Tue, 02 Sep 2014 12:48:45 -0700
8055494: Add C2 x86 intrinsic for BigInteger::multiplyToLen() method
Summary: Add new C2 intrinsic for BigInteger::multiplyToLen() on x86 in 64-bit VM.
Reviewed-by: roland
1 /*
2 * Copyright (c) 1997, 2013, 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 "asm/macroAssembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterGenerator.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "oops/arrayOop.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "runtime/arguments.hpp"
39 #include "runtime/deoptimization.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubRoutines.hpp"
43 #include "runtime/synchronizer.hpp"
44 #include "runtime/timer.hpp"
45 #include "runtime/vframeArray.hpp"
46 #include "utilities/debug.hpp"
47 #include "utilities/macros.hpp"
49 #define __ _masm->
52 #ifndef CC_INTERP
53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
54 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
57 //------------------------------------------------------------------------------------------------------------------------
59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
60 address entry = __ pc();
62 // Note: There should be a minimal interpreter frame set up when stack
63 // overflow occurs since we check explicitly for it now.
64 //
65 #ifdef ASSERT
66 { Label L;
67 __ lea(rax, Address(rbp,
68 frame::interpreter_frame_monitor_block_top_offset * wordSize));
69 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
70 // (stack grows negative)
71 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
72 __ stop ("interpreter frame not set up");
73 __ bind(L);
74 }
75 #endif // ASSERT
76 // Restore bcp under the assumption that the current frame is still
77 // interpreted
78 __ restore_bcp();
80 // expression stack must be empty before entering the VM if an exception
81 // happened
82 __ empty_expression_stack();
83 __ empty_FPU_stack();
84 // throw exception
85 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
86 return entry;
87 }
89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
90 address entry = __ pc();
91 // expression stack must be empty before entering the VM if an exception happened
92 __ empty_expression_stack();
93 __ empty_FPU_stack();
94 // setup parameters
95 // ??? convention: expect aberrant index in register rbx,
96 __ lea(rax, ExternalAddress((address)name));
97 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
98 return entry;
99 }
101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
102 address entry = __ pc();
103 // object is at TOS
104 __ pop(rax);
105 // expression stack must be empty before entering the VM if an exception
106 // happened
107 __ empty_expression_stack();
108 __ empty_FPU_stack();
109 __ call_VM(noreg,
110 CAST_FROM_FN_PTR(address,
111 InterpreterRuntime::throw_ClassCastException),
112 rax);
113 return entry;
114 }
116 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
117 assert(!pass_oop || message == NULL, "either oop or message but not both");
118 address entry = __ pc();
119 if (pass_oop) {
120 // object is at TOS
121 __ pop(rbx);
122 }
123 // expression stack must be empty before entering the VM if an exception happened
124 __ empty_expression_stack();
125 __ empty_FPU_stack();
126 // setup parameters
127 __ lea(rax, ExternalAddress((address)name));
128 if (pass_oop) {
129 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
130 } else {
131 if (message != NULL) {
132 __ lea(rbx, ExternalAddress((address)message));
133 } else {
134 __ movptr(rbx, NULL_WORD);
135 }
136 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
137 }
138 // throw exception
139 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
140 return entry;
141 }
144 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
145 address entry = __ pc();
146 // NULL last_sp until next java call
147 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
148 __ dispatch_next(state);
149 return entry;
150 }
153 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
154 address entry = __ pc();
156 #ifdef COMPILER2
157 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
158 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
159 for (int i = 1; i < 8; i++) {
160 __ ffree(i);
161 }
162 } else if (UseSSE < 2) {
163 __ empty_FPU_stack();
164 }
165 #endif
166 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
167 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
168 } else {
169 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
170 }
172 // In SSE mode, interpreter returns FP results in xmm0 but they need
173 // to end up back on the FPU so it can operate on them.
