Mercurial > jdk8-mips64-public > hotspot / file revision
src/cpu/x86/vm/templateInterpreter_x86_32.cpp@e9ff18c4ace7
src/cpu/x86/vm/templateInterpreter_x86_32.cpp
Wed, 02 Jun 2010 22:45:42 -0700
- author
- jrose
- date
- Wed, 02 Jun 2010 22:45:42 -0700
- changeset 1934
- e9ff18c4ace7
- parent 1907
- c18cbe5936b8
- parent 1920
- ab102d5d923e
- child 2138
- d5d065957597
- permissions
- -rw-r--r--
Merge
1 /*
2 * Copyright (c) 1997, 2010, 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 "incls/_precompiled.incl"
26 #include "incls/_templateInterpreter_x86_32.cpp.incl"
28 #define __ _masm->
31 #ifndef CC_INTERP
32 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
33 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
34 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
36 //------------------------------------------------------------------------------------------------------------------------
38 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
39 address entry = __ pc();
41 // Note: There should be a minimal interpreter frame set up when stack
42 // overflow occurs since we check explicitly for it now.
43 //
44 #ifdef ASSERT
45 { Label L;
46 __ lea(rax, Address(rbp,
47 frame::interpreter_frame_monitor_block_top_offset * wordSize));
48 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
49 // (stack grows negative)
50 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
51 __ stop ("interpreter frame not set up");
52 __ bind(L);
53 }
54 #endif // ASSERT
55 // Restore bcp under the assumption that the current frame is still
56 // interpreted
57 __ restore_bcp();
59 // expression stack must be empty before entering the VM if an exception
60 // happened
61 __ empty_expression_stack();
62 __ empty_FPU_stack();
63 // throw exception
64 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
65 return entry;
66 }
68 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
69 address entry = __ pc();
70 // expression stack must be empty before entering the VM if an exception happened
71 __ empty_expression_stack();
72 __ empty_FPU_stack();
73 // setup parameters
74 // ??? convention: expect aberrant index in register rbx,
75 __ lea(rax, ExternalAddress((address)name));
76 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
77 return entry;
78 }
80 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
81 address entry = __ pc();
82 // object is at TOS
83 __ pop(rax);
84 // expression stack must be empty before entering the VM if an exception
85 // happened
86 __ empty_expression_stack();
87 __ empty_FPU_stack();
88 __ call_VM(noreg,
89 CAST_FROM_FN_PTR(address,
90 InterpreterRuntime::throw_ClassCastException),
91 rax);
92 return entry;
93 }
95 // Arguments are: required type at TOS+4, failing object (or NULL) at TOS.
96 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
97 address entry = __ pc();
99 __ pop(rbx); // actual failing object is at TOS
100 __ pop(rax); // required type is at TOS+4
102 __ verify_oop(rbx);
103 __ verify_oop(rax);
105 // Various method handle types use interpreter registers as temps.
106 __ restore_bcp();
107 __ restore_locals();
109 // Expression stack must be empty before entering the VM for an exception.
110 __ empty_expression_stack();
111 __ empty_FPU_stack();
112 __ call_VM(noreg,
113 CAST_FROM_FN_PTR(address,
114 InterpreterRuntime::throw_WrongMethodTypeException),
115 // pass required type, failing object (or NULL)
116 rax, rbx);
117 return entry;
118 }
121 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
122 assert(!pass_oop || message == NULL, "either oop or message but not both");
123 address entry = __ pc();
124 if (pass_oop) {
125 // object is at TOS
126 __ pop(rbx);
127 }
128 // expression stack must be empty before entering the VM if an exception happened
129 __ empty_expression_stack();
130 __ empty_FPU_stack();
131 // setup parameters
132 __ lea(rax, ExternalAddress((address)name));
133 if (pass_oop) {
134 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
135 } else {
136 if (message != NULL) {
137 __ lea(rbx, ExternalAddress((address)message));
138 } else {
139 __ movptr(rbx, NULL_WORD);
140 }
141 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
142 }
143 // throw exception
144 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
145 return entry;
146 }
149 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
150 address entry = __ pc();
151 // NULL last_sp until next java call
152 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
153 __ dispatch_next(state);
154 return entry;
155 }
158 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
159 TosState incoming_state = state;
161 Label interpreter_entry;
162 address compiled_entry = __ pc();
164 #ifdef COMPILER2
165 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
166 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
167 for (int i = 1; i < 8; i++) {
168 __ ffree(i);
169 }
170 } else if (UseSSE < 2) {
171 __ empty_FPU_stack();
172 }
173 #endif
174 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
175 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
176 } else {
177 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
178 }
180 __ jmp(interpreter_entry, relocInfo::none);
181 // emit a sentinel we can test for when converting an interpreter
182 // entry point to a compiled entry point.
183 __ a_long(Interpreter::return_sentinel);
184 __ a_long((int)compiled_entry);
185 address entry = __ pc();
186 __ bind(interpreter_entry);
188 // In SSE mode, interpreter returns FP results in xmm0 but they need
189 // to end up back on the FPU so it can operate on them.
190 if (incoming_state == ftos && UseSSE >= 1) {
191 __ subptr(rsp, wordSize);
192 __ movflt(Address(rsp, 0), xmm0);
193 __ fld_s(Address(rsp, 0));
194 __ addptr(rsp, wordSize);
195 } else if (incoming_state == dtos && UseSSE >= 2) {
196 __ subptr(rsp, 2*wordSize);
197 __ movdbl(Address(rsp, 0), xmm0);
198 __ fld_d(Address(rsp, 0));
199 __ addptr(rsp, 2*wordSize);
200 }
202 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
204 // Restore stack bottom in case i2c adjusted stack
205 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
206 // and NULL it as marker that rsp is now tos until next java call
207 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
209 __ restore_bcp();
210 __ restore_locals();
212 Label L_got_cache, L_giant_index;
213 if (EnableInvokeDynamic) {
214 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
215 __ jcc(Assembler::equal, L_giant_index);
216 }
217 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
218 __ bind(L_got_cache);
219 __ movl(rbx, Address(rbx, rcx,
220 Address::times_ptr, constantPoolCacheOopDesc::base_offset() +
221 ConstantPoolCacheEntry::flags_offset()));
222 __ andptr(rbx, 0xFF);
223 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
224 __ dispatch_next(state, step);
226 // out of the main line of code...
