Mercurial > hg > graal-compiler
annotate src/cpu/sparc/vm/interp_masm_sparc.cpp @ 7183:b2dbd323c668
8003848: Make ConstMethod::generic_signature_index optional and move Method::_max_stack to ConstMethod.
Summary: Make ConstMethod::generic_signature_index optional and move Method::_max_stack to ConstMethod.
Reviewed-by: bdelsart, sspitsyn, coleenp
author | jiangli |
---|---|
date | Tue, 27 Nov 2012 17:03:56 -0500 |
parents | 69fb89ec6fa7 |
children | 5505fbbae3d3 |
rev | line source |
---|---|
0 | 1 /* |
6123
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
2 * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved. |
0 | 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 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1506
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1506
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1506
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 25 #include "precompiled.hpp" |
26 #include "interp_masm_sparc.hpp" | |
27 #include "interpreter/interpreter.hpp" | |
28 #include "interpreter/interpreterRuntime.hpp" | |
29 #include "oops/arrayOop.hpp" | |
30 #include "oops/markOop.hpp" | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
31 #include "oops/methodData.hpp" |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
32 #include "oops/method.hpp" |
1972 | 33 #include "prims/jvmtiExport.hpp" |
34 #include "prims/jvmtiRedefineClassesTrace.hpp" | |
35 #include "prims/jvmtiThreadState.hpp" | |
36 #include "runtime/basicLock.hpp" | |
37 #include "runtime/biasedLocking.hpp" | |
38 #include "runtime/sharedRuntime.hpp" | |
39 #ifdef TARGET_OS_FAMILY_linux | |
40 # include "thread_linux.inline.hpp" | |
41 #endif | |
42 #ifdef TARGET_OS_FAMILY_solaris | |
43 # include "thread_solaris.inline.hpp" | |
44 #endif | |
0 | 45 |
46 #ifndef CC_INTERP | |
47 #ifndef FAST_DISPATCH | |
48 #define FAST_DISPATCH 1 | |
49 #endif | |
50 #undef FAST_DISPATCH | |
51 | |
52 // Implementation of InterpreterMacroAssembler | |
53 | |
54 // This file specializes the assember with interpreter-specific macros | |
55 | |
727 | 56 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); |
57 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); | |
0 | 58 |
59 #else // CC_INTERP | |
60 #ifndef STATE | |
61 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name)) | |
62 #endif // STATE | |
63 | |
64 #endif // CC_INTERP | |
65 | |
66 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { | |
67 // Note: this algorithm is also used by C1's OSR entry sequence. | |
68 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). | |
69 assert_different_registers(args_size, locals_size); | |
70 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. | |
71 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words | |
72 // Use br/mov combination because it works on both V8 and V9 and is | |
73 // faster. | |
74 Label skip_move; | |
75 br(Assembler::negative, true, Assembler::pt, skip_move); | |
76 delayed()->mov(G0, delta); | |
77 bind(skip_move); | |
78 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) | |
79 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes | |
80 } | |
81 | |
82 #ifndef CC_INTERP | |
83 | |
84 // Dispatch code executed in the prolog of a bytecode which does not do it's | |
85 // own dispatch. The dispatch address is computed and placed in IdispatchAddress | |
86 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { | |
87 assert_not_delayed(); | |
88 #ifdef FAST_DISPATCH | |
89 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since | |
90 // they both use I2. | |
91 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); | |
92 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
93 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); | |
94 // add offset to correct dispatch table | |
95 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
96 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr | |
97 #else | |
727 | 98 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode |
0 | 99 // dispatch table to use |
727 | 100 AddressLiteral tbl(Interpreter::dispatch_table(state)); |
101 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
102 set(tbl, G3_scratch); // compute addr of table | |
103 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr | |
0 | 104 #endif |
105 } | |
106 | |
107 | |
108 // Dispatch code executed in the epilog of a bytecode which does not do it's | |
109 // own dispatch. The dispatch address in IdispatchAddress is used for the | |
110 // dispatch. | |
111 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { | |
112 assert_not_delayed(); | |
113 verify_FPU(1, state); | |
114 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
115 jmp( IdispatchAddress, 0 ); | |
116 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); | |
117 else delayed()->nop(); | |
118 } | |
119 | |
120 | |
121 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { | |
122 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) | |
123 assert_not_delayed(); | |
124 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
125 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr); | |
126 } | |
127 | |
128 | |
129 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { | |
130 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) | |
131 assert_not_delayed(); | |
132 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
133 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); | |
134 } | |
135 | |
136 | |
137 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { | |
138 // load current bytecode | |
139 assert_not_delayed(); | |
140 ldub( Lbcp, 0, Lbyte_code); // load next bytecode | |
141 dispatch_base(state, table); | |
142 } | |
143 | |
144 | |
145 void InterpreterMacroAssembler::call_VM_leaf_base( | |
146 Register java_thread, | |
147 address entry_point, | |
148 int number_of_arguments | |
149 ) { | |
150 if (!java_thread->is_valid()) | |
151 java_thread = L7_thread_cache; | |
152 // super call | |
153 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); | |
154 } | |
155 | |
156 | |
157 void InterpreterMacroAssembler::call_VM_base( | |
158 Register oop_result, | |
159 Register java_thread, | |
160 Register last_java_sp, | |
161 address entry_point, | |
162 int number_of_arguments, | |
163 bool check_exception | |
164 ) { | |
165 if (!java_thread->is_valid()) | |
166 java_thread = L7_thread_cache; | |
167 // See class ThreadInVMfromInterpreter, which assumes that the interpreter | |
168 // takes responsibility for setting its own thread-state on call-out. | |
169 // However, ThreadInVMfromInterpreter resets the state to "in_Java". | |
170 | |
171 //save_bcp(); // save bcp | |
172 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); | |
173 //restore_bcp(); // restore bcp | |
174 //restore_locals(); // restore locals pointer | |
175 } | |
176 | |
177 | |
178 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { | |
179 if (JvmtiExport::can_pop_frame()) { | |
180 Label L; | |
181 | |
182 // Check the "pending popframe condition" flag in the current thread | |
727 | 183 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg); |
0 | 184 |
185 // Initiate popframe handling only if it is not already being processed. If the flag | |
186 // has the popframe_processing bit set, it means that this code is called *during* popframe | |
187 // handling - we don't want to reenter. | |
188 btst(JavaThread::popframe_pending_bit, scratch_reg); | |
189 br(zero, false, pt, L); | |
190 delayed()->nop(); | |
191 btst(JavaThread::popframe_processing_bit, scratch_reg); | |
192 br(notZero, false, pt, L); | |
193 delayed()->nop(); | |
194 | |
195 // Call Interpreter::remove_activation_preserving_args_entry() to get the | |
196 // address of the same-named entrypoint in the generated interpreter code. | |
197 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); | |
198 | |
199 // Jump to Interpreter::_remove_activation_preserving_args_entry | |
200 jmpl(O0, G0, G0); | |
201 delayed()->nop(); | |
202 bind(L); | |
203 } | |
204 } | |
205 | |
206 | |
207 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { | |
208 Register thr_state = G4_scratch; | |
727 | 209 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); |
210 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset()); | |
211 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset()); | |
212 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset()); | |
0 | 213 switch (state) { |
214 case ltos: ld_long(val_addr, Otos_l); break; | |
215 case atos: ld_ptr(oop_addr, Otos_l); | |
216 st_ptr(G0, oop_addr); break; | |
217 case btos: // fall through | |
218 case ctos: // fall through | |
219 case stos: // fall through | |
220 case itos: ld(val_addr, Otos_l1); break; | |
221 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; | |
222 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; | |
223 case vtos: /* nothing to do */ break; | |
224 default : ShouldNotReachHere(); | |
225 } | |
226 // Clean up tos value in the jvmti thread state | |
227 or3(G0, ilgl, G3_scratch); | |
228 stw(G3_scratch, tos_addr); | |
229 st_long(G0, val_addr); | |
230 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
231 } | |
232 | |
233 | |
234 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { | |
235 if (JvmtiExport::can_force_early_return()) { | |
236 Label L; | |
237 Register thr_state = G3_scratch; | |
727 | 238 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); |
3839 | 239 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; |
0 | 240 |
241 // Initiate earlyret handling only if it is not already being processed. | |
242 // If the flag has the earlyret_processing bit set, it means that this code | |
243 // is called *during* earlyret handling - we don't want to reenter. | |
727 | 244 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch); |
3839 | 245 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L); |
0 | 246 |
247 // Call Interpreter::remove_activation_early_entry() to get the address of the | |
248 // same-named entrypoint in the generated interpreter code | |
727 | 249 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1); |
0 | 250 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); |
251 | |
252 // Jump to Interpreter::_remove_activation_early_entry | |
253 jmpl(O0, G0, G0); | |
254 delayed()->nop(); | |
255 bind(L); | |
256 } | |
257 } | |
258 | |
259 | |
1295 | 260 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { |
0 | 261 mov(arg_1, O0); |
1295 | 262 mov(arg_2, O1); |
263 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2); | |
0 | 264 } |
265 #endif /* CC_INTERP */ | |
266 | |
267 | |
268 #ifndef CC_INTERP | |
269 | |
270 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { | |
271 assert_not_delayed(); | |
272 dispatch_Lbyte_code(state, table); | |
273 } | |
274 | |
275 | |
276 void InterpreterMacroAssembler::dispatch_normal(TosState state) { | |
277 dispatch_base(state, Interpreter::normal_table(state)); | |
278 } | |
279 | |
280 | |
281 void InterpreterMacroAssembler::dispatch_only(TosState state) { | |
282 dispatch_base(state, Interpreter::dispatch_table(state)); | |
283 } | |
284 | |
285 | |
286 // common code to dispatch and dispatch_only | |
287 // dispatch value in Lbyte_code and increment Lbcp | |
288 | |
289 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) { | |
290 verify_FPU(1, state); | |
291 // %%%%% maybe implement +VerifyActivationFrameSize here | |
292 //verify_thread(); //too slow; we will just verify on method entry & exit | |
293 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
294 #ifdef FAST_DISPATCH | |
295 if (table == Interpreter::dispatch_table(state)) { | |
296 // use IdispatchTables | |
297 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); | |
298 // add offset to correct dispatch table | |
299 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
300 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr | |
301 } else { | |
302 #endif | |
303 // dispatch table to use | |
727 | 304 AddressLiteral tbl(table); |
0 | 305 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize |
727 | 306 set(tbl, G3_scratch); // compute addr of table |
0 | 307 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr |
308 #ifdef FAST_DISPATCH | |
309 } | |
310 #endif | |
311 jmp( G3_scratch, 0 ); | |
312 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); | |
313 else delayed()->nop(); | |
314 } | |
315 | |
316 | |
317 // Helpers for expression stack | |
318 | |
319 // Longs and doubles are Category 2 computational types in the | |
320 // JVM specification (section 3.11.1) and take 2 expression stack or | |
