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) 1997, 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_32.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
43 #ifdef CC_INTERP
44 void InterpreterMacroAssembler::get_method(Register reg) {
45 movptr(reg, Address(rbp, -(sizeof(BytecodeInterpreter) + 2 * wordSize)));
46 movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
47 }
48 #endif // CC_INTERP
51 #ifndef CC_INTERP
52 void InterpreterMacroAssembler::call_VM_leaf_base(
53 address entry_point,
54 int number_of_arguments
55 ) {
56 // interpreter specific
57 //
58 // Note: No need to save/restore bcp & locals (rsi & rdi) pointer
59 // since these are callee saved registers and no blocking/
60 // GC can happen in leaf calls.
61 // Further Note: DO NOT save/restore bcp/locals. If a caller has
62 // already saved them so that it can use rsi/rdi as temporaries
63 // then a save/restore here will DESTROY the copy the caller
64 // saved! There used to be a save_bcp() that only happened in
65 // the ASSERT path (no restore_bcp). Which caused bizarre failures
66 // when jvm built with ASSERTs.
67 #ifdef ASSERT
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: last_sp != NULL");
72 bind(L);
73 }
74 #endif
75 // super call
76 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
77 // interpreter specific
79 // Used to ASSERT that rsi/rdi were equal to frame's bcp/locals
80 // but since they may not have been saved (and we don't want to
81 // save them here (see note above) the assert is invalid.
82 }
85 void InterpreterMacroAssembler::call_VM_base(
86 Register oop_result,
87 Register java_thread,
88 Register last_java_sp,
89 address entry_point,
90 int number_of_arguments,
91 bool check_exceptions
92 ) {
93 #ifdef ASSERT
94 { Label L;
95 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
96 jcc(Assembler::equal, L);
97 stop("InterpreterMacroAssembler::call_VM_base: last_sp != NULL");
98 bind(L);
99 }
100 #endif /* ASSERT */
101 // interpreter specific
102 //
103 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
104 // really make a difference for these runtime calls, since they are
105 // slow anyway. Btw., bcp must be saved/restored since it may change
106 // due to GC.
107 assert(java_thread == noreg , "not expecting a precomputed java thread");
108 save_bcp();
109 // super call
110 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
111 // interpreter specific
112 restore_bcp();
113 restore_locals();
114 }
117 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
118 if (JvmtiExport::can_pop_frame()) {
119 Label L;
120 // Initiate popframe handling only if it is not already being processed. If the flag
121 // has the popframe_processing bit set, it means that this code is called *during* popframe
122 // handling - we don't want to reenter.
123 Register pop_cond = java_thread; // Not clear if any other register is available...
124 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
125 testl(pop_cond, JavaThread::popframe_pending_bit);
126 jcc(Assembler::zero, L);
127 testl(pop_cond, JavaThread::popframe_processing_bit);
128 jcc(Assembler::notZero, L);
129 // Call Interpreter::remove_activation_preserving_args_entry() to get the
130 // address of the same-named entrypoint in the generated interpreter code.
131 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
132 jmp(rax);
133 bind(L);
134 get_thread(java_thread);
135 }
136 }
139 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
140 get_thread(rcx);
141 movl(rcx, Address(rcx, JavaThread::jvmti_thread_state_offset()));
142 const Address tos_addr (rcx, JvmtiThreadState::earlyret_tos_offset());
143 const Address oop_addr (rcx, JvmtiThreadState::earlyret_oop_offset());
144 const Address val_addr (rcx, JvmtiThreadState::earlyret_value_offset());
145 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
146 + in_ByteSize(wordSize));
147 switch (state) {
148 case atos: movptr(rax, oop_addr);
149 movptr(oop_addr, NULL_WORD);
150 verify_oop(rax, state); break;
151 case ltos:
152 movl(rdx, val_addr1); // fall through
153 case btos: // fall through
154 case ctos: // fall through
155 case stos: // fall through
156 case itos: movl(rax, val_addr); break;
157 case ftos: fld_s(val_addr); break;
158 case dtos: fld_d(val_addr); break;
159 case vtos: /* nothing to do */ break;
160 default : ShouldNotReachHere();
161 }
162 // Clean up tos value in the thread object
163 movl(tos_addr, (int32_t) ilgl);
164 movptr(val_addr, NULL_WORD);
165 NOT_LP64(movptr(val_addr1, NULL_WORD));
166 }
169 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
170 if (JvmtiExport::can_force_early_return()) {
171 Label L;
172 Register tmp = java_thread;
173 movptr(tmp, Address(tmp, JavaThread::jvmti_thread_state_offset()));
174 testptr(tmp, tmp);
175 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
177 // Initiate earlyret handling only if it is not already being processed.
178 // If the flag has the earlyret_processing bit set, it means that this code
179 // is called *during* earlyret handling - we don't want to reenter.
180 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
181 cmpl(tmp, JvmtiThreadState::earlyret_pending);
182 jcc(Assembler::notEqual, L);
184 // Call Interpreter::remove_activation_early_entry() to get the address of the
185 // same-named entrypoint in the generated interpreter code.