227 if (EnableInvokeDynamic) {
228 __ bind(L_giant_index);
229 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
230 __ jmp(L_got_cache);
231 }
233 return entry;
234 }
237 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
238 address entry = __ pc();
240 // In SSE mode, FP results are in xmm0
241 if (state == ftos && UseSSE > 0) {
242 __ subptr(rsp, wordSize);
243 __ movflt(Address(rsp, 0), xmm0);
244 __ fld_s(Address(rsp, 0));
245 __ addptr(rsp, wordSize);
246 } else if (state == dtos && UseSSE >= 2) {
247 __ subptr(rsp, 2*wordSize);
248 __ movdbl(Address(rsp, 0), xmm0);
249 __ fld_d(Address(rsp, 0));
250 __ addptr(rsp, 2*wordSize);
251 }
253 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
255 // The stack is not extended by deopt but we must NULL last_sp as this
256 // entry is like a "return".
257 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
258 __ restore_bcp();
259 __ restore_locals();
260 // handle exceptions
261 { Label L;
262 const Register thread = rcx;
263 __ get_thread(thread);
264 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
265 __ jcc(Assembler::zero, L);
266 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
267 __ should_not_reach_here();
268 __ bind(L);
269 }
270 __ dispatch_next(state, step);
271 return entry;
272 }
275 int AbstractInterpreter::BasicType_as_index(BasicType type) {
276 int i = 0;
277 switch (type) {
278 case T_BOOLEAN: i = 0; break;
279 case T_CHAR : i = 1; break;
280 case T_BYTE : i = 2; break;
281 case T_SHORT : i = 3; break;
282 case T_INT : // fall through
283 case T_LONG : // fall through
284 case T_VOID : i = 4; break;
285 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
286 case T_DOUBLE : i = 6; break;
287 case T_OBJECT : // fall through
288 case T_ARRAY : i = 7; break;
289 default : ShouldNotReachHere();
290 }
291 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
292 return i;
293 }
296 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
297 address entry = __ pc();
298 switch (type) {
299 case T_BOOLEAN: __ c2bool(rax); break;
300 case T_CHAR : __ andptr(rax, 0xFFFF); break;
301 case T_BYTE : __ sign_extend_byte (rax); break;
302 case T_SHORT : __ sign_extend_short(rax); break;
303 case T_INT : /* nothing to do */ break;
304 case T_DOUBLE :
305 case T_FLOAT :
306 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
307 __ pop(t); // remove return address first
308 // Must return a result for interpreter or compiler. In SSE
309 // mode, results are returned in xmm0 and the FPU stack must
310 // be empty.
311 if (type == T_FLOAT && UseSSE >= 1) {
312 // Load ST0
313 __ fld_d(Address(rsp, 0));
314 // Store as float and empty fpu stack
315 __ fstp_s(Address(rsp, 0));
316 // and reload
317 __ movflt(xmm0, Address(rsp, 0));
318 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
319 __ movdbl(xmm0, Address(rsp, 0));
320 } else {
321 // restore ST0
322 __ fld_d(Address(rsp, 0));
323 }
324 // and pop the temp
325 __ addptr(rsp, 2 * wordSize);
326 __ push(t); // restore return address
327 }
328 break;
329 case T_OBJECT :
330 // retrieve result from frame
331 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
332 // and verify it
333 __ verify_oop(rax);
334 break;
335 default : ShouldNotReachHere();
336 }
337 __ ret(0); // return from result handler
338 return entry;
339 }
341 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
342 address entry = __ pc();
343 __ push(state);
344 __ call_VM(noreg, runtime_entry);
345 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
346 return entry;
347 }
350 // Helpers for commoning out cases in the various type of method entries.
351 //
353 // increment invocation count & check for overflow
354 //
355 // Note: checking for negative value instead of overflow
356 // so we have a 'sticky' overflow test
357 //
358 // rbx,: method
359 // rcx: invocation counter
360 //
361 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
363 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
364 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset());
366 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
367 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
368 }
369 // Update standard invocation counters
370 __ movl(rax, backedge_counter); // load backedge counter
372 __ incrementl(rcx, InvocationCounter::count_increment);
373 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
375 __ movl(invocation_counter, rcx); // save invocation count
376 __ addl(rcx, rax); // add both counters
378 // profile_method is non-null only for interpreted method so
379 // profile_method != NULL == !native_call
380 // BytecodeInterpreter only calls for native so code is elided.
382 if (ProfileInterpreter && profile_method != NULL) {
383 // Test to see if we should create a method data oop
384 __ cmp32(rcx,
385 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
386 __ jcc(Assembler::less, *profile_method_continue);
388 // if no method data exists, go to profile_method
389 __ test_method_data_pointer(rax, *profile_method);
390 }
392 __ cmp32(rcx,
393 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
394 __ jcc(Assembler::aboveEqual, *overflow);
396 }
398 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
400 // Asm interpreter on entry
401 // rdi - locals
402 // rsi - bcp
403 // rbx, - method
404 // rdx - cpool
405 // rbp, - interpreter frame
407 // C++ interpreter on entry
408 // rsi - new interpreter state pointer
409 // rbp - interpreter frame pointer
410 // rbx - method
412 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
413 // rbx, - method
414 // rcx - rcvr (assuming there is one)
415 // top of stack return address of interpreter caller
416 // rsp - sender_sp
418 // C++ interpreter only
419 // rsi - previous interpreter state pointer
421 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
423 // InterpreterRuntime::frequency_counter_overflow takes one argument
424 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
425 // The call returns the address of the verified entry point for the method or NULL
426 // if the compilation did not complete (either went background or bailed out).
427 __ movptr(rax, (intptr_t)false);
428 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
430 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
432 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
433 // and jump to the interpreted entry.
434 __ jmp(*do_continue, relocInfo::none);
436 }
438 void InterpreterGenerator::generate_stack_overflow_check(void) {
439 // see if we've got enough room on the stack for locals plus overhead.
440 // the expression stack grows down incrementally, so the normal guard
441 // page mechanism will work for that.
442 //
443 // Registers live on entry:
444 //
445 // Asm interpreter
446 // rdx: number of additional locals this frame needs (what we must check)
447 // rbx,: methodOop
449 // destroyed on exit
450 // rax,
452 // NOTE: since the additional locals are also always pushed (wasn't obvious in
453 // generate_method_entry) so the guard should work for them too.
454 //
456 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
457 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
459 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
460 // be sure to change this if you add/subtract anything to/from the overhead area
461 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
463 const int page_size = os::vm_page_size();
465 Label after_frame_check;
467 // see if the frame is greater than one page in size. If so,
468 // then we need to verify there is enough stack space remaining
469 // for the additional locals.
