr - const offset in index | |
1030 // convention: move aberrant index into G3_scratch for exception message | |
1031 mov(index, G3_scratch); | |
1032 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); | |
1033 | |
1034 // add offset if didn't do it in delay slot | |
1035 if (index_shift > 0) add(array, index, res); // addr - const offset in index | |
1036 } | |
1037 | |
1038 | |
1039 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { | |
1040 assert_not_delayed(); | |
1041 | |
1042 // pop array | |
1043 pop_ptr(array); | |
1044 | |
1045 // check array | |
1046 index_check_without_pop(array, index, index_shift, tmp, res); | |
1047 } | |
1048 | |
1049 | |
1050 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { | |
1051 ld_ptr(Lmethod, in_bytes(methodOopDesc::constants_offset()), Rdst); | |
1052 } | |
1053 | |
1054 | |
1055 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { | |
1056 get_constant_pool(Rdst); | |
1057 ld_ptr(Rdst, constantPoolOopDesc::cache_offset_in_bytes(), Rdst); | |
1058 } | |
1059 | |
1060 | |
1061 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { | |
1062 get_constant_pool(Rcpool); | |
1063 ld_ptr(Rcpool, constantPoolOopDesc::tags_offset_in_bytes(), Rtags); | |
1064 } | |
1065 | |
1066 | |
1067 // unlock if synchronized method | |
1068 // | |
1069 // Unlock the receiver if this is a synchronized method. | |
1070 // Unlock any Java monitors from syncronized blocks. | |
1071 // | |
1072 // If there are locked Java monitors | |
1073 // If throw_monitor_exception | |
1074 // throws IllegalMonitorStateException | |
1075 // Else if install_monitor_exception | |
1076 // installs IllegalMonitorStateException | |
1077 // Else | |
1078 // no error processing | |
1079 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, | |
1080 bool throw_monitor_exception, | |
1081 bool install_monitor_exception) { | |
1082 Label unlocked, unlock, no_unlock; | |
1083 | |
1084 // get the value of _do_not_unlock_if_synchronized into G1_scratch | |
1085 const Address do_not_unlock_if_synchronized(G2_thread, 0, | |
1086 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); | |
1087 ldbool(do_not_unlock_if_synchronized, G1_scratch); | |
1088 stbool(G0, do_not_unlock_if_synchronized); // reset the flag | |
1089 | |
1090 // check if synchronized method | |
1091 const Address access_flags(Lmethod, 0, in_bytes(methodOopDesc::access_flags_offset())); | |
1092 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
1093 push(state); // save tos | |
1094 ld(access_flags, G3_scratch); | |
1095 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); | |
1096 br( zero, false, pt, unlocked); | |
1097 delayed()->nop(); | |
1098 | |
1099 // Don't unlock anything if the _do_not_unlock_if_synchronized flag | |
1100 // is set. | |
1101 tstbool(G1_scratch); | |
1102 br(Assembler::notZero, false, pn, no_unlock); | |
1103 delayed()->nop(); | |
1104 | |
1105 // BasicObjectLock will be first in list, since this is a synchronized method. However, need | |
1106 // to check that the object has not been unlocked by an explicit monitorexit bytecode. | |
1107 | |
1108 //Intel: if (throw_monitor_exception) ... else ... | |
1109 // Entry already unlocked, need to throw exception | |
1110 //... | |
1111 | |
1112 // pass top-most monitor elem | |
1113 add( top_most_monitor(), O1 ); | |
1114 | |
1115 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); | |
1116 br_notnull(G3_scratch, false, pt, unlock); | |
1117 delayed()->nop(); | |
1118 | |
1119 if (throw_monitor_exception) { | |
1120 // Entry already unlocked need to throw an exception | |
1121 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
1122 should_not_reach_here(); | |
1123 } else { | |
1124 // Monitor already unlocked during a stack unroll. | |
1125 // If requested, install an illegal_monitor_state_exception. | |
1126 // Continue with stack unrolling. | |
1127 if (install_monitor_exception) { | |
1128 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); | |
1129 } | |
1130 ba(false, unlocked); | |
1131 delayed()->nop(); | |
1132 } | |
1133 | |
1134 bind(unlock); | |
1135 | |
1136 unlock_object(O1); | |
1137 | |
1138 bind(unlocked); | |
1139 | |
1140 // I0, I1: Might contain return value | |
1141 | |
1142 // Check that all monitors are unlocked | |
1143 { Label loop, exception, entry, restart; | |
1144 | |
1145 Register Rmptr = O0; | |
1146 Register Rtemp = O1; | |
1147 Register Rlimit = Lmonitors; | |
1148 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
1149 assert( (delta & LongAlignmentMask) == 0, | |
1150 "sizeof BasicObjectLock must be even number of doublewords"); | |
1151 | |
1152 #ifdef ASSERT | |
1153 add(top_most_monitor(), Rmptr, delta); | |
1154 { Label L; | |
1155 // ensure that Rmptr starts out above (or at) Rlimit | |
1156 cmp(Rmptr, Rlimit); | |
1157 brx(Assembler::greaterEqualUnsigned, false, pn, L); | |
1158 delayed()->nop(); | |
1159 stop("monitor stack has negative size"); | |
1160 bind(L); | |
1161 } | |
1162 #endif | |
1163 bind(restart); | |
1164 ba(false, entry); | |
1165 delayed()-> | |
1166 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry | |
1167 | |
1168 // Entry is still locked, need to throw exception | |
1169 bind(exception); | |
1170 if (throw_monitor_exception) { | |
1171 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
1172 should_not_reach_here(); | |
1173 } else { | |
1174 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. | |
1175 // Unlock does not block, so don't have to worry about the frame | |
1176 unlock_object(Rmptr); | |
1177 if (install_monitor_exception) { | |
1178 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); | |
1179 } | |
1180 ba(false, restart); | |
1181 delayed()->nop(); | |
1182 } | |
1183 | |
1184 bind(loop); | |
1185 cmp(Rtemp, G0); // check if current entry is used | |
1186 brx(Assembler::notEqual, false, pn, exception); | |
1187 delayed()-> | |
1188 dec(Rmptr, delta); // otherwise advance to next entry | |
1189 #ifdef ASSERT | |
1190 { Label L; | |
1191 // ensure that Rmptr has not somehow stepped below Rlimit | |
1192 cmp(Rmptr, Rlimit); | |
1193 brx(Assembler::greaterEqualUnsigned, false, pn, L); | |
1194 delayed()->nop(); | |
1195 stop("ran off the end of the monitor stack"); | |
1196 bind(L); | |
1197 } | |
1198 #endif | |
1199 bind(entry); | |
1200 cmp(Rmptr, Rlimit); // check if bottom reached | |
1201 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry | |
1202 delayed()-> | |
1203 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); | |
1204 } | |
1205 | |
1206 bind(no_unlock); | |
1207 pop(state); | |
1208 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
1209 } | |
1210 | |
1211 | |
1212 // remove activation | |
1213 // | |
1214 // Unlock the receiver if this is a synchronized method. | |
1215 // Unlock any Java monitors from syncronized blocks. | |
1216 // Remove the activation from the stack. | |
1217 // | |
1218 // If there are locked Java monitors | |
1219 // If throw_monitor_exception | |
1220 // throws IllegalMonitorStateException | |
1221 // Else if install_monitor_exception | |
1222 // installs IllegalMonitorStateException | |
1223 // Else | |
1224 // no error processing | |
1225 void InterpreterMacroAssembler::remove_activation(TosState state, | |
1226 bool throw_monitor_exception, | |
1227 bool install_monitor_exception) { | |
1228 | |
1229 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); | |
1230 | |
1231 // save result (push state before jvmti call and pop it afterwards) and notify jvmti | |
1232 notify_method_exit(false, state, NotifyJVMTI); | |