186 get_thread(java_thread);
187 movptr(tmp, Address(java_thread, JavaThread::jvmti_thread_state_offset()));
188 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
189 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
190 jmp(rax);
191 bind(L);
192 get_thread(java_thread);
193 }
194 }
197 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
198 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
199 movl(reg, Address(rsi, bcp_offset));
200 bswapl(reg);
201 shrl(reg, 16);
202 }
205 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register reg, int bcp_offset, size_t index_size) {
206 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
207 if (index_size == sizeof(u2)) {
208 load_unsigned_short(reg, Address(rsi, bcp_offset));
209 } else if (index_size == sizeof(u4)) {
210 assert(EnableInvokeDynamic, "giant index used only for JSR 292");
211 movl(reg, Address(rsi, bcp_offset));
212 // Check if the secondary index definition is still ~x, otherwise
213 // we have to change the following assembler code to calculate the
214 // plain index.
215 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
216 notl(reg); // convert to plain index
217 } else if (index_size == sizeof(u1)) {
218 load_unsigned_byte(reg, Address(rsi, bcp_offset));
219 } else {
220 ShouldNotReachHere();
221 }
222 }
225 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register index,
226 int bcp_offset, 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 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
232 shlptr(index, 2); // convert from field index to ConstantPoolCacheEntry index
233 }
236 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
237 Register index,
238 Register bytecode,
239 int byte_no,
240 int bcp_offset,
241 size_t index_size) {
242 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
243 movptr(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
244 const int shift_count = (1 + byte_no) * BitsPerByte;
245 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
246 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
247 "correct shift count");
248 shrptr(bytecode, shift_count);
249 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
250 andptr(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
251 }
254 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
255 int bcp_offset, size_t index_size) {
256 assert(cache != tmp, "must use different register");
257 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
258 assert(sizeof(ConstantPoolCacheEntry) == 4*wordSize, "adjust code below");
259 // convert from field index to ConstantPoolCacheEntry index
260 // and from word offset to byte offset
261 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
262 shll(tmp, 2 + LogBytesPerWord);
263 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
264 // skip past the header
265 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
266 addptr(cache, tmp); // construct pointer to cache entry
267 }
269 // Load object from cpool->resolved_references(index)
270 void InterpreterMacroAssembler::load_resolved_reference_at_index(
271 Register result, Register index) {
272 assert_different_registers(result, index);
273 // convert from field index to resolved_references() index and from
274 // word index to byte offset. Since this is a java object, it can be compressed
275 Register tmp = index; // reuse
276 shll(tmp, LogBytesPerHeapOop);
278 get_constant_pool(result);
279 // load pointer for resolved_references[] objArray
280 movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
281 // JNIHandles::resolve(obj);
282 movptr(result, Address(result, 0));
283 // Add in the index
284 addptr(result, tmp);
285 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
286 }
288 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
289 // a subtype of super_klass. EAX holds the super_klass. Blows ECX.
290 // Resets EDI to locals. Register sub_klass cannot be any of the above.
291 void InterpreterMacroAssembler::gen_subtype_check( Register Rsub_klass, Label &ok_is_subtype ) {
292 assert( Rsub_klass != rax, "rax, holds superklass" );
293 assert( Rsub_klass != rcx, "used as a temp" );
294 assert( Rsub_klass != rdi, "used as a temp, restored from locals" );
296 // Profile the not-null value's klass.
297 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
299 // Do the check.
300 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
302 // Profile the failure of the check.
303 profile_typecheck_failed(rcx); // blows rcx
304 }
306 void InterpreterMacroAssembler::f2ieee() {
307 if (IEEEPrecision) {
308 fstp_s(Address(rsp, 0));
309 fld_s(Address(rsp, 0));
310 }
311 }
314 void InterpreterMacroAssembler::d2ieee() {
315 if (IEEEPrecision) {
316 fstp_d(Address(rsp, 0));
317 fld_d(Address(rsp, 0));
318 }
319 }
321 // Java Expression Stack
323 void InterpreterMacroAssembler::pop_ptr(Register r) {
324 pop(r);
325 }
327 void InterpreterMacroAssembler::pop_i(Register r) {
328 pop(r);
329 }
331 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
332 pop(lo);
333 pop(hi);
334 }
336 void InterpreterMacroAssembler::pop_f() {
337 fld_s(Address(rsp, 0));
338 addptr(rsp, 1 * wordSize);
339 }
341 void InterpreterMacroAssembler::pop_d() {
342 fld_d(Address(rsp, 0));
343 addptr(rsp, 2 * wordSize);
344 }
347 void InterpreterMacroAssembler::pop(TosState state) {
348 switch (state) {
349 case atos: pop_ptr(rax); break;
350 case btos: // fall through
351 case ctos: // fall through
352 case stos: // fall through
353 case itos: pop_i(rax); break;
354 case ltos: pop_l(rax, rdx); break;
355 case ftos: pop_f(); break;
356 case dtos: pop_d(); break;
357 case vtos: /* nothing to do */ break;
358 default : ShouldNotReachHere();
359 }
360 verify_oop(rax, state);
361 }
363 void InterpreterMacroAssembler::push_ptr(Register r) {
364 push(r);
365 }
367 void InterpreterMacroAssembler::push_i(Register r) {
368 push(r);
369 }
371 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
372 push(hi);
373 push(lo);
374 }
376 void InterpreterMacroAssembler::push_f() {
377 // Do not schedule for no AGI! Never write beyond rsp!