174 if (state == ftos && UseSSE >= 1) {
175 __ subptr(rsp, wordSize);
176 __ movflt(Address(rsp, 0), xmm0);
177 __ fld_s(Address(rsp, 0));
178 __ addptr(rsp, wordSize);
179 } else if (state == dtos && UseSSE >= 2) {
180 __ subptr(rsp, 2*wordSize);
181 __ movdbl(Address(rsp, 0), xmm0);
182 __ fld_d(Address(rsp, 0));
183 __ addptr(rsp, 2*wordSize);
184 }
186 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
188 // Restore stack bottom in case i2c adjusted stack
189 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
190 // and NULL it as marker that rsp is now tos until next java call
191 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
193 __ restore_bcp();
194 __ restore_locals();
196 if (state == atos) {
197 Register mdp = rbx;
198 Register tmp = rcx;
199 __ profile_return_type(mdp, rax, tmp);
200 }
202 const Register cache = rbx;
203 const Register index = rcx;
204 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
206 const Register flags = cache;
207 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
208 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
209 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
210 __ dispatch_next(state, step);
212 return entry;
213 }
216 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
217 address entry = __ pc();
219 // In SSE mode, FP results are in xmm0
220 if (state == ftos && UseSSE > 0) {
221 __ subptr(rsp, wordSize);
222 __ movflt(Address(rsp, 0), xmm0);
223 __ fld_s(Address(rsp, 0));
224 __ addptr(rsp, wordSize);
225 } else if (state == dtos && UseSSE >= 2) {
226 __ subptr(rsp, 2*wordSize);
227 __ movdbl(Address(rsp, 0), xmm0);
228 __ fld_d(Address(rsp, 0));
229 __ addptr(rsp, 2*wordSize);
230 }
232 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
234 // The stack is not extended by deopt but we must NULL last_sp as this
235 // entry is like a "return".
236 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
237 __ restore_bcp();
238 __ restore_locals();
239 // handle exceptions
240 { Label L;
241 const Register thread = rcx;
242 __ get_thread(thread);
243 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
244 __ jcc(Assembler::zero, L);
245 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
246 __ should_not_reach_here();
247 __ bind(L);
248 }
249 __ dispatch_next(state, step);
250 return entry;
251 }
254 int AbstractInterpreter::BasicType_as_index(BasicType type) {
255 int i = 0;
256 switch (type) {
257 case T_BOOLEAN: i = 0; break;
258 case T_CHAR : i = 1; break;
259 case T_BYTE : i = 2; break;
260 case T_SHORT : i = 3; break;
261 case T_INT : // fall through
262 case T_LONG : // fall through
263 case T_VOID : i = 4; break;
264 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
265 case T_DOUBLE : i = 6; break;
266 case T_OBJECT : // fall through
267 case T_ARRAY : i = 7; break;
268 default : ShouldNotReachHere();
269 }
270 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
271 return i;
272 }
275 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
276 address entry = __ pc();
277 switch (type) {
278 case T_BOOLEAN: __ c2bool(rax); break;
279 case T_CHAR : __ andptr(rax, 0xFFFF); break;
280 case T_BYTE : __ sign_extend_byte (rax); break;
281 case T_SHORT : __ sign_extend_short(rax); break;
282 case T_INT : /* nothing to do */ break;
283 case T_DOUBLE :
284 case T_FLOAT :
285 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
286 __ pop(t); // remove return address first
287 // Must return a result for interpreter or compiler. In SSE
288 // mode, results are returned in xmm0 and the FPU stack must
289 // be empty.
290 if (type == T_FLOAT && UseSSE >= 1) {
291 // Load ST0
292 __ fld_d(Address(rsp, 0));
293 // Store as float and empty fpu stack
294 __ fstp_s(Address(rsp, 0));
295 // and reload
296 __ movflt(xmm0, Address(rsp, 0));
297 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
298 __ movdbl(xmm0, Address(rsp, 0));
299 } else {
300 // restore ST0
301 __ fld_d(Address(rsp, 0));
302 }
303 // and pop the temp
304 __ addptr(rsp, 2 * wordSize);
305 __ push(t); // restore return address
306 }
307 break;
308 case T_OBJECT :
309 // retrieve result from frame
310 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
311 // and verify it
312 __ verify_oop(rax);
313 break;
314 default : ShouldNotReachHere();
315 }
316 __ ret(0); // return from result handler
317 return entry;
318 }
320 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
321 address entry = __ pc();
322 __ push(state);
323 __ call_VM(noreg, runtime_entry);
324 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
325 return entry;
326 }
329 // Helpers for commoning out cases in the various type of method entries.
330 //
332 // increment invocation count & check for overflow
333 //
334 // Note: checking for negative value instead of overflow
335 // so we have a 'sticky' overflow test
336 //
337 // rbx,: method
338 // rcx: invocation counter
339 //
340 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
341 Label done;
342 // Note: In tiered we increment either counters in MethodCounters* or in MDO
343 // depending if we're profiling or not.
344 if (TieredCompilation) {
345 int increment = InvocationCounter::count_increment;
346 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
347 Label no_mdo;
348 if (ProfileInterpreter) {
349 // Are we profiling?