321 // local slots. | |
322 // Aligning them on 32 bit with tagged stacks is hard because the code generated | |
323 // for the dup* bytecodes depends on what types are already on the stack. | |
324 // If the types are split into the two stack/local slots, that is much easier | |
325 // (and we can use 0 for non-reference tags). | |
326 | |
327 // Known good alignment in _LP64 but unknown otherwise | |
328 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { | |
329 assert_not_delayed(); | |
330 | |
331 #ifdef _LP64 | |
332 ldf(FloatRegisterImpl::D, r1, offset, d); | |
333 #else | |
334 ldf(FloatRegisterImpl::S, r1, offset, d); | |
1506 | 335 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor()); |
0 | 336 #endif |
337 } | |
338 | |
339 // Known good alignment in _LP64 but unknown otherwise | |
340 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { | |
341 assert_not_delayed(); | |
342 | |
343 #ifdef _LP64 | |
344 stf(FloatRegisterImpl::D, d, r1, offset); | |
345 // store something more useful here | |
1506 | 346 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) |
0 | 347 #else |
348 stf(FloatRegisterImpl::S, d, r1, offset); | |
1506 | 349 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize); |
0 | 350 #endif |
351 } | |
352 | |
353 | |
354 // Known good alignment in _LP64 but unknown otherwise | |
355 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { | |
356 assert_not_delayed(); | |
357 #ifdef _LP64 | |
358 ldx(r1, offset, rd); | |
359 #else | |
360 ld(r1, offset, rd); | |
1506 | 361 ld(r1, offset + Interpreter::stackElementSize, rd->successor()); |
0 | 362 #endif |
363 } | |
364 | |
365 // Known good alignment in _LP64 but unknown otherwise | |
366 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { | |
367 assert_not_delayed(); | |
368 | |
369 #ifdef _LP64 | |
370 stx(l, r1, offset); | |
371 // store something more useful here | |
1506 | 372 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) |
0 | 373 #else |
374 st(l, r1, offset); | |
1506 | 375 st(l->successor(), r1, offset + Interpreter::stackElementSize); |
0 | 376 #endif |
377 } | |
378 | |
379 void InterpreterMacroAssembler::pop_i(Register r) { | |
380 assert_not_delayed(); | |
381 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
1506 | 382 inc(Lesp, Interpreter::stackElementSize); |
0 | 383 debug_only(verify_esp(Lesp)); |
384 } | |
385 | |
386 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { | |
387 assert_not_delayed(); | |
388 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
1506 | 389 inc(Lesp, Interpreter::stackElementSize); |
0 | 390 debug_only(verify_esp(Lesp)); |
391 } | |
392 | |
393 void InterpreterMacroAssembler::pop_l(Register r) { | |
394 assert_not_delayed(); | |
395 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
1506 | 396 inc(Lesp, 2*Interpreter::stackElementSize); |
0 | 397 debug_only(verify_esp(Lesp)); |
398 } | |
399 | |
400 | |
401 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { | |
402 assert_not_delayed(); | |
403 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); | |
1506 | 404 inc(Lesp, Interpreter::stackElementSize); |
0 | 405 debug_only(verify_esp(Lesp)); |
406 } | |
407 | |
408 | |
409 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { | |
410 assert_not_delayed(); | |
411 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); | |
1506 | 412 inc(Lesp, 2*Interpreter::stackElementSize); |
0 | 413 debug_only(verify_esp(Lesp)); |
414 } | |
415 | |
416 | |
417 void InterpreterMacroAssembler::push_i(Register r) { | |
418 assert_not_delayed(); | |
419 debug_only(verify_esp(Lesp)); | |
1506 | 420 st(r, Lesp, 0); |
421 dec(Lesp, Interpreter::stackElementSize); | |
0 | 422 } |
423 | |
424 void InterpreterMacroAssembler::push_ptr(Register r) { | |
425 assert_not_delayed(); | |
1506 | 426 st_ptr(r, Lesp, 0); |
427 dec(Lesp, Interpreter::stackElementSize); | |
0 | 428 } |
429 | |
430 // remember: our convention for longs in SPARC is: | |
431 // O0 (Otos_l1) has high-order part in first word, | |
432 // O1 (Otos_l2) has low-order part in second word | |
433 | |
434 void InterpreterMacroAssembler::push_l(Register r) { | |
435 assert_not_delayed(); | |
436 debug_only(verify_esp(Lesp)); | |
1506 | 437 // Longs are stored in memory-correct order, even if unaligned. |
438 int offset = -Interpreter::stackElementSize; | |
0 | 439 store_unaligned_long(r, Lesp, offset); |
1506 | 440 dec(Lesp, 2 * Interpreter::stackElementSize); |
0 | 441 } |
442 | |
443 | |
444 void InterpreterMacroAssembler::push_f(FloatRegister f) { | |
445 assert_not_delayed(); | |
446 debug_only(verify_esp(Lesp)); | |
1506 | 447 stf(FloatRegisterImpl::S, f, Lesp, 0); |
448 dec(Lesp, Interpreter::stackElementSize); | |
0 | 449 } |
450 | |
451 | |
452 void InterpreterMacroAssembler::push_d(FloatRegister d) { | |
453 assert_not_delayed(); | |
454 debug_only(verify_esp(Lesp)); | |
1506 | 455 // Longs are stored in memory-correct order, even if unaligned. |
456 int offset = -Interpreter::stackElementSize; | |
0 | 457 store_unaligned_double(d, Lesp, offset); |
1506 | 458 dec(Lesp, 2 * Interpreter::stackElementSize); |
0 | 459 } |
460 | |
461 | |
462 void InterpreterMacroAssembler::push(TosState state) { | |
463 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
464 switch (state) { | |
465 case atos: push_ptr(); break; | |
466 case btos: push_i(); break; | |
467 case ctos: | |
468 case stos: push_i(); break; | |
469 case itos: push_i(); break; | |
470 case ltos: push_l(); break; | |
471 case ftos: push_f(); break; | |
472 case dtos: push_d(); break; | |
473 case vtos: /* nothing to do */ break; | |
474 default : ShouldNotReachHere(); | |
475 } | |
476 } | |
477 | |
478 | |
479 void InterpreterMacroAssembler::pop(TosState state) { | |
480 switch (state) { | |
481 case atos: pop_ptr(); break; | |
482 case btos: pop_i(); break; | |
483 case ctos: | |
484 case stos: pop_i(); break; | |
485 case itos: pop_i(); break; | |
486 case ltos: pop_l(); break; | |
487 case ftos: pop_f(); break; | |
488 case dtos: pop_d(); break; | |
489 case vtos: /* nothing to do */ break; | |
490 default : ShouldNotReachHere(); | |
491 } | |
492 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
493 } | |
494 | |
495 | |
1506 | 496 // Helpers for swap and dup |
497 void InterpreterMacroAssembler::load_ptr(int n, Register val) { | |
0 | 498 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); |
499 } | |
1506 | 500 void InterpreterMacroAssembler::store_ptr(int n, Register val) { |
0 | 501 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); |
502 } | |
503 | |
504 | |
505 void InterpreterMacroAssembler::load_receiver(Register param_count, | |
506 Register recv) { | |
1506 | 507 sll(param_count, Interpreter::logStackElementSize, param_count); |
6266
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
508 ld_ptr(Lesp, param_count, recv); // gets receiver oop |
0 | 509 } |
510 | |
511 void InterpreterMacroAssembler::empty_expression_stack() { | |
512 // Reset Lesp. | |
513 sub( Lmonitors, wordSize, Lesp ); | |
514 | |
515 // Reset SP by subtracting more space from Lesp. | |
516 Label done; | |
727 | 517 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!"); |
0 | 518 |
519 // A native does not need to do this, since its callee does not change SP. | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
520 ld(Lmethod, Method::access_flags_offset(), Gframe_size); // Load access flags. |
0 | 521 btst(JVM_ACC_NATIVE, Gframe_size); |
522 br(Assembler::notZero, false, Assembler::pt, done); | |
523 delayed()->nop(); | |
524 | |
525 // Compute max expression stack+register save area | |
7183
b2dbd323c668
8003848: Make ConstMethod::generic_signature_index optional and move Method::_max_stack to ConstMethod.
jiangli
parents:
6805
diff
changeset
|
526 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size); |
b2dbd323c668
8003848: Make ConstMethod::generic_signature_index optional and move Method::_max_stack to ConstMethod.
jiangli
parents:
6805
diff
changeset
|
527 lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size); // Load max stack. |
0 | 528 add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size ); |
529 | |
530 // | |
531 // now set up a stack frame with the size computed above | |
532 // | |
533 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below | |
534 sll( Gframe_size, LogBytesPerWord, Gframe_size ); | |
535 sub( Lesp, Gframe_size, Gframe_size ); | |
536 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary | |
537 debug_only(verify_sp(Gframe_size, G4_scratch)); | |
538 #ifdef _LP64 | |
539 sub(Gframe_size, STACK_BIAS, Gframe_size ); | |
540 #endif | |
541 mov(Gframe_size, SP); | |
542 | |
543 bind(done); | |
544 } | |
545 | |
546 | |
547 #ifdef ASSERT | |
548 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { | |
549 Label Bad, OK; | |
550 | |
551 // Saved SP must be aligned. | |
552 #ifdef _LP64 | |
553 btst(2*BytesPerWord-1, Rsp); | |
554 #else | |
555 btst(LongAlignmentMask, Rsp); | |
556 #endif | |
557 br(Assembler::notZero, false, Assembler::pn, Bad); | |
558 delayed()->nop(); | |
559 | |
560 // Saved SP, plus register window size, must not be above FP. | |
561 add(Rsp, frame::register_save_words * wordSize, Rtemp); | |
562 #ifdef _LP64 | |
563 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP | |
564 #endif | |
3839 | 565 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad); |
0 | 566 |
567 // Saved SP must not be ridiculously below current SP. | |
568 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); | |
569 set(maxstack, Rtemp); | |
570 sub(SP, Rtemp, Rtemp); | |
571 #ifdef _LP64 | |
572 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp | |
573 #endif | |
3839 | 574 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad); |
575 | |
576 ba_short(OK); | |
0 | 577 |
578 bind(Bad); | |
579 stop("on return to interpreted call, restored SP is corrupted"); | |
580 | |
581 bind(OK); | |
582 } | |
583 | |
584 | |
585 void InterpreterMacroAssembler::verify_esp(Register Resp) { | |
586 // about to read or write Resp[0] | |
587 // make sure it is not in the monitors or the register save area | |
588 Label OK1, OK2; | |
589 | |
590 cmp(Resp, Lmonitors); | |
591 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); | |
592 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); | |
593 stop("too many pops: Lesp points into monitor area"); | |
594 bind(OK1); | |
595 #ifdef _LP64 | |
596 sub(Resp, STACK_BIAS, Resp); | |
597 #endif | |
598 cmp(Resp, SP); | |
599 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); | |
600 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); | |
601 stop("too many pushes: Lesp points into register window"); | |
602 bind(OK2); | |
603 } | |
604 #endif // ASSERT | |
605 | |
606 // Load compiled (i2c) or interpreter entry when calling from interpreted and | |
607 // do the call. Centralized so that all interpreter calls will do the same actions. | |
608 // If jvmti single stepping is on for a thread we must not call compiled code. | |
609 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { | |
610 | |
611 // Assume we want to go compiled if available | |
612 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
613 ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target); |
0 | 614 |
615 if (JvmtiExport::can_post_interpreter_events()) { | |
616 // JVMTI events, such as single-stepping, are implemented partly by avoiding running | |
617 // compiled code in threads for which the event is enabled. Check here for | |
618 // interp_only_mode if these events CAN be enabled. | |
619 verify_thread(); | |
620 Label skip_compiled_code; | |
621 | |
727 | 622 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
0 | 623 ld(interp_only, scratch); |
3839 | 624 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
625 delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target); |
0 | 626 bind(skip_compiled_code); |
627 } | |
628 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
629 // the i2c_adapters need Method* in G5_method (right? %%%) |
0 | 630 // do the call |
631 #ifdef ASSERT | |
632 { | |
633 Label ok; | |
3839 | 634 br_notnull_short(target, Assembler::pt, ok); |
0 | 635 stop("null entry point"); |
636 bind(ok); | |
637 } | |
638 #endif // ASSERT | |
639 | |
640 // Adjust Rret first so Llast_SP can be same as Rret | |
641 add(Rret, -frame::pc_return_offset, O7); | |
642 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer | |
643 // Record SP so we can remove any stack space allocated by adapter transition | |
644 jmp(target, 0); | |