378 subptr(rsp, 1 * wordSize);
379 fstp_s(Address(rsp, 0));
380 }
382 void InterpreterMacroAssembler::push_d(Register r) {
383 // Do not schedule for no AGI! Never write beyond rsp!
384 subptr(rsp, 2 * wordSize);
385 fstp_d(Address(rsp, 0));
386 }
389 void InterpreterMacroAssembler::push(TosState state) {
390 verify_oop(rax, state);
391 switch (state) {
392 case atos: push_ptr(rax); break;
393 case btos: // fall through
394 case ctos: // fall through
395 case stos: // fall through
396 case itos: push_i(rax); break;
397 case ltos: push_l(rax, rdx); break;
398 case ftos: push_f(); break;
399 case dtos: push_d(rax); break;
400 case vtos: /* nothing to do */ break;
401 default : ShouldNotReachHere();
402 }
403 }
406 // Helpers for swap and dup
407 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
408 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
409 }
411 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
412 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
413 }
415 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
416 // set sender sp
417 lea(rsi, Address(rsp, wordSize));
418 // record last_sp
419 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rsi);
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 get_thread(temp);
434 // interp_only is an int, on little endian it is sufficient to test the byte only
435 // Is a cmpl faster?
436 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
437 jccb(Assembler::zero, run_compiled_code);
438 jmp(Address(method, Method::interpreter_entry_offset()));
439 bind(run_compiled_code);
440 }
442 jmp(Address(method, Method::from_interpreted_offset()));
444 }
447 // The following two routines provide a hook so that an implementation
448 // can schedule the dispatch in two parts. Intel does not do this.
449 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
450 // Nothing Intel-specific to be done here.
451 }
453 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
454 dispatch_next(state, step);
455 }
457 void InterpreterMacroAssembler::dispatch_base(TosState state, 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 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
465 cmpptr(rcx, min_frame_size);
466 jcc(Assembler::greaterEqual, L);
467 stop("broken stack frame");
468 bind(L);
469 }
470 if (verifyoop) verify_oop(rax, state);
471 Address index(noreg, rbx, Address::times_ptr);
472 ExternalAddress tbl((address)table);
473 ArrayAddress dispatch(tbl, index);
474 jump(dispatch);
475 }
478 void InterpreterMacroAssembler::dispatch_only(TosState state) {
479 dispatch_base(state, Interpreter::dispatch_table(state));
480 }
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 rsi to prevent AGI)
494 load_unsigned_byte(rbx, Address(rsi, step));
495 // advance rsi
496 increment(rsi, step);
497 dispatch_base(state, Interpreter::dispatch_table(state));
498 }
501 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
502 // load current bytecode
503 load_unsigned_byte(rbx, Address(rsi, 0));
504 dispatch_base(state, table);
505 }
507 // remove activation
508 //
509 // Unlock the receiver if this is a synchronized method.
510 // Unlock any Java monitors from syncronized blocks.
511 // Remove the activation from the stack.
512 //
513 // If there are locked Java monitors
514 // If throw_monitor_exception
515 // throws IllegalMonitorStateException
516 // Else if install_monitor_exception
517 // installs IllegalMonitorStateException
518 // Else
519 // no error processing
520 void InterpreterMacroAssembler::remove_activation(TosState state, Register ret_addr,
521 bool throw_monitor_exception,
522 bool install_monitor_exception,
523 bool notify_jvmdi) {
524 // Note: Registers rax, rdx and FPU ST(0) may be in use for the result
525 // check if synchronized method
526 Label unlocked, unlock, no_unlock;
528 get_thread(rcx);
529 const Address do_not_unlock_if_synchronized(rcx,
530 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
532 movbool(rbx, do_not_unlock_if_synchronized);
533 mov(rdi,rbx);
534 movbool(do_not_unlock_if_synchronized, false); // reset the flag
536 movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); // get method access flags
537 movl(rcx, Address(rbx, Method::access_flags_offset()));
539 testl(rcx, JVM_ACC_SYNCHRONIZED);
540 jcc(Assembler::zero, unlocked);
542 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
543 // is set.
544 mov(rcx,rdi);
545 testbool(rcx);
546 jcc(Assembler::notZero, no_unlock);
548 // unlock monitor
549 push(state); // save result
551 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
552 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
553 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
554 lea (rdx, monitor); // address of first monitor
556 movptr (rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
557 testptr(rax, rax);
558 jcc (Assembler::notZero, unlock);
560 pop(state);
561 if (throw_monitor_exception) {
562 empty_FPU_stack(); // remove possible return value from FPU-stack, otherwise stack could overflow
564 // Entry already unlocked, need to throw exception
565 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
566 should_not_reach_here();
567 } else {
568 // Monitor already unlocked during a stack unroll.