350 __ movptr(rax, Address(rbx, Method::method_data_offset()));
351 __ testptr(rax, rax);
352 __ jccb(Assembler::zero, no_mdo);
353 // Increment counter in the MDO
354 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
355 in_bytes(InvocationCounter::counter_offset()));
356 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
357 __ jmp(done);
358 }
359 __ bind(no_mdo);
360 // Increment counter in MethodCounters
361 const Address invocation_counter(rax,
362 MethodCounters::invocation_counter_offset() +
363 InvocationCounter::counter_offset());
365 __ get_method_counters(rbx, rax, done);
366 __ increment_mask_and_jump(invocation_counter, increment, mask,
367 rcx, false, Assembler::zero, overflow);
368 __ bind(done);
369 } else {
370 const Address backedge_counter (rax,
371 MethodCounters::backedge_counter_offset() +
372 InvocationCounter::counter_offset());
373 const Address invocation_counter(rax,
374 MethodCounters::invocation_counter_offset() +
375 InvocationCounter::counter_offset());
377 __ get_method_counters(rbx, rax, done);
379 if (ProfileInterpreter) {
380 __ incrementl(Address(rax,
381 MethodCounters::interpreter_invocation_counter_offset()));
382 }
384 // Update standard invocation counters
385 __ movl(rcx, invocation_counter);
386 __ incrementl(rcx, InvocationCounter::count_increment);
387 __ movl(invocation_counter, rcx); // save invocation count
389 __ movl(rax, backedge_counter); // load backedge counter
390 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
392 __ addl(rcx, rax); // add both counters
394 // profile_method is non-null only for interpreted method so
395 // profile_method != NULL == !native_call
396 // BytecodeInterpreter only calls for native so code is elided.
398 if (ProfileInterpreter && profile_method != NULL) {
399 // Test to see if we should create a method data oop
400 __ cmp32(rcx,
401 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
402 __ jcc(Assembler::less, *profile_method_continue);
404 // if no method data exists, go to profile_method
405 __ test_method_data_pointer(rax, *profile_method);
406 }
408 __ cmp32(rcx,
409 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
410 __ jcc(Assembler::aboveEqual, *overflow);
411 __ bind(done);
412 }
413 }
415 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
417 // Asm interpreter on entry
418 // rdi - locals
419 // rsi - bcp
420 // rbx, - method
421 // rdx - cpool
422 // rbp, - interpreter frame
424 // C++ interpreter on entry
425 // rsi - new interpreter state pointer
426 // rbp - interpreter frame pointer
427 // rbx - method
429 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
430 // rbx, - method
431 // rcx - rcvr (assuming there is one)
432 // top of stack return address of interpreter caller
433 // rsp - sender_sp
435 // C++ interpreter only
436 // rsi - previous interpreter state pointer
438 // InterpreterRuntime::frequency_counter_overflow takes one argument
439 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
440 // The call returns the address of the verified entry point for the method or NULL
441 // if the compilation did not complete (either went background or bailed out).
442 __ movptr(rax, (intptr_t)false);
443 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
445 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
447 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
448 // and jump to the interpreted entry.
449 __ jmp(*do_continue, relocInfo::none);
451 }
453 void InterpreterGenerator::generate_stack_overflow_check(void) {
454 // see if we've got enough room on the stack for locals plus overhead.
455 // the expression stack grows down incrementally, so the normal guard
456 // page mechanism will work for that.
457 //
458 // Registers live on entry:
459 //
460 // Asm interpreter
461 // rdx: number of additional locals this frame needs (what we must check)
462 // rbx,: Method*
464 // destroyed on exit
465 // rax,
467 // NOTE: since the additional locals are also always pushed (wasn't obvious in
468 // generate_method_entry) so the guard should work for them too.
469 //
471 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
472 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
474 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
475 // be sure to change this if you add/subtract anything to/from the overhead area
476 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
478 const int page_size = os::vm_page_size();
480 Label after_frame_check;
482 // see if the frame is greater than one page in size. If so,
483 // then we need to verify there is enough stack space remaining
484 // for the additional locals.