470 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
471 __ jcc(Assembler::belowEqual, after_frame_check);
473 // compute rsp as if this were going to be the last frame on
474 // the stack before the red zone
476 Label after_frame_check_pop;
478 __ push(rsi);
480 const Register thread = rsi;
482 __ get_thread(thread);
484 const Address stack_base(thread, Thread::stack_base_offset());
485 const Address stack_size(thread, Thread::stack_size_offset());
487 // locals + overhead, in bytes
488 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
490 #ifdef ASSERT
491 Label stack_base_okay, stack_size_okay;
492 // verify that thread stack base is non-zero
493 __ cmpptr(stack_base, (int32_t)NULL_WORD);
494 __ jcc(Assembler::notEqual, stack_base_okay);
495 __ stop("stack base is zero");
496 __ bind(stack_base_okay);
497 // verify that thread stack size is non-zero
498 __ cmpptr(stack_size, 0);
499 __ jcc(Assembler::notEqual, stack_size_okay);
500 __ stop("stack size is zero");
501 __ bind(stack_size_okay);
502 #endif
504 // Add stack base to locals and subtract stack size
505 __ addptr(rax, stack_base);
506 __ subptr(rax, stack_size);
508 // Use the maximum number of pages we might bang.
509 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
510 (StackRedPages+StackYellowPages);
511 __ addptr(rax, max_pages * page_size);
513 // check against the current stack bottom
514 __ cmpptr(rsp, rax);
515 __ jcc(Assembler::above, after_frame_check_pop);
517 __ pop(rsi); // get saved bcp / (c++ prev state ).
519 __ pop(rax); // get return address
520 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
522 // all done with frame size check
523 __ bind(after_frame_check_pop);
524 __ pop(rsi);
526 __ bind(after_frame_check);
527 }
529 // Allocate monitor and lock method (asm interpreter)
530 // rbx, - methodOop
531 //
532 void InterpreterGenerator::lock_method(void) {
533 // synchronize method
534 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
535 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
536 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
538 #ifdef ASSERT
539 { Label L;
540 __ movl(rax, access_flags);
541 __ testl(rax, JVM_ACC_SYNCHRONIZED);
542 __ jcc(Assembler::notZero, L);
543 __ stop("method doesn't need synchronization");
544 __ bind(L);
545 }
546 #endif // ASSERT
547 // get synchronization object
548 { Label done;
549 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
550 __ movl(rax, access_flags);
551 __ testl(rax, JVM_ACC_STATIC);
552 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
553 __ jcc(Assembler::zero, done);
554 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
555 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
556 __ movptr(rax, Address(rax, mirror_offset));
557 __ bind(done);
558 }
559 // add space for monitor & lock
560 __ subptr(rsp, entry_size); // add space for a monitor entry
561 __ movptr(monitor_block_top, rsp); // set new monitor block top
562 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
563 __ mov(rdx, rsp); // object address
564 __ lock_object(rdx);
565 }
567 //
568 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
569 // and for native methods hence the shared code.
571 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
572 // initialize fixed part of activation frame
573 __ push(rax); // save return address
574 __ enter(); // save old & set new rbp,
577 __ push(rsi); // set sender sp
578 __ push((int32_t)NULL_WORD); // leave last_sp as null
579 __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop
580 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
581 __ push(rbx); // save methodOop
582 if (ProfileInterpreter) {
583 Label method_data_continue;
584 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
585 __ testptr(rdx, rdx);
586 __ jcc(Assembler::zero, method_data_continue);
587 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
588 __ bind(method_data_continue);
589 __ push(rdx); // set the mdp (method data pointer)
590 } else {
591 __ push(0);
592 }
594 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
595 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
596 __ push(rdx); // set constant pool cache
597 __ push(rdi); // set locals pointer
598 if (native_call) {
599 __ push(0); // no bcp
600 } else {
601 __ push(rsi); // set bcp
602 }
603 __ push(0); // reserve word for pointer to expression stack bottom
604 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
605 }
607 // End of helpers
609 //
610 // Various method entries
611 //------------------------------------------------------------------------------------------------------------------------
612 //
613 //
615 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
617 address InterpreterGenerator::generate_accessor_entry(void) {
619 // rbx,: methodOop
620 // rcx: receiver (preserve for slow entry into asm interpreter)
622 // rsi: senderSP must preserved for slow path, set SP to it on fast path
624 address entry_point = __ pc();
625 Label xreturn_path;
627 // do fastpath for resolved accessor methods
628 if (UseFastAccessorMethods) {
629 Label slow_path;
630 // If we need a safepoint check, generate full interpreter entry.
631 ExternalAddress state(SafepointSynchronize::address_of_state());
632 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
633 SafepointSynchronize::_not_synchronized);
635 __ jcc(Assembler::notEqual, slow_path);
636 // ASM/C++ Interpreter
637 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
638 // Note: We can only use this code if the getfield has been resolved
639 // and if we don't have a null-pointer exception => check for
640 // these conditions first and use slow path if necessary.
641 // rbx,: method
642 // rcx: receiver
643 __ movptr(rax, Address(rsp, wordSize));
645 // check if local 0 != NULL and read field
646 __ testptr(rax, rax);
647 __ jcc(Assembler::zero, slow_path);
649 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
650 // read first instruction word and extract bytecode @ 1 and index @ 2
651 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
652 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
653 // Shift codes right to get the index on the right.
654 // The bytecode fetched looks like <index><0xb4><0x2a>
655 __ shrl(rdx, 2*BitsPerByte);
656 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
657 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
659 // rax,: local 0
660 // rbx,: method
661 // rcx: receiver - do not destroy since it is needed for slow path!
662 // rcx: scratch
663 // rdx: constant pool cache index
664 // rdi: constant pool cache
665 // rsi: sender sp
667 // check if getfield has been resolved and read constant pool cache entry
668 // check the validity of the cache entry by testing whether _indices field
669 // contains Bytecode::_getfield in b1 byte.
670 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
671 __ movl(rcx,
672 Address(rdi,
673 rdx,
674 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
675 __ shrl(rcx, 2*BitsPerByte);
676 __ andl(rcx, 0xFF);
677 __ cmpl(rcx, Bytecodes::_getfield);
678 __ jcc(Assembler::notEqual, slow_path);
680 // Note: constant pool entry is not valid before bytecode is resolved
681 __ movptr(rcx,
682 Address(rdi,
683 rdx,
684 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
685 __ movl(rdx,
686 Address(rdi,
687 rdx,
688 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
690 Label notByte, notShort, notChar;
691 const Address field_address (rax, rcx, Address::times_1);
693 // Need to differentiate between igetfield, agetfield, bgetfield etc.
694 // because they are different sizes.