1233 | |
1234 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
1235 verify_oop(Lmethod); | |
1236 verify_thread(); | |
1237 | |
1238 // return tos | |
1239 assert(Otos_l1 == Otos_i, "adjust code below"); | |
1240 switch (state) { | |
1241 #ifdef _LP64 | |
1242 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 | |
1243 #else | |
1244 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1 | |
1245 #endif | |
1246 case btos: // fall through | |
1247 case ctos: | |
1248 case stos: // fall through | |
1249 case atos: // fall through | |
1250 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 | |
1251 case ftos: // fall through | |
1252 case dtos: // fall through | |
1253 case vtos: /* nothing to do */ break; | |
1254 default : ShouldNotReachHere(); | |
1255 } | |
1256 | |
1257 #if defined(COMPILER2) && !defined(_LP64) | |
1258 if (state == ltos) { | |
1259 // C2 expects long results in G1 we can't tell if we're returning to interpreted | |
1260 // or compiled so just be safe use G1 and O0/O1 | |
1261 | |
1262 // Shift bits into high (msb) of G1 | |
1263 sllx(Otos_l1->after_save(), 32, G1); | |
1264 // Zero extend low bits | |
1265 srl (Otos_l2->after_save(), 0, Otos_l2->after_save()); | |
1266 or3 (Otos_l2->after_save(), G1, G1); | |
1267 } | |
1268 #endif /* COMPILER2 */ | |
1269 | |
1270 } | |
1271 #endif /* CC_INTERP */ | |
1272 | |
1273 | |
1274 // Lock object | |
1275 // | |
1276 // Argument - lock_reg points to the BasicObjectLock to be used for locking, | |
1277 // it must be initialized with the object to lock | |
1278 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { | |
1279 if (UseHeavyMonitors) { | |
1280 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); | |
1281 } | |
1282 else { | |
1283 Register obj_reg = Object; | |
1284 Register mark_reg = G4_scratch; | |
1285 Register temp_reg = G1_scratch; | |
1286 Address lock_addr = Address(lock_reg, 0, BasicObjectLock::lock_offset_in_bytes()); | |
1287 Address mark_addr = Address(obj_reg, 0, oopDesc::mark_offset_in_bytes()); | |
1288 Label done; | |
1289 | |
1290 Label slow_case; | |
1291 | |
1292 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); | |
1293 | |
1294 // load markOop from object into mark_reg | |
1295 ld_ptr(mark_addr, mark_reg); | |
1296 | |
1297 if (UseBiasedLocking) { | |
1298 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); | |
1299 } | |
1300 | |
1301 // get the address of basicLock on stack that will be stored in the object | |
1302 // we need a temporary register here as we do not want to clobber lock_reg | |
1303 // (cas clobbers the destination register) | |
1304 mov(lock_reg, temp_reg); | |
1305 // set mark reg to be (markOop of object | UNLOCK_VALUE) | |
1306 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); | |
1307 // initialize the box (Must happen before we update the object mark!) | |
1308 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); | |
1309 // compare and exchange object_addr, markOop | 1, stack address of basicLock | |
1310 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
1311 casx_under_lock(mark_addr.base(), mark_reg, temp_reg, | |
1312 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
1313 | |
1314 // if the compare and exchange succeeded we are done (we saw an unlocked object) | |
1315 cmp(mark_reg, temp_reg); | |
1316 brx(Assembler::equal, true, Assembler::pt, done); | |
1317 delayed()->nop(); | |
1318 | |
1319 // We did not see an unlocked object so try the fast recursive case | |
1320 | |
1321 // Check if owner is self by comparing the value in the markOop of object | |
1322 // with the stack pointer | |
1323 sub(temp_reg, SP, temp_reg); | |
1324 #ifdef _LP64 | |
1325 sub(temp_reg, STACK_BIAS, temp_reg); | |
1326 #endif | |
1327 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
1328 | |
1329 // Composite "andcc" test: | |
1330 // (a) %sp -vs- markword proximity check, and, | |
1331 // (b) verify mark word LSBs == 0 (Stack-locked). | |
1332 // | |
1333 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) | |
1334 // Note that the page size used for %sp proximity testing is arbitrary and is | |
1335 // unrelated to the actual MMU page size. We use a 'logical' page size of | |
1336 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate | |
1337 // field of the andcc instruction. | |
1338 andcc (temp_reg, 0xFFFFF003, G0) ; | |
1339 | |
1340 // if condition is true we are done and hence we can store 0 in the displaced | |
1341 // header indicating it is a recursive lock and be done | |
1342 brx(Assembler::zero, true, Assembler::pt, done); | |
1343 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); | |
1344 | |
1345 // none of the above fast optimizations worked so we have to get into the | |
1346 // slow case of monitor enter | |
1347 bind(slow_case); | |
1348 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); | |
1349 | |
1350 bind(done); | |
1351 } | |
1352 } | |
1353 | |
1354 // Unlocks an object. Used in monitorexit bytecode and remove_activation. | |
1355 // | |
1356 // Argument - lock_reg points to the BasicObjectLock for lock | |
1357 // Throw IllegalMonitorException if object is not locked by current thread | |
1358 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { | |
1359 if (UseHeavyMonitors) { | |
1360 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); | |
1361 } else { | |
1362 Register obj_reg = G3_scratch; | |
1363 Register mark_reg = G4_scratch; | |
1364 Register displaced_header_reg = G1_scratch; | |
1365 Address lock_addr = Address(lock_reg, 0, BasicObjectLock::lock_offset_in_bytes()); | |
1366 Address lockobj_addr = Address(lock_reg, 0, BasicObjectLock::obj_offset_in_bytes()); | |
1367 Address mark_addr = Address(obj_reg, 0, oopDesc::mark_offset_in_bytes()); | |
1368 Label done; | |
1369 | |
1370 if (UseBiasedLocking) { | |
1371 // load the object out of the BasicObjectLock | |
1372 ld_ptr(lockobj_addr, obj_reg); | |
1373 biased_locking_exit(mark_addr, mark_reg, done, true); | |
1374 st_ptr(G0, lockobj_addr); // free entry | |
1375 } | |
1376 | |
1377 // Test first if we are in the fast recursive case | |
1378 ld_ptr(lock_addr, displaced_header_reg, BasicLock::displaced_header_offset_in_bytes()); | |
1379 br_null(displaced_header_reg, true, Assembler::pn, done); | |
1380 delayed()->st_ptr(G0, lockobj_addr); // free entry | |
1381 | |
1382 // See if it is still a light weight lock, if so we just unlock | |
1383 // the object and we are done | |
1384 | |
1385 if (!UseBiasedLocking) { | |
1386 // load the object out of the BasicObjectLock | |
1387 ld_ptr(lockobj_addr, obj_reg); | |
1388 } | |
1389 | |
1390 // we have the displaced header in displaced_header_reg | |
1391 // we expect to see the stack address of the basicLock in case the | |
1392 // lock is still a light weight lock (lock_reg) | |
1393 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
1394 casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg, | |
1395 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
1396 cmp(lock_reg, displaced_header_reg); | |
1397 brx(Assembler::equal, true, Assembler::pn, done); | |
1398 delayed()->st_ptr(G0, lockobj_addr); // free entry | |
1399 | |
1400 // The lock has been converted into a heavy lock and hence | |
1401 // we need to get into the slow case | |
1402 | |
1403 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); | |
1404 | |
1405 bind(done); | |
1406 } | |
1407 } | |
1408 | |
1409 #ifndef CC_INTERP | |
1410 | |
1411 // Get the method data pointer from the methodOop and set the | |