485 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
486 __ jcc(Assembler::belowEqual, after_frame_check);
488 // compute rsp as if this were going to be the last frame on
489 // the stack before the red zone
491 Label after_frame_check_pop;
493 __ push(rsi);
495 const Register thread = rsi;
497 __ get_thread(thread);
499 const Address stack_base(thread, Thread::stack_base_offset());
500 const Address stack_size(thread, Thread::stack_size_offset());
502 // locals + overhead, in bytes
503 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
505 #ifdef ASSERT
506 Label stack_base_okay, stack_size_okay;
507 // verify that thread stack base is non-zero
508 __ cmpptr(stack_base, (int32_t)NULL_WORD);
509 __ jcc(Assembler::notEqual, stack_base_okay);
510 __ stop("stack base is zero");
511 __ bind(stack_base_okay);
512 // verify that thread stack size is non-zero
513 __ cmpptr(stack_size, 0);
514 __ jcc(Assembler::notEqual, stack_size_okay);
515 __ stop("stack size is zero");
516 __ bind(stack_size_okay);
517 #endif
519 // Add stack base to locals and subtract stack size
520 __ addptr(rax, stack_base);
521 __ subptr(rax, stack_size);
523 // Use the maximum number of pages we might bang.
524 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
525 (StackRedPages+StackYellowPages);
526 __ addptr(rax, max_pages * page_size);
528 // check against the current stack bottom
529 __ cmpptr(rsp, rax);
530 __ jcc(Assembler::above, after_frame_check_pop);
532 __ pop(rsi); // get saved bcp / (c++ prev state ).
534 // Restore sender's sp as SP. This is necessary if the sender's
535 // frame is an extended compiled frame (see gen_c2i_adapter())
536 // and safer anyway in case of JSR292 adaptations.
538 __ pop(rax); // return address must be moved if SP is changed
539 __ mov(rsp, rsi);
540 __ push(rax);
542 // Note: the restored frame is not necessarily interpreted.
543 // Use the shared runtime version of the StackOverflowError.
544 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
545 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
546 // all done with frame size check
547 __ bind(after_frame_check_pop);
548 __ pop(rsi);
550 __ bind(after_frame_check);
551 }
553 // Allocate monitor and lock method (asm interpreter)
554 // rbx, - Method*
555 //
556 void InterpreterGenerator::lock_method(void) {
557 // synchronize method
558 const Address access_flags (rbx, Method::access_flags_offset());
559 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
560 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
562 #ifdef ASSERT
563 { Label L;
564 __ movl(rax, access_flags);
565 __ testl(rax, JVM_ACC_SYNCHRONIZED);
566 __ jcc(Assembler::notZero, L);
567 __ stop("method doesn't need synchronization");
568 __ bind(L);
569 }
570 #endif // ASSERT
571 // get synchronization object
572 { Label done;
573 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
574 __ movl(rax, access_flags);
575 __ testl(rax, JVM_ACC_STATIC);
576 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
577 __ jcc(Assembler::zero, done);
578 __ movptr(rax, Address(rbx, Method::const_offset()));
579 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
580 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
581 __ movptr(rax, Address(rax, mirror_offset));
582 __ bind(done);
583 }
584 // add space for monitor & lock
585 __ subptr(rsp, entry_size); // add space for a monitor entry
586 __ movptr(monitor_block_top, rsp); // set new monitor block top
587 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
588 __ mov(rdx, rsp); // object address
589 __ lock_object(rdx);
590 }
592 //
593 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
594 // and for native methods hence the shared code.
596 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
597 // initialize fixed part of activation frame
598 __ push(rax); // save return address
599 __ enter(); // save old & set new rbp,
602 __ push(rsi); // set sender sp
603 __ push((int32_t)NULL_WORD); // leave last_sp as null
604 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
605 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
606 __ push(rbx); // save Method*
607 if (ProfileInterpreter) {
608 Label method_data_continue;
609 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
610 __ testptr(rdx, rdx);
611 __ jcc(Assembler::zero, method_data_continue);
612 __ addptr(rdx, in_bytes(MethodData::data_offset()));
613 __ bind(method_data_continue);
614 __ push(rdx); // set the mdp (method data pointer)
615 } else {
616 __ push(0);
617 }
619 __ movptr(rdx, Address(rbx, Method::const_offset()));
620 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
621 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
622 __ push(rdx); // set constant pool cache
623 __ push(rdi); // set locals pointer
624 if (native_call) {
625 __ push(0); // no bcp
626 } else {
627 __ push(rsi); // set bcp
628 }
629 __ push(0); // reserve word for pointer to expression stack bottom
630 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
631 }
633 // End of helpers
635 //
636 // Various method entries
637 //------------------------------------------------------------------------------------------------------------------------
638 //
639 //
641 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
643 address InterpreterGenerator::generate_accessor_entry(void) {
645 // rbx,: Method*
646 // rcx: receiver (preserve for slow entry into asm interpreter)
648 // rsi: senderSP must preserved for slow path, set SP to it on fast path
650 address entry_point = __ pc();
651 Label xreturn_path;
653 // do fastpath for resolved accessor methods
654 if (UseFastAccessorMethods) {
655 Label slow_path;
656 // If we need a safepoint check, generate full interpreter entry.