695 // Use the type from the constant pool cache
696 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
697 // Make sure we don't need to mask rdx for tosBits after the above shift
698 ConstantPoolCacheEntry::verify_tosBits();
699 __ cmpl(rdx, btos);
700 __ jcc(Assembler::notEqual, notByte);
701 __ load_signed_byte(rax, field_address);
702 __ jmp(xreturn_path);
704 __ bind(notByte);
705 __ cmpl(rdx, stos);
706 __ jcc(Assembler::notEqual, notShort);
707 __ load_signed_short(rax, field_address);
708 __ jmp(xreturn_path);
710 __ bind(notShort);
711 __ cmpl(rdx, ctos);
712 __ jcc(Assembler::notEqual, notChar);
713 __ load_unsigned_short(rax, field_address);
714 __ jmp(xreturn_path);
716 __ bind(notChar);
717 #ifdef ASSERT
718 Label okay;
719 __ cmpl(rdx, atos);
720 __ jcc(Assembler::equal, okay);
721 __ cmpl(rdx, itos);
722 __ jcc(Assembler::equal, okay);
723 __ stop("what type is this?");
724 __ bind(okay);
725 #endif // ASSERT
726 // All the rest are a 32 bit wordsize
727 // This is ok for now. Since fast accessors should be going away
728 __ movptr(rax, field_address);
730 __ bind(xreturn_path);
732 // _ireturn/_areturn
733 __ pop(rdi); // get return address
734 __ mov(rsp, rsi); // set sp to sender sp
735 __ jmp(rdi);
737 // generate a vanilla interpreter entry as the slow path
738 __ bind(slow_path);
740 (void) generate_normal_entry(false);
741 return entry_point;
742 }
743 return NULL;
745 }
747 //
748 // Interpreter stub for calling a native method. (asm interpreter)
749 // This sets up a somewhat different looking stack for calling the native method
750 // than the typical interpreter frame setup.
751 //
753 address InterpreterGenerator::generate_native_entry(bool synchronized) {
754 // determine code generation flags
755 bool inc_counter = UseCompiler || CountCompiledCalls;
757 // rbx,: methodOop
758 // rsi: sender sp
759 // rsi: previous interpreter state (C++ interpreter) must preserve
760 address entry_point = __ pc();
763 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
764 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
765 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
767 // get parameter size (always needed)
768 __ load_unsigned_short(rcx, size_of_parameters);
770 // native calls don't need the stack size check since they have no expression stack
771 // and the arguments are already on the stack and we only add a handful of words
772 // to the stack
774 // rbx,: methodOop
775 // rcx: size of parameters
776 // rsi: sender sp
778 __ pop(rax); // get return address
779 // for natives the size of locals is zero
781 // compute beginning of parameters (rdi)
782 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
785 // add 2 zero-initialized slots for native calls
786 // NULL result handler
787 __ push((int32_t)NULL_WORD);
788 // NULL oop temp (mirror or jni oop result)
789 __ push((int32_t)NULL_WORD);
791 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
792 // initialize fixed part of activation frame
794 generate_fixed_frame(true);
796 // make sure method is native & not abstract
797 #ifdef ASSERT
798 __ movl(rax, access_flags);
799 {
800 Label L;
801 __ testl(rax, JVM_ACC_NATIVE);
802 __ jcc(Assembler::notZero, L);
803 __ stop("tried to execute non-native method as native");
804 __ bind(L);
805 }
806 { Label L;
807 __ testl(rax, JVM_ACC_ABSTRACT);
808 __ jcc(Assembler::zero, L);
809 __ stop("tried to execute abstract method in interpreter");
810 __ bind(L);
811 }
812 #endif
814 // Since at this point in the method invocation the exception handler
815 // would try to exit the monitor of synchronized methods which hasn't
816 // been entered yet, we set the thread local variable
817 // _do_not_unlock_if_synchronized to true. The remove_activation will
818 // check this flag.
820 __ get_thread(rax);
821 const Address do_not_unlock_if_synchronized(rax,
822 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
823 __ movbool(do_not_unlock_if_synchronized, true);
825 // increment invocation count & check for overflow
826 Label invocation_counter_overflow;
827 if (inc_counter) {
828 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
829 }
831 Label continue_after_compile;
832 __ bind(continue_after_compile);
834 bang_stack_shadow_pages(true);
836 // reset the _do_not_unlock_if_synchronized flag
837 __ get_thread(rax);
838 __ movbool(do_not_unlock_if_synchronized, false);
840 // check for synchronized methods
841 // Must happen AFTER invocation_counter check and stack overflow check,
842 // so method is not locked if overflows.
843 //
844 if (synchronized) {
845 lock_method();
846 } else {
847 // no synchronization necessary
848 #ifdef ASSERT
849 { Label L;
850 __ movl(rax, access_flags);
851 __ testl(rax, JVM_ACC_SYNCHRONIZED);
852 __ jcc(Assembler::zero, L);
853 __ stop("method needs synchronization");
854 __ bind(L);
855 }
856 #endif
857 }
859 // start execution
860 #ifdef ASSERT
861 { Label L;
862 const Address monitor_block_top (rbp,
863 frame::interpreter_frame_monitor_block_top_offset * wordSize);
864 __ movptr(rax, monitor_block_top);
865 __ cmpptr(rax, rsp);
866 __ jcc(Assembler::equal, L);
867 __ stop("broken stack frame setup in interpreter");
868 __ bind(L);
869 }
870 #endif
872 // jvmti/dtrace support
873 __ notify_method_entry();
875 // work registers
876 const Register method = rbx;
877 const Register thread = rdi;
878 const Register t = rcx;
880 // allocate space for parameters
881 __ get_method(method);
882 __ verify_oop(method);
883 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset()));
884 __ shlptr(t, Interpreter::logStackElementSize);
885 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
886 __ subptr(rsp, t);
887 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
889 // get signature handler
890 { Label L;
891 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
892 __ testptr(t, t);
893 __ jcc(Assembler::notZero, L);
894 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
895 __ get_method(method);
896 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
897 __ bind(L);
898 }
900 // call signature handler
901 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
902 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
903 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
904 // The generated handlers do not touch RBX (the method oop).
905 // However, large signatures cannot be cached and are generated
906 // each time here. The slow-path generator will blow RBX
907 // sometime, so we must reload it after the call.
908 __ call(t);
909 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
911 // result handler is in rax,
912 // set result handler
913 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
915 // pass mirror handle if static call
916 { Label L;
917 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
918 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
919 __ testl(t, JVM_ACC_STATIC);
920 __ jcc(Assembler::zero, L);
921 // get mirror
922 __ movptr(t, Address(method, methodOopDesc:: constants_offset()));
923 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
924 __ movptr(t, Address(t, mirror_offset));
925 // copy mirror into activation frame
926 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
927 // pass handle to mirror
928 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
929 __ movptr(Address(rsp, wordSize), t);
930 __ bind(L);
931 }
933 // get native function entry point
934 { Label L;
935 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
936 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
937 __ cmpptr(rax, unsatisfied.addr());
938 __ jcc(Assembler::notEqual, L);
939 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
940 __ get_method(method);
941 __ verify_oop(method);
942 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
943 __ bind(L);
944 }
946 // pass JNIEnv
947 __ get_thread(thread);
948 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
949 __ movptr(Address(rsp, 0), t);
951 // set_last_Java_frame_before_call
952 // It is enough that the pc()
953 // points into the right code segment. It does not have to be the correct return pc.