569 // If requested, install an illegal_monitor_state_exception.
570 // Continue with stack unrolling.
571 if (install_monitor_exception) {
572 empty_FPU_stack(); // remove possible return value from FPU-stack, otherwise stack could overflow
573 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
574 }
575 jmp(unlocked);
576 }
578 bind(unlock);
579 unlock_object(rdx);
580 pop(state);
582 // Check that for block-structured locking (i.e., that all locked objects has been unlocked)
583 bind(unlocked);
585 // rax, rdx: Might contain return value
587 // Check that all monitors are unlocked
588 {
589 Label loop, exception, entry, restart;
590 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
591 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
592 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
594 bind(restart);
595 movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry
596 lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
597 jmp(entry);
599 // Entry already locked, need to throw exception
600 bind(exception);
602 if (throw_monitor_exception) {
603 empty_FPU_stack(); // remove possible return value from FPU-stack, otherwise stack could overflow
605 // Throw exception
606 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
607 should_not_reach_here();
608 } else {
609 // Stack unrolling. Unlock object and install illegal_monitor_exception
610 // Unlock does not block, so don't have to worry about the frame
612 push(state);
613 mov(rdx, rcx);
614 unlock_object(rdx);
615 pop(state);
617 if (install_monitor_exception) {
618 empty_FPU_stack(); // remove possible return value from FPU-stack, otherwise stack could overflow
619 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
620 }
622 jmp(restart);
623 }
625 bind(loop);
626 cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used
627 jcc(Assembler::notEqual, exception);
629 addptr(rcx, entry_size); // otherwise advance to next entry
630 bind(entry);
631 cmpptr(rcx, rbx); // check if bottom reached
632 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
633 }
635 bind(no_unlock);
637 // jvmti support
638 if (notify_jvmdi) {
639 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
640 } else {
641 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
642 }
644 // remove activation
645 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
646 leave(); // remove frame anchor
647 pop(ret_addr); // get return address
648 mov(rsp, rbx); // set sp to sender sp
649 if (UseSSE) {
650 // float and double are returned in xmm register in SSE-mode
651 if (state == ftos && UseSSE >= 1) {
652 subptr(rsp, wordSize);
653 fstp_s(Address(rsp, 0));
654 movflt(xmm0, Address(rsp, 0));
655 addptr(rsp, wordSize);
656 } else if (state == dtos && UseSSE >= 2) {
657 subptr(rsp, 2*wordSize);
658 fstp_d(Address(rsp, 0));
659 movdbl(xmm0, Address(rsp, 0));
660 addptr(rsp, 2*wordSize);
661 }
662 }
663 }
665 #endif /* !CC_INTERP */
668 // Lock object
669 //
670 // Argument: rdx : Points to BasicObjectLock to be used for locking. Must
671 // be initialized with object to lock
672 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
673 assert(lock_reg == rdx, "The argument is only for looks. It must be rdx");
675 if (UseHeavyMonitors) {
676 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
677 } else {
679 Label done;
681 const Register swap_reg = rax; // Must use rax, for cmpxchg instruction
682 const Register obj_reg = rcx; // Will contain the oop
684 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
685 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
686 const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes();
688 Label slow_case;
690 // Load object pointer into obj_reg %rcx
691 movptr(obj_reg, Address(lock_reg, obj_offset));
693 if (UseBiasedLocking) {
694 // Note: we use noreg for the temporary register since it's hard
695 // to come up with a free register on all incoming code paths
696 biased_locking_enter(lock_reg, obj_reg, swap_reg, noreg, false, done, &slow_case);
697 }
699 // Load immediate 1 into swap_reg %rax,
700 movptr(swap_reg, (int32_t)1);
702 // Load (object->mark() | 1) into swap_reg %rax,
703 orptr(swap_reg, Address(obj_reg, 0));
705 // Save (object->mark() | 1) into BasicLock's displaced header
706 movptr(Address(lock_reg, mark_offset), swap_reg);
708 assert(lock_offset == 0, "displached header must be first word in BasicObjectLock");
709 if (os::is_MP()) {
710 lock();
711 }
712 cmpxchgptr(lock_reg, Address(obj_reg, 0));
713 if (PrintBiasedLockingStatistics) {
714 cond_inc32(Assembler::zero,
715 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
716 }
717 jcc(Assembler::zero, done);
719 // Test if the oopMark is an obvious stack pointer, i.e.,
720 // 1) (mark & 3) == 0, and
721 // 2) rsp <= mark < mark + os::pagesize()
722 //
723 // These 3 tests can be done by evaluating the following
724 // expression: ((mark - rsp) & (3 - os::vm_page_size())),
725 // assuming both stack pointer and pagesize have their
726 // least significant 2 bits clear.