1412 // specified register to its value. | |
1413 | |
1414 void InterpreterMacroAssembler::set_method_data_pointer_offset(Register Roff) { | |
1415 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1416 Label get_continue; | |
1417 | |
1418 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr); | |
1419 test_method_data_pointer(get_continue); | |
1420 add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr); | |
1421 if (Roff != noreg) | |
1422 // Roff contains a method data index ("mdi"). It defaults to zero. | |
1423 add(ImethodDataPtr, Roff, ImethodDataPtr); | |
1424 bind(get_continue); | |
1425 } | |
1426 | |
1427 // Set the method data pointer for the current bcp. | |
1428 | |
1429 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { | |
1430 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1431 Label zero_continue; | |
1432 | |
1433 // Test MDO to avoid the call if it is NULL. | |
1434 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr); | |
1435 test_method_data_pointer(zero_continue); | |
1436 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); | |
1437 set_method_data_pointer_offset(O0); | |
1438 bind(zero_continue); | |
1439 } | |
1440 | |
1441 // Test ImethodDataPtr. If it is null, continue at the specified label | |
1442 | |
1443 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { | |
1444 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1445 #ifdef _LP64 | |
1446 bpr(Assembler::rc_z, false, Assembler::pn, ImethodDataPtr, zero_continue); | |
1447 #else | |
1448 tst(ImethodDataPtr); | |
1449 br(Assembler::zero, false, Assembler::pn, zero_continue); | |
1450 #endif | |
1451 delayed()->nop(); | |
1452 } | |
1453 | |
1454 void InterpreterMacroAssembler::verify_method_data_pointer() { | |
1455 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1456 #ifdef ASSERT | |
1457 Label verify_continue; | |
1458 test_method_data_pointer(verify_continue); | |
1459 | |
1460 // If the mdp is valid, it will point to a DataLayout header which is | |
1461 // consistent with the bcp. The converse is highly probable also. | |
1462 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); | |
1463 ld_ptr(Address(Lmethod, 0, in_bytes(methodOopDesc::const_offset())), O5); | |
1464 add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), G3_scratch); | |
1465 add(G3_scratch, O5, G3_scratch); | |
1466 cmp(Lbcp, G3_scratch); | |
1467 brx(Assembler::equal, false, Assembler::pt, verify_continue); | |
1468 | |
1469 Register temp_reg = O5; | |
1470 delayed()->mov(ImethodDataPtr, temp_reg); | |
1471 // %%% should use call_VM_leaf here? | |
1472 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); | |
1473 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); | |
1474 Address d_save(FP, 0, -sizeof(jdouble) + STACK_BIAS); | |
1475 stf(FloatRegisterImpl::D, Ftos_d, d_save); | |
1476 mov(temp_reg->after_save(), O2); | |
1477 save_thread(L7_thread_cache); | |
1478 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); | |
1479 delayed()->nop(); | |
1480 restore_thread(L7_thread_cache); | |
1481 ldf(FloatRegisterImpl::D, d_save, Ftos_d); | |
1482 restore(); | |
1483 bind(verify_continue); | |
1484 #endif // ASSERT | |
1485 } | |
1486 | |
1487 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, | |
1488 Register cur_bcp, | |
1489 Register Rtmp, | |
1490 Label &profile_continue) { | |
1491 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1492 // Control will flow to "profile_continue" if the counter is less than the | |
1493 // limit or if we call profile_method() | |
1494 | |
1495 Label done; | |
1496 | |
1497 // if no method data exists, and the counter is high enough, make one | |
1498 #ifdef _LP64 | |
1499 bpr(Assembler::rc_nz, false, Assembler::pn, ImethodDataPtr, done); | |
1500 #else | |
1501 tst(ImethodDataPtr); | |
1502 br(Assembler::notZero, false, Assembler::pn, done); | |
1503 #endif | |
1504 | |
1505 // Test to see if we should create a method data oop | |
1506 Address profile_limit(Rtmp, (address)&InvocationCounter::InterpreterProfileLimit); | |
1507 #ifdef _LP64 | |
1508 delayed()->nop(); | |
1509 sethi(profile_limit); | |
1510 #else | |
1511 delayed()->sethi(profile_limit); | |
1512 #endif | |
1513 ld(profile_limit, Rtmp); | |
1514 cmp(invocation_count, Rtmp); | |
1515 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); | |
1516 delayed()->nop(); | |
1517 | |
1518 // Build it now. | |
1519 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), cur_bcp); | |
1520 set_method_data_pointer_offset(O0); | |
1521 ba(false, profile_continue); | |
1522 delayed()->nop(); | |
1523 bind(done); | |
1524 } | |
1525 | |
1526 // Store a value at some constant offset from the method data pointer. | |
1527 | |
1528 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { | |
1529 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1530 st_ptr(value, ImethodDataPtr, constant); | |
1531 } | |
1532 | |
1533 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, | |
1534 Register bumped_count, | |
1535 bool decrement) { | |
1536 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1537 | |
1538 // Load the counter. | |
1539 ld_ptr(counter, bumped_count); | |
1540 | |
1541 if (decrement) { | |
1542 // Decrement the register. Set condition codes. | |
1543 subcc(bumped_count, DataLayout::counter_increment, bumped_count); | |
1544 | |
1545 // If the decrement causes the counter to overflow, stay negative | |
1546 Label L; | |
1547 brx(Assembler::negative, true, Assembler::pn, L); | |
1548 | |
1549 // Store the decremented counter, if it is still negative. | |
1550 delayed()->st_ptr(bumped_count, counter); | |
1551 bind(L); | |
1552 } else { | |
1553 // Increment the register. Set carry flag. | |
1554 addcc(bumped_count, DataLayout::counter_increment, bumped_count); | |
1555 | |
1556 // If the increment causes the counter to overflow, pull back by 1. | |
1557 assert(DataLayout::counter_increment == 1, "subc works"); | |
1558 subc(bumped_count, G0, bumped_count); | |
1559 | |
1560 // Store the incremented counter. | |
1561 st_ptr(bumped_count, counter); | |
1562 } | |
1563 } | |
1564 | |
1565 // Increment the value at some constant offset from the method data pointer. | |
1566 | |
1567 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, | |
1568 Register bumped_count, | |
1569 bool decrement) { | |
1570 // Locate the counter at a fixed offset from the mdp: | |
1571 Address counter(ImethodDataPtr, 0, constant); | |
1572 increment_mdp_data_at(counter, bumped_count, decrement); | |
1573 } | |
1574 | |
1575 // Increment the value at some non-fixed (reg + constant) offset from | |
1576 // the method data pointer. | |
1577 | |
1578 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, | |
1579 int constant, | |
1580 Register bumped_count, | |
1581 Register scratch2, | |
1582 bool decrement) { | |
1583 // Add the constant to reg to get the offset. | |
1584 add(ImethodDataPtr, reg, scratch2); | |
1585 Address counter(scratch2, 0, constant); | |
1586 increment_mdp_data_at(counter, bumped_count, decrement); | |
1587 } | |
1588 | |
1589 // Set a flag value at the current method data pointer position. | |
1590 // Updates a single byte of the header, to avoid races with other header bits. | |
1591 | |
1592 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, | |
1593 Register scratch) { | |
1594 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1595 // Load the data header | |
1596 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); | |
1597 | |
1598 // Set the flag | |
1599 or3(scratch, flag_constant, scratch); | |
1600 | |