645 delayed()->mov(SP, Llast_SP); | |
646 } | |
647 | |
648 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { | |
649 assert_not_delayed(); | |
650 | |
651 Label not_taken; | |
652 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); | |
653 else br (cc, false, Assembler::pn, not_taken); | |
654 delayed()->nop(); | |
655 | |
656 TemplateTable::branch(false,false); | |
657 | |
658 bind(not_taken); | |
659 | |
660 profile_not_taken_branch(G3_scratch); | |
661 } | |
662 | |
663 | |
664 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( | |
665 int bcp_offset, | |
666 Register Rtmp, | |
667 Register Rdst, | |
668 signedOrNot is_signed, | |
669 setCCOrNot should_set_CC ) { | |
670 assert(Rtmp != Rdst, "need separate temp register"); | |
671 assert_not_delayed(); | |
672 switch (is_signed) { | |
673 default: ShouldNotReachHere(); | |
674 | |
675 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte | |
676 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte | |
677 } | |
678 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte | |
679 sll( Rdst, BitsPerByte, Rdst); | |
680 switch (should_set_CC ) { | |
681 default: ShouldNotReachHere(); | |
682 | |
683 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; | |
684 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; | |
685 } | |
686 } | |
687 | |
688 | |
689 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( | |
690 int bcp_offset, | |
691 Register Rtmp, | |
692 Register Rdst, | |
693 setCCOrNot should_set_CC ) { | |
694 assert(Rtmp != Rdst, "need separate temp register"); | |
695 assert_not_delayed(); | |
696 add( Lbcp, bcp_offset, Rtmp); | |
697 andcc( Rtmp, 3, G0); | |
698 Label aligned; | |
699 switch (should_set_CC ) { | |
700 default: ShouldNotReachHere(); | |
701 | |
702 case set_CC: break; | |
703 case dont_set_CC: break; | |
704 } | |
705 | |
706 br(Assembler::zero, true, Assembler::pn, aligned); | |
707 #ifdef _LP64 | |
708 delayed()->ldsw(Rtmp, 0, Rdst); | |
709 #else | |
710 delayed()->ld(Rtmp, 0, Rdst); | |
711 #endif | |
712 | |
713 ldub(Lbcp, bcp_offset + 3, Rdst); | |
714 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); | |
715 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); | |
716 #ifdef _LP64 | |
717 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); | |
718 #else | |
719 // Unsigned load is faster than signed on some implementations | |
720 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); | |
721 #endif | |
722 or3(Rtmp, Rdst, Rdst ); | |
723 | |
724 bind(aligned); | |
725 if (should_set_CC == set_CC) tst(Rdst); | |
726 } | |
727 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
728 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index, |
1565 | 729 int bcp_offset, size_t index_size) { |
1503 | 730 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); |
1565 | 731 if (index_size == sizeof(u2)) { |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
732 get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned); |
1565 | 733 } else if (index_size == sizeof(u4)) { |
2416
38fea01eb669
6817525: turn on method handle functionality by default for JSR 292
twisti
parents:
2118
diff
changeset
|
734 assert(EnableInvokeDynamic, "giant index used only for JSR 292"); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
735 get_4_byte_integer_at_bcp(bcp_offset, temp, index); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
736 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
737 xor3(index, -1, index); // convert to plain index |
1565 | 738 } else if (index_size == sizeof(u1)) { |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
739 ldub(Lbcp, bcp_offset, index); |
1565 | 740 } else { |
741 ShouldNotReachHere(); | |
1503 | 742 } |
743 } | |
744 | |
745 | |
746 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp, | |
1565 | 747 int bcp_offset, size_t index_size) { |
0 | 748 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); |
749 assert_different_registers(cache, tmp); | |
750 assert_not_delayed(); | |
1565 | 751 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size); |
1503 | 752 // convert from field index to ConstantPoolCacheEntry index and from |
753 // word index to byte offset | |
0 | 754 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); |
755 add(LcpoolCache, tmp, cache); | |
756 } | |
757 | |
758 | |
3852
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
759 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, |
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
760 Register temp, |
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
761 Register bytecode, |
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
762 int byte_no, |
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
763 int bcp_offset, |
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
764 size_t index_size) { |
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
765 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
766 ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode); |
3852
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
767 const int shift_count = (1 + byte_no) * BitsPerByte; |
6266
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
768 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || |
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
769 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), |
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
770 "correct shift count"); |
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
771 srl(bytecode, shift_count, bytecode); |
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
772 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); |
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
773 and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode); |
3852
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
774 } |
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
775 |
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
776 |
1503 | 777 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, |
1565 | 778 int bcp_offset, size_t index_size) { |
0 | 779 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); |
780 assert_different_registers(cache, tmp); | |
781 assert_not_delayed(); | |
1565 | 782 if (index_size == sizeof(u2)) { |
783 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); | |
784 } else { | |
785 ShouldNotReachHere(); // other sizes not supported here | |
786 } | |
0 | 787 // convert from field index to ConstantPoolCacheEntry index |
788 // and from word index to byte offset | |
789 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); | |
790 // skip past the header | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
791 add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp); |
0 | 792 // construct pointer to cache entry |
793 add(LcpoolCache, tmp, cache); | |
794 } | |
795 | |
796 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
797 // Load object from cpool->resolved_references(index) |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
798 void InterpreterMacroAssembler::load_resolved_reference_at_index( |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
799 Register result, Register index) { |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
800 assert_different_registers(result, index); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
801 assert_not_delayed(); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
802 // convert from field index to resolved_references() index and from |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
803 // word index to byte offset. Since this is a java object, it can be compressed |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
804 Register tmp = index; // reuse |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
805 sll(index, LogBytesPerHeapOop, tmp); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
806 get_constant_pool(result); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
807 // load pointer for resolved_references[] objArray |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
808 ld_ptr(result, ConstantPool::resolved_references_offset_in_bytes(), result); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
809 // JNIHandles::resolve(result) |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
810 ld_ptr(result, 0, result); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
811 // Add in the index |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
812 add(result, tmp, result); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
813 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
814 } |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
815 |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
816 |
0 | 817 // 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
|
818 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. |
0 | 819 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, |
820 Register Rsuper_klass, | |
821 Register Rtmp1, | |
822 Register Rtmp2, | |
823 Register Rtmp3, | |
824 Label &ok_is_subtype ) { | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
825 Label not_subtype; |
0 | 826 |
827 // Profile the not-null value's klass. | |
828 profile_typecheck(Rsub_klass, Rtmp1); | |
829 | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
830 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
831 Rtmp1, Rtmp2, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
832 &ok_is_subtype, ¬_subtype, NULL); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
833 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
834 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
835 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
836 &ok_is_subtype, NULL); |
0 | 837 |
838 bind(not_subtype); | |
839 profile_typecheck_failed(Rtmp1); | |
840 } | |
841 | |
842 // Separate these two to allow for delay slot in middle | |
843 // These are used to do a test and full jump to exception-throwing code. | |
844 | |
845 // %%%%% Could possibly reoptimize this by testing to see if could use | |
846 // a single conditional branch (i.e. if span is small enough. | |
847 // If you go that route, than get rid of the split and give up | |
848 // on the delay-slot hack. | |
849 | |
850 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, | |
851 Label& ok ) { | |
852 assert_not_delayed(); | |
853 br(ok_condition, true, pt, ok); | |
854 // DELAY SLOT | |
855 } | |
856 | |
857 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, | |
858 Label& ok ) { | |
859 assert_not_delayed(); | |
860 bp( ok_condition, true, Assembler::xcc, pt, ok); | |
861 // DELAY SLOT | |
862 } | |
863 | |
864 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, | |
865 Label& ok ) { | |
866 assert_not_delayed(); | |
867 brx(ok_condition, true, pt, ok); | |
868 // DELAY SLOT | |
869 } | |
870 | |
871 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, | |
872 Register Rscratch, | |
873 Label& ok ) { | |
874 assert(throw_entry_point != NULL, "entry point must be generated by now"); | |
727 | 875 AddressLiteral dest(throw_entry_point); |
876 jump_to(dest, Rscratch); | |
0 | 877 delayed()->nop(); |
878 bind(ok); | |
879 } | |
880 | |
881 | |
882 // And if you cannot use the delay slot, here is a shorthand: | |
883 | |
884 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, | |
885 address throw_entry_point, | |
886 Register Rscratch ) { | |
887 Label ok; | |
888 if (ok_condition != never) { | |
889 throw_if_not_1_icc( ok_condition, ok); | |
890 delayed()->nop(); | |
891 } | |
892 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
893 } | |
894 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, | |
895 address throw_entry_point, | |
896 Register Rscratch ) { | |
897 Label ok; | |
898 if (ok_condition != never) { | |
899 throw_if_not_1_xcc( ok_condition, ok); | |
900 delayed()->nop(); | |
901 } | |
902 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
903 } | |
904 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, | |
905 address throw_entry_point, | |
906 Register Rscratch ) { | |
907 Label ok; | |
908 if (ok_condition != never) { | |
909 throw_if_not_1_x( ok_condition, ok); | |
910 delayed()->nop(); | |
911 } | |
912 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
913 } | |
914 | |
915 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res | |
916 // Note: res is still shy of address by array offset into object. | |
917 | |
918 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { | |
919 assert_not_delayed(); | |
920 | |
921 verify_oop(array); | |
922 #ifdef _LP64 | |
