Thu, 03 Jan 2013 16:30:47 -0800
8005544: Use 256bit YMM registers in arraycopy stubs on x86
Summary: Use YMM registers in arraycopy and array_fill stubs.
Reviewed-by: roland, twisti
1 /*
2 * Copyright (c) 2003, 2012, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "interp_masm_x86_64.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "oops/arrayOop.hpp"
30 #include "oops/markOop.hpp"
31 #include "oops/methodData.hpp"
32 #include "oops/method.hpp"
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 #include "runtime/thread.inline.hpp"
42 // Implementation of InterpreterMacroAssembler
44 #ifdef CC_INTERP
45 void InterpreterMacroAssembler::get_method(Register reg) {
46 movptr(reg, Address(rbp, -((int)sizeof(BytecodeInterpreter) + 2 * wordSize)));
47 movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
48 }
49 #endif // CC_INTERP
51 #ifndef CC_INTERP
53 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
54 int number_of_arguments) {
55 // interpreter specific
56 //
57 // Note: No need to save/restore bcp & locals (r13 & r14) pointer
58 // since these are callee saved registers and no blocking/
59 // GC can happen in leaf calls.
60 // Further Note: DO NOT save/restore bcp/locals. If a caller has
61 // already saved them so that it can use esi/edi as temporaries
62 // then a save/restore here will DESTROY the copy the caller
63 // saved! There used to be a save_bcp() that only happened in
64 // the ASSERT path (no restore_bcp). Which caused bizarre failures
65 // when jvm built with ASSERTs.
66 #ifdef ASSERT
67 {
68 Label L;
69 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
70 jcc(Assembler::equal, L);
71 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
72 " last_sp != NULL");
73 bind(L);
74 }
75 #endif
76 // super call
77 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
78 // interpreter specific
79 // Used to ASSERT that r13/r14 were equal to frame's bcp/locals
80 // but since they may not have been saved (and we don't want to
81 // save thme here (see note above) the assert is invalid.
82 }
84 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
85 Register java_thread,
86 Register last_java_sp,
87 address entry_point,
88 int number_of_arguments,
89 bool check_exceptions) {
90 // interpreter specific
91 //
92 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
93 // really make a difference for these runtime calls, since they are
94 // slow anyway. Btw., bcp must be saved/restored since it may change
95 // due to GC.
96 // assert(java_thread == noreg , "not expecting a precomputed java thread");
97 save_bcp();
98 #ifdef ASSERT
99 {
100 Label L;
101 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
102 jcc(Assembler::equal, L);
103 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
104 " last_sp != NULL");
105 bind(L);
106 }
107 #endif /* ASSERT */
108 // super call
109 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
110 entry_point, number_of_arguments,
111 check_exceptions);
112 // interpreter specific
113 restore_bcp();
114 restore_locals();
115 }
118 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
119 if (JvmtiExport::can_pop_frame()) {
120 Label L;
121 // Initiate popframe handling only if it is not already being
122 // processed. If the flag has the popframe_processing bit set, it
123 // means that this code is called *during* popframe handling - we
124 // don't want to reenter.
125 // This method is only called just after the call into the vm in
126 // call_VM_base, so the arg registers are available.
127 movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
128 testl(c_rarg0, JavaThread::popframe_pending_bit);
129 jcc(Assembler::zero, L);
130 testl(c_rarg0, JavaThread::popframe_processing_bit);
131 jcc(Assembler::notZero, L);
132 // Call Interpreter::remove_activation_preserving_args_entry() to get the
133 // address of the same-named entrypoint in the generated interpreter code.
134 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
135 jmp(rax);
136 bind(L);
137 }
138 }
141 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
142 movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
143 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
144 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
145 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
146 switch (state) {
147 case atos: movptr(rax, oop_addr);
148 movptr(oop_addr, (int32_t)NULL_WORD);
149 verify_oop(rax, state); break;
150 case ltos: movptr(rax, val_addr); break;
151 case btos: // fall through
152 case ctos: // fall through
153 case stos: // fall through
154 case itos: movl(rax, val_addr); break;
155 case ftos: movflt(xmm0, val_addr); break;
156 case dtos: movdbl(xmm0, val_addr); break;
157 case vtos: /* nothing to do */ break;
158 default : ShouldNotReachHere();
159 }
160 // Clean up tos value in the thread object
161 movl(tos_addr, (int) ilgl);
162 movl(val_addr, (int32_t) NULL_WORD);
163 }
166 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
167 if (JvmtiExport::can_force_early_return()) {
168 Label L;
169 movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
170 testptr(c_rarg0, c_rarg0);
171 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
173 // Initiate earlyret handling only if it is not already being processed.
174 // If the flag has the earlyret_processing bit set, it means that this code
175 // is called *during* earlyret handling - we don't want to reenter.
176 movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
177 cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
178 jcc(Assembler::notEqual, L);
180 // Call Interpreter::remove_activation_early_entry() to get the address of the
181 // same-named entrypoint in the generated interpreter code.