727 // NOTE: the oopMark is in swap_reg %rax, as the result of cmpxchg
728 subptr(swap_reg, rsp);
729 andptr(swap_reg, 3 - os::vm_page_size());
731 // Save the test result, for recursive case, the result is zero
732 movptr(Address(lock_reg, mark_offset), swap_reg);
734 if (PrintBiasedLockingStatistics) {
735 cond_inc32(Assembler::zero,
736 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
737 }
738 jcc(Assembler::zero, done);
740 bind(slow_case);
742 // Call the runtime routine for slow case
743 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
745 bind(done);
746 }
747 }
750 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
751 //
752 // Argument: rdx : Points to BasicObjectLock structure for lock
753 // Throw an IllegalMonitorException if object is not locked by current thread
754 //
755 // Uses: rax, rbx, rcx, rdx
756 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
757 assert(lock_reg == rdx, "The argument is only for looks. It must be rdx");
759 if (UseHeavyMonitors) {
760 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
761 } else {
762 Label done;
764 const Register swap_reg = rax; // Must use rax, for cmpxchg instruction
765 const Register header_reg = rbx; // Will contain the old oopMark
766 const Register obj_reg = rcx; // Will contain the oop
768 save_bcp(); // Save in case of exception
770 // Convert from BasicObjectLock structure to object and BasicLock structure
771 // Store the BasicLock address into %rax,
772 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
774 // Load oop into obj_reg(%rcx)
775 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes ()));
777 // Free entry
778 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD);
780 if (UseBiasedLocking) {
781 biased_locking_exit(obj_reg, header_reg, done);
782 }
784 // Load the old header from BasicLock structure
785 movptr(header_reg, Address(swap_reg, BasicLock::displaced_header_offset_in_bytes()));
787 // Test for recursion
788 testptr(header_reg, header_reg);
790 // zero for recursive case
791 jcc(Assembler::zero, done);
793 // Atomic swap back the old header
794 if (os::is_MP()) lock();
795 cmpxchgptr(header_reg, Address(obj_reg, 0));
797 // zero for recursive case
798 jcc(Assembler::zero, done);
800 // Call the runtime routine for slow case.
801 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), obj_reg); // restore obj
802 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
804 bind(done);
806 restore_bcp();
807 }
808 }
811 #ifndef CC_INTERP
813 // Test ImethodDataPtr. If it is null, continue at the specified label
814 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
815 assert(ProfileInterpreter, "must be profiling interpreter");
816 movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
817 testptr(mdp, mdp);
818 jcc(Assembler::zero, zero_continue);
819 }
822 // Set the method data pointer for the current bcp.
823 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
824 assert(ProfileInterpreter, "must be profiling interpreter");
825 Label set_mdp;
826 push(rax);
827 push(rbx);
829 get_method(rbx);
830 // Test MDO to avoid the call if it is NULL.
831 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
832 testptr(rax, rax);
833 jcc(Assembler::zero, set_mdp);
834 // rbx,: method
835 // rsi: bcp
836 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, rsi);
837 // rax,: mdi
838 // mdo is guaranteed to be non-zero here, we checked for it before the call.
839 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
840 addptr(rbx, in_bytes(MethodData::data_offset()));
841 addptr(rax, rbx);
842 bind(set_mdp);
843 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
844 pop(rbx);
845 pop(rax);
846 }
848 void InterpreterMacroAssembler::verify_method_data_pointer() {
849 assert(ProfileInterpreter, "must be profiling interpreter");
850 #ifdef ASSERT
851 Label verify_continue;
852 push(rax);
853 push(rbx);
854 push(rcx);
855 push(rdx);
856 test_method_data_pointer(rcx, verify_continue); // If mdp is zero, continue
857 get_method(rbx);
859 // If the mdp is valid, it will point to a DataLayout header which is
860 // consistent with the bcp. The converse is highly probable also.
861 load_unsigned_short(rdx, Address(rcx, in_bytes(DataLayout::bci_offset())));
862 addptr(rdx, Address(rbx, Method::const_offset()));
863 lea(rdx, Address(rdx, ConstMethod::codes_offset()));
864 cmpptr(rdx, rsi);
865 jcc(Assembler::equal, verify_continue);
866 // rbx,: method
867 // rsi: bcp
868 // rcx: mdp
869 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), rbx, rsi, rcx);
870 bind(verify_continue);
871 pop(rdx);
872 pop(rcx);
873 pop(rbx);
874 pop(rax);
875 #endif // ASSERT
876 }
879 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) {
880 // %%% this seems to be used to store counter data which is surely 32bits
881 // however 64bit side stores 64 bits which seems wrong
882 assert(ProfileInterpreter, "must be profiling interpreter");
883 Address data(mdp_in, constant);
884 movptr(data, value);
885 }
888 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
889 int constant,
890 bool decrement) {
891 // Counter address
892 Address data(mdp_in, constant);
894 increment_mdp_data_at(data, decrement);
895 }
898 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
899 bool decrement) {
901 assert( DataLayout::counter_increment==1, "flow-free idiom only works with 1" );
902 assert(ProfileInterpreter, "must be profiling interpreter");
904 // %%% 64bit treats this as 64 bit which seems unlikely
905 if (decrement) {
906 // Decrement the register. Set condition codes.