657 ExternalAddress state(SafepointSynchronize::address_of_state());
658 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
659 SafepointSynchronize::_not_synchronized);
661 __ jcc(Assembler::notEqual, slow_path);
662 // ASM/C++ Interpreter
663 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
664 // Note: We can only use this code if the getfield has been resolved
665 // and if we don't have a null-pointer exception => check for
666 // these conditions first and use slow path if necessary.
667 // rbx,: method
668 // rcx: receiver
669 __ movptr(rax, Address(rsp, wordSize));
671 // check if local 0 != NULL and read field
672 __ testptr(rax, rax);
673 __ jcc(Assembler::zero, slow_path);
675 // read first instruction word and extract bytecode @ 1 and index @ 2
676 __ movptr(rdx, Address(rbx, Method::const_offset()));
677 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
678 __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
679 // Shift codes right to get the index on the right.
680 // The bytecode fetched looks like <index><0xb4><0x2a>
681 __ shrl(rdx, 2*BitsPerByte);
682 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
683 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
685 // rax,: local 0
686 // rbx,: method
687 // rcx: receiver - do not destroy since it is needed for slow path!
688 // rcx: scratch
689 // rdx: constant pool cache index
690 // rdi: constant pool cache
691 // rsi: sender sp
693 // check if getfield has been resolved and read constant pool cache entry
694 // check the validity of the cache entry by testing whether _indices field
695 // contains Bytecode::_getfield in b1 byte.
696 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
697 __ movl(rcx,
698 Address(rdi,
699 rdx,
700 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
701 __ shrl(rcx, 2*BitsPerByte);
702 __ andl(rcx, 0xFF);
703 __ cmpl(rcx, Bytecodes::_getfield);
704 __ jcc(Assembler::notEqual, slow_path);
706 // Note: constant pool entry is not valid before bytecode is resolved
707 __ movptr(rcx,
708 Address(rdi,
709 rdx,
710 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
711 __ movl(rdx,
712 Address(rdi,
713 rdx,
714 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
716 Label notByte, notShort, notChar;
717 const Address field_address (rax, rcx, Address::times_1);
719 // Need to differentiate between igetfield, agetfield, bgetfield etc.
720 // because they are different sizes.
721 // Use the type from the constant pool cache
722 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
723 // Make sure we don't need to mask rdx after the above shift
724 ConstantPoolCacheEntry::verify_tos_state_shift();
725 __ cmpl(rdx, btos);
726 __ jcc(Assembler::notEqual, notByte);
727 __ load_signed_byte(rax, field_address);
728 __ jmp(xreturn_path);
730 __ bind(notByte);
731 __ cmpl(rdx, stos);
732 __ jcc(Assembler::notEqual, notShort);
733 __ load_signed_short(rax, field_address);
734 __ jmp(xreturn_path);
736 __ bind(notShort);
737 __ cmpl(rdx, ctos);
738 __ jcc(Assembler::notEqual, notChar);
739 __ load_unsigned_short(rax, field_address);
740 __ jmp(xreturn_path);
742 __ bind(notChar);
743 #ifdef ASSERT
744 Label okay;
745 __ cmpl(rdx, atos);
746 __ jcc(Assembler::equal, okay);
747 __ cmpl(rdx, itos);
748 __ jcc(Assembler::equal, okay);
749 __ stop("what type is this?");
750 __ bind(okay);
751 #endif // ASSERT
752 // All the rest are a 32 bit wordsize
753 // This is ok for now. Since fast accessors should be going away
754 __ movptr(rax, field_address);
756 __ bind(xreturn_path);
758 // _ireturn/_areturn
759 __ pop(rdi); // get return address
760 __ mov(rsp, rsi); // set sp to sender sp
761 __ jmp(rdi);
763 // generate a vanilla interpreter entry as the slow path
764 __ bind(slow_path);
766 (void) generate_normal_entry(false);
767 return entry_point;
768 }
769 return NULL;
771 }
773 // Method entry for java.lang.ref.Reference.get.
774 address InterpreterGenerator::generate_Reference_get_entry(void) {
775 #if INCLUDE_ALL_GCS
776 // Code: _aload_0, _getfield, _areturn
777 // parameter size = 1
778 //
779 // The code that gets generated by this routine is split into 2 parts:
780 // 1. The "intrinsified" code for G1 (or any SATB based GC),
781 // 2. The slow path - which is an expansion of the regular method entry.