1601 // Store the modified header. | |
1602 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); | |
1603 } | |
1604 | |
1605 // Test the location at some offset from the method data pointer. | |
1606 // If it is not equal to value, branch to the not_equal_continue Label. | |
1607 // Set condition codes to match the nullness of the loaded value. | |
1608 | |
1609 void InterpreterMacroAssembler::test_mdp_data_at(int offset, | |
1610 Register value, | |
1611 Label& not_equal_continue, | |
1612 Register scratch) { | |
1613 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1614 ld_ptr(ImethodDataPtr, offset, scratch); | |
1615 cmp(value, scratch); | |
1616 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); | |
1617 delayed()->tst(scratch); | |
1618 } | |
1619 | |
1620 // Update the method data pointer by the displacement located at some fixed | |
1621 // offset from the method data pointer. | |
1622 | |
1623 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, | |
1624 Register scratch) { | |
1625 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1626 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); | |
1627 add(ImethodDataPtr, scratch, ImethodDataPtr); | |
1628 } | |
1629 | |
1630 // Update the method data pointer by the displacement located at the | |
1631 // offset (reg + offset_of_disp). | |
1632 | |
1633 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, | |
1634 int offset_of_disp, | |
1635 Register scratch) { | |
1636 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1637 add(reg, offset_of_disp, scratch); | |
1638 ld_ptr(ImethodDataPtr, scratch, scratch); | |
1639 add(ImethodDataPtr, scratch, ImethodDataPtr); | |
1640 } | |
1641 | |
1642 // Update the method data pointer by a simple constant displacement. | |
1643 | |
1644 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { | |
1645 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1646 add(ImethodDataPtr, constant, ImethodDataPtr); | |
1647 } | |
1648 | |
1649 // Update the method data pointer for a _ret bytecode whose target | |
1650 // was not among our cached targets. | |
1651 | |
1652 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, | |
1653 Register return_bci) { | |
1654 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1655 push(state); | |
1656 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile | |
1657 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); | |
1658 ld_ptr(l_tmp, return_bci); | |
1659 pop(state); | |
1660 } | |
1661 | |
1662 // Count a taken branch in the bytecodes. | |
1663 | |
1664 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { | |
1665 if (ProfileInterpreter) { | |
1666 Label profile_continue; | |
1667 | |
1668 // If no method data exists, go to profile_continue. | |
1669 test_method_data_pointer(profile_continue); | |
1670 | |
1671 // We are taking a branch. Increment the taken count. | |
1672 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); | |
1673 | |
1674 // The method data pointer needs to be updated to reflect the new target. | |
1675 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); | |
1676 bind (profile_continue); | |
1677 } | |
1678 } | |
1679 | |
1680 | |
1681 // Count a not-taken branch in the bytecodes. | |
1682 | |
1683 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { | |
1684 if (ProfileInterpreter) { | |
1685 Label profile_continue; | |
1686 | |
1687 // If no method data exists, go to profile_continue. | |
1688 test_method_data_pointer(profile_continue); | |
1689 | |
1690 // We are taking a branch. Increment the not taken count. | |
1691 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); | |
1692 | |
1693 // The method data pointer needs to be updated to correspond to the | |
1694 // next bytecode. | |
1695 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); | |
1696 bind (profile_continue); | |
1697 } | |
1698 } | |
1699 | |
1700 | |
1701 // Count a non-virtual call in the bytecodes. | |
1702 | |
1703 void InterpreterMacroAssembler::profile_call(Register scratch) { | |
1704 if (ProfileInterpreter) { | |
1705 Label profile_continue; | |
1706 | |
1707 // If no method data exists, go to profile_continue. | |
1708 test_method_data_pointer(profile_continue); | |
1709 | |
1710 // We are making a call. Increment the count. | |
1711 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
1712 | |
1713 // The method data pointer needs to be updated to reflect the new target. | |
1714 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); | |
1715 bind (profile_continue); | |
1716 } | |
1717 } | |
1718 | |
1719 | |
1720 // Count a final call in the bytecodes. | |
1721 | |
1722 void InterpreterMacroAssembler::profile_final_call(Register scratch) { | |
1723 if (ProfileInterpreter) { | |
1724 Label profile_continue; | |
1725 | |
1726 // If no method data exists, go to profile_continue. | |
1727 test_method_data_pointer(profile_continue); | |
1728 | |
1729 // We are making a call. Increment the count. | |
1730 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
1731 | |
1732 // The method data pointer needs to be updated to reflect the new target. | |
1733 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); | |
1734 bind (profile_continue); | |
1735 } | |
1736 } | |
1737 | |
1738 | |
1739 // Count a virtual call in the bytecodes. | |
1740 | |
1741 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, | |
1742 Register scratch) { | |
1743 if (ProfileInterpreter) { | |
1744 Label profile_continue; | |
1745 | |
1746 // If no method data exists, go to profile_continue. | |
1747 test_method_data_pointer(profile_continue); | |
1748 | |
1749 // We are making a call. Increment the count. | |
1750 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
1751 | |
1752 // Record the receiver type. | |
1753 record_klass_in_profile(receiver, scratch); | |
1754 | |
1755 // The method data pointer needs to be updated to reflect the new target. | |
1756 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); | |
1757 bind (profile_continue); | |
1758 } | |
1759 } | |
1760 | |
1761 void InterpreterMacroAssembler::record_klass_in_profile_helper( | |
1762 Register receiver, Register scratch, | |
1763 int start_row, Label& done) { | |
1764 int last_row = VirtualCallData::row_limit() - 1; | |
1765 assert(start_row <= last_row, "must be work left to do"); | |
1766 // Test this row for both the receiver and for null. | |
1767 // Take any of three different outcomes: | |
1768 // 1. found receiver => increment count and goto done | |
1769 // 2. found null => keep looking for case 1, maybe allocate this cell | |
1770 // 3. found something else => keep looking for cases 1 and 2 | |
1771 // Case 3 is handled by a recursive call. | |
1772 for (int row = start_row; row <= last_row; row++) { | |
1773 Label next_test; | |
1774 bool test_for_null_also = (row == start_row); | |
1775 | |
1776 // See if the receiver is receiver[n]. | |
1777 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); | |
1778 test_mdp_data_at(recvr_offset, receiver, next_test, scratch); | |
1779 | |
1780 // The receiver is receiver[n]. Increment count[n]. | |
1781 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); | |
1782 increment_mdp_data_at(count_offset, scratch); | |
1783 ba(false, done); | |
1784 delayed()->nop(); | |
1785 bind(next_test); | |
1786 | |
1787 if (test_for_null_also) { | |
1788 // Failed the equality check on receiver[n]... Test for null. | |
1789 if (start_row == last_row) { | |
1790 // The only thing left to do is handle the null case. | |
1791 brx(Assembler::notZero, false, Assembler::pt, done); | |