954 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
956 // change thread state
957 #ifdef ASSERT
958 { Label L;
959 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
960 __ cmpl(t, _thread_in_Java);
961 __ jcc(Assembler::equal, L);
962 __ stop("Wrong thread state in native stub");
963 __ bind(L);
964 }
965 #endif
967 // Change state to native
968 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
969 __ call(rax);
971 // result potentially in rdx:rax or ST0
973 // Either restore the MXCSR register after returning from the JNI Call
974 // or verify that it wasn't changed.
975 if (VM_Version::supports_sse()) {
976 if (RestoreMXCSROnJNICalls) {
977 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
978 }
979 else if (CheckJNICalls ) {
980 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
981 }
982 }
984 // Either restore the x87 floating pointer control word after returning
985 // from the JNI call or verify that it wasn't changed.
986 if (CheckJNICalls) {
987 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
988 }
990 // save potential result in ST(0) & rdx:rax
991 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
992 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
993 // It is safe to do this push because state is _thread_in_native and return address will be found
994 // via _last_native_pc and not via _last_jave_sp
996 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
997 // If the order changes or anything else is added to the stack the code in
998 // interpreter_frame_result will have to be changed.
1000 { Label L;
1001 Label push_double;
1002 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1003 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1004 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1005 float_handler.addr());
1006 __ jcc(Assembler::equal, push_double);
1007 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1008 double_handler.addr());
1009 __ jcc(Assembler::notEqual, L);
1010 __ bind(push_double);
1011 __ push(dtos);
1012 __ bind(L);
1013 }
1014 __ push(ltos);
1016 // change thread state
1017 __ get_thread(thread);
1018 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1019 if(os::is_MP()) {
1020 if (UseMembar) {
1021 // Force this write out before the read below
1022 __ membar(Assembler::Membar_mask_bits(
1023 Assembler::LoadLoad | Assembler::LoadStore |
1024 Assembler::StoreLoad | Assembler::StoreStore));
1025 } else {
1026 // Write serialization page so VM thread can do a pseudo remote membar.
1027 // We use the current thread pointer to calculate a thread specific
1028 // offset to write to within the page. This minimizes bus traffic
1029 // due to cache line collision.
1030 __ serialize_memory(thread, rcx);
1031 }
1032 }
1034 if (AlwaysRestoreFPU) {
1035 // Make sure the control word is correct.
1036 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1037 }
1039 // check for safepoint operation in progress and/or pending suspend requests
1040 { Label Continue;
1042 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1043 SafepointSynchronize::_not_synchronized);
1045 Label L;
1046 __ jcc(Assembler::notEqual, L);
1047 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1048 __ jcc(Assembler::equal, Continue);
1049 __ bind(L);
1051 // Don't use call_VM as it will see a possible pending exception and forward it
1052 // and never return here preventing us from clearing _last_native_pc down below.
1053 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1054 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1055 // by hand.
1056 //
1057 __ push(thread);
1058 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1059 JavaThread::check_special_condition_for_native_trans)));
1060 __ increment(rsp, wordSize);
1061 __ get_thread(thread);
1063 __ bind(Continue);
1064 }
1066 // change thread state
1067 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1069 __ reset_last_Java_frame(thread, true, true);
1071 // reset handle block
1072 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1073 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1075 // If result was an oop then unbox and save it in the frame
1076 { Label L;
1077 Label no_oop, store_result;
1078 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1079 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1080 handler.addr());
1081 __ jcc(Assembler::notEqual, no_oop);
1082 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1083 __ pop(ltos);
1084 __ testptr(rax, rax);
1085 __ jcc(Assembler::zero, store_result);
1086 // unbox
1087 __ movptr(rax, Address(rax, 0));
1088 __ bind(store_result);
1089 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1090 // keep stack depth as expected by pushing oop which will eventually be discarded
1091 __ push(ltos);
1092 __ bind(no_oop);
1093 }
1095 {
1096 Label no_reguard;
1097 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1098 __ jcc(Assembler::notEqual, no_reguard);
1100 __ pusha();
1101 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1102 __ popa();
1104 __ bind(no_reguard);
1105 }
1107 // restore rsi to have legal interpreter frame,
1108 // i.e., bci == 0 <=> rsi == code_base()
1109 // Can't call_VM until bcp is within reasonable.
1110 __ get_method(method); // method is junk from thread_in_native to now.
1111 __ verify_oop(method);
1112 __ movptr(rsi, Address(method,methodOopDesc::const_offset())); // get constMethodOop
1113 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
1115 // handle exceptions (exception handling will handle unlocking!)
1116 { Label L;
1117 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1118 __ jcc(Assembler::zero, L);
1119 // Note: At some point we may want to unify this with the code used in call_VM_base();
1120 // i.e., we should use the StubRoutines::forward_exception code. For now this
1121 // doesn't work here because the rsp is not correctly set at this point.
1122 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1123 __ should_not_reach_here();
1124 __ bind(L);
1125 }
1127 // do unlocking if necessary
1128 { Label L;
1129 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1130 __ testl(t, JVM_ACC_SYNCHRONIZED);
1131 __ jcc(Assembler::zero, L);
1132 // the code below should be shared with interpreter macro assembler implementation
1133 { Label unlock;
1134 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1135 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1136 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1138 __ lea(rdx, monitor); // address of first monitor
1140 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1141 __ testptr(t, t);
1142 __ jcc(Assembler::notZero, unlock);
1144 // Entry already unlocked, need to throw exception
1145 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1146 __ should_not_reach_here();
1148 __ bind(unlock);
1149 __ unlock_object(rdx);
1150 }
1151 __ bind(L);
1152 }
1154 // jvmti/dtrace support
1155 // Note: This must happen _after_ handling/throwing any exceptions since
1156 // the exception handler code notifies the runtime of method exits
1157 // too. If this happens before, method entry/exit notifications are
1158 // not properly paired (was bug - gri 11/22/99).