923 // sign extend since tos (index) can be a 32bit value | |
924 sra(index, G0, index); | |
925 #endif // _LP64 | |
926 | |
927 // check array | |
928 Label ptr_ok; | |
929 tst(array); | |
930 throw_if_not_1_x( notZero, ptr_ok ); | |
931 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index | |
932 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); | |
933 | |
934 Label index_ok; | |
935 cmp(index, tmp); | |
936 throw_if_not_1_icc( lessUnsigned, index_ok ); | |
937 if (index_shift > 0) delayed()->sll(index, index_shift, index); | |
938 else delayed()->add(array, index, res); // addr - const offset in index | |
939 // convention: move aberrant index into G3_scratch for exception message | |
940 mov(index, G3_scratch); | |
941 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); | |
942 | |
943 // add offset if didn't do it in delay slot | |
944 if (index_shift > 0) add(array, index, res); // addr - const offset in index | |
945 } | |
946 | |
947 | |
948 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { | |
949 assert_not_delayed(); | |
950 | |
951 // pop array | |
952 pop_ptr(array); | |
953 | |
954 // check array | |
955 index_check_without_pop(array, index, index_shift, tmp, res); | |
956 } | |
957 | |
958 | |
6123
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
959 void InterpreterMacroAssembler::get_const(Register Rdst) { |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
960 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst); |
6123
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
961 } |
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
962 |
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
963 |
0 | 964 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { |
6123
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
965 get_const(Rdst); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
966 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); |
0 | 967 } |
968 | |
969 | |
970 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { | |
971 get_constant_pool(Rdst); | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
972 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); |
0 | 973 } |
974 | |
975 | |
976 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { | |
977 get_constant_pool(Rcpool); | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
978 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags); |
0 | 979 } |
980 | |
981 | |
982 // unlock if synchronized method | |
983 // | |
984 // Unlock the receiver if this is a synchronized method. | |
985 // Unlock any Java monitors from syncronized blocks. | |
986 // | |
987 // If there are locked Java monitors | |
988 // If throw_monitor_exception | |
989 // throws IllegalMonitorStateException | |
990 // Else if install_monitor_exception | |
991 // installs IllegalMonitorStateException | |
992 // Else | |
993 // no error processing | |
994 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, | |
995 bool throw_monitor_exception, | |
996 bool install_monitor_exception) { | |
997 Label unlocked, unlock, no_unlock; | |
998 | |
999 // get the value of _do_not_unlock_if_synchronized into G1_scratch | |
727 | 1000 const Address do_not_unlock_if_synchronized(G2_thread, |
1001 JavaThread::do_not_unlock_if_synchronized_offset()); | |
0 | 1002 ldbool(do_not_unlock_if_synchronized, G1_scratch); |
1003 stbool(G0, do_not_unlock_if_synchronized); // reset the flag | |
1004 | |
1005 // check if synchronized method | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
1006 const Address access_flags(Lmethod, Method::access_flags_offset()); |
0 | 1007 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); |
1008 push(state); // save tos | |
727 | 1009 ld(access_flags, G3_scratch); // Load access flags. |
0 | 1010 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); |
727 | 1011 br(zero, false, pt, unlocked); |
0 | 1012 delayed()->nop(); |
1013 | |
1014 // Don't unlock anything if the _do_not_unlock_if_synchronized flag | |
1015 // is set. | |
3839 | 1016 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock); |
0 | 1017 delayed()->nop(); |
1018 | |
1019 // BasicObjectLock will be first in list, since this is a synchronized method. However, need | |
1020 // to check that the object has not been unlocked by an explicit monitorexit bytecode. | |
1021 | |
1022 //Intel: if (throw_monitor_exception) ... else ... | |
1023 // Entry already unlocked, need to throw exception | |
1024 //... | |
1025 | |
1026 // pass top-most monitor elem | |
1027 add( top_most_monitor(), O1 ); | |
1028 | |
1029 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); | |
3839 | 1030 br_notnull_short(G3_scratch, pt, unlock); |
0 | 1031 |
1032 if (throw_monitor_exception) { | |
1033 // Entry already unlocked need to throw an exception | |
1034 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
1035 should_not_reach_here(); | |
1036 } else { | |
1037 // Monitor already unlocked during a stack unroll. | |
1038 // If requested, install an illegal_monitor_state_exception. | |
1039 // Continue with stack unrolling. | |
1040 if (install_monitor_exception) { | |
1041 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); | |
1042 } | |
3839 | 1043 ba_short(unlocked); |
0 | 1044 } |
1045 | |
1046 bind(unlock); | |
1047 | |
1048 unlock_object(O1); | |
1049 | |
1050 bind(unlocked); | |
1051 | |
1052 // I0, I1: Might contain return value | |
1053 | |
1054 // Check that all monitors are unlocked | |
1055 { Label loop, exception, entry, restart; | |
1056 | |
1057 Register Rmptr = O0; | |
1058 Register Rtemp = O1; | |
1059 Register Rlimit = Lmonitors; | |
1060 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
1061 assert( (delta & LongAlignmentMask) == 0, | |
1062 "sizeof BasicObjectLock must be even number of doublewords"); | |
1063 | |
1064 #ifdef ASSERT | |
1065 add(top_most_monitor(), Rmptr, delta); | |
1066 { Label L; | |
1067 // ensure that Rmptr starts out above (or at) Rlimit | |
3839 | 1068 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); |
0 | 1069 stop("monitor stack has negative size"); |
1070 bind(L); | |
1071 } | |
1072 #endif | |
1073 bind(restart); | |
3839 | 1074 ba(entry); |
0 | 1075 delayed()-> |
1076 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry | |
1077 | |
1078 // Entry is still locked, need to throw exception | |
1079 bind(exception); | |
1080 if (throw_monitor_exception) { | |
1081 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
1082 should_not_reach_here(); | |
1083 } else { | |
1084 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. | |
1085 // Unlock does not block, so don't have to worry about the frame | |
1086 unlock_object(Rmptr); | |
1087 if (install_monitor_exception) { | |
1088 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); | |
1089 } | |
3839 | 1090 ba_short(restart); |
0 | 1091 } |
1092 | |
1093 bind(loop); | |
1094 cmp(Rtemp, G0); // check if current entry is used | |
1095 brx(Assembler::notEqual, false, pn, exception); | |
1096 delayed()-> | |
1097 dec(Rmptr, delta); // otherwise advance to next entry | |
1098 #ifdef ASSERT | |
1099 { Label L; | |
1100 // ensure that Rmptr has not somehow stepped below Rlimit | |
3839 | 1101 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); |
0 | 1102 stop("ran off the end of the monitor stack"); |
1103 bind(L); | |
1104 } | |
1105 #endif | |
1106 bind(entry); | |
1107 cmp(Rmptr, Rlimit); // check if bottom reached | |
1108 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry | |
1109 delayed()-> | |
1110 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); | |
1111 } | |
1112 | |
1113 bind(no_unlock); | |
1114 pop(state); | |
1115 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
1116 } | |
1117 | |
1118 | |
1119 // remove activation | |
1120 // | |
1121 // Unlock the receiver if this is a synchronized method. | |
1122 // Unlock any Java monitors from syncronized blocks. | |
1123 // Remove the activation from the stack. | |
1124 // | |
1125 // If there are locked Java monitors | |
1126 // If throw_monitor_exception | |
1127 // throws IllegalMonitorStateException | |
1128 // Else if install_monitor_exception | |
1129 // installs IllegalMonitorStateException | |
1130 // Else | |
1131 // no error processing | |
1132 void InterpreterMacroAssembler::remove_activation(TosState state, | |
1133 bool throw_monitor_exception, | |
1134 bool install_monitor_exception) { | |
1135 | |
1136 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); | |
1137 | |
1138 // save result (push state before jvmti call and pop it afterwards) and notify jvmti | |
1139 notify_method_exit(false, state, NotifyJVMTI); | |
1140 | |
1141 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
1142 verify_thread(); | |
1143 | |
1144 // return tos | |
1145 assert(Otos_l1 == Otos_i, "adjust code below"); | |
1146 switch (state) { | |
1147 #ifdef _LP64 | |
1148 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 | |
1149 #else | |
1150 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1 | |
1151 #endif | |
1152 case btos: // fall through | |
1153 case ctos: | |
1154 case stos: // fall through | |
1155 case atos: // fall through | |
1156 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 | |
1157 case ftos: // fall through | |
1158 case dtos: // fall through | |
1159 case vtos: /* nothing to do */ break; | |
1160 default : ShouldNotReachHere(); | |
1161 } | |
1162 | |
1163 #if defined(COMPILER2) && !defined(_LP64) | |
1164 if (state == ltos) { | |
1165 // C2 expects long results in G1 we can't tell if we're returning to interpreted | |
1166 // or compiled so just be safe use G1 and O0/O1 | |
1167 | |
1168 // Shift bits into high (msb) of G1 | |
1169 sllx(Otos_l1->after_save(), 32, G1); | |
1170 // Zero extend low bits | |
1171 srl (Otos_l2->after_save(), 0, Otos_l2->after_save()); | |
1172 or3 (Otos_l2->after_save(), G1, G1); | |
1173 } | |
1174 #endif /* COMPILER2 */ | |
1175 | |
1176 } | |
1177 #endif /* CC_INTERP */ | |
1178 | |
1179 | |
1180 // Lock object | |
1181 // | |
1182 // Argument - lock_reg points to the BasicObjectLock to be used for locking, | |
1183 // it must be initialized with the object to lock | |
1184 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { | |
1185 if (UseHeavyMonitors) { | |
1186 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); | |
1187 } | |
1188 else { | |
1189 Register obj_reg = Object; | |
1190 Register mark_reg = G4_scratch; | |
1191 Register temp_reg = G1_scratch; | |
727 | 1192 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); |
1193 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); | |
0 | 1194 Label done; |
1195 | |
1196 Label slow_case; | |
1197 | |
1198 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); | |
1199 | |
1200 // load markOop from object into mark_reg | |
1201 ld_ptr(mark_addr, mark_reg); | |
1202 | |
1203 if (UseBiasedLocking) { | |
1204 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); | |
1205 } | |
1206 | |
1207 // get the address of basicLock on stack that will be stored in the object | |
1208 // we need a temporary register here as we do not want to clobber lock_reg | |
1209 // (cas clobbers the destination register) | |
1210 mov(lock_reg, temp_reg); | |
1211 // set mark reg to be (markOop of object | UNLOCK_VALUE) | |
1212 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); | |
1213 // initialize the box (Must happen before we update the object mark!) | |
1214 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); | |
1215 // compare and exchange object_addr, markOop | 1, stack address of basicLock | |
1216 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
1217 casx_under_lock(mark_addr.base(), mark_reg, temp_reg, | |
1218 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
1219 | |
1220 // if the compare and exchange succeeded we are done (we saw an unlocked object) | |
3839 | 1221 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done); |
0 | 1222 |
1223 // We did not see an unlocked object so try the fast recursive case | |
1224 | |
1225 // Check if owner is self by comparing the value in the markOop of object | |
1226 // with the stack pointer | |
1227 sub(temp_reg, SP, temp_reg); | |
1228 #ifdef _LP64 | |
1229 sub(temp_reg, STACK_BIAS, temp_reg); | |
1230 #endif | |
1231 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
1232 | |
1233 // Composite "andcc" test: | |
1234 // (a) %sp -vs- markword proximity check, and, | |
1235 // (b) verify mark word LSBs == 0 (Stack-locked). | |
1236 // | |
1237 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) | |