1792 delayed()->nop(); | |
1793 break; | |
1794 } | |
1795 // Since null is rare, make it be the branch-taken case. | |
1796 Label found_null; | |
1797 brx(Assembler::zero, false, Assembler::pn, found_null); | |
1798 delayed()->nop(); | |
1799 | |
1800 // Put all the "Case 3" tests here. | |
1801 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done); | |
1802 | |
1803 // Found a null. Keep searching for a matching receiver, | |
1804 // but remember that this is an empty (unused) slot. | |
1805 bind(found_null); | |
1806 } | |
1807 } | |
1808 | |
1809 // In the fall-through case, we found no matching receiver, but we | |
1810 // observed the receiver[start_row] is NULL. | |
1811 | |
1812 // Fill in the receiver field and increment the count. | |
1813 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); | |
1814 set_mdp_data_at(recvr_offset, receiver); | |
1815 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); | |
1816 mov(DataLayout::counter_increment, scratch); | |
1817 set_mdp_data_at(count_offset, scratch); | |
1818 ba(false, done); | |
1819 delayed()->nop(); | |
1820 } | |
1821 | |
1822 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, | |
1823 Register scratch) { | |
1824 assert(ProfileInterpreter, "must be profiling"); | |
1825 Label done; | |
1826 | |
1827 record_klass_in_profile_helper(receiver, scratch, 0, done); | |
1828 | |
1829 bind (done); | |
1830 } | |
1831 | |
1832 | |
1833 // Count a ret in the bytecodes. | |
1834 | |
1835 void InterpreterMacroAssembler::profile_ret(TosState state, | |
1836 Register return_bci, | |
1837 Register scratch) { | |
1838 if (ProfileInterpreter) { | |
1839 Label profile_continue; | |
1840 uint row; | |
1841 | |
1842 // If no method data exists, go to profile_continue. | |
1843 test_method_data_pointer(profile_continue); | |
1844 | |
1845 // Update the total ret count. | |
1846 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
1847 | |
1848 for (row = 0; row < RetData::row_limit(); row++) { | |
1849 Label next_test; | |
1850 | |
1851 // See if return_bci is equal to bci[n]: | |
1852 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), | |
1853 return_bci, next_test, scratch); | |
1854 | |
1855 // return_bci is equal to bci[n]. Increment the count. | |
1856 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); | |
1857 | |
1858 // The method data pointer needs to be updated to reflect the new target. | |
1859 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); | |
1860 ba(false, profile_continue); | |
1861 delayed()->nop(); | |
1862 bind(next_test); | |
1863 } | |
1864 | |
1865 update_mdp_for_ret(state, return_bci); | |
1866 | |
1867 bind (profile_continue); | |
1868 } | |
1869 } | |
1870 | |
1871 // Profile an unexpected null in the bytecodes. | |
1872 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { | |
1873 if (ProfileInterpreter) { | |
1874 Label profile_continue; | |
1875 | |
1876 // If no method data exists, go to profile_continue. | |
1877 test_method_data_pointer(profile_continue); | |
1878 | |
1879 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); | |
1880 | |
1881 // The method data pointer needs to be updated. | |
1882 int mdp_delta = in_bytes(BitData::bit_data_size()); | |
1883 if (TypeProfileCasts) { | |
1884 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); | |
1885 } | |
1886 update_mdp_by_constant(mdp_delta); | |
1887 | |
1888 bind (profile_continue); | |
1889 } | |
1890 } | |
1891 | |
1892 void InterpreterMacroAssembler::profile_typecheck(Register klass, | |
1893 Register scratch) { | |
1894 if (ProfileInterpreter) { | |
1895 Label profile_continue; | |
1896 | |
1897 // If no method data exists, go to profile_continue. | |
1898 test_method_data_pointer(profile_continue); | |
1899 | |
1900 int mdp_delta = in_bytes(BitData::bit_data_size()); | |
1901 if (TypeProfileCasts) { | |
1902 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); | |
1903 | |
1904 // Record the object type. | |
1905 record_klass_in_profile(klass, scratch); | |
1906 } | |
1907 | |
1908 // The method data pointer needs to be updated. | |
1909 update_mdp_by_constant(mdp_delta); | |
1910 | |
1911 bind (profile_continue); | |
1912 } | |
1913 } | |
1914 | |
1915 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { | |
1916 if (ProfileInterpreter && TypeProfileCasts) { | |
1917 Label profile_continue; | |
1918 | |
1919 // If no method data exists, go to profile_continue. | |
1920 test_method_data_pointer(profile_continue); | |
1921 | |
1922 int count_offset = in_bytes(CounterData::count_offset()); | |
1923 // Back up the address, since we have already bumped the mdp. | |
1924 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); | |
1925 | |
1926 // *Decrement* the counter. We expect to see zero or small negatives. | |
1927 increment_mdp_data_at(count_offset, scratch, true); | |
1928 | |
1929 bind (profile_continue); | |
1930 } | |
1931 } | |
1932 | |
1933 // Count the default case of a switch construct. | |
1934 | |
1935 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { | |
1936 if (ProfileInterpreter) { | |
1937 Label profile_continue; | |
1938 | |
1939 // If no method data exists, go to profile_continue. | |
1940 test_method_data_pointer(profile_continue); | |
1941 | |
1942 // Update the default case count | |
1943 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), | |
1944 scratch); | |
1945 | |
1946 // The method data pointer needs to be updated. | |
1947 update_mdp_by_offset( | |
1948 in_bytes(MultiBranchData::default_displacement_offset()), | |
1949 scratch); | |
1950 | |
1951 bind (profile_continue); | |
1952 } | |
1953 } | |
1954 | |
1955 // Count the index'th case of a switch construct. | |
1956 | |
1957 void InterpreterMacroAssembler::profile_switch_case(Register index, | |
1958 Register scratch, | |
1959 Register scratch2, | |
1960 Register scratch3) { | |
1961 if (ProfileInterpreter) { | |
1962 Label profile_continue; | |
1963 | |
1964 // If no method data exists, go to profile_continue. | |
1965 test_method_data_pointer(profile_continue); | |
1966 | |
1967 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() | |
1968 set(in_bytes(MultiBranchData::per_case_size()), scratch); | |
1969 smul(index, scratch, scratch); | |
1970 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); | |
1971 | |
1972 // Update the case count | |
1973 increment_mdp_data_at(scratch, | |
1974 in_bytes(MultiBranchData::relative_count_offset()), | |
1975 scratch2, | |
1976 scratch3); | |
1977 | |
1978 // The method data pointer needs to be updated. | |
1979 update_mdp_by_offset(scratch, | |
1980 in_bytes(MultiBranchData::relative_displacement_offset()), | |
1981 scratch2); | |
1982 | |
1983 bind (profile_continue); | |
1984 } | |
1985 } | |
1986 | |
1987 // add a InterpMonitorElem to stack (see frame_sparc.hpp) | |
1988 | |
1989 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, | |
1990 Register Rtemp, | |
1991 Register Rtemp2 ) { | |
1992 | |
1993 Register Rlimit = Lmonitors; | |
1994 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
1995 assert( (delta & LongAlignmentMask) == 0, | |
1996 "sizeof BasicObjectLock must be even number of doublewords"); | |
1997 | |
1998 sub( SP, delta, SP); | |
1999 sub( Lesp, delta, Lesp); | |
2000 sub( Lmonitors, delta, Lmonitors); | |
2001 | |
2002 if (!stack_is_empty) { | |
2003 | |
2004 // must copy stack contents down | |
2005 | |
2006 Label start_copying, next; | |
2007 | |
2008 // untested("monitor stack expansion"); | |
2009 compute_stack_base(Rtemp); | |