182 movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
183 movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
184 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
185 jmp(rax);
186 bind(L);
187 }
188 }
191 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
192 Register reg,
193 int bcp_offset) {
194 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
195 movl(reg, Address(r13, bcp_offset));
196 bswapl(reg);
197 shrl(reg, 16);
198 }
201 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
202 int bcp_offset,
203 size_t index_size) {
204 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
205 if (index_size == sizeof(u2)) {
206 load_unsigned_short(index, Address(r13, bcp_offset));
207 } else if (index_size == sizeof(u4)) {
208 assert(EnableInvokeDynamic, "giant index used only for JSR 292");
209 movl(index, Address(r13, bcp_offset));
210 // Check if the secondary index definition is still ~x, otherwise
211 // we have to change the following assembler code to calculate the
212 // plain index.
213 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
214 notl(index); // convert to plain index
215 } else if (index_size == sizeof(u1)) {
216 load_unsigned_byte(index, Address(r13, bcp_offset));
217 } else {
218 ShouldNotReachHere();
219 }
220 }
223 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
224 Register index,
225 int bcp_offset,
226 size_t index_size) {
227 assert_different_registers(cache, index);
228 get_cache_index_at_bcp(index, bcp_offset, index_size);
229 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
230 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
231 // convert from field index to ConstantPoolCacheEntry index
232 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
233 shll(index, 2);
234 }
237 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
238 Register index,
239 Register bytecode,
240 int byte_no,
241 int bcp_offset,
242 size_t index_size) {
243 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
244 // We use a 32-bit load here since the layout of 64-bit words on
245 // little-endian machines allow us that.
246 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
247 const int shift_count = (1 + byte_no) * BitsPerByte;
248 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
249 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
250 "correct shift count");
251 shrl(bytecode, shift_count);
252 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
253 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
254 }
257 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
258 Register tmp,
259 int bcp_offset,
260 size_t index_size) {
261 assert(cache != tmp, "must use different register");
262 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
263 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
264 // convert from field index to ConstantPoolCacheEntry index
265 // and from word offset to byte offset
266 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
267 shll(tmp, 2 + LogBytesPerWord);
268 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
269 // skip past the header
270 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
271 addptr(cache, tmp); // construct pointer to cache entry
272 }
274 // Load object from cpool->resolved_references(index)
275 void InterpreterMacroAssembler::load_resolved_reference_at_index(
276 Register result, Register index) {
277 assert_different_registers(result, index);
278 // convert from field index to resolved_references() index and from
279 // word index to byte offset. Since this is a java object, it can be compressed
280 Register tmp = index; // reuse
281 shll(tmp, LogBytesPerHeapOop);
283 get_constant_pool(result);
284 // load pointer for resolved_references[] objArray
285 movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
286 // JNIHandles::resolve(obj);
287 movptr(result, Address(result, 0));
288 // Add in the index
289 addptr(result, tmp);
290 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
291 }
293 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
294 // subtype of super_klass.
295 //
296 // Args:
297 // rax: superklass
298 // Rsub_klass: subklass
299 //
300 // Kills:
301 // rcx, rdi
302 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
303 Label& ok_is_subtype) {
304 assert(Rsub_klass != rax, "rax holds superklass");
305 assert(Rsub_klass != r14, "r14 holds locals");
306 assert(Rsub_klass != r13, "r13 holds bcp");
307 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
308 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
310 // Profile the not-null value's klass.
311 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
313 // Do the check.
314 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
316 // Profile the failure of the check.
317 profile_typecheck_failed(rcx); // blows rcx
318 }
322 // Java Expression Stack
324 void InterpreterMacroAssembler::pop_ptr(Register r) {
325 pop(r);
326 }
328 void InterpreterMacroAssembler::pop_i(Register r) {
329 // XXX can't use pop currently, upper half non clean
330 movl(r, Address(rsp, 0));
331 addptr(rsp, wordSize);
332 }
334 void InterpreterMacroAssembler::pop_l(Register r) {
335 movq(r, Address(rsp, 0));
336 addptr(rsp, 2 * Interpreter::stackElementSize);
337 }
339 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
340 movflt(r, Address(rsp, 0));
341 addptr(rsp, wordSize);
342 }
344 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
345 movdbl(r, Address(rsp, 0));
346 addptr(rsp, 2 * Interpreter::stackElementSize);
347 }
349 void InterpreterMacroAssembler::push_ptr(Register r) {
350 push(r);
351 }
353 void InterpreterMacroAssembler::push_i(Register r) {
354 push(r);
355 }
357 void InterpreterMacroAssembler::push_l(Register r) {
358 subptr(rsp, 2 * wordSize);
359 movq(Address(rsp, 0), r);
360 }
362 void InterpreterMacroAssembler::push_f(XMMRegister r) {
363 subptr(rsp, wordSize);
364 movflt(Address(rsp, 0), r);
365 }
367 void InterpreterMacroAssembler::push_d(XMMRegister r) {
368 subptr(rsp, 2 * wordSize);
369 movdbl(Address(rsp, 0), r);
370 }
372 void InterpreterMacroAssembler::pop(TosState state) {
373 switch (state) {
374 case atos: pop_ptr(); break;
375 case btos:
376 case ctos:
377 case stos:
378 case itos: pop_i(); break;
379 case ltos: pop_l(); break;
380 case ftos: pop_f(); break;
381 case dtos: pop_d(); break;
382 case vtos: /* nothing to do */ break;
383 default: ShouldNotReachHere();
384 }
385 verify_oop(rax, state);
386 }
388 void InterpreterMacroAssembler::push(TosState state) {
389 verify_oop(rax, state);
390 switch (state) {
391 case atos: push_ptr(); break;
392 case btos:
393 case ctos:
394 case stos:
395 case itos: push_i(); break;
396 case ltos: push_l(); break;
397 case ftos: push_f(); break;
398 case dtos: push_d(); break;
399 case vtos: /* nothing to do */ break;
400 default : ShouldNotReachHere();
401 }
402 }
405 // Helpers for swap and dup
406 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
407 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
408 }
410 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
411 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
412 }
415 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
416 // set sender sp
417 lea(r13, Address(rsp, wordSize));
418 // record last_sp
419 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
420 }
423 // Jump to from_interpreted entry of a call unless single stepping is possible
424 // in this thread in which case we must call the i2i entry
425 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
426 prepare_to_jump_from_interpreted();
428 if (JvmtiExport::can_post_interpreter_events()) {
429 Label run_compiled_code;
430 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
431 // compiled code in threads for which the event is enabled. Check here for
432 // interp_only_mode if these events CAN be enabled.