1238 // Note that the page size used for %sp proximity testing is arbitrary and is | |
1239 // unrelated to the actual MMU page size. We use a 'logical' page size of | |
1240 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate | |
1241 // field of the andcc instruction. | |
1242 andcc (temp_reg, 0xFFFFF003, G0) ; | |
1243 | |
1244 // if condition is true we are done and hence we can store 0 in the displaced | |
1245 // header indicating it is a recursive lock and be done | |
1246 brx(Assembler::zero, true, Assembler::pt, done); | |
1247 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); | |
1248 | |
1249 // none of the above fast optimizations worked so we have to get into the | |
1250 // slow case of monitor enter | |
1251 bind(slow_case); | |
1252 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); | |
1253 | |
1254 bind(done); | |
1255 } | |
1256 } | |
1257 | |
1258 // Unlocks an object. Used in monitorexit bytecode and remove_activation. | |
1259 // | |
1260 // Argument - lock_reg points to the BasicObjectLock for lock | |
1261 // Throw IllegalMonitorException if object is not locked by current thread | |
1262 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { | |
1263 if (UseHeavyMonitors) { | |
1264 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); | |
1265 } else { | |
1266 Register obj_reg = G3_scratch; | |
1267 Register mark_reg = G4_scratch; | |
1268 Register displaced_header_reg = G1_scratch; | |
727 | 1269 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); |
1270 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); | |
0 | 1271 Label done; |
1272 | |
1273 if (UseBiasedLocking) { | |
1274 // load the object out of the BasicObjectLock | |
1275 ld_ptr(lockobj_addr, obj_reg); | |
1276 biased_locking_exit(mark_addr, mark_reg, done, true); | |
1277 st_ptr(G0, lockobj_addr); // free entry | |
1278 } | |
1279 | |
1280 // Test first if we are in the fast recursive case | |
727 | 1281 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); |
1282 ld_ptr(lock_addr, displaced_header_reg); | |
0 | 1283 br_null(displaced_header_reg, true, Assembler::pn, done); |
1284 delayed()->st_ptr(G0, lockobj_addr); // free entry | |
1285 | |
1286 // See if it is still a light weight lock, if so we just unlock | |
1287 // the object and we are done | |
1288 | |
1289 if (!UseBiasedLocking) { | |
1290 // load the object out of the BasicObjectLock | |
1291 ld_ptr(lockobj_addr, obj_reg); | |
1292 } | |
1293 | |
1294 // we have the displaced header in displaced_header_reg | |
1295 // we expect to see the stack address of the basicLock in case the | |
1296 // lock is still a light weight lock (lock_reg) | |
1297 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
1298 casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg, | |
1299 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
1300 cmp(lock_reg, displaced_header_reg); | |
1301 brx(Assembler::equal, true, Assembler::pn, done); | |
1302 delayed()->st_ptr(G0, lockobj_addr); // free entry | |
1303 | |
1304 // The lock has been converted into a heavy lock and hence | |
1305 // we need to get into the slow case | |
1306 | |
1307 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); | |
1308 | |
1309 bind(done); | |
1310 } | |
1311 } | |
1312 | |
1313 #ifndef CC_INTERP | |
1314 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
1315 // Get the method data pointer from the Method* and set the |
0 | 1316 // specified register to its value. |
1317 | |
2118
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1318 void InterpreterMacroAssembler::set_method_data_pointer() { |
0 | 1319 assert(ProfileInterpreter, "must be profiling interpreter"); |
1320 Label get_continue; | |
1321 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
1322 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); |
0 | 1323 test_method_data_pointer(get_continue); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
1324 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); |
0 | 1325 bind(get_continue); |
1326 } | |
1327 | |
1328 // Set the method data pointer for the current bcp. | |
1329 | |
1330 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { | |
1331 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1332 Label zero_continue; | |
1333 | |
1334 // Test MDO to avoid the call if it is NULL. | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
1335 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr); |
0 | 1336 test_method_data_pointer(zero_continue); |
1337 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
1338 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr); |
2118
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1339 add(ImethodDataPtr, O0, ImethodDataPtr); |
0 | 1340 bind(zero_continue); |
1341 } | |
1342 | |
1343 // Test ImethodDataPtr. If it is null, continue at the specified label | |
1344 | |
1345 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { | |
1346 assert(ProfileInterpreter, "must be profiling interpreter"); | |
3839 | 1347 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue); |
0 | 1348 } |
1349 | |
1350 void InterpreterMacroAssembler::verify_method_data_pointer() { | |
1351 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1352 #ifdef ASSERT | |
1353 Label verify_continue; | |
1354 test_method_data_pointer(verify_continue); | |
1355 | |
1356 // If the mdp is valid, it will point to a DataLayout header which is | |
1357 // consistent with the bcp. The converse is highly probable also. | |
1358 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
1359 ld_ptr(Lmethod, Method::const_offset(), O5); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
1360 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch); |
0 | 1361 add(G3_scratch, O5, G3_scratch); |
1362 cmp(Lbcp, G3_scratch); | |
1363 brx(Assembler::equal, false, Assembler::pt, verify_continue); | |
1364 | |
1365 Register temp_reg = O5; | |
1366 delayed()->mov(ImethodDataPtr, temp_reg); | |
1367 // %%% should use call_VM_leaf here? | |
1368 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); | |
1369 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); | |
727 | 1370 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); |
0 | 1371 stf(FloatRegisterImpl::D, Ftos_d, d_save); |
1372 mov(temp_reg->after_save(), O2); | |
1373 save_thread(L7_thread_cache); | |
1374 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); | |
1375 delayed()->nop(); | |
1376 restore_thread(L7_thread_cache); | |
1377 ldf(FloatRegisterImpl::D, d_save, Ftos_d); | |
1378 restore(); | |
1379 bind(verify_continue); | |
1380 #endif // ASSERT | |
1381 } | |
1382 | |
1383 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, | |
1384 Register Rtmp, | |
1385 Label &profile_continue) { | |
1386 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1387 // Control will flow to "profile_continue" if the counter is less than the | |
1388 // limit or if we call profile_method() | |
1389 | |
1390 Label done; | |
1391 | |
1392 // if no method data exists, and the counter is high enough, make one | |
3839 | 1393 br_notnull_short(ImethodDataPtr, Assembler::pn, done); |
0 | 1394 |
1395 // Test to see if we should create a method data oop | |
727 | 1396 AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit); |
1397 sethi(profile_limit, Rtmp); | |
1398 ld(Rtmp, profile_limit.low10(), Rtmp); | |
6805
69fb89ec6fa7
7198084: NPG: distance is too big for short branches in test_invocation_counter_for_mdp()
kvn
parents:
6725
diff
changeset
|
1399 cmp(invocation_count, Rtmp); |
69fb89ec6fa7
7198084: NPG: distance is too big for short branches in test_invocation_counter_for_mdp()
kvn
parents:
6725
diff
changeset
|
1400 // Use long branches because call_VM() code and following code generated by |
69fb89ec6fa7
7198084: NPG: distance is too big for short branches in test_invocation_counter_for_mdp()
kvn
parents:
6725
diff
changeset
|
1401 // test_backedge_count_for_osr() is large in debug VM. |
69fb89ec6fa7
7198084: NPG: distance is too big for short branches in test_invocation_counter_for_mdp()
kvn
parents:
6725
diff
changeset
|
1402 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); |
69fb89ec6fa7
7198084: NPG: distance is too big for short branches in test_invocation_counter_for_mdp()
kvn
parents:
6725
diff
changeset
|
1403 delayed()->nop(); |
0 | 1404 |
1405 // Build it now. | |
2118
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1406 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); |
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1407 set_method_data_pointer_for_bcp(); |
6805
69fb89ec6fa7
7198084: NPG: distance is too big for short branches in test_invocation_counter_for_mdp()
kvn
parents:
6725
diff
changeset
|
1408 ba(profile_continue); |
69fb89ec6fa7
7198084: NPG: distance is too big for short branches in test_invocation_counter_for_mdp()
kvn
parents:
6725
diff
changeset
|
1409 delayed()->nop(); |
0 | 1410 bind(done); |
1411 } | |
1412 | |
1413 // Store a value at some constant offset from the method data pointer. | |
1414 | |
1415 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { | |
1416 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1417 st_ptr(value, ImethodDataPtr, constant); | |
1418 } | |
1419 | |
1420 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, | |
1421 Register bumped_count, | |
1422 bool decrement) { | |
1423 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1424 | |
1425 // Load the counter. | |
1426 ld_ptr(counter, bumped_count); | |
1427 | |
1428 if (decrement) { | |
1429 // Decrement the register. Set condition codes. | |
1430 subcc(bumped_count, DataLayout::counter_increment, bumped_count); | |
1431 | |
1432 // If the decrement causes the counter to overflow, stay negative | |
1433 Label L; | |
1434 brx(Assembler::negative, true, Assembler::pn, L); | |
1435 | |
1436 // Store the decremented counter, if it is still negative. | |
1437 delayed()->st_ptr(bumped_count, counter); | |
1438 bind(L); | |
1439 } else { | |
1440 // Increment the register. Set carry flag. | |
1441 addcc(bumped_count, DataLayout::counter_increment, bumped_count); | |
1442 | |
1443 // If the increment causes the counter to overflow, pull back by 1. | |
1444 assert(DataLayout::counter_increment == 1, "subc works"); | |
1445 subc(bumped_count, G0, bumped_count); | |
1446 | |
1447 // Store the incremented counter. | |
1448 st_ptr(bumped_count, counter); | |
1449 } | |
1450 } | |
1451 | |
1452 // Increment the value at some constant offset from the method data pointer. | |
1453 | |
1454 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, | |
1455 Register bumped_count, | |
1456 bool decrement) { | |
1457 // Locate the counter at a fixed offset from the mdp: | |
727 | 1458 Address counter(ImethodDataPtr, constant); |
0 | 1459 increment_mdp_data_at(counter, bumped_count, decrement); |
1460 } | |
1461 | |
1462 // Increment the value at some non-fixed (reg + constant) offset from | |
1463 // the method data pointer. | |
1464 | |
1465 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, | |
1466 int constant, | |
1467 Register bumped_count, | |
1468 Register scratch2, | |
1469 bool decrement) { | |
1470 // Add the constant to reg to get the offset. | |
1471 add(ImethodDataPtr, reg, scratch2); | |
727 | 1472 Address counter(scratch2, constant); |
0 | 1473 increment_mdp_data_at(counter, bumped_count, decrement); |
1474 } | |
1475 | |
1476 // Set a flag value at the current method data pointer position. | |
1477 // Updates a single byte of the header, to avoid races with other header bits. | |
1478 | |
1479 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, | |
1480 Register scratch) { | |
1481 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1482 // Load the data header | |
1483 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); | |
1484 | |
1485 // Set the flag | |
1486 or3(scratch, flag_constant, scratch); | |
1487 | |
1488 // Store the modified header. | |
1489 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); | |
1490 } | |
1491 | |
1492 // Test the location at some offset from the method data pointer. | |
1493 // If it is not equal to value, branch to the not_equal_continue Label. | |
1494 // Set condition codes to match the nullness of the loaded value. | |
1495 | |
1496 void InterpreterMacroAssembler::test_mdp_data_at(int offset, | |
1497 Register value, | |
1498 Label& not_equal_continue, | |
1499 Register scratch) { | |
1500 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1501 ld_ptr(ImethodDataPtr, offset, scratch); | |
1502 cmp(value, scratch); | |
1503 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); | |