782 //
783 // Notes:-
784 // * In the G1 code we do not check whether we need to block for
785 // a safepoint. If G1 is enabled then we must execute the specialized
786 // code for Reference.get (except when the Reference object is null)
787 // so that we can log the value in the referent field with an SATB
788 // update buffer.
789 // If the code for the getfield template is modified so that the
790 // G1 pre-barrier code is executed when the current method is
791 // Reference.get() then going through the normal method entry
792 // will be fine.
793 // * The G1 code below can, however, check the receiver object (the instance
794 // of java.lang.Reference) and jump to the slow path if null. If the
795 // Reference object is null then we obviously cannot fetch the referent
796 // and so we don't need to call the G1 pre-barrier. Thus we can use the
797 // regular method entry code to generate the NPE.
798 //
799 // This code is based on generate_accessor_enty.
801 // rbx,: Method*
802 // rcx: receiver (preserve for slow entry into asm interpreter)
804 // rsi: senderSP must preserved for slow path, set SP to it on fast path
806 address entry = __ pc();
808 const int referent_offset = java_lang_ref_Reference::referent_offset;
809 guarantee(referent_offset > 0, "referent offset not initialized");
811 if (UseG1GC) {
812 Label slow_path;
814 // Check if local 0 != NULL
815 // If the receiver is null then it is OK to jump to the slow path.
816 __ movptr(rax, Address(rsp, wordSize));
817 __ testptr(rax, rax);
818 __ jcc(Assembler::zero, slow_path);
820 // rax: local 0 (must be preserved across the G1 barrier call)
821 //
822 // rbx: method (at this point it's scratch)
823 // rcx: receiver (at this point it's scratch)
824 // rdx: scratch
825 // rdi: scratch
826 //
827 // rsi: sender sp
829 // Preserve the sender sp in case the pre-barrier
830 // calls the runtime
831 __ push(rsi);
833 // Load the value of the referent field.
834 const Address field_address(rax, referent_offset);
835 __ movptr(rax, field_address);
837 // Generate the G1 pre-barrier code to log the value of
838 // the referent field in an SATB buffer.
839 __ get_thread(rcx);
840 __ g1_write_barrier_pre(noreg /* obj */,
841 rax /* pre_val */,
842 rcx /* thread */,
843 rbx /* tmp */,
844 true /* tosca_save */,
845 true /* expand_call */);
847 // _areturn
848 __ pop(rsi); // get sender sp
849 __ pop(rdi); // get return address
850 __ mov(rsp, rsi); // set sp to sender sp
851 __ jmp(rdi);
853 __ bind(slow_path);
854 (void) generate_normal_entry(false);
856 return entry;
857 }
858 #endif // INCLUDE_ALL_GCS
860 // If G1 is not enabled then attempt to go through the accessor entry point
861 // Reference.get is an accessor
862 return generate_accessor_entry();
863 }
865 /**
866 * Method entry for static native methods:
867 * int java.util.zip.CRC32.update(int crc, int b)
868 */
869 address InterpreterGenerator::generate_CRC32_update_entry() {
870 if (UseCRC32Intrinsics) {
871 address entry = __ pc();
873 // rbx,: Method*
874 // rsi: senderSP must preserved for slow path, set SP to it on fast path
875 // rdx: scratch
876 // rdi: scratch
878 Label slow_path;
879 // If we need a safepoint check, generate full interpreter entry.
880 ExternalAddress state(SafepointSynchronize::address_of_state());
881 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
882 SafepointSynchronize::_not_synchronized);
883 __ jcc(Assembler::notEqual, slow_path);
885 // We don't generate local frame and don't align stack because
886 // we call stub code and there is no safepoint on this path.
888 // Load parameters
889 const Register crc = rax; // crc
890 const Register val = rdx; // source java byte value
891 const Register tbl = rdi; // scratch
893 // Arguments are reversed on java expression stack
894 __ movl(val, Address(rsp, wordSize)); // byte value
895 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
897 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
898 __ notl(crc); // ~crc
899 __ update_byte_crc32(crc, val, tbl);
900 __ notl(crc); // ~crc
901 // result in rax
903 // _areturn
904 __ pop(rdi); // get return address
905 __ mov(rsp, rsi); // set sp to sender sp
906 __ jmp(rdi);
908 // generate a vanilla native entry as the slow path
909 __ bind(slow_path);
911 (void) generate_native_entry(false);
913 return entry;
914 }
915 return generate_native_entry(false);
916 }
918 /**
919 * Method entry for static native methods:
920 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
921 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
922 */
923 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
924 if (UseCRC32Intrinsics) {
925 address entry = __ pc();
927 // rbx,: Method*
928 // rsi: senderSP must preserved for slow path, set SP to it on fast path
929 // rdx: scratch
930 // rdi: scratch
932 Label slow_path;
933 // If we need a safepoint check, generate full interpreter entry.