1159 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1161 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1162 __ pop(ltos);
1163 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1164 __ call(t);
1166 // remove activation
1167 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1168 __ leave(); // remove frame anchor
1169 __ pop(rdi); // get return address
1170 __ mov(rsp, t); // set sp to sender sp
1171 __ jmp(rdi);
1173 if (inc_counter) {
1174 // Handle overflow of counter and compile method
1175 __ bind(invocation_counter_overflow);
1176 generate_counter_overflow(&continue_after_compile);
1177 }
1179 return entry_point;
1180 }
1182 //
1183 // Generic interpreted method entry to (asm) interpreter
1184 //
1185 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1186 // determine code generation flags
1187 bool inc_counter = UseCompiler || CountCompiledCalls;
1189 // rbx,: methodOop
1190 // rsi: sender sp
1191 address entry_point = __ pc();
1194 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
1195 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
1196 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
1197 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
1199 // get parameter size (always needed)
1200 __ load_unsigned_short(rcx, size_of_parameters);
1202 // rbx,: methodOop
1203 // rcx: size of parameters
1205 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1207 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1208 __ subl(rdx, rcx); // rdx = no. of additional locals
1210 // see if we've got enough room on the stack for locals plus overhead.
1211 generate_stack_overflow_check();
1213 // get return address
1214 __ pop(rax);
1216 // compute beginning of parameters (rdi)
1217 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1219 // rdx - # of additional locals
1220 // allocate space for locals
1221 // explicitly initialize locals
1222 {
1223 Label exit, loop;
1224 __ testl(rdx, rdx);
1225 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1226 __ bind(loop);
1227 __ push((int32_t)NULL_WORD); // initialize local variables
1228 __ decrement(rdx); // until everything initialized
1229 __ jcc(Assembler::greater, loop);
1230 __ bind(exit);
1231 }
1233 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
1234 // initialize fixed part of activation frame
1235 generate_fixed_frame(false);
1237 // make sure method is not native & not abstract
1238 #ifdef ASSERT
1239 __ movl(rax, access_flags);
1240 {
1241 Label L;
1242 __ testl(rax, JVM_ACC_NATIVE);
1243 __ jcc(Assembler::zero, L);
1244 __ stop("tried to execute native method as non-native");
1245 __ bind(L);
1246 }
1247 { Label L;
1248 __ testl(rax, JVM_ACC_ABSTRACT);
1249 __ jcc(Assembler::zero, L);
1250 __ stop("tried to execute abstract method in interpreter");
1251 __ bind(L);
1252 }
1253 #endif
1255 // Since at this point in the method invocation the exception handler
1256 // would try to exit the monitor of synchronized methods which hasn't
1257 // been entered yet, we set the thread local variable
1258 // _do_not_unlock_if_synchronized to true. The remove_activation will
1259 // check this flag.
1261 __ get_thread(rax);
1262 const Address do_not_unlock_if_synchronized(rax,
1263 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1264 __ movbool(do_not_unlock_if_synchronized, true);
1266 // increment invocation count & check for overflow
1267 Label invocation_counter_overflow;
1268 Label profile_method;
1269 Label profile_method_continue;
1270 if (inc_counter) {
1271 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1272 if (ProfileInterpreter) {
1273 __ bind(profile_method_continue);
1274 }
1275 }
1276 Label continue_after_compile;
1277 __ bind(continue_after_compile);
1279 bang_stack_shadow_pages(false);
1281 // reset the _do_not_unlock_if_synchronized flag
1282 __ get_thread(rax);
1283 __ movbool(do_not_unlock_if_synchronized, false);
1285 // check for synchronized methods
1286 // Must happen AFTER invocation_counter check and stack overflow check,
1287 // so method is not locked if overflows.
1288 //
1289 if (synchronized) {
1290 // Allocate monitor and lock method
1291 lock_method();
1292 } else {
1293 // no synchronization necessary
1294 #ifdef ASSERT
1295 { Label L;
1296 __ movl(rax, access_flags);
1297 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1298 __ jcc(Assembler::zero, L);
1299 __ stop("method needs synchronization");
1300 __ bind(L);
1301 }
1302 #endif
1303 }
1305 // start execution
1306 #ifdef ASSERT
1307 { Label L;
1308 const Address monitor_block_top (rbp,
1309 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1310 __ movptr(rax, monitor_block_top);
1311 __ cmpptr(rax, rsp);
1312 __ jcc(Assembler::equal, L);
1313 __ stop("broken stack frame setup in interpreter");
1314 __ bind(L);
1315 }
1316 #endif
1318 // jvmti support
1319 __ notify_method_entry();
1321 __ dispatch_next(vtos);
1323 // invocation counter overflow
1324 if (inc_counter) {
1325 if (ProfileInterpreter) {
1326 // We have decided to profile this method in the interpreter
1327 __ bind(profile_method);
1329 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi, true);
1331 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
1332 __ movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1333 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1334 __ test_method_data_pointer(rax, profile_method_continue);
1335 __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
1336 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1337 __ jmp(profile_method_continue);
1338 }
1339 // Handle overflow of counter and compile method
1340 __ bind(invocation_counter_overflow);
1341 generate_counter_overflow(&continue_after_compile);
1342 }
1344 return entry_point;
1345 }
1347 //------------------------------------------------------------------------------------------------------------------------
1348 // Entry points
1349 //
1350 // Here we generate the various kind of entries into the interpreter.
1351 // The two main entry type are generic bytecode methods and native call method.
1352 // These both come in synchronized and non-synchronized versions but the
1353 // frame layout they create is very similar. The other method entry
1354 // types are really just special purpose entries that are really entry
1355 // and interpretation all in one. These are for trivial methods like
1356 // accessor, empty, or special math methods.
1357 //
1358 // When control flow reaches any of the entry types for the interpreter
1359 // the following holds ->
1360 //
1361 // Arguments:
1362 //
1363 // rbx,: methodOop
1364 // rcx: receiver
1365 //
1366 //
1367 // Stack layout immediately at entry
1368 //
1369 // [ return address ] <--- rsp
1370 // [ parameter n ]
1371 // ...
1372 // [ parameter 1 ]
1373 // [ expression stack ] (caller's java expression stack)
1375 // Assuming that we don't go to one of the trivial specialized
1376 // entries the stack will look like below when we are ready to execute
1377 // the first bytecode (or call the native routine). The register usage
1378 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1379 //
1380 // local variables follow incoming parameters immediately; i.e.
1381 // the return address is moved to the end of the locals).
1382 //
1383 // [ monitor entry ] <--- rsp
1384 // ...
1385 // [ monitor entry ]
1386 // [ expr. stack bottom ]
1387 // [ saved rsi ]
1388 // [ current rdi ]
1389 // [ methodOop ]
1390 // [ saved rbp, ] <--- rbp,
1391 // [ return address ]
1392 // [ local variable m ]
1393 // ...
1394 // [ local variable 1 ]
1395 // [ parameter n ]
1396 // ...