2010 ba( false, start_copying ); | |
2011 delayed()->cmp( Rtemp, Rlimit); // done? duplicated below | |
2012 | |
2013 // note: must copy from low memory upwards | |
2014 // On entry to loop, | |
2015 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) | |
2016 // Loop mutates Rtemp | |
2017 | |
2018 bind( next); | |
2019 | |
2020 st_ptr(Rtemp2, Rtemp, 0); | |
2021 inc(Rtemp, wordSize); | |
2022 cmp(Rtemp, Rlimit); // are we done? (duplicated above) | |
2023 | |
2024 bind( start_copying ); | |
2025 | |
2026 brx( notEqual, true, pn, next ); | |
2027 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); | |
2028 | |
2029 // done copying stack | |
2030 } | |
2031 } | |
2032 | |
2033 // Locals | |
2034 #ifdef ASSERT | |
2035 void InterpreterMacroAssembler::verify_local_tag(frame::Tag t, | |
2036 Register base, | |
2037 Register scratch, | |
2038 int n) { | |
2039 if (TaggedStackInterpreter) { | |
2040 Label ok, long_ok; | |
2041 // Use dst for scratch | |
2042 assert_different_registers(base, scratch); | |
2043 ld_ptr(base, Interpreter::local_tag_offset_in_bytes(n), scratch); | |
2044 if (t == frame::TagCategory2) { | |
2045 cmp(scratch, G0); | |
2046 brx(Assembler::equal, false, Assembler::pt, long_ok); | |
2047 delayed()->ld_ptr(base, Interpreter::local_tag_offset_in_bytes(n+1), scratch); | |
2048 stop("local long/double tag value bad"); | |
2049 bind(long_ok); | |
2050 // compare second half tag | |
2051 cmp(scratch, G0); | |
2052 } else if (t == frame::TagValue) { | |
2053 cmp(scratch, G0); | |
2054 } else { | |
2055 assert_different_registers(O3, base, scratch); | |
2056 mov(t, O3); | |
2057 cmp(scratch, O3); | |
2058 } | |
2059 brx(Assembler::equal, false, Assembler::pt, ok); | |
2060 delayed()->nop(); | |
2061 // Also compare if the local value is zero, then the tag might | |
2062 // not have been set coming from deopt. | |
2063 ld_ptr(base, Interpreter::local_offset_in_bytes(n), scratch); | |
2064 cmp(scratch, G0); | |
2065 brx(Assembler::equal, false, Assembler::pt, ok); | |
2066 delayed()->nop(); | |
2067 stop("Local tag value is bad"); | |
2068 bind(ok); | |
2069 } | |
2070 } | |
2071 #endif // ASSERT | |
2072 | |
2073 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { | |
2074 assert_not_delayed(); | |
2075 sll(index, Interpreter::logStackElementSize(), index); | |
2076 sub(Llocals, index, index); | |
2077 debug_only(verify_local_tag(frame::TagReference, index, dst)); | |
2078 ld_ptr(index, Interpreter::value_offset_in_bytes(), dst); | |
2079 // Note: index must hold the effective address--the iinc template uses it | |
2080 } | |
2081 | |
2082 // Just like access_local_ptr but the tag is a returnAddress | |
2083 void InterpreterMacroAssembler::access_local_returnAddress(Register index, | |
2084 Register dst ) { | |
2085 assert_not_delayed(); | |
2086 sll(index, Interpreter::logStackElementSize(), index); | |
2087 sub(Llocals, index, index); | |
2088 debug_only(verify_local_tag(frame::TagValue, index, dst)); | |
2089 ld_ptr(index, Interpreter::value_offset_in_bytes(), dst); | |
2090 } | |
2091 | |
2092 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { | |
2093 assert_not_delayed(); | |
2094 sll(index, Interpreter::logStackElementSize(), index); | |
2095 sub(Llocals, index, index); | |
2096 debug_only(verify_local_tag(frame::TagValue, index, dst)); | |
2097 ld(index, Interpreter::value_offset_in_bytes(), dst); | |
2098 // Note: index must hold the effective address--the iinc template uses it | |
2099 } | |
2100 | |
2101 | |
2102 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { | |
2103 assert_not_delayed(); | |
2104 sll(index, Interpreter::logStackElementSize(), index); | |
2105 sub(Llocals, index, index); | |
2106 debug_only(verify_local_tag(frame::TagCategory2, index, dst)); | |
2107 // First half stored at index n+1 (which grows down from Llocals[n]) | |
2108 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); | |
2109 } | |
2110 | |
2111 | |
2112 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { | |
2113 assert_not_delayed(); | |
2114 sll(index, Interpreter::logStackElementSize(), index); | |
2115 sub(Llocals, index, index); | |
2116 debug_only(verify_local_tag(frame::TagValue, index, G1_scratch)); | |
2117 ldf(FloatRegisterImpl::S, index, Interpreter::value_offset_in_bytes(), dst); | |
2118 } | |
2119 | |
2120 | |
2121 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { | |
2122 assert_not_delayed(); | |
2123 sll(index, Interpreter::logStackElementSize(), index); | |
2124 sub(Llocals, index, index); | |
2125 debug_only(verify_local_tag(frame::TagCategory2, index, G1_scratch)); | |
2126 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); | |
2127 } | |
2128 | |
2129 | |
2130 #ifdef ASSERT | |
2131 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { | |
2132 Label L; | |
2133 | |
2134 assert(Rindex != Rscratch, "Registers cannot be same"); | |
2135 assert(Rindex != Rscratch1, "Registers cannot be same"); | |
2136 assert(Rlimit != Rscratch, "Registers cannot be same"); | |
2137 assert(Rlimit != Rscratch1, "Registers cannot be same"); | |
2138 assert(Rscratch1 != Rscratch, "Registers cannot be same"); | |
2139 | |
2140 // untested("reg area corruption"); | |
2141 add(Rindex, offset, Rscratch); | |
2142 add(Rlimit, 64 + STACK_BIAS, Rscratch1); | |
2143 cmp(Rscratch, Rscratch1); | |
2144 brx(Assembler::greaterEqualUnsigned, false, pn, L); | |
2145 delayed()->nop(); | |
2146 stop("regsave area is being clobbered"); | |
2147 bind(L); | |
2148 } | |
2149 #endif // ASSERT | |
2150 | |
2151 void InterpreterMacroAssembler::tag_local(frame::Tag t, | |
2152 Register base, | |
2153 Register src, | |
2154 int n) { | |
2155 if (TaggedStackInterpreter) { | |
2156 // have to store zero because local slots can be reused (rats!) | |
2157 if (t == frame::TagValue) { | |
2158 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n)); | |
2159 } else if (t == frame::TagCategory2) { | |
2160 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n)); | |
2161 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n+1)); | |
2162 } else { | |
2163 // assert that we don't stomp the value in 'src' | |
2164 // O3 is arbitrary because it's not used. | |
2165 assert_different_registers(src, base, O3); | |
2166 mov( t, O3); | |
2167 st_ptr(O3, base, Interpreter::local_tag_offset_in_bytes(n)); | |
2168 } | |
2169 } | |
2170 } | |
2171 | |
2172 | |
2173 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { | |
2174 assert_not_delayed(); | |
2175 sll(index, Interpreter::logStackElementSize(), index); | |
2176 sub(Llocals, index, index); | |
2177 debug_only(check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch);) | |
2178 tag_local(frame::TagValue, index, src); | |
2179 st(src, index, Interpreter::value_offset_in_bytes()); | |
2180 } | |
2181 | |
2182 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src, | |
2183 Register tag ) { | |
2184 assert_not_delayed(); | |
2185 sll(index, Interpreter::logStackElementSize(), index); | |
2186 sub(Llocals, index, index); | |
2187 #ifdef ASSERT | |
2188 check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch); | |
2189 #endif | |
2190 st_ptr(src, index, Interpreter::value_offset_in_bytes()); | |
2191 // Store tag register directly | |
2192 if (TaggedStackInterpreter) { | |
2193 st_ptr(tag, index, Interpreter::tag_offset_in_bytes()); | |
2194 } | |
2195 } | |
2196 | |
2197 | |
2198 | |
2199 void InterpreterMacroAssembler::store_local_ptr( int n, Register src, | |
2200 Register tag ) { | |
2201 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); | |