934 ExternalAddress state(SafepointSynchronize::address_of_state());
935 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
936 SafepointSynchronize::_not_synchronized);
937 __ jcc(Assembler::notEqual, slow_path);
939 // We don't generate local frame and don't align stack because
940 // we call stub code and there is no safepoint on this path.
942 // Load parameters
943 const Register crc = rax; // crc
944 const Register buf = rdx; // source java byte array address
945 const Register len = rdi; // length
947 // Arguments are reversed on java expression stack
948 __ movl(len, Address(rsp, wordSize)); // Length
949 // Calculate address of start element
950 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
951 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
952 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
953 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
954 } else {
955 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
956 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
957 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
958 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
959 }
961 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
962 // result in rax
964 // _areturn
965 __ pop(rdi); // get return address
966 __ mov(rsp, rsi); // set sp to sender sp
967 __ jmp(rdi);
969 // generate a vanilla native entry as the slow path
970 __ bind(slow_path);
972 (void) generate_native_entry(false);
974 return entry;
975 }
976 return generate_native_entry(false);
977 }
979 //
980 // Interpreter stub for calling a native method. (asm interpreter)
981 // This sets up a somewhat different looking stack for calling the native method
982 // than the typical interpreter frame setup.
983 //
985 address InterpreterGenerator::generate_native_entry(bool synchronized) {
986 // determine code generation flags
987 bool inc_counter = UseCompiler || CountCompiledCalls;
989 // rbx,: Method*
990 // rsi: sender sp
991 // rsi: previous interpreter state (C++ interpreter) must preserve
992 address entry_point = __ pc();
994 const Address constMethod (rbx, Method::const_offset());
995 const Address access_flags (rbx, Method::access_flags_offset());
996 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
998 // get parameter size (always needed)
999 __ movptr(rcx, constMethod);
1000 __ load_unsigned_short(rcx, size_of_parameters);
1002 // native calls don't need the stack size check since they have no expression stack
1003 // and the arguments are already on the stack and we only add a handful of words
1004 // to the stack
1006 // rbx,: Method*
1007 // rcx: size of parameters
1008 // rsi: sender sp
1010 __ pop(rax); // get return address
1011 // for natives the size of locals is zero
1013 // compute beginning of parameters (rdi)
1014 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1017 // add 2 zero-initialized slots for native calls
1018 // NULL result handler
1019 __ push((int32_t)NULL_WORD);
1020 // NULL oop temp (mirror or jni oop result)
1021 __ push((int32_t)NULL_WORD);
1023 // initialize fixed part of activation frame
1024 generate_fixed_frame(true);
1026 // make sure method is native & not abstract
1027 #ifdef ASSERT
1028 __ movl(rax, access_flags);
1029 {
1030 Label L;
1031 __ testl(rax, JVM_ACC_NATIVE);
1032 __ jcc(Assembler::notZero, L);
1033 __ stop("tried to execute non-native method as native");
1034 __ bind(L);
1035 }
1036 { Label L;
1037 __ testl(rax, JVM_ACC_ABSTRACT);
1038 __ jcc(Assembler::zero, L);
1039 __ stop("tried to execute abstract method in interpreter");
1040 __ bind(L);
1041 }
1042 #endif
1044 // Since at this point in the method invocation the exception handler
1045 // would try to exit the monitor of synchronized methods which hasn't
1046 // been entered yet, we set the thread local variable
1047 // _do_not_unlock_if_synchronized to true. The remove_activation will
1048 // check this flag.
1050 __ get_thread(rax);
1051 const Address do_not_unlock_if_synchronized(rax,
1052 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1053 __ movbool(do_not_unlock_if_synchronized, true);
1055 // increment invocation count & check for overflow
1056 Label invocation_counter_overflow;
1057 if (inc_counter) {
1058 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1059 }
1061 Label continue_after_compile;
1062 __ bind(continue_after_compile);
1064 bang_stack_shadow_pages(true);
1066 // reset the _do_not_unlock_if_synchronized flag
1067 __ get_thread(rax);
1068 __ movbool(do_not_unlock_if_synchronized, false);
1070 // check for synchronized methods
1071 // Must happen AFTER invocation_counter check and stack overflow check,
1072 // so method is not locked if overflows.