433 // interp_only is an int, on little endian it is sufficient to test the byte only
434 // Is a cmpl faster?
435 cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
436 jccb(Assembler::zero, run_compiled_code);
437 jmp(Address(method, Method::interpreter_entry_offset()));
438 bind(run_compiled_code);
439 }
441 jmp(Address(method, Method::from_interpreted_offset()));
443 }
446 // The following two routines provide a hook so that an implementation
447 // can schedule the dispatch in two parts. amd64 does not do this.
448 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
449 // Nothing amd64 specific to be done here
450 }
452 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
453 dispatch_next(state, step);
454 }
456 void InterpreterMacroAssembler::dispatch_base(TosState state,
457 address* table,
458 bool verifyoop) {
459 verify_FPU(1, state);
460 if (VerifyActivationFrameSize) {
461 Label L;
462 mov(rcx, rbp);
463 subptr(rcx, rsp);
464 int32_t min_frame_size =
465 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
466 wordSize;
467 cmpptr(rcx, (int32_t)min_frame_size);
468 jcc(Assembler::greaterEqual, L);
469 stop("broken stack frame");
470 bind(L);
471 }
472 if (verifyoop) {
473 verify_oop(rax, state);
474 }
475 lea(rscratch1, ExternalAddress((address)table));
476 jmp(Address(rscratch1, rbx, Address::times_8));
477 }
479 void InterpreterMacroAssembler::dispatch_only(TosState state) {
480 dispatch_base(state, Interpreter::dispatch_table(state));
481 }
483 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
484 dispatch_base(state, Interpreter::normal_table(state));
485 }
487 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
488 dispatch_base(state, Interpreter::normal_table(state), false);
489 }
492 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
493 // load next bytecode (load before advancing r13 to prevent AGI)
494 load_unsigned_byte(rbx, Address(r13, step));
495 // advance r13
496 increment(r13, step);
497 dispatch_base(state, Interpreter::dispatch_table(state));
498 }
500 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
501 // load current bytecode
502 load_unsigned_byte(rbx, Address(r13, 0));
503 dispatch_base(state, table);
504 }
506 // remove activation
507 //
508 // Unlock the receiver if this is a synchronized method.
509 // Unlock any Java monitors from syncronized blocks.
510 // Remove the activation from the stack.
511 //
512 // If there are locked Java monitors
513 // If throw_monitor_exception
514 // throws IllegalMonitorStateException
515 // Else if install_monitor_exception
516 // installs IllegalMonitorStateException
517 // Else
518 // no error processing
519 void InterpreterMacroAssembler::remove_activation(
520 TosState state,
521 Register ret_addr,
522 bool throw_monitor_exception,
523 bool install_monitor_exception,
524 bool notify_jvmdi) {
525 // Note: Registers rdx xmm0 may be in use for the
526 // result check if synchronized method
527 Label unlocked, unlock, no_unlock;
529 // get the value of _do_not_unlock_if_synchronized into rdx
530 const Address do_not_unlock_if_synchronized(r15_thread,
531 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
532 movbool(rdx, do_not_unlock_if_synchronized);
533 movbool(do_not_unlock_if_synchronized, false); // reset the flag
535 // get method access flags
536 movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
537 movl(rcx, Address(rbx, Method::access_flags_offset()));
538 testl(rcx, JVM_ACC_SYNCHRONIZED);
539 jcc(Assembler::zero, unlocked);
541 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
542 // is set.
543 testbool(rdx);
544 jcc(Assembler::notZero, no_unlock);
546 // unlock monitor
547 push(state); // save result
549 // BasicObjectLock will be first in list, since this is a
550 // synchronized method. However, need to check that the object has
551 // not been unlocked by an explicit monitorexit bytecode.
552 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
553 wordSize - (int) sizeof(BasicObjectLock));
554 // We use c_rarg1 so that if we go slow path it will be the correct
555 // register for unlock_object to pass to VM directly
556 lea(c_rarg1, monitor); // address of first monitor
558 movptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
559 testptr(rax, rax);
560 jcc(Assembler::notZero, unlock);
562 pop(state);
563 if (throw_monitor_exception) {
564 // Entry already unlocked, need to throw exception
565 call_VM(noreg, CAST_FROM_FN_PTR(address,
566 InterpreterRuntime::throw_illegal_monitor_state_exception));
567 should_not_reach_here();
568 } else {
569 // Monitor already unlocked during a stack unroll. If requested,
570 // install an illegal_monitor_state_exception. Continue with
571 // stack unrolling.