1397 // [ parameter 1 ] <--- rdi
1399 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1400 // determine code generation flags
1401 bool synchronized = false;
1402 address entry_point = NULL;
1404 switch (kind) {
1405 case Interpreter::zerolocals : break;
1406 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1407 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
1408 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
1409 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
1410 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
1411 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
1412 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
1414 case Interpreter::java_lang_math_sin : // fall thru
1415 case Interpreter::java_lang_math_cos : // fall thru
1416 case Interpreter::java_lang_math_tan : // fall thru
1417 case Interpreter::java_lang_math_abs : // fall thru
1418 case Interpreter::java_lang_math_log : // fall thru
1419 case Interpreter::java_lang_math_log10 : // fall thru
1420 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1421 default : ShouldNotReachHere(); break;
1422 }
1424 if (entry_point) return entry_point;
1426 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1428 }
1430 // These should never be compiled since the interpreter will prefer
1431 // the compiled version to the intrinsic version.
1432 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1433 switch (method_kind(m)) {
1434 case Interpreter::java_lang_math_sin : // fall thru
1435 case Interpreter::java_lang_math_cos : // fall thru
1436 case Interpreter::java_lang_math_tan : // fall thru
1437 case Interpreter::java_lang_math_abs : // fall thru
1438 case Interpreter::java_lang_math_log : // fall thru
1439 case Interpreter::java_lang_math_log10 : // fall thru
1440 case Interpreter::java_lang_math_sqrt :
1441 return false;
1442 default:
1443 return true;
1444 }
1445 }
1447 // How much stack a method activation needs in words.
1448 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1450 const int stub_code = 4; // see generate_call_stub
1451 // Save space for one monitor to get into the interpreted method in case
1452 // the method is synchronized
1453 int monitor_size = method->is_synchronized() ?
1454 1*frame::interpreter_frame_monitor_size() : 0;
1456 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1457 // be sure to change this if you add/subtract anything to/from the overhead area
1458 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1460 const int extra_stack = methodOopDesc::extra_stack_entries();
1461 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1462 Interpreter::stackElementWords;
1463 return overhead_size + method_stack + stub_code;
1464 }
1466 // asm based interpreter deoptimization helpers
1468 int AbstractInterpreter::layout_activation(methodOop method,
1469 int tempcount,
1470 int popframe_extra_args,
1471 int moncount,
1472 int callee_param_count,
1473 int callee_locals,
1474 frame* caller,
1475 frame* interpreter_frame,
1476 bool is_top_frame) {
1477 // Note: This calculation must exactly parallel the frame setup
1478 // in AbstractInterpreterGenerator::generate_method_entry.
1479 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1480 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1481 // as determined by a previous call to this method.
1482 // It is also guaranteed to be walkable even though it is in a skeletal state
1483 // NOTE: return size is in words not bytes
1485 // fixed size of an interpreter frame:
1486 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1487 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1488 Interpreter::stackElementWords;
1490 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1492 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1493 // Since the callee parameters already account for the callee's params we only need to account for
1494 // the extra locals.
1497 int size = overhead +
1498 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1499 (moncount*frame::interpreter_frame_monitor_size()) +
1500 tempcount*Interpreter::stackElementWords + popframe_extra_args;
1502 if (interpreter_frame != NULL) {
1503 #ifdef ASSERT
1504 if (!EnableMethodHandles)
1505 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1506 // Probably, since deoptimization doesn't work yet.
1507 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1508 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1509 #endif
1511 interpreter_frame->interpreter_frame_set_method(method);
1512 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1513 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1514 // and sender_sp is fp+8
1515 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1517 interpreter_frame->interpreter_frame_set_locals(locals);
1518 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1519 BasicObjectLock* monbot = montop - moncount;
1520 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1522 // Set last_sp
1523 intptr_t* rsp = (intptr_t*) monbot -
1524 tempcount*Interpreter::stackElementWords -
1525 popframe_extra_args;
1526 interpreter_frame->interpreter_frame_set_last_sp(rsp);
1528 // All frames but the initial (oldest) interpreter frame we fill in have a
1529 // value for sender_sp that allows walking the stack but isn't
1530 // truly correct. Correct the value here.
1532 if (extra_locals != 0 &&
1533 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1534 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1535 }
1536 *interpreter_frame->interpreter_frame_cache_addr() =
1537 method->constants()->cache();
1538 }
1539 return size;
1540 }
1543 //------------------------------------------------------------------------------------------------------------------------
1544 // Exceptions
1546 void TemplateInterpreterGenerator::generate_throw_exception() {
1547 // Entry point in previous activation (i.e., if the caller was interpreted)
1548 Interpreter::_rethrow_exception_entry = __ pc();
1549 const Register thread = rcx;
1551 // Restore sp to interpreter_frame_last_sp even though we are going
1552 // to empty the expression stack for the exception processing.
1553 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1554 // rax,: exception
1555 // rdx: return address/pc that threw exception
1556 __ restore_bcp(); // rsi points to call/send
1557 __ restore_locals();
1559 // Entry point for exceptions thrown within interpreter code
1560 Interpreter::_throw_exception_entry = __ pc();
1561 // expression stack is undefined here
1562 // rax,: exception
1563 // rsi: exception bcp
1564 __ verify_oop(rax);
1566 // expression stack must be empty before entering the VM in case of an exception
1567 __ empty_expression_stack();
1568 __ empty_FPU_stack();
1569 // find exception handler address and preserve exception oop
1570 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1571 // rax,: exception handler entry point
1572 // rdx: preserved exception oop
1573 // rsi: bcp for exception handler
1574 __ push_ptr(rdx); // push exception which is now the only value on the stack
1575 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1577 // If the exception is not handled in the current frame the frame is removed and
1578 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1579 //
1580 // Note: At this point the bci is still the bxi for the instruction which caused
1581 // the exception and the expression stack is empty. Thus, for any VM calls
1582 // at this point, GC will find a legal oop map (with empty expression stack).
1584 // In current activation
1585 // tos: exception
1586 // rsi: exception bcp
1588 //
1589 // JVMTI PopFrame support
1590 //
1592 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1593 __ empty_expression_stack();
1594 __ empty_FPU_stack();
1595 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1596 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1597 // popframe handling cycles.
1598 __ get_thread(thread);
1599 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1600 __ orl(rdx, JavaThread::popframe_processing_bit);
1601 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1603 {
1604 // Check to see whether we are returning to a deoptimized frame.
1605 // (The PopFrame call ensures that the caller of the popped frame is
1606 // either interpreted or compiled and deoptimizes it if compiled.)
1607 // In this case, we can't call dispatch_next() after the frame is
1608 // popped, but instead must save the incoming arguments and restore
1609 // them after deoptimization has occurred.
1610 //
1611 // Note that we don't compare the return PC against the
1612 // deoptimization blob's unpack entry because of the presence of
1613 // adapter frames in C2.