1504 delayed()->tst(scratch); | |
1505 } | |
1506 | |
1507 // Update the method data pointer by the displacement located at some fixed | |
1508 // offset from the method data pointer. | |
1509 | |
1510 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, | |
1511 Register scratch) { | |
1512 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1513 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); | |
1514 add(ImethodDataPtr, scratch, ImethodDataPtr); | |
1515 } | |
1516 | |
1517 // Update the method data pointer by the displacement located at the | |
1518 // offset (reg + offset_of_disp). | |
1519 | |
1520 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, | |
1521 int offset_of_disp, | |
1522 Register scratch) { | |
1523 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1524 add(reg, offset_of_disp, scratch); | |
1525 ld_ptr(ImethodDataPtr, scratch, scratch); | |
1526 add(ImethodDataPtr, scratch, ImethodDataPtr); | |
1527 } | |
1528 | |
1529 // Update the method data pointer by a simple constant displacement. | |
1530 | |
1531 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { | |
1532 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1533 add(ImethodDataPtr, constant, ImethodDataPtr); | |
1534 } | |
1535 | |
1536 // Update the method data pointer for a _ret bytecode whose target | |
1537 // was not among our cached targets. | |
1538 | |
1539 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, | |
1540 Register return_bci) { | |
1541 assert(ProfileInterpreter, "must be profiling interpreter"); | |
1542 push(state); | |
1543 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile | |
1544 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); | |
1545 ld_ptr(l_tmp, return_bci); | |
1546 pop(state); | |
1547 } | |
1548 | |
1549 // Count a taken branch in the bytecodes. | |
1550 | |
1551 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { | |
1552 if (ProfileInterpreter) { | |
1553 Label profile_continue; | |
1554 | |
1555 // If no method data exists, go to profile_continue. | |
1556 test_method_data_pointer(profile_continue); | |
1557 | |
1558 // We are taking a branch. Increment the taken count. | |
1559 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); | |
1560 | |
1561 // The method data pointer needs to be updated to reflect the new target. | |
1562 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); | |
1563 bind (profile_continue); | |
1564 } | |
1565 } | |
1566 | |
1567 | |
1568 // Count a not-taken branch in the bytecodes. | |
1569 | |
1570 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { | |
1571 if (ProfileInterpreter) { | |
1572 Label profile_continue; | |
1573 | |
1574 // If no method data exists, go to profile_continue. | |
1575 test_method_data_pointer(profile_continue); | |
1576 | |
1577 // We are taking a branch. Increment the not taken count. | |
1578 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); | |
1579 | |
1580 // The method data pointer needs to be updated to correspond to the | |
1581 // next bytecode. | |
1582 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); | |
1583 bind (profile_continue); | |
1584 } | |
1585 } | |
1586 | |
1587 | |
1588 // Count a non-virtual call in the bytecodes. | |
1589 | |
1590 void InterpreterMacroAssembler::profile_call(Register scratch) { | |
1591 if (ProfileInterpreter) { | |
1592 Label profile_continue; | |
1593 | |
1594 // If no method data exists, go to profile_continue. | |
1595 test_method_data_pointer(profile_continue); | |
1596 | |
1597 // We are making a call. Increment the count. | |
1598 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
1599 | |
1600 // The method data pointer needs to be updated to reflect the new target. | |
1601 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); | |
1602 bind (profile_continue); | |
1603 } | |
1604 } | |
1605 | |
1606 | |
1607 // Count a final call in the bytecodes. | |
1608 | |
1609 void InterpreterMacroAssembler::profile_final_call(Register scratch) { | |
1610 if (ProfileInterpreter) { | |
1611 Label profile_continue; | |
1612 | |
1613 // If no method data exists, go to profile_continue. | |
1614 test_method_data_pointer(profile_continue); | |
1615 | |
1616 // We are making a call. Increment the count. | |
1617 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
1618 | |
1619 // The method data pointer needs to be updated to reflect the new target. | |
1620 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); | |
1621 bind (profile_continue); | |
1622 } | |
1623 } | |
1624 | |
1625 | |
1626 // Count a virtual call in the bytecodes. | |
1627 | |
1628 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, | |
1503 | 1629 Register scratch, |
1630 bool receiver_can_be_null) { | |
0 | 1631 if (ProfileInterpreter) { |
1632 Label profile_continue; | |
1633 | |
1634 // If no method data exists, go to profile_continue. | |
1635 test_method_data_pointer(profile_continue); | |
1636 | |
1503 | 1637 |
1638 Label skip_receiver_profile; | |
1639 if (receiver_can_be_null) { | |
1640 Label not_null; | |
3839 | 1641 br_notnull_short(receiver, Assembler::pt, not_null); |
1503 | 1642 // We are making a call. Increment the count for null receiver. |
1643 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
3839 | 1644 ba_short(skip_receiver_profile); |
1503 | 1645 bind(not_null); |
1646 } | |
1647 | |
0 | 1648 // Record the receiver type. |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1649 record_klass_in_profile(receiver, scratch, true); |
1503 | 1650 bind(skip_receiver_profile); |
0 | 1651 |
1652 // The method data pointer needs to be updated to reflect the new target. | |
1653 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); | |
1654 bind (profile_continue); | |
1655 } | |
1656 } | |
1657 | |
1658 void InterpreterMacroAssembler::record_klass_in_profile_helper( | |
1659 Register receiver, Register scratch, | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1660 int start_row, Label& done, bool is_virtual_call) { |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1661 if (TypeProfileWidth == 0) { |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1662 if (is_virtual_call) { |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1663 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1664 } |
967
6918603297f7
6858208: jvm crash when specifying TypeProfileWidth=0 on jdk 6.0
poonam
parents:
727
diff
changeset
|
1665 return; |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1666 } |
967
6918603297f7
6858208: jvm crash when specifying TypeProfileWidth=0 on jdk 6.0
poonam
parents:
727
diff
changeset
|
1667 |
0 | 1668 int last_row = VirtualCallData::row_limit() - 1; |
1669 assert(start_row <= last_row, "must be work left to do"); | |
1670 // Test this row for both the receiver and for null. | |
1671 // Take any of three different outcomes: | |
1672 // 1. found receiver => increment count and goto done | |
1673 // 2. found null => keep looking for case 1, maybe allocate this cell | |
1674 // 3. found something else => keep looking for cases 1 and 2 | |
1675 // Case 3 is handled by a recursive call. | |
1676 for (int row = start_row; row <= last_row; row++) { | |
1677 Label next_test; | |
1678 bool test_for_null_also = (row == start_row); | |
1679 | |
1680 // See if the receiver is receiver[n]. | |
1681 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); | |
1682 test_mdp_data_at(recvr_offset, receiver, next_test, scratch); | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1683 // delayed()->tst(scratch); |
0 | 1684 |
1685 // The receiver is receiver[n]. Increment count[n]. | |
1686 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); | |
1687 increment_mdp_data_at(count_offset, scratch); | |
3839 | 1688 ba_short(done); |
0 | 1689 bind(next_test); |
1690 | |
1691 if (test_for_null_also) { | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1692 Label found_null; |
0 | 1693 // Failed the equality check on receiver[n]... Test for null. |
1694 if (start_row == last_row) { | |
1695 // The only thing left to do is handle the null case. | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1696 if (is_virtual_call) { |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1697 brx(Assembler::zero, false, Assembler::pn, found_null); |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1698 delayed()->nop(); |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1699 // Receiver did not match any saved receiver and there is no empty row for it. |
1251
576e77447e3c
6923002: assert(false,"this call site should not be polymorphic")
kvn
parents:
1206
diff
changeset
|
1700 // Increment total counter to indicate polymorphic case. |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1701 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); |
3839 | 1702 ba_short(done); |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1703 bind(found_null); |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1704 } else { |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1705 brx(Assembler::notZero, false, Assembler::pt, done); |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1706 delayed()->nop(); |
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1707 } |
0 | 1708 break; |
1709 } | |
1710 // Since null is rare, make it be the branch-taken case. | |
1711 brx(Assembler::zero, false, Assembler::pn, found_null); | |
1712 delayed()->nop(); | |
1713 | |
1714 // Put all the "Case 3" tests here. | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1715 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call); |
0 | 1716 |
1717 // Found a null. Keep searching for a matching receiver, | |
1718 // but remember that this is an empty (unused) slot. | |
1719 bind(found_null); | |
1720 } | |
1721 } | |
1722 | |
1723 // In the fall-through case, we found no matching receiver, but we | |
1724 // observed the receiver[start_row] is NULL. | |
1725 | |
1726 // Fill in the receiver field and increment the count. | |
1727 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); | |
1728 set_mdp_data_at(recvr_offset, receiver); | |
1729 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); | |
1730 mov(DataLayout::counter_increment, scratch); | |
1731 set_mdp_data_at(count_offset, scratch); | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1732 if (start_row > 0) { |
3839 | 1733 ba_short(done); |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1734 } |
0 | 1735 } |
1736 | |
1737 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1738 Register scratch, bool is_virtual_call) { |
0 | 1739 assert(ProfileInterpreter, "must be profiling"); |
1740 Label done; | |
1741 | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1742 record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call); |
0 | 1743 |
1744 bind (done); | |
1745 } | |
1746 | |
1747 | |
1748 // Count a ret in the bytecodes. | |
1749 | |
1750 void InterpreterMacroAssembler::profile_ret(TosState state, | |
1751 Register return_bci, | |
1752 Register scratch) { | |
1753 if (ProfileInterpreter) { | |
1754 Label profile_continue; | |
1755 uint row; | |
1756 | |
1757 // If no method data exists, go to profile_continue. | |
1758 test_method_data_pointer(profile_continue); | |
1759 | |
1760 // Update the total ret count. | |
1761 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
1762 | |
1763 for (row = 0; row < RetData::row_limit(); row++) { | |
1764 Label next_test; | |
1765 | |
1766 // See if return_bci is equal to bci[n]: | |
1767 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), | |
1768 return_bci, next_test, scratch); | |
1769 | |
1770 // return_bci is equal to bci[n]. Increment the count. | |
1771 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); | |
1772 | |
1773 // The method data pointer needs to be updated to reflect the new target. | |
1774 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); | |
3839 | 1775 ba_short(profile_continue); |
0 | 1776 bind(next_test); |
1777 } | |
1778 | |
1779 update_mdp_for_ret(state, return_bci); | |
1780 | |
1781 bind (profile_continue); | |
1782 } | |
1783 } | |
1784 | |
1785 // Profile an unexpected null in the bytecodes. | |
1786 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { | |
1787 if (ProfileInterpreter) { | |
1788 Label profile_continue; | |
1789 | |
1790 // If no method data exists, go to profile_continue. | |
1791 test_method_data_pointer(profile_continue); | |
1792 | |
1793 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); | |
1794 | |
1795 // The method data pointer needs to be updated. | |
1796 int mdp_delta = in_bytes(BitData::bit_data_size()); | |