1073 //
1074 if (synchronized) {
1075 lock_method();
1076 } else {
1077 // no synchronization necessary
1078 #ifdef ASSERT
1079 { Label L;
1080 __ movl(rax, access_flags);
1081 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1082 __ jcc(Assembler::zero, L);
1083 __ stop("method needs synchronization");
1084 __ bind(L);
1085 }
1086 #endif
1087 }
1089 // start execution
1090 #ifdef ASSERT
1091 { Label L;
1092 const Address monitor_block_top (rbp,
1093 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1094 __ movptr(rax, monitor_block_top);
1095 __ cmpptr(rax, rsp);
1096 __ jcc(Assembler::equal, L);
1097 __ stop("broken stack frame setup in interpreter");
1098 __ bind(L);
1099 }
1100 #endif
1102 // jvmti/dtrace support
1103 __ notify_method_entry();
1105 // work registers
1106 const Register method = rbx;
1107 const Register thread = rdi;
1108 const Register t = rcx;
1110 // allocate space for parameters
1111 __ get_method(method);
1112 __ movptr(t, Address(method, Method::const_offset()));
1113 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1115 __ shlptr(t, Interpreter::logStackElementSize);
1116 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
1117 __ subptr(rsp, t);
1118 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1120 // get signature handler
1121 { Label L;
1122 __ movptr(t, Address(method, Method::signature_handler_offset()));
1123 __ testptr(t, t);
1124 __ jcc(Assembler::notZero, L);
1125 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1126 __ get_method(method);
1127 __ movptr(t, Address(method, Method::signature_handler_offset()));
1128 __ bind(L);
1129 }
1131 // call signature handler
1132 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1133 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1134 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1135 // The generated handlers do not touch RBX (the method oop).
1136 // However, large signatures cannot be cached and are generated
1137 // each time here. The slow-path generator will blow RBX
1138 // sometime, so we must reload it after the call.
1139 __ call(t);
1140 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1142 // result handler is in rax,
1143 // set result handler
1144 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1146 // pass mirror handle if static call
1147 { Label L;
1148 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1149 __ movl(t, Address(method, Method::access_flags_offset()));
1150 __ testl(t, JVM_ACC_STATIC);
1151 __ jcc(Assembler::zero, L);
1152 // get mirror
1153 __ movptr(t, Address(method, Method:: const_offset()));
1154 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1155 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1156 __ movptr(t, Address(t, mirror_offset));
1157 // copy mirror into activation frame
1158 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1159 // pass handle to mirror
1160 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1161 __ movptr(Address(rsp, wordSize), t);
1162 __ bind(L);
1163 }
1165 // get native function entry point
1166 { Label L;
1167 __ movptr(rax, Address(method, Method::native_function_offset()));
1168 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1169 __ cmpptr(rax, unsatisfied.addr());
1170 __ jcc(Assembler::notEqual, L);
1171 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1172 __ get_method(method);
1173 __ movptr(rax, Address(method, Method::native_function_offset()));
1174 __ bind(L);
1175 }
1177 // pass JNIEnv
1178 __ get_thread(thread);
1179 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1180 __ movptr(Address(rsp, 0), t);
1182 // set_last_Java_frame_before_call
1183 // It is enough that the pc()
1184 // points into the right code segment. It does not have to be the correct return pc.
1185 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1187 // change thread state
1188 #ifdef ASSERT
1189 { Label L;
1190 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1191 __ cmpl(t, _thread_in_Java);
1192 __ jcc(Assembler::equal, L);
1193 __ stop("Wrong thread state in native stub");
1194 __ bind(L);
1195 }
1196 #endif
1198 // Change state to native
1199 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1200 __ call(rax);
1202 // result potentially in rdx:rax or ST0
1204 // Verify or restore cpu control state after JNI call
1205 __ restore_cpu_control_state_after_jni();
1207 // save potential result in ST(0) & rdx:rax
1208 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1209 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1210 // It is safe to do this push because state is _thread_in_native and return address will be found
1211 // via _last_native_pc and not via _last_jave_sp
1213 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1214 // If the order changes or anything else is added to the stack the code in
1215 // interpreter_frame_result will have to be changed.
1217 { Label L;
1218 Label push_double;
1219 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1220 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1221 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1222 float_handler.addr());
1223 __ jcc(Assembler::equal, push_double);
1224 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1225 double_handler.addr());
1226 __ jcc(