572 if (install_monitor_exception) {
573 call_VM(noreg, CAST_FROM_FN_PTR(address,
574 InterpreterRuntime::new_illegal_monitor_state_exception));
575 }
576 jmp(unlocked);
577 }
579 bind(unlock);
580 unlock_object(c_rarg1);
581 pop(state);
583 // Check that for block-structured locking (i.e., that all locked
584 // objects has been unlocked)
585 bind(unlocked);
587 // rax: Might contain return value
589 // Check that all monitors are unlocked
590 {
591 Label loop, exception, entry, restart;
592 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
593 const Address monitor_block_top(
594 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
595 const Address monitor_block_bot(
596 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
598 bind(restart);
599 // We use c_rarg1 so that if we go slow path it will be the correct
600 // register for unlock_object to pass to VM directly
601 movptr(c_rarg1, monitor_block_top); // points to current entry, starting
602 // with top-most entry
603 lea(rbx, monitor_block_bot); // points to word before bottom of
604 // monitor block
605 jmp(entry);
607 // Entry already locked, need to throw exception
608 bind(exception);
610 if (throw_monitor_exception) {
611 // Throw exception
612 MacroAssembler::call_VM(noreg,
613 CAST_FROM_FN_PTR(address, InterpreterRuntime::
614 throw_illegal_monitor_state_exception));
615 should_not_reach_here();
616 } else {
617 // Stack unrolling. Unlock object and install illegal_monitor_exception.
618 // Unlock does not block, so don't have to worry about the frame.
619 // We don't have to preserve c_rarg1 since we are going to throw an exception.
621 push(state);
622 unlock_object(c_rarg1);
623 pop(state);
625 if (install_monitor_exception) {
626 call_VM(noreg, CAST_FROM_FN_PTR(address,
627 InterpreterRuntime::
628 new_illegal_monitor_state_exception));
629 }
631 jmp(restart);
632 }
634 bind(loop);
635 // check if current entry is used
636 cmpptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
637 jcc(Assembler::notEqual, exception);
639 addptr(c_rarg1, entry_size); // otherwise advance to next entry
640 bind(entry);
641 cmpptr(c_rarg1, rbx); // check if bottom reached
642 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
643 }
645 bind(no_unlock);
647 // jvmti support
648 if (notify_jvmdi) {
649 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
650 } else {
651 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
652 }
654 // remove activation
655 // get sender sp
656 movptr(rbx,
657 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
658 leave(); // remove frame anchor
659 pop(ret_addr); // get return address
660 mov(rsp, rbx); // set sp to sender sp
661 }
663 #endif // C_INTERP
665 // Lock object
666 //
667 // Args:
668 // c_rarg1: BasicObjectLock to be used for locking
669 //
670 // Kills:
671 // rax
672 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
673 // rscratch1, rscratch2 (scratch regs)
674 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
675 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
677 if (UseHeavyMonitors) {
678 call_VM(noreg,
679 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
680 lock_reg);
681 } else {
682 Label done;
684 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
685 const Register obj_reg = c_rarg3; // Will contain the oop
687 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
688 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
689 const int mark_offset = lock_offset +
690 BasicLock::displaced_header_offset_in_bytes();
692 Label slow_case;
694 // Load object pointer into obj_reg %c_rarg3
695 movptr(obj_reg, Address(lock_reg, obj_offset));
697 if (UseBiasedLocking) {
698 biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
699 }
701 // Load immediate 1 into swap_reg %rax
702 movl(swap_reg, 1);
704 // Load (object->mark() | 1) into swap_reg %rax
705 orptr(swap_reg, Address(obj_reg, 0));
707 // Save (object->mark() | 1) into BasicLock's displaced header
708 movptr(Address(lock_reg, mark_offset), swap_reg);
710 assert(lock_offset == 0,
711 "displached header must be first word in BasicObjectLock");
713 if (os::is_MP()) lock();
714 cmpxchgptr(lock_reg, Address(obj_reg, 0));
715 if (PrintBiasedLockingStatistics) {
716 cond_inc32(Assembler::zero,
717 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
718 }
719 jcc(Assembler::zero, done);
721 // Test if the oopMark is an obvious stack pointer, i.e.,
722 // 1) (mark & 7) == 0, and
723 // 2) rsp <= mark < mark + os::pagesize()
724 //
725 // These 3 tests can be done by evaluating the following
726 // expression: ((mark - rsp) & (7 - os::vm_page_size())),
727 // assuming both stack pointer and pagesize have their
728 // least significant 3 bits clear.
729 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
730 subptr(swap_reg, rsp);
731 andptr(swap_reg, 7 - os::vm_page_size());
733 // Save the test result, for recursive case, the result is zero
734 movptr(Address(lock_reg, mark_offset), swap_reg);
736 if (PrintBiasedLockingStatistics) {
737 cond_inc32(Assembler::zero,
738 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
739 }
740 jcc(Assembler::zero, done);
742 bind(slow_case);
744 // Call the runtime routine for slow case
745 call_VM(noreg,
746 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
747 lock_reg);
749 bind(done);
750 }
751 }
754 // Unlocks an object. Used in monitorexit bytecode and
755 // remove_activation. Throws an IllegalMonitorException if object is
756 // not locked by current thread.