1797 if (TypeProfileCasts) { | |
1798 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); | |
1799 } | |
1800 update_mdp_by_constant(mdp_delta); | |
1801 | |
1802 bind (profile_continue); | |
1803 } | |
1804 } | |
1805 | |
1806 void InterpreterMacroAssembler::profile_typecheck(Register klass, | |
1807 Register scratch) { | |
1808 if (ProfileInterpreter) { | |
1809 Label profile_continue; | |
1810 | |
1811 // If no method data exists, go to profile_continue. | |
1812 test_method_data_pointer(profile_continue); | |
1813 | |
1814 int mdp_delta = in_bytes(BitData::bit_data_size()); | |
1815 if (TypeProfileCasts) { | |
1816 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); | |
1817 | |
1818 // Record the object type. | |
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1819 record_klass_in_profile(klass, scratch, false); |
0 | 1820 } |
1821 | |
1822 // The method data pointer needs to be updated. | |
1823 update_mdp_by_constant(mdp_delta); | |
1824 | |
1825 bind (profile_continue); | |
1826 } | |
1827 } | |
1828 | |
1829 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { | |
1830 if (ProfileInterpreter && TypeProfileCasts) { | |
1831 Label profile_continue; | |
1832 | |
1833 // If no method data exists, go to profile_continue. | |
1834 test_method_data_pointer(profile_continue); | |
1835 | |
1836 int count_offset = in_bytes(CounterData::count_offset()); | |
1837 // Back up the address, since we have already bumped the mdp. | |
1838 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); | |
1839 | |
1840 // *Decrement* the counter. We expect to see zero or small negatives. | |
1841 increment_mdp_data_at(count_offset, scratch, true); | |
1842 | |
1843 bind (profile_continue); | |
1844 } | |
1845 } | |
1846 | |
1847 // Count the default case of a switch construct. | |
1848 | |
1849 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { | |
1850 if (ProfileInterpreter) { | |
1851 Label profile_continue; | |
1852 | |
1853 // If no method data exists, go to profile_continue. | |
1854 test_method_data_pointer(profile_continue); | |
1855 | |
1856 // Update the default case count | |
1857 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), | |
1858 scratch); | |
1859 | |
1860 // The method data pointer needs to be updated. | |
1861 update_mdp_by_offset( | |
1862 in_bytes(MultiBranchData::default_displacement_offset()), | |
1863 scratch); | |
1864 | |
1865 bind (profile_continue); | |
1866 } | |
1867 } | |
1868 | |
1869 // Count the index'th case of a switch construct. | |
1870 | |
1871 void InterpreterMacroAssembler::profile_switch_case(Register index, | |
1872 Register scratch, | |
1873 Register scratch2, | |
1874 Register scratch3) { | |
1875 if (ProfileInterpreter) { | |
1876 Label profile_continue; | |
1877 | |
1878 // If no method data exists, go to profile_continue. | |
1879 test_method_data_pointer(profile_continue); | |
1880 | |
1881 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() | |
1882 set(in_bytes(MultiBranchData::per_case_size()), scratch); | |
1883 smul(index, scratch, scratch); | |
1884 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); | |
1885 | |
1886 // Update the case count | |
1887 increment_mdp_data_at(scratch, | |
1888 in_bytes(MultiBranchData::relative_count_offset()), | |
1889 scratch2, | |
1890 scratch3); | |
1891 | |
1892 // The method data pointer needs to be updated. | |
1893 update_mdp_by_offset(scratch, | |
1894 in_bytes(MultiBranchData::relative_displacement_offset()), | |
1895 scratch2); | |
1896 | |
1897 bind (profile_continue); | |
1898 } | |
1899 } | |
1900 | |
1901 // add a InterpMonitorElem to stack (see frame_sparc.hpp) | |
1902 | |
1903 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, | |
1904 Register Rtemp, | |
1905 Register Rtemp2 ) { | |
1906 | |
1907 Register Rlimit = Lmonitors; | |
1908 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
1909 assert( (delta & LongAlignmentMask) == 0, | |
1910 "sizeof BasicObjectLock must be even number of doublewords"); | |
1911 | |
1912 sub( SP, delta, SP); | |
1913 sub( Lesp, delta, Lesp); | |
1914 sub( Lmonitors, delta, Lmonitors); | |
1915 | |
1916 if (!stack_is_empty) { | |
1917 | |
1918 // must copy stack contents down | |
1919 | |
1920 Label start_copying, next; | |
1921 | |
1922 // untested("monitor stack expansion"); | |
1923 compute_stack_base(Rtemp); | |
3839 | 1924 ba(start_copying); |
1925 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below | |
0 | 1926 |
1927 // note: must copy from low memory upwards | |
1928 // On entry to loop, | |
1929 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) | |
1930 // Loop mutates Rtemp | |
1931 | |
1932 bind( next); | |
1933 | |
1934 st_ptr(Rtemp2, Rtemp, 0); | |
1935 inc(Rtemp, wordSize); | |
1936 cmp(Rtemp, Rlimit); // are we done? (duplicated above) | |
1937 | |
1938 bind( start_copying ); | |
1939 | |
1940 brx( notEqual, true, pn, next ); | |
1941 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); | |
1942 | |
1943 // done copying stack | |
1944 } | |
1945 } | |
1946 | |
1947 // Locals | |
1948 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { | |
1949 assert_not_delayed(); | |
1506 | 1950 sll(index, Interpreter::logStackElementSize, index); |
0 | 1951 sub(Llocals, index, index); |
1506 | 1952 ld_ptr(index, 0, dst); |
0 | 1953 // Note: index must hold the effective address--the iinc template uses it |
1954 } | |
1955 | |
1956 // Just like access_local_ptr but the tag is a returnAddress | |
1957 void InterpreterMacroAssembler::access_local_returnAddress(Register index, | |
1958 Register dst ) { | |
1959 assert_not_delayed(); | |
1506 | 1960 sll(index, Interpreter::logStackElementSize, index); |
0 | 1961 sub(Llocals, index, index); |
1506 | 1962 ld_ptr(index, 0, dst); |
0 | 1963 } |
1964 | |
1965 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { | |
1966 assert_not_delayed(); | |
1506 | 1967 sll(index, Interpreter::logStackElementSize, index); |
0 | 1968 sub(Llocals, index, index); |
1506 | 1969 ld(index, 0, dst); |
0 | 1970 // Note: index must hold the effective address--the iinc template uses it |
1971 } | |
1972 | |
1973 | |
1974 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { | |
1975 assert_not_delayed(); | |
1506 | 1976 sll(index, Interpreter::logStackElementSize, index); |
0 | 1977 sub(Llocals, index, index); |
1978 // First half stored at index n+1 (which grows down from Llocals[n]) | |
1979 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); | |
1980 } | |
1981 | |
1982 | |
1983 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { | |
1984 assert_not_delayed(); | |
1506 | 1985 sll(index, Interpreter::logStackElementSize, index); |
0 | 1986 sub(Llocals, index, index); |
1506 | 1987 ldf(FloatRegisterImpl::S, index, 0, dst); |
0 | 1988 } |
1989 | |
1990 | |
1991 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { | |
1992 assert_not_delayed(); | |
1506 | 1993 sll(index, Interpreter::logStackElementSize, index); |
0 | 1994 sub(Llocals, index, index); |
1995 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); | |
1996 } | |
1997 | |
1998 | |
1999 #ifdef ASSERT | |
2000 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { | |
2001 Label L; | |
2002 | |
2003 assert(Rindex != Rscratch, "Registers cannot be same"); | |
2004 assert(Rindex != Rscratch1, "Registers cannot be same"); | |
2005 assert(Rlimit != Rscratch, "Registers cannot be same"); | |
2006 assert(Rlimit != Rscratch1, "Registers cannot be same"); | |
2007 assert(Rscratch1 != Rscratch, "Registers cannot be same"); | |
2008 | |
2009 // untested("reg area corruption"); | |
2010 add(Rindex, offset, Rscratch); | |
2011 add(Rlimit, 64 + STACK_BIAS, Rscratch1); | |
3839 | 2012 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L); |
0 | 2013 stop("regsave area is being clobbered"); |
2014 bind(L); | |
2015 } | |
2016 #endif // ASSERT | |
2017 | |
2018 | |
2019 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { | |
2020 assert_not_delayed(); | |
1506 | 2021 sll(index, Interpreter::logStackElementSize, index); |
0 | 2022 sub(Llocals, index, index); |
1506 | 2023 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);) |
2024 st(src, index, 0); | |
0 | 2025 } |
2026 | |
1506 | 2027 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) { |
0 | 2028 assert_not_delayed(); |
1506 | 2029 sll(index, Interpreter::logStackElementSize, index); |
0 | 2030 sub(Llocals, index, index); |
1506 | 2031 #ifdef ASSERT |
2032 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); | |
2033 #endif | |
2034 st_ptr(src, index, 0); | |
0 | 2035 } |
2036 | |
2037 | |
2038 | |
1506 | 2039 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) { |
2040 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); | |
0 | 2041 } |
2042 | |
2043 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { | |
2044 assert_not_delayed(); | |
1506 | 2045 sll(index, Interpreter::logStackElementSize, index); |
0 | 2046 sub(Llocals, index, index); |
1506 | 2047 #ifdef ASSERT |
0 | 2048 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); |
1506 | 2049 #endif |
0 | 2050 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 |
2051 } | |
2052 | |
2053 | |
2054 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { | |
2055 assert_not_delayed(); | |
1506 | 2056 sll(index, Interpreter::logStackElementSize, index); |
0 | 2057 sub(Llocals, index, index); |
1506 | 2058 #ifdef ASSERT |
2059 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); | |
2060 #endif | |
2061 stf(FloatRegisterImpl::S, src, index, 0); | |
0 | 2062 } |
2063 | |
2064 | |
2065 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { | |
2066 assert_not_delayed(); | |
1506 | 2067 sll(index, Interpreter::logStackElementSize, index); |
0 | 2068 sub(Llocals, index, index); |
1506 | 2069 #ifdef ASSERT |
0 | 2070 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); |
1506 | 2071 #endif |
0 | 2072 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); |
2073 } | |
2074 | |
2075 | |
2076 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { | |
2077 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
2078 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong); | |
2079 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; | |
2080 } | |
2081 | |
2082 | |
2083 Address InterpreterMacroAssembler::top_most_monitor() { | |
727 | 2084 return Address(FP, top_most_monitor_byte_offset()); |
0 | 2085 } |
2086 | |
2087 | |
2088 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { | |
2089 add( Lesp, wordSize, Rdest ); | |
2090 } | |
2091 | |
2092 #endif /* CC_INTERP */ | |
2093 | |
2094 void InterpreterMacroAssembler::increment_invocation_counter( Register Rtmp, Register Rtmp2 ) { | |
2095 assert(UseCompiler, "incrementing must be useful"); | |
2096 #ifdef CC_INTERP | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2097 Address inv_counter(G5_method, Method::invocation_counter_offset() + |
727 | 2098 InvocationCounter::counter_offset()); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2099 Address be_counter (G5_method, Method::backedge_counter_offset() + |
727 | 2100 InvocationCounter::counter_offset()); |
0 | 2101 #else |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2102 Address inv_counter(Lmethod, Method::invocation_counter_offset() + |
727 | 2103 InvocationCounter::counter_offset()); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2104 Address be_counter (Lmethod, Method::backedge_counter_offset() + |
727 | 2105 InvocationCounter::counter_offset()); |
0 | 2106 #endif /* CC_INTERP */ |
2107 int delta = InvocationCounter::count_increment; | |
2108 | |
2109 // Load each counter in a register | |
2110 ld( inv_counter, Rtmp ); | |
2111 ld( be_counter, Rtmp2 ); | |
2112 | |
2113 assert( is_simm13( delta ), " delta too large."); | |
2114 | |
2115 // Add the delta to the invocation counter and store the result | |
2116 add( Rtmp, delta, Rtmp ); | |
2117 | |
2118 // Mask the backedge counter | |
2119 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); | |
2120 | |
2121 // Store value | |
2122 st( Rtmp, inv_counter); | |
2123 | |
2124 // Add invocation counter + backedge counter | |
2125 add( Rtmp, Rtmp2, Rtmp); | |
2126 | |
2127 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! | |
2128 } | |
2129 | |
2130 void InterpreterMacroAssembler::increment_backedge_counter( Register Rtmp, Register Rtmp2 ) { | |
2131 assert(UseCompiler, "incrementing must be useful"); | |
2132 #ifdef CC_INTERP | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2133 Address be_counter (G5_method, Method::backedge_counter_offset() + |