2202 if (TaggedStackInterpreter) { | |
2203 st_ptr(tag, Llocals, Interpreter::local_tag_offset_in_bytes(n)); | |
2204 } | |
2205 } | |
2206 | |
2207 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { | |
2208 assert_not_delayed(); | |
2209 sll(index, Interpreter::logStackElementSize(), index); | |
2210 sub(Llocals, index, index); | |
2211 #ifdef ASSERT | |
2212 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); | |
2213 #endif | |
2214 tag_local(frame::TagCategory2, index, src); | |
2215 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 | |
2216 } | |
2217 | |
2218 | |
2219 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { | |
2220 assert_not_delayed(); | |
2221 sll(index, Interpreter::logStackElementSize(), index); | |
2222 sub(Llocals, index, index); | |
2223 #ifdef ASSERT | |
2224 check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch); | |
2225 #endif | |
2226 tag_local(frame::TagValue, index, G1_scratch); | |
2227 stf(FloatRegisterImpl::S, src, index, Interpreter::value_offset_in_bytes()); | |
2228 } | |
2229 | |
2230 | |
2231 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { | |
2232 assert_not_delayed(); | |
2233 sll(index, Interpreter::logStackElementSize(), index); | |
2234 sub(Llocals, index, index); | |
2235 #ifdef ASSERT | |
2236 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); | |
2237 #endif | |
2238 tag_local(frame::TagCategory2, index, G1_scratch); | |
2239 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); | |
2240 } | |
2241 | |
2242 | |
2243 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { | |
2244 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
2245 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong); | |
2246 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; | |
2247 } | |
2248 | |
2249 | |
2250 Address InterpreterMacroAssembler::top_most_monitor() { | |
2251 return Address(FP, 0, top_most_monitor_byte_offset()); | |
2252 } | |
2253 | |
2254 | |
2255 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { | |
2256 add( Lesp, wordSize, Rdest ); | |
2257 } | |
2258 | |
2259 #endif /* CC_INTERP */ | |
2260 | |
2261 void InterpreterMacroAssembler::increment_invocation_counter( Register Rtmp, Register Rtmp2 ) { | |
2262 assert(UseCompiler, "incrementing must be useful"); | |
2263 #ifdef CC_INTERP | |
2264 Address inv_counter(G5_method, 0, in_bytes(methodOopDesc::invocation_counter_offset() | |
2265 + InvocationCounter::counter_offset())); | |
2266 Address be_counter(G5_method, 0, in_bytes(methodOopDesc::backedge_counter_offset() | |
2267 + InvocationCounter::counter_offset())); | |
2268 #else | |
2269 Address inv_counter(Lmethod, 0, in_bytes(methodOopDesc::invocation_counter_offset() | |
2270 + InvocationCounter::counter_offset())); | |
2271 Address be_counter(Lmethod, 0, in_bytes(methodOopDesc::backedge_counter_offset() | |
2272 + InvocationCounter::counter_offset())); | |
2273 #endif /* CC_INTERP */ | |
2274 int delta = InvocationCounter::count_increment; | |
2275 | |
2276 // Load each counter in a register | |
2277 ld( inv_counter, Rtmp ); | |
2278 ld( be_counter, Rtmp2 ); | |
2279 | |
2280 assert( is_simm13( delta ), " delta too large."); | |
2281 | |
2282 // Add the delta to the invocation counter and store the result | |
2283 add( Rtmp, delta, Rtmp ); | |
2284 | |
2285 // Mask the backedge counter | |
2286 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); | |
2287 | |
2288 // Store value | |
2289 st( Rtmp, inv_counter); | |
2290 | |
2291 // Add invocation counter + backedge counter | |
2292 add( Rtmp, Rtmp2, Rtmp); | |
2293 | |
2294 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! | |
2295 } | |
2296 | |
2297 void InterpreterMacroAssembler::increment_backedge_counter( Register Rtmp, Register Rtmp2 ) { | |
2298 assert(UseCompiler, "incrementing must be useful"); | |
2299 #ifdef CC_INTERP | |
2300 Address be_counter(G5_method, 0, in_bytes(methodOopDesc::backedge_counter_offset() | |
2301 + InvocationCounter::counter_offset())); | |
2302 Address inv_counter(G5_method, 0, in_bytes(methodOopDesc::invocation_counter_offset() | |
2303 + InvocationCounter::counter_offset())); | |
2304 #else | |
2305 Address be_counter(Lmethod, 0, in_bytes(methodOopDesc::backedge_counter_offset() | |
2306 + InvocationCounter::counter_offset())); | |
2307 Address inv_counter(Lmethod, 0, in_bytes(methodOopDesc::invocation_counter_offset() | |
2308 + InvocationCounter::counter_offset())); | |
2309 #endif /* CC_INTERP */ | |
2310 int delta = InvocationCounter::count_increment; | |
2311 // Load each counter in a register | |
2312 ld( be_counter, Rtmp ); | |
2313 ld( inv_counter, Rtmp2 ); | |
2314 | |
2315 // Add the delta to the backedge counter | |
2316 add( Rtmp, delta, Rtmp ); | |
2317 | |
2318 // Mask the invocation counter, add to backedge counter | |
2319 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); | |
2320 | |
2321 // and store the result to memory | |
2322 st( Rtmp, be_counter ); | |
2323 | |
2324 // Add backedge + invocation counter | |
2325 add( Rtmp, Rtmp2, Rtmp ); | |
2326 | |
2327 // Note that this macro must leave backedge_count + invocation_count in Rtmp! | |
2328 } | |
2329 | |
2330 #ifndef CC_INTERP | |
2331 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, | |
2332 Register branch_bcp, | |
2333 Register Rtmp ) { | |
2334 Label did_not_overflow; | |
2335 Label overflow_with_error; | |
2336 assert_different_registers(backedge_count, Rtmp, branch_bcp); | |
2337 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); | |
2338 | |
2339 Address limit(Rtmp, address(&InvocationCounter::InterpreterBackwardBranchLimit)); | |
2340 load_contents(limit, Rtmp); | |
2341 cmp(backedge_count, Rtmp); | |
2342 br(Assembler::lessUnsigned, false, Assembler::pt, did_not_overflow); | |
2343 delayed()->nop(); | |
2344 | |
2345 // When ProfileInterpreter is on, the backedge_count comes from the | |
2346 // methodDataOop, which value does not get reset on the call to | |
2347 // frequency_counter_overflow(). To avoid excessive calls to the overflow | |
2348 // routine while the method is being compiled, add a second test to make sure | |
2349 // the overflow function is called only once every overflow_frequency. | |
2350 if (ProfileInterpreter) { | |
2351 const int overflow_frequency = 1024; | |
2352 andcc(backedge_count, overflow_frequency-1, Rtmp); | |
2353 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); | |
2354 delayed()->nop(); | |
2355 } | |
2356 | |
2357 // overflow in loop, pass branch bytecode | |
2358 set(6,Rtmp); | |
2359 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); | |
2360 | |
2361 // Was an OSR adapter generated? | |
2362 // O0 = osr nmethod | |
2363 tst(O0); | |
2364 brx(Assembler::zero, false, Assembler::pn, overflow_with_error); | |
2365 delayed()->nop(); | |
2366 | |
2367 // Has the nmethod been invalidated already? | |
2368 ld(O0, nmethod::entry_bci_offset(), O2); | |
2369 cmp(O2, InvalidOSREntryBci); | |
2370 br(Assembler::equal, false, Assembler::pn, overflow_with_error); | |
2371 delayed()->nop(); | |
2372 | |
2373 // migrate the interpreter frame off of the stack | |
2374 | |
2375 mov(G2_thread, L7); | |
2376 // save nmethod | |
2377 mov(O0, L6); | |
2378 set_last_Java_frame(SP, noreg); | |
2379 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); | |
2380 reset_last_Java_frame(); | |
2381 mov(L7, G2_thread); | |
2382 | |
2383 // move OSR nmethod to I1 | |
2384 mov(L6, I1); | |
2385 | |
2386 // OSR buffer to I0 | |
2387 mov(O0, I0); | |
2388 | |
2389 // remove the interpreter frame | |
2390 restore(I5_savedSP, 0, SP); | |