757 //
758 // Args:
759 // c_rarg1: BasicObjectLock for lock
760 //
761 // Kills:
762 // rax
763 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
764 // rscratch1, rscratch2 (scratch regs)
765 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
766 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
768 if (UseHeavyMonitors) {
769 call_VM(noreg,
770 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
771 lock_reg);
772 } else {
773 Label done;
775 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
776 const Register header_reg = c_rarg2; // Will contain the old oopMark
777 const Register obj_reg = c_rarg3; // Will contain the oop
779 save_bcp(); // Save in case of exception
781 // Convert from BasicObjectLock structure to object and BasicLock
782 // structure Store the BasicLock address into %rax
783 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
785 // Load oop into obj_reg(%c_rarg3)
786 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
788 // Free entry
789 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
791 if (UseBiasedLocking) {
792 biased_locking_exit(obj_reg, header_reg, done);
793 }
795 // Load the old header from BasicLock structure
796 movptr(header_reg, Address(swap_reg,
797 BasicLock::displaced_header_offset_in_bytes()));
799 // Test for recursion
800 testptr(header_reg, header_reg);
802 // zero for recursive case
803 jcc(Assembler::zero, done);
805 // Atomic swap back the old header
806 if (os::is_MP()) lock();
807 cmpxchgptr(header_reg, Address(obj_reg, 0));
809 // zero for recursive case
810 jcc(Assembler::zero, done);
812 // Call the runtime routine for slow case.
813 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
814 obj_reg); // restore obj
815 call_VM(noreg,
816 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
817 lock_reg);
819 bind(done);
821 restore_bcp();
822 }
823 }
825 #ifndef CC_INTERP
827 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
828 Label& zero_continue) {
829 assert(ProfileInterpreter, "must be profiling interpreter");
830 movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
831 testptr(mdp, mdp);
832 jcc(Assembler::zero, zero_continue);
833 }
836 // Set the method data pointer for the current bcp.
837 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
838 assert(ProfileInterpreter, "must be profiling interpreter");
839 Label set_mdp;
840 push(rax);
841 push(rbx);
843 get_method(rbx);
844 // Test MDO to avoid the call if it is NULL.
845 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
846 testptr(rax, rax);
847 jcc(Assembler::zero, set_mdp);
848 // rbx: method
849 // r13: bcp
850 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
851 // rax: mdi
852 // mdo is guaranteed to be non-zero here, we checked for it before the call.
853 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
854 addptr(rbx, in_bytes(MethodData::data_offset()));
855 addptr(rax, rbx);
856 bind(set_mdp);
857 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
858 pop(rbx);
859 pop(rax);
860 }
862 void InterpreterMacroAssembler::verify_method_data_pointer() {
863 assert(ProfileInterpreter, "must be profiling interpreter");
864 #ifdef ASSERT
865 Label verify_continue;
866 push(rax);
867 push(rbx);
868 push(c_rarg3);
869 push(c_rarg2);
870 test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
871 get_method(rbx);
873 // If the mdp is valid, it will point to a DataLayout header which is
874 // consistent with the bcp. The converse is highly probable also.
875 load_unsigned_short(c_rarg2,
876 Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
877 addptr(c_rarg2, Address(rbx, Method::const_offset()));
878 lea(c_rarg2, Address(c_rarg2, ConstMethod::codes_offset()));
879 cmpptr(c_rarg2, r13);
880 jcc(Assembler::equal, verify_continue);
881 // rbx: method
882 // r13: bcp
883 // c_rarg3: mdp
884 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
885 rbx, r13, c_rarg3);
886 bind(verify_continue);
887 pop(c_rarg2);
888 pop(c_rarg3);
889 pop(rbx);
890 pop(rax);
891 #endif // ASSERT
892 }
895 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
896 int constant,
897 Register value) {
898 assert(ProfileInterpreter, "must be profiling interpreter");
899 Address data(mdp_in, constant);
900 movptr(data, value);
901 }
904 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
905 int constant,
906 bool decrement) {
907 // Counter address
908 Address data(mdp_in, constant);
910 increment_mdp_data_at(data, decrement);
911 }
913 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
914 bool decrement) {
915 assert(ProfileInterpreter, "must be profiling interpreter");
916 // %%% this does 64bit counters at best it is wasting space
917 // at worst it is a rare bug when counters overflow
919 if (decrement) {
920 // Decrement the register. Set condition codes.
921 addptr(data, (int32_t) -DataLayout::counter_increment);
922 // If the decrement causes the counter to overflow, stay negative
923 Label L;
924 jcc(Assembler::negative, L);
925 addptr(data, (int32_t) DataLayout::counter_increment);
926 bind(L);
927 } else {
928 assert(DataLayout::counter_increment == 1,
929 "flow-free idiom only works with 1");
930 // Increment the register. Set carry flag.
931 addptr(data, DataLayout::counter_increment);
932 // If the increment causes the counter to overflow, pull back by 1.