727 | 2134 InvocationCounter::counter_offset()); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2135 Address inv_counter(G5_method, Method::invocation_counter_offset() + |
727 | 2136 InvocationCounter::counter_offset()); |
0 | 2137 #else |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2138 Address be_counter (Lmethod, Method::backedge_counter_offset() + |
727 | 2139 InvocationCounter::counter_offset()); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2140 Address inv_counter(Lmethod, Method::invocation_counter_offset() + |
727 | 2141 InvocationCounter::counter_offset()); |
0 | 2142 #endif /* CC_INTERP */ |
2143 int delta = InvocationCounter::count_increment; | |
2144 // Load each counter in a register | |
2145 ld( be_counter, Rtmp ); | |
2146 ld( inv_counter, Rtmp2 ); | |
2147 | |
2148 // Add the delta to the backedge counter | |
2149 add( Rtmp, delta, Rtmp ); | |
2150 | |
2151 // Mask the invocation counter, add to backedge counter | |
2152 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); | |
2153 | |
2154 // and store the result to memory | |
2155 st( Rtmp, be_counter ); | |
2156 | |
2157 // Add backedge + invocation counter | |
2158 add( Rtmp, Rtmp2, Rtmp ); | |
2159 | |
2160 // Note that this macro must leave backedge_count + invocation_count in Rtmp! | |
2161 } | |
2162 | |
2163 #ifndef CC_INTERP | |
2164 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, | |
2165 Register branch_bcp, | |
2166 Register Rtmp ) { | |
2167 Label did_not_overflow; | |
2168 Label overflow_with_error; | |
2169 assert_different_registers(backedge_count, Rtmp, branch_bcp); | |
2170 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); | |
2171 | |
727 | 2172 AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit); |
0 | 2173 load_contents(limit, Rtmp); |
3839 | 2174 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow); |
0 | 2175 |
2176 // When ProfileInterpreter is on, the backedge_count comes from the | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2177 // MethodData*, which value does not get reset on the call to |
0 | 2178 // frequency_counter_overflow(). To avoid excessive calls to the overflow |
2179 // routine while the method is being compiled, add a second test to make sure | |
2180 // the overflow function is called only once every overflow_frequency. | |
2181 if (ProfileInterpreter) { | |
2182 const int overflow_frequency = 1024; | |
2183 andcc(backedge_count, overflow_frequency-1, Rtmp); | |
2184 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); | |
2185 delayed()->nop(); | |
2186 } | |
2187 | |
2188 // overflow in loop, pass branch bytecode | |
2189 set(6,Rtmp); | |
2190 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); | |
2191 | |
2192 // Was an OSR adapter generated? | |
2193 // O0 = osr nmethod | |
3839 | 2194 br_null_short(O0, Assembler::pn, overflow_with_error); |
0 | 2195 |
2196 // Has the nmethod been invalidated already? | |
2197 ld(O0, nmethod::entry_bci_offset(), O2); | |
3839 | 2198 cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, overflow_with_error); |
0 | 2199 |
2200 // migrate the interpreter frame off of the stack | |
2201 | |
2202 mov(G2_thread, L7); | |
2203 // save nmethod | |
2204 mov(O0, L6); | |
2205 set_last_Java_frame(SP, noreg); | |
2206 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); | |
2207 reset_last_Java_frame(); | |
2208 mov(L7, G2_thread); | |
2209 | |
2210 // move OSR nmethod to I1 | |
2211 mov(L6, I1); | |
2212 | |
2213 // OSR buffer to I0 | |
2214 mov(O0, I0); | |
2215 | |
2216 // remove the interpreter frame | |
2217 restore(I5_savedSP, 0, SP); | |
2218 | |
2219 // Jump to the osr code. | |
2220 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); | |
2221 jmp(O2, G0); | |
2222 delayed()->nop(); | |
2223 | |
2224 bind(overflow_with_error); | |
2225 | |
2226 bind(did_not_overflow); | |
2227 } | |
2228 | |
2229 | |
2230 | |
2231 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { | |
2232 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } | |
2233 } | |
2234 | |
2235 | |
2236 // local helper function for the verify_oop_or_return_address macro | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2237 static bool verify_return_address(Method* m, int bci) { |
0 | 2238 #ifndef PRODUCT |
2239 address pc = (address)(m->constMethod()) | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2240 + in_bytes(ConstMethod::codes_offset()) + bci; |
0 | 2241 // assume it is a valid return address if it is inside m and is preceded by a jsr |
2242 if (!m->contains(pc)) return false; | |
2243 address jsr_pc; | |
2244 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); | |
2245 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; | |
2246 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); | |
2247 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; | |
2248 #endif // PRODUCT | |
2249 return false; | |
2250 } | |
2251 | |
2252 | |
2253 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { | |
2254 if (!VerifyOops) return; | |
2255 // the VM documentation for the astore[_wide] bytecode allows | |
2256 // the TOS to be not only an oop but also a return address | |
2257 Label test; | |
2258 Label skip; | |
2259 // See if it is an address (in the current method): | |
2260 | |
2261 mov(reg, Rtmp); | |
2262 const int log2_bytecode_size_limit = 16; | |
2263 srl(Rtmp, log2_bytecode_size_limit, Rtmp); | |
3839 | 2264 br_notnull_short( Rtmp, pt, test ); |
0 | 2265 |
2266 // %%% should use call_VM_leaf here? | |
2267 save_frame_and_mov(0, Lmethod, O0, reg, O1); | |
2268 save_thread(L7_thread_cache); | |
2269 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); | |
2270 delayed()->nop(); | |
2271 restore_thread(L7_thread_cache); | |
2272 br_notnull( O0, false, pt, skip ); | |
2273 delayed()->restore(); | |
2274 | |
2275 // Perform a more elaborate out-of-line call | |
2276 // Not an address; verify it: | |
2277 bind(test); | |
2278 verify_oop(reg); | |
2279 bind(skip); | |
2280 } | |
2281 | |
2282 | |
2283 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { | |
2284 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); | |
2285 } | |
2286 #endif /* CC_INTERP */ | |
2287 | |
2288 // Inline assembly for: | |
2289 // | |
2290 // if (thread is in interp_only_mode) { | |
2291 // InterpreterRuntime::post_method_entry(); | |
2292 // } | |
2293 // if (DTraceMethodProbes) { | |
605 | 2294 // SharedRuntime::dtrace_method_entry(method, receiver); |
0 | 2295 // } |
610
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2296 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { |
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2297 // SharedRuntime::rc_trace_method_entry(method, receiver); |
0 | 2298 // } |
2299 | |
2300 void InterpreterMacroAssembler::notify_method_entry() { | |
2301 | |
2302 // C++ interpreter only uses this for native methods. | |
2303 | |
2304 // Whenever JVMTI puts a thread in interp_only_mode, method | |
2305 // entry/exit events are sent for that thread to track stack | |
2306 // depth. If it is possible to enter interp_only_mode we add | |
2307 // the code to check if the event should be sent. | |
2308 if (JvmtiExport::can_post_interpreter_events()) { | |
2309 Label L; | |
2310 Register temp_reg = O5; | |
727 | 2311 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
0 | 2312 ld(interp_only, temp_reg); |
3839 | 2313 cmp_and_br_short(temp_reg, 0, equal, pt, L); |
0 | 2314 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); |
2315 bind(L); | |
2316 } | |
2317 | |
2318 { | |
2319 Register temp_reg = O5; | |
2320 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); | |
2321 call_VM_leaf(noreg, | |
2322 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), | |
2323 G2_thread, Lmethod); | |
2324 } | |
610
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2325 |
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2326 // RedefineClasses() tracing support for obsolete method entry |
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2327 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { |
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2328 call_VM_leaf(noreg, |
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2329 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), |
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2330 G2_thread, Lmethod); |
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2331 } |
0 | 2332 } |
2333 | |
2334 | |
2335 // Inline assembly for: | |
2336 // | |
2337 // if (thread is in interp_only_mode) { | |
2338 // // save result | |
2339 // InterpreterRuntime::post_method_exit(); | |
2340 // // restore result | |
2341 // } | |
2342 // if (DTraceMethodProbes) { | |
2343 // SharedRuntime::dtrace_method_exit(thread, method); | |
2344 // } | |
2345 // | |
2346 // Native methods have their result stored in d_tmp and l_tmp | |
2347 // Java methods have their result stored in the expression stack | |
2348 | |
2349 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, | |
2350 TosState state, | |
2351 NotifyMethodExitMode mode) { | |
2352 // C++ interpreter only uses this for native methods. | |
2353 | |
2354 // Whenever JVMTI puts a thread in interp_only_mode, method | |
2355 // entry/exit events are sent for that thread to track stack | |
2356 // depth. If it is possible to enter interp_only_mode we add | |
2357 // the code to check if the event should be sent. | |
2358 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { | |
2359 Label L; | |
2360 Register temp_reg = O5; | |
727 | 2361 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
0 | 2362 ld(interp_only, temp_reg); |
3839 | 2363 cmp_and_br_short(temp_reg, 0, equal, pt, L); |
0 | 2364 |
2365 // Note: frame::interpreter_frame_result has a dependency on how the | |
2366 // method result is saved across the call to post_method_exit. For | |
2367 // native methods it assumes the result registers are saved to | |
2368 // l_scratch and d_scratch. If this changes then the interpreter_frame_result | |
2369 // implementation will need to be updated too. | |
2370 | |
2371 save_return_value(state, is_native_method); | |
2372 call_VM(noreg, | |
2373 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); | |
2374 restore_return_value(state, is_native_method); | |
2375 bind(L); | |
2376 } | |
2377 | |
2378 { | |
2379 Register temp_reg = O5; | |
2380 // Dtrace notification | |
2381 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); | |
2382 save_return_value(state, is_native_method); | |
2383 call_VM_leaf( | |
2384 noreg, | |
2385 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), | |
2386 G2_thread, Lmethod); | |
2387 restore_return_value(state, is_native_method); | |
2388 } | |
2389 } | |
2390 | |
2391 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { | |
2392 #ifdef CC_INTERP | |
2393 // result potentially in O0/O1: save it across calls | |
2394 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult)); | |
2395 #ifdef _LP64 | |
2396 stx(O0, STATE(_native_lresult)); | |
2397 #else | |
2398 std(O0, STATE(_native_lresult)); | |
2399 #endif | |
2400 #else // CC_INTERP | |
2401 if (is_native_call) { | |
2402 stf(FloatRegisterImpl::D, F0, d_tmp); | |
2403 #ifdef _LP64 | |
2404 stx(O0, l_tmp); | |
2405 #else | |
2406 std(O0, l_tmp); | |
2407 #endif | |
2408 } else { | |
2409 push(state); | |
2410 } | |
2411 #endif // CC_INTERP | |
2412 } | |
2413 | |
2414 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { | |
2415 #ifdef CC_INTERP | |
2416 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0); | |
2417 #ifdef _LP64 | |
2418 ldx(STATE(_native_lresult), O0); | |
2419 #else | |
2420 ldd(STATE(_native_lresult), O0); | |
2421 #endif | |
2422 #else // CC_INTERP | |
2423 if (is_native_call) { | |
2424 ldf(FloatRegisterImpl::D, d_tmp, F0); | |
2425 #ifdef _LP64 | |
2426 ldx(l_tmp, O0); | |
2427 #else | |
2428 ldd(l_tmp, O0); | |
2429 #endif | |
2430 } else { | |
2431 pop(state); | |
2432 } | |
2433 #endif // CC_INTERP | |
2434 } | |
1783 | 2435 |
2436 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. | |
2437 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, | |
2438 int increment, int mask, | |
2439 Register scratch1, Register scratch2, | |
2440 Condition cond, Label *where) { | |
2441 ld(counter_addr, scratch1); | |
2442 add(scratch1, increment, scratch1); | |
2443 if (is_simm13(mask)) { | |
2444 andcc(scratch1, mask, G0); | |
2445 } else { | |
2446 set(mask, scratch2); | |
2447 andcc(scratch1, scratch2, G0); | |
2448 } | |
2449 br(cond, false, Assembler::pn, *where); | |
2450 delayed()->st(scratch1, counter_addr); | |
2451 } |