2391 | |
2392 // Jump to the osr code. | |
2393 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); | |
2394 jmp(O2, G0); | |
2395 delayed()->nop(); | |
2396 | |
2397 bind(overflow_with_error); | |
2398 | |
2399 bind(did_not_overflow); | |
2400 } | |
2401 | |
2402 | |
2403 | |
2404 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { | |
2405 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } | |
2406 } | |
2407 | |
2408 | |
2409 // local helper function for the verify_oop_or_return_address macro | |
2410 static bool verify_return_address(methodOopDesc* m, int bci) { | |
2411 #ifndef PRODUCT | |
2412 address pc = (address)(m->constMethod()) | |
2413 + in_bytes(constMethodOopDesc::codes_offset()) + bci; | |
2414 // assume it is a valid return address if it is inside m and is preceded by a jsr | |
2415 if (!m->contains(pc)) return false; | |
2416 address jsr_pc; | |
2417 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); | |
2418 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; | |
2419 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); | |
2420 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; | |
2421 #endif // PRODUCT | |
2422 return false; | |
2423 } | |
2424 | |
2425 | |
2426 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { | |
2427 if (!VerifyOops) return; | |
2428 // the VM documentation for the astore[_wide] bytecode allows | |
2429 // the TOS to be not only an oop but also a return address | |
2430 Label test; | |
2431 Label skip; | |
2432 // See if it is an address (in the current method): | |
2433 | |
2434 mov(reg, Rtmp); | |
2435 const int log2_bytecode_size_limit = 16; | |
2436 srl(Rtmp, log2_bytecode_size_limit, Rtmp); | |
2437 br_notnull( Rtmp, false, pt, test ); | |
2438 delayed()->nop(); | |
2439 | |
2440 // %%% should use call_VM_leaf here? | |
2441 save_frame_and_mov(0, Lmethod, O0, reg, O1); | |
2442 save_thread(L7_thread_cache); | |
2443 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); | |
2444 delayed()->nop(); | |
2445 restore_thread(L7_thread_cache); | |
2446 br_notnull( O0, false, pt, skip ); | |
2447 delayed()->restore(); | |
2448 | |
2449 // Perform a more elaborate out-of-line call | |
2450 // Not an address; verify it: | |
2451 bind(test); | |
2452 verify_oop(reg); | |
2453 bind(skip); | |
2454 } | |
2455 | |
2456 | |
2457 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { | |
2458 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); | |
2459 } | |
2460 #endif /* CC_INTERP */ | |
2461 | |
2462 // Inline assembly for: | |
2463 // | |
2464 // if (thread is in interp_only_mode) { | |
2465 // InterpreterRuntime::post_method_entry(); | |
2466 // } | |
2467 // if (DTraceMethodProbes) { | |
2468 // SharedRuntime::dtrace_method_entry(method, reciever); | |
2469 // } | |
2470 | |
2471 void InterpreterMacroAssembler::notify_method_entry() { | |
2472 | |
2473 // C++ interpreter only uses this for native methods. | |
2474 | |
2475 // Whenever JVMTI puts a thread in interp_only_mode, method | |
2476 // entry/exit events are sent for that thread to track stack | |
2477 // depth. If it is possible to enter interp_only_mode we add | |
2478 // the code to check if the event should be sent. | |
2479 if (JvmtiExport::can_post_interpreter_events()) { | |
2480 Label L; | |
2481 Register temp_reg = O5; | |
2482 | |
2483 const Address interp_only (G2_thread, 0, in_bytes(JavaThread::interp_only_mode_offset())); | |
2484 | |
2485 ld(interp_only, temp_reg); | |
2486 tst(temp_reg); | |
2487 br(zero, false, pt, L); | |
2488 delayed()->nop(); | |
2489 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); | |
2490 bind(L); | |
2491 } | |
2492 | |
2493 { | |
2494 Register temp_reg = O5; | |
2495 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); | |
2496 call_VM_leaf(noreg, | |
2497 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), | |
2498 G2_thread, Lmethod); | |
2499 } | |
2500 } | |
2501 | |
2502 | |
2503 // Inline assembly for: | |
2504 // | |
2505 // if (thread is in interp_only_mode) { | |
2506 // // save result | |
2507 // InterpreterRuntime::post_method_exit(); | |
2508 // // restore result | |
2509 // } | |
2510 // if (DTraceMethodProbes) { | |
2511 // SharedRuntime::dtrace_method_exit(thread, method); | |
2512 // } | |
2513 // | |
2514 // Native methods have their result stored in d_tmp and l_tmp | |
2515 // Java methods have their result stored in the expression stack | |
2516 | |
2517 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, | |
2518 TosState state, | |
2519 NotifyMethodExitMode mode) { | |
2520 // C++ interpreter only uses this for native methods. | |
2521 | |
2522 // Whenever JVMTI puts a thread in interp_only_mode, method | |
2523 // entry/exit events are sent for that thread to track stack | |
2524 // depth. If it is possible to enter interp_only_mode we add | |
2525 // the code to check if the event should be sent. | |
2526 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { | |
2527 Label L; | |
2528 Register temp_reg = O5; | |
2529 | |
2530 const Address interp_only (G2_thread, 0, in_bytes(JavaThread::interp_only_mode_offset())); | |
2531 | |
2532 ld(interp_only, temp_reg); | |
2533 tst(temp_reg); | |
2534 br(zero, false, pt, L); | |
2535 delayed()->nop(); | |
2536 | |
2537 // Note: frame::interpreter_frame_result has a dependency on how the | |
2538 // method result is saved across the call to post_method_exit. For | |
2539 // native methods it assumes the result registers are saved to | |
2540 // l_scratch and d_scratch. If this changes then the interpreter_frame_result | |
2541 // implementation will need to be updated too. | |
2542 | |
2543 save_return_value(state, is_native_method); | |
2544 call_VM(noreg, | |
2545 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); | |
2546 restore_return_value(state, is_native_method); | |
2547 bind(L); | |
2548 } | |
2549 | |
2550 { | |
2551 Register temp_reg = O5; | |
2552 // Dtrace notification | |
2553 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); | |
2554 save_return_value(state, is_native_method); | |
2555 call_VM_leaf( | |
2556 noreg, | |
2557 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), | |
2558 G2_thread, Lmethod); | |
2559 restore_return_value(state, is_native_method); | |
2560 } | |
2561 } | |
2562 | |
2563 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { | |
2564 #ifdef CC_INTERP | |
2565 // result potentially in O0/O1: save it across calls | |
2566 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult)); | |
2567 #ifdef _LP64 | |
2568 stx(O0, STATE(_native_lresult)); | |
2569 #else | |
2570 std(O0, STATE(_native_lresult)); | |
2571 #endif | |
2572 #else // CC_INTERP | |
2573 if (is_native_call) { | |
2574 stf(FloatRegisterImpl::D, F0, d_tmp); | |
2575 #ifdef _LP64 | |
2576 stx(O0, l_tmp); | |
2577 #else | |
2578 std(O0, l_tmp); | |
2579 #endif | |
2580 } else { | |
2581 push(state); | |
2582 } | |
2583 #endif // CC_INTERP | |
2584 } | |
2585 | |
2586 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { | |
2587 #ifdef CC_INTERP | |
2588 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0); | |
2589 #ifdef _LP64 | |
2590 ldx(STATE(_native_lresult), O0); | |
2591 #else | |
2592 ldd(STATE(_native_lresult), O0); | |
2593 #endif | |
2594 #else // CC_INTERP | |
2595 if (is_native_call) { | |
2596 ldf(FloatRegisterImpl::D, d_tmp, F0); | |
2597 #ifdef _LP64 | |
2598 ldx(l_tmp, O0); | |
2599 #else | |
2600 ldd(l_tmp, O0); | |
2601 #endif | |
2602 } else { | |
2603 pop(state); | |
2604 } | |
2605 #endif // CC_INTERP | |
2606 } |