933 sbbptr(data, (int32_t)0);
934 }
935 }
938 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
939 Register reg,
940 int constant,
941 bool decrement) {
942 Address data(mdp_in, reg, Address::times_1, constant);
944 increment_mdp_data_at(data, decrement);
945 }
947 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
948 int flag_byte_constant) {
949 assert(ProfileInterpreter, "must be profiling interpreter");
950 int header_offset = in_bytes(DataLayout::header_offset());
951 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
952 // Set the flag
953 orl(Address(mdp_in, header_offset), header_bits);
954 }
958 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
959 int offset,
960 Register value,
961 Register test_value_out,
962 Label& not_equal_continue) {
963 assert(ProfileInterpreter, "must be profiling interpreter");
964 if (test_value_out == noreg) {
965 cmpptr(value, Address(mdp_in, offset));
966 } else {
967 // Put the test value into a register, so caller can use it:
968 movptr(test_value_out, Address(mdp_in, offset));
969 cmpptr(test_value_out, value);
970 }
971 jcc(Assembler::notEqual, not_equal_continue);
972 }
975 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
976 int offset_of_disp) {
977 assert(ProfileInterpreter, "must be profiling interpreter");
978 Address disp_address(mdp_in, offset_of_disp);
979 addptr(mdp_in, disp_address);
980 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
981 }
984 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
985 Register reg,
986 int offset_of_disp) {
987 assert(ProfileInterpreter, "must be profiling interpreter");
988 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
989 addptr(mdp_in, disp_address);
990 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
991 }
994 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
995 int constant) {
996 assert(ProfileInterpreter, "must be profiling interpreter");
997 addptr(mdp_in, constant);
998 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
999 }
1002 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1003 assert(ProfileInterpreter, "must be profiling interpreter");
1004 push(return_bci); // save/restore across call_VM
1005 call_VM(noreg,
1006 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1007 return_bci);
1008 pop(return_bci);
1009 }
1012 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1013 Register bumped_count) {
1014 if (ProfileInterpreter) {
1015 Label profile_continue;
1017 // If no method data exists, go to profile_continue.
1018 // Otherwise, assign to mdp
1019 test_method_data_pointer(mdp, profile_continue);
1021 // We are taking a branch. Increment the taken count.
1022 // We inline increment_mdp_data_at to return bumped_count in a register
1023 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1024 Address data(mdp, in_bytes(JumpData::taken_offset()));
1025 movptr(bumped_count, data);
1026 assert(DataLayout::counter_increment == 1,
1027 "flow-free idiom only works with 1");
1028 addptr(bumped_count, DataLayout::counter_increment);
1029 sbbptr(bumped_count, 0);
1030 movptr(data, bumped_count); // Store back out
1032 // The method data pointer needs to be updated to reflect the new target.
1033 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1034 bind(profile_continue);
1035 }
1036 }
1039 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1040 if (ProfileInterpreter) {
1041 Label profile_continue;
1043 // If no method data exists, go to profile_continue.
1044 test_method_data_pointer(mdp, profile_continue);
1046 // We are taking a branch. Increment the not taken count.
1047 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1049 // The method data pointer needs to be updated to correspond to
1050 // the next bytecode
1051 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1052 bind(profile_continue);
1053 }
1054 }
1057 void InterpreterMacroAssembler::profile_call(Register mdp) {
1058 if (ProfileInterpreter) {
1059 Label profile_continue;
1061 // If no method data exists, go to profile_continue.
1062 test_method_data_pointer(mdp, profile_continue);
1064 // We are making a call. Increment the count.
1065 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1067 // The method data pointer needs to be updated to reflect the new target.
1068 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1069 bind(profile_continue);
1070 }
1071 }
1074 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1075 if (ProfileInterpreter) {
1076 Label profile_continue;
1078 // If no method data exists, go to profile_continue.
1079 test_method_data_pointer(mdp, profile_continue);
1081 // We are making a call. Increment the count.
1082 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1084 // The method data pointer needs to be updated to reflect the new target.
1085 update_mdp_by_constant(mdp,
1086 in_bytes(VirtualCallData::
1087 virtual_call_data_size()));
1088 bind(profile_continue);
1089 }
1090 }
1093 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1094 Register mdp,
1095 Register reg2,
1096 bool receiver_can_be_null) {
1097 if (ProfileInterpreter) {
1098 Label profile_continue;
1100 // If no method data exists, go to profile_continue.
1101 test_method_data_pointer(mdp, profile_continue);
1103 Label skip_receiver_profile;
1104 if (receiver_can_be_null) {
1105 Label not_null;
1106 testptr(receiver, receiver);
1107 jccb(Assembler::notZero, not_null);
1108 // We are making a call. Increment the count for null receiver.
1109 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1110 jmp(skip_receiver_profile);
1111 bind(not_null);
1112 }
1114 // Record the receiver type.
1115 record_klass_in_profile(receiver, mdp, reg2, true);
1116 bind(skip_receiver_profile);
1118 // The method data pointer needs to be updated to reflect the new target.
1119 update_mdp_by_constant(mdp,
1120 in_bytes(VirtualCallData::
1121 virtual_call_data_size()));
1122 bind(profile_continue);
1123 }
1124 }
1126 // This routine creates a state machine for updating the multi-row
1127 // type profile at a virtual call site (or other type-sensitive bytecode).
1128 // The machine visits each row (of receiver/count) until the receiver type
1129 // is found, or until it runs out of rows. At the same time, it remembers
1130 // the location of the first empty row. (An empty row records null for its
1131 // receiver, and can be allocated for a newly-observed receiver type.)