1614 Label caller_not_deoptimized;
1615 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1616 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1617 __ testl(rax, rax);
1618 __ jcc(Assembler::notZero, caller_not_deoptimized);
1620 // Compute size of arguments for saving when returning to deoptimized caller
1621 __ get_method(rax);
1622 __ verify_oop(rax);
1623 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset())));
1624 __ shlptr(rax, Interpreter::logStackElementSize);
1625 __ restore_locals();
1626 __ subptr(rdi, rax);
1627 __ addptr(rdi, wordSize);
1628 // Save these arguments
1629 __ get_thread(thread);
1630 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1632 __ remove_activation(vtos, rdx,
1633 /* throw_monitor_exception */ false,
1634 /* install_monitor_exception */ false,
1635 /* notify_jvmdi */ false);
1637 // Inform deoptimization that it is responsible for restoring these arguments
1638 __ get_thread(thread);
1639 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1641 // Continue in deoptimization handler
1642 __ jmp(rdx);
1644 __ bind(caller_not_deoptimized);
1645 }
1647 __ remove_activation(vtos, rdx,
1648 /* throw_monitor_exception */ false,
1649 /* install_monitor_exception */ false,
1650 /* notify_jvmdi */ false);
1652 // Finish with popframe handling
1653 // A previous I2C followed by a deoptimization might have moved the
1654 // outgoing arguments further up the stack. PopFrame expects the
1655 // mutations to those outgoing arguments to be preserved and other
1656 // constraints basically require this frame to look exactly as
1657 // though it had previously invoked an interpreted activation with
1658 // no space between the top of the expression stack (current
1659 // last_sp) and the top of stack. Rather than force deopt to
1660 // maintain this kind of invariant all the time we call a small
1661 // fixup routine to move the mutated arguments onto the top of our
1662 // expression stack if necessary.
1663 __ mov(rax, rsp);
1664 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1665 __ get_thread(thread);
1666 // PC must point into interpreter here
1667 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1668 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1669 __ get_thread(thread);
1670 __ reset_last_Java_frame(thread, true, true);
1671 // Restore the last_sp and null it out
1672 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1673 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1675 __ restore_bcp();
1676 __ restore_locals();
1677 // The method data pointer was incremented already during
1678 // call profiling. We have to restore the mdp for the current bcp.
1679 if (ProfileInterpreter) {
1680 __ set_method_data_pointer_for_bcp();
1681 }
1683 // Clear the popframe condition flag
1684 __ get_thread(thread);
1685 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1687 __ dispatch_next(vtos);
1688 // end of PopFrame support
1690 Interpreter::_remove_activation_entry = __ pc();
1692 // preserve exception over this code sequence
1693 __ pop_ptr(rax);
1694 __ get_thread(thread);
1695 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1696 // remove the activation (without doing throws on illegalMonitorExceptions)
1697 __ remove_activation(vtos, rdx, false, true, false);
1698 // restore exception
1699 __ get_thread(thread);
1700 __ movptr(rax, Address(thread, JavaThread::vm_result_offset()));
1701 __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
1702 __ verify_oop(rax);
1704 // Inbetween activations - previous activation type unknown yet
1705 // compute continuation point - the continuation point expects
1706 // the following registers set up:
1707 //
1708 // rax: exception
1709 // rdx: return address/pc that threw exception
1710 // rsp: expression stack of caller
1711 // rbp: rbp, of caller
1712 __ push(rax); // save exception
1713 __ push(rdx); // save return address
1714 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1715 __ mov(rbx, rax); // save exception handler
1716 __ pop(rdx); // restore return address
1717 __ pop(rax); // restore exception
1718 // Note that an "issuing PC" is actually the next PC after the call
1719 __ jmp(rbx); // jump to exception handler of caller
1720 }
1723 //
1724 // JVMTI ForceEarlyReturn support
1725 //
1726 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1727 address entry = __ pc();
1728 const Register thread = rcx;
1730 __ restore_bcp();
1731 __ restore_locals();
1732 __ empty_expression_stack();
1733 __ empty_FPU_stack();
1734 __ load_earlyret_value(state);
1736 __ get_thread(thread);
1737 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1738 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1740 // Clear the earlyret state
1741 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1743 __ remove_activation(state, rsi,
1744 false, /* throw_monitor_exception */
1745 false, /* install_monitor_exception */
1746 true); /* notify_jvmdi */
1747 __ jmp(rsi);
1748 return entry;
1749 } // end of ForceEarlyReturn support
1752 //------------------------------------------------------------------------------------------------------------------------
1753 // Helper for vtos entry point generation
1755 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
1756 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1757 Label L;
1758 fep = __ pc(); __ push(ftos); __ jmp(L);
1759 dep = __ pc(); __ push(dtos); __ jmp(L);
1760 lep = __ pc(); __ push(ltos); __ jmp(L);
1761 aep = __ pc(); __ push(atos); __ jmp(L);
1762 bep = cep = sep = // fall through
1763 iep = __ pc(); __ push(itos); // fall through
1764 vep = __ pc(); __ bind(L); // fall through
1765 generate_and_dispatch(t);
1766 }
1768 //------------------------------------------------------------------------------------------------------------------------
1769 // Generation of individual instructions
1771 // helpers for generate_and_dispatch
1775 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1776 : TemplateInterpreterGenerator(code) {
1777 generate_all(); // down here so it can be "virtual"
1778 }
1780 //------------------------------------------------------------------------------------------------------------------------
1782 // Non-product code
1783 #ifndef PRODUCT
1784 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1785 address entry = __ pc();
1787 // prepare expression stack
1788 __ pop(rcx); // pop return address so expression stack is 'pure'
1789 __ push(state); // save tosca
1791 // pass tosca registers as arguments & call tracer
1792 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1793 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1794 __ pop(state); // restore tosca
1796 // return
1797 __ jmp(rcx);
1799 return entry;
1800 }
1803 void TemplateInterpreterGenerator::count_bytecode() {
1804 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1805 }
1808 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1809 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1810 }
1813 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1814 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1815 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1816 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1817 ExternalAddress table((address) BytecodePairHistogram::_counters);
1818 Address index(noreg, rbx, Address::times_4);
1819 __ incrementl(ArrayAddress(table, index));
1820 }
1823 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1824 // Call a little run-time stub to avoid blow-up for each bytecode.
1825 // The run-time runtime saves the right registers, depending on
1826 // the tosca in-state for the given template.
1827 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1828 "entry must have been generated");
1829 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1830 }
1833 void TemplateInterpreterGenerator::stop_interpreter_at() {
1834 Label L;
1835 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1836 StopInterpreterAt);
1837 __ jcc(Assembler::notEqual, L);
1838 __ int3();
1839 __ bind(L);
1840 }
1841 #endif // !PRODUCT
1842 #endif // CC_INTERP