907 addl(data, -DataLayout::counter_increment);
908 // If the decrement causes the counter to overflow, stay negative
909 Label L;
910 jcc(Assembler::negative, L);
911 addl(data, DataLayout::counter_increment);
912 bind(L);
913 } else {
914 assert(DataLayout::counter_increment == 1,
915 "flow-free idiom only works with 1");
916 // Increment the register. Set carry flag.
917 addl(data, DataLayout::counter_increment);
918 // If the increment causes the counter to overflow, pull back by 1.
919 sbbl(data, 0);
920 }
921 }
924 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
925 Register reg,
926 int constant,
927 bool decrement) {
928 Address data(mdp_in, reg, Address::times_1, constant);
930 increment_mdp_data_at(data, decrement);
931 }
934 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, int flag_byte_constant) {
935 assert(ProfileInterpreter, "must be profiling interpreter");
936 int header_offset = in_bytes(DataLayout::header_offset());
937 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
938 // Set the flag
939 orl(Address(mdp_in, header_offset), header_bits);
940 }
944 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
945 int offset,
946 Register value,
947 Register test_value_out,
948 Label& not_equal_continue) {
949 assert(ProfileInterpreter, "must be profiling interpreter");
950 if (test_value_out == noreg) {
951 cmpptr(value, Address(mdp_in, offset));
952 } else {
953 // Put the test value into a register, so caller can use it:
954 movptr(test_value_out, Address(mdp_in, offset));
955 cmpptr(test_value_out, value);
956 }
957 jcc(Assembler::notEqual, not_equal_continue);
958 }
961 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) {
962 assert(ProfileInterpreter, "must be profiling interpreter");
963 Address disp_address(mdp_in, offset_of_disp);
964 addptr(mdp_in,disp_address);
965 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
966 }
969 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, Register reg, int offset_of_disp) {
970 assert(ProfileInterpreter, "must be profiling interpreter");
971 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
972 addptr(mdp_in, disp_address);
973 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
974 }
977 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
978 assert(ProfileInterpreter, "must be profiling interpreter");
979 addptr(mdp_in, constant);
980 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
981 }
984 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
985 assert(ProfileInterpreter, "must be profiling interpreter");
986 push(return_bci); // save/restore across call_VM
987 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
988 pop(return_bci);
989 }
992 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
993 if (ProfileInterpreter) {
994 Label profile_continue;
996 // If no method data exists, go to profile_continue.
997 // Otherwise, assign to mdp
998 test_method_data_pointer(mdp, profile_continue);
1000 // We are taking a branch. Increment the taken count.
1001 // We inline increment_mdp_data_at to return bumped_count in a register
1002 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1003 Address data(mdp, in_bytes(JumpData::taken_offset()));
1005 // %%% 64bit treats these cells as 64 bit but they seem to be 32 bit
1006 movl(bumped_count,data);
1007 assert( DataLayout::counter_increment==1, "flow-free idiom only works with 1" );
1008 addl(bumped_count, DataLayout::counter_increment);
1009 sbbl(bumped_count, 0);
1010 movl(data,bumped_count); // Store back out
1012 // The method data pointer needs to be updated to reflect the new target.
1013 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1014 bind (profile_continue);
1015 }
1016 }
1019 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1020 if (ProfileInterpreter) {
1021 Label profile_continue;
1023 // If no method data exists, go to profile_continue.
1024 test_method_data_pointer(mdp, profile_continue);
1026 // We are taking a branch. Increment the not taken count.
1027 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1029 // The method data pointer needs to be updated to correspond to the next bytecode
1030 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1031 bind (profile_continue);
1032 }
1033 }
1036 void InterpreterMacroAssembler::profile_call(Register mdp) {
1037 if (ProfileInterpreter) {
1038 Label profile_continue;
1040 // If no method data exists, go to profile_continue.
1041 test_method_data_pointer(mdp, profile_continue);
1043 // We are making a call. Increment the count.
1044 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1046 // The method data pointer needs to be updated to reflect the new target.
1047 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1048 bind (profile_continue);
1049 }
1050 }
1053 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1054 if (ProfileInterpreter) {
1055 Label profile_continue;
1057 // If no method data exists, go to profile_continue.
1058 test_method_data_pointer(mdp, profile_continue);
1060 // We are making a call. Increment the count.
1061 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1063 // The method data pointer needs to be updated to reflect the new target.
1064 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1065 bind (profile_continue);
1066 }
1067 }
1070 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, Register mdp,
1071 Register reg2,
1072 bool receiver_can_be_null) {
1073 if (ProfileInterpreter) {
1074 Label profile_continue;
1076 // If no method data exists, go to profile_continue.
1077 test_method_data_pointer(mdp, profile_continue);
1079 Label skip_receiver_profile;
1080 if (receiver_can_be_null) {
1081 Label not_null;
1082 testptr(receiver, receiver);
1083 jccb(Assembler::notZero, not_null);
1084 // We are making a call. Increment the count for null receiver.
1085 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1086 jmp(skip_receiver_profile);
1087 bind(not_null);
1088 }
1090 // Record the receiver type.
1091 record_klass_in_profile(receiver, mdp, reg2, true);
1092 bind(skip_receiver_profile);
1094 // The method data pointer needs to be updated to reflect the new target.