1132 // Because there are two degrees of freedom in the state, a simple linear
1133 // search will not work; it must be a decision tree. Hence this helper
1134 // function is recursive, to generate the required tree structured code.
1135 // It's the interpreter, so we are trading off code space for speed.
1136 // See below for example code.
1137 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1138 Register receiver, Register mdp,
1139 Register reg2, int start_row,
1140 Label& done, bool is_virtual_call) {
1141 if (TypeProfileWidth == 0) {
1142 if (is_virtual_call) {
1143 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1144 }
1145 return;
1146 }
1148 int last_row = VirtualCallData::row_limit() - 1;
1149 assert(start_row <= last_row, "must be work left to do");
1150 // Test this row for both the receiver and for null.
1151 // Take any of three different outcomes:
1152 // 1. found receiver => increment count and goto done
1153 // 2. found null => keep looking for case 1, maybe allocate this cell
1154 // 3. found something else => keep looking for cases 1 and 2
1155 // Case 3 is handled by a recursive call.
1156 for (int row = start_row; row <= last_row; row++) {
1157 Label next_test;
1158 bool test_for_null_also = (row == start_row);
1160 // See if the receiver is receiver[n].
1161 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1162 test_mdp_data_at(mdp, recvr_offset, receiver,
1163 (test_for_null_also ? reg2 : noreg),
1164 next_test);
1165 // (Reg2 now contains the receiver from the CallData.)
1167 // The receiver is receiver[n]. Increment count[n].
1168 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1169 increment_mdp_data_at(mdp, count_offset);
1170 jmp(done);
1171 bind(next_test);
1173 if (test_for_null_also) {
1174 Label found_null;
1175 // Failed the equality check on receiver[n]... Test for null.
1176 testptr(reg2, reg2);
1177 if (start_row == last_row) {
1178 // The only thing left to do is handle the null case.
1179 if (is_virtual_call) {
1180 jccb(Assembler::zero, found_null);
1181 // Receiver did not match any saved receiver and there is no empty row for it.
1182 // Increment total counter to indicate polymorphic case.
1183 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1184 jmp(done);
1185 bind(found_null);
1186 } else {
1187 jcc(Assembler::notZero, done);
1188 }
1189 break;
1190 }
1191 // Since null is rare, make it be the branch-taken case.
1192 jcc(Assembler::zero, found_null);
1194 // Put all the "Case 3" tests here.
1195 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1197 // Found a null. Keep searching for a matching receiver,
1198 // but remember that this is an empty (unused) slot.
1199 bind(found_null);
1200 }
1201 }
1203 // In the fall-through case, we found no matching receiver, but we
1204 // observed the receiver[start_row] is NULL.
1206 // Fill in the receiver field and increment the count.
1207 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1208 set_mdp_data_at(mdp, recvr_offset, receiver);
1209 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1210 movl(reg2, DataLayout::counter_increment);
1211 set_mdp_data_at(mdp, count_offset, reg2);
1212 if (start_row > 0) {
1213 jmp(done);
1214 }
1215 }
1217 // Example state machine code for three profile rows:
1218 // // main copy of decision tree, rooted at row[1]
1219 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1220 // if (row[0].rec != NULL) {
1221 // // inner copy of decision tree, rooted at row[1]
1222 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1223 // if (row[1].rec != NULL) {
1224 // // degenerate decision tree, rooted at row[2]
1225 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1226 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1227 // row[2].init(rec); goto done;
1228 // } else {
1229 // // remember row[1] is empty
1230 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1231 // row[1].init(rec); goto done;
1232 // }
1233 // } else {
1234 // // remember row[0] is empty
1235 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1236 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1237 // row[0].init(rec); goto done;
1238 // }
1239 // done:
1241 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1242 Register mdp, Register reg2,
1243 bool is_virtual_call) {
1244 assert(ProfileInterpreter, "must be profiling");
1245 Label done;
1247 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1249 bind (done);
1250 }
1252 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1253 Register mdp) {
1254 if (ProfileInterpreter) {
1255 Label profile_continue;
1256 uint row;
1258 // If no method data exists, go to profile_continue.
1259 test_method_data_pointer(mdp, profile_continue);
1261 // Update the total ret count.
1262 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1264 for (row = 0; row < RetData::row_limit(); row++) {
1265 Label next_test;
1267 // See if return_bci is equal to bci[n]:
1268 test_mdp_data_at(mdp,
1269 in_bytes(RetData::bci_offset(row)),
1270 return_bci, noreg,
1271 next_test);
1273 // return_bci is equal to bci[n]. Increment the count.
1274 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1276 // The method data pointer needs to be updated to reflect the new target.
1277 update_mdp_by_offset(mdp,
1278 in_bytes(RetData::bci_displacement_offset(row)));
1279 jmp(profile_continue);
1280 bind(next_test);
1281 }
1283 update_mdp_for_ret(return_bci);
1285 bind(profile_continue);
1286 }
1287 }
1290 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1291 if (ProfileInterpreter) {
1292 Label profile_continue;
1294 // If no method data exists, go to profile_continue.
1295 test_method_data_pointer(mdp, profile_continue);
1297 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1299 // The method data pointer needs to be updated.
1300 int mdp_delta = in_bytes(BitData::bit_data_size());
1301 if (TypeProfileCasts) {
1302 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1303 }
1304 update_mdp_by_constant(mdp, mdp_delta);
1306 bind(profile_continue);
1307 }
1308 }
1311 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1312 if (ProfileInterpreter && TypeProfileCasts) {
1313 Label profile_continue;
1315 // If no method data exists, go to profile_continue.
1316 test_method_data_pointer(mdp, profile_continue);
1318 int count_offset = in_bytes(CounterData::count_offset());
1319 // Back up the address, since we have already bumped the mdp.
1320 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1322 // *Decrement* the counter. We expect to see zero or small negatives.
1323 increment_mdp_data_at(mdp, count_offset, true);
1325 bind (profile_continue);
1326 }
1327 }
1330 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1331 if (ProfileInterpreter) {
1332 Label profile_continue;
1334 // If no method data exists, go to profile_continue.
1335 test_method_data_pointer(mdp, profile_continue);
1337 // The method data pointer needs to be updated.
1338 int mdp_delta = in_bytes(BitData::bit_data_size());
1339 if (TypeProfileCasts) {
1340 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1342 // Record the object type.
1343 record_klass_in_profile(klass, mdp, reg2, false);
1344 }
1345 update_mdp_by_constant(mdp, mdp_delta);
1347 bind(profile_continue);
1348 }
1349 }
1352 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1353 if (ProfileInterpreter) {
1354 Label profile_continue;
1356 // If no method data exists, go to profile_continue.
1357 test_method_data_pointer(mdp, profile_continue);
1359 // Update the default case count
1360 increment_mdp_data_at(mdp,
1361 in_bytes(MultiBranchData::default_count_offset()));
1363 // The method data pointer needs to be updated.
1364 update_mdp_by_offset(mdp,
1365 in_bytes(MultiBranchData::
1366 default_displacement_offset()));
1368 bind(profile_continue);
1369 }
1370 }
1373 void InterpreterMacroAssembler::profile_switch_case(Register index,
1374 Register mdp,
1375 Register reg2) {
1376 if (ProfileInterpreter) {
1377 Label profile_continue;
1379 // If no method data exists, go to profile_continue.
1380 test_method_data_pointer(mdp, profile_continue);
1382 // Build the base (index * per_case_size_in_bytes()) +
1383 // case_array_offset_in_bytes()
1384 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1385 imulptr(index, reg2); // XXX l ?
1386 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1388 // Update the case count
1389 increment_mdp_data_at(mdp,
1390 index,
1391 in_bytes(MultiBranchData::relative_count_offset()));
1393 // The method data pointer needs to be updated.
1394 update_mdp_by_offset(mdp,
1395 index,
1396 in_bytes(MultiBranchData::
1397 relative_displacement_offset()));
1399 bind(profile_continue);
1400 }
1401 }
1405 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1406 if (state == atos) {
1407 MacroAssembler::verify_oop(reg);
1408 }
1409 }
1411 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1412 }
1413 #endif // !CC_INTERP
1416 void InterpreterMacroAssembler::notify_method_entry() {
1417 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1418 // track stack depth. If it is possible to enter interp_only_mode we add
1419 // the code to check if the event should be sent.
1420 if (JvmtiExport::can_post_interpreter_events()) {
1421 Label L;
1422 movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1423 testl(rdx, rdx);
1424 jcc(Assembler::zero, L);
1425 call_VM(noreg, CAST_FROM_FN_PTR(address,
1426 InterpreterRuntime::post_method_entry));
1427 bind(L);
1428 }
1430 {
1431 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1432 get_method(c_rarg1);
1433 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1434 r15_thread, c_rarg1);
1435 }
1437 // RedefineClasses() tracing support for obsolete method entry
1438 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1439 get_method(c_rarg1);
1440 call_VM_leaf(
1441 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1442 r15_thread, c_rarg1);
1443 }
1444 }
1447 void InterpreterMacroAssembler::notify_method_exit(
1448 TosState state, NotifyMethodExitMode mode) {
1449 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1450 // track stack depth. If it is possible to enter interp_only_mode we add
1451 // the code to check if the event should be sent.
1452 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1453 Label L;
1454 // Note: frame::interpreter_frame_result has a dependency on how the
1455 // method result is saved across the call to post_method_exit. If this
1456 // is changed then the interpreter_frame_result implementation will
1457 // need to be updated too.
1459 // For c++ interpreter the result is always stored at a known location in the frame
1460 // template interpreter will leave it on the top of the stack.
1461 NOT_CC_INTERP(push(state);)
1462 movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1463 testl(rdx, rdx);
1464 jcc(Assembler::zero, L);
1465 call_VM(noreg,
1466 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1467 bind(L);
1468 NOT_CC_INTERP(pop(state));
1469 }
1471 {
1472 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1473 NOT_CC_INTERP(push(state));
1474 get_method(c_rarg1);
1475 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1476 r15_thread, c_rarg1);
1477 NOT_CC_INTERP(pop(state));
1478 }
1479 }
1481 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1482 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1483 int increment, int mask,
1484 Register scratch, bool preloaded,
1485 Condition cond, Label* where) {
1486 if (!preloaded) {
1487 movl(scratch, counter_addr);
1488 }
1489 incrementl(scratch, increment);
1490 movl(counter_addr, scratch);
1491 andl(scratch, mask);
1492 jcc(cond, *where);
1493 }