1095 update_mdp_by_constant(mdp,
1096 in_bytes(VirtualCallData::
1097 virtual_call_data_size()));
1098 bind(profile_continue);
1099 }
1100 }
1103 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1104 Register receiver, Register mdp,
1105 Register reg2, int start_row,
1106 Label& done, bool is_virtual_call) {
1107 if (TypeProfileWidth == 0) {
1108 if (is_virtual_call) {
1109 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1110 }
1111 return;
1112 }
1114 int last_row = VirtualCallData::row_limit() - 1;
1115 assert(start_row <= last_row, "must be work left to do");
1116 // Test this row for both the receiver and for null.
1117 // Take any of three different outcomes:
1118 // 1. found receiver => increment count and goto done
1119 // 2. found null => keep looking for case 1, maybe allocate this cell
1120 // 3. found something else => keep looking for cases 1 and 2
1121 // Case 3 is handled by a recursive call.
1122 for (int row = start_row; row <= last_row; row++) {
1123 Label next_test;
1124 bool test_for_null_also = (row == start_row);
1126 // See if the receiver is receiver[n].
1127 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1128 test_mdp_data_at(mdp, recvr_offset, receiver,
1129 (test_for_null_also ? reg2 : noreg),
1130 next_test);
1131 // (Reg2 now contains the receiver from the CallData.)
1133 // The receiver is receiver[n]. Increment count[n].
1134 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1135 increment_mdp_data_at(mdp, count_offset);
1136 jmp(done);
1137 bind(next_test);
1139 if (row == start_row) {
1140 Label found_null;
1141 // Failed the equality check on receiver[n]... Test for null.
1142 testptr(reg2, reg2);
1143 if (start_row == last_row) {
1144 // The only thing left to do is handle the null case.
1145 if (is_virtual_call) {
1146 jccb(Assembler::zero, found_null);
1147 // Receiver did not match any saved receiver and there is no empty row for it.
1148 // Increment total counter to indicate polymorphic case.
1149 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1150 jmp(done);
1151 bind(found_null);
1152 } else {
1153 jcc(Assembler::notZero, done);
1154 }
1155 break;
1156 }
1157 // Since null is rare, make it be the branch-taken case.
1158 jcc(Assembler::zero, found_null);
1160 // Put all the "Case 3" tests here.
1161 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1163 // Found a null. Keep searching for a matching receiver,
1164 // but remember that this is an empty (unused) slot.
1165 bind(found_null);
1166 }
1167 }
1169 // In the fall-through case, we found no matching receiver, but we
1170 // observed the receiver[start_row] is NULL.
1172 // Fill in the receiver field and increment the count.
1173 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1174 set_mdp_data_at(mdp, recvr_offset, receiver);
1175 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1176 movptr(reg2, (intptr_t)DataLayout::counter_increment);
1177 set_mdp_data_at(mdp, count_offset, reg2);
1178 if (start_row > 0) {
1179 jmp(done);
1180 }
1181 }
1183 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1184 Register mdp, Register reg2,
1185 bool is_virtual_call) {
1186 assert(ProfileInterpreter, "must be profiling");
1187 Label done;
1189 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1191 bind (done);
1192 }
1194 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1195 if (ProfileInterpreter) {
1196 Label profile_continue;
1197 uint row;
1199 // If no method data exists, go to profile_continue.
1200 test_method_data_pointer(mdp, profile_continue);
1202 // Update the total ret count.
1203 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1205 for (row = 0; row < RetData::row_limit(); row++) {
1206 Label next_test;
1208 // See if return_bci is equal to bci[n]:
1209 test_mdp_data_at(mdp, in_bytes(RetData::bci_offset(row)), return_bci,
1210 noreg, next_test);
1212 // return_bci is equal to bci[n]. Increment the count.
1213 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1215 // The method data pointer needs to be updated to reflect the new target.
1216 update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)));
1217 jmp(profile_continue);
1218 bind(next_test);
1219 }
1221 update_mdp_for_ret(return_bci);
1223 bind (profile_continue);
1224 }
1225 }
1228 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1229 if (ProfileInterpreter) {
1230 Label profile_continue;
1232 // If no method data exists, go to profile_continue.
1233 test_method_data_pointer(mdp, profile_continue);
1235 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1237 // The method data pointer needs to be updated.
1238 int mdp_delta = in_bytes(BitData::bit_data_size());
1239 if (TypeProfileCasts) {
1240 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1241 }
1242 update_mdp_by_constant(mdp, mdp_delta);
1244 bind (profile_continue);
1245 }
1246 }
1249 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1250 if (ProfileInterpreter && TypeProfileCasts) {
1251 Label profile_continue;
1253 // If no method data exists, go to profile_continue.
1254 test_method_data_pointer(mdp, profile_continue);
1256 int count_offset = in_bytes(CounterData::count_offset());
1257 // Back up the address, since we have already bumped the mdp.
1258 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1260 // *Decrement* the counter. We expect to see zero or small negatives.
1261 increment_mdp_data_at(mdp, count_offset, true);
1263 bind (profile_continue);
1264 }
1265 }
1268 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2)
1269 {
1270 if (ProfileInterpreter) {
1271 Label profile_continue;
1273 // If no method data exists, go to profile_continue.
1274 test_method_data_pointer(mdp, profile_continue);
1276 // The method data pointer needs to be updated.
1277 int mdp_delta = in_bytes(BitData::bit_data_size());
1278 if (TypeProfileCasts) {
1279 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1281 // Record the object type.
1282 record_klass_in_profile(klass, mdp, reg2, false);
1283 assert(reg2 == rdi, "we know how to fix this blown reg");
1284 restore_locals(); // Restore EDI
1285 }
1286 update_mdp_by_constant(mdp, mdp_delta);
1288 bind(profile_continue);
1289 }
1290 }
1293 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1294 if (ProfileInterpreter) {
1295 Label profile_continue;
1297 // If no method data exists, go to profile_continue.
1298 test_method_data_pointer(mdp, profile_continue);
1300 // Update the default case count
1301 increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()));
1303 // The method data pointer needs to be updated.
1304 update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()));
1306 bind (profile_continue);
1307 }
1308 }
1311 void InterpreterMacroAssembler::profile_switch_case(Register index, Register mdp, Register reg2) {
1312 if (ProfileInterpreter) {
1313 Label profile_continue;
1315 // If no method data exists, go to profile_continue.
1316 test_method_data_pointer(mdp, profile_continue);
1318 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1319 movptr(reg2, (intptr_t)in_bytes(MultiBranchData::per_case_size()));
1320 // index is positive and so should have correct value if this code were
1321 // used on 64bits
1322 imulptr(index, reg2);
1323 addptr(index, in_bytes(MultiBranchData::case_array_offset()));
1325 // Update the case count
1326 increment_mdp_data_at(mdp, index, in_bytes(MultiBranchData::relative_count_offset()));
1328 // The method data pointer needs to be updated.
1329 update_mdp_by_offset(mdp, index, in_bytes(MultiBranchData::relative_displacement_offset()));
1331 bind (profile_continue);
1332 }
1333 }
1335 #endif // !CC_INTERP
1339 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1340 if (state == atos) MacroAssembler::verify_oop(reg);
1341 }
1344 #ifndef CC_INTERP
1345 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1346 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
1347 }
1349 #endif /* CC_INTERP */
1352 void InterpreterMacroAssembler::notify_method_entry() {
1353 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1354 // track stack depth. If it is possible to enter interp_only_mode we add
1355 // the code to check if the event should be sent.
1356 if (JvmtiExport::can_post_interpreter_events()) {
1357 Label L;
1358 get_thread(rcx);
1359 movl(rcx, Address(rcx, JavaThread::interp_only_mode_offset()));
1360 testl(rcx,rcx);
1361 jcc(Assembler::zero, L);
1362 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
1363 bind(L);
1364 }
1366 {
1367 SkipIfEqual skip_if(this, &DTraceMethodProbes, 0);
1368 get_thread(rcx);
1369 get_method(rbx);
1370 call_VM_leaf(
1371 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), rcx, rbx);
1372 }
1374 // RedefineClasses() tracing support for obsolete method entry
1375 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1376 get_thread(rcx);
1377 get_method(rbx);
1378 call_VM_leaf(
1379 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1380 rcx, rbx);
1381 }
1382 }
1385 void InterpreterMacroAssembler::notify_method_exit(
1386 TosState state, NotifyMethodExitMode mode) {
1387 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1388 // track stack depth. If it is possible to enter interp_only_mode we add
1389 // the code to check if the event should be sent.
1390 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1391 Label L;
1392 // Note: frame::interpreter_frame_result has a dependency on how the
1393 // method result is saved across the call to post_method_exit. If this
1394 // is changed then the interpreter_frame_result implementation will
1395 // need to be updated too.
1397 // For c++ interpreter the result is always stored at a known location in the frame
1398 // template interpreter will leave it on the top of the stack.
1399 NOT_CC_INTERP(push(state);)
1400 get_thread(rcx);
1401 movl(rcx, Address(rcx, JavaThread::interp_only_mode_offset()));
1402 testl(rcx,rcx);
1403 jcc(Assembler::zero, L);
1404 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1405 bind(L);
1406 NOT_CC_INTERP(pop(state);)
1407 }
1409 {
1410 SkipIfEqual skip_if(this, &DTraceMethodProbes, 0);
1411 NOT_CC_INTERP(push(state));
1412 get_thread(rbx);
1413 get_method(rcx);
1414 call_VM_leaf(
1415 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1416 rbx, rcx);
1417 NOT_CC_INTERP(pop(state));
1418 }
1419 }
1421 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1422 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1423 int increment, int mask,
1424 Register scratch, bool preloaded,
1425 Condition cond, Label* where) {
1426 if (!preloaded) {
1427 movl(scratch, counter_addr);
1428 }
1429 incrementl(scratch, increment);
1430 movl(counter_addr, scratch);
1431 andl(scratch, mask);
1432 jcc(cond, *where);
1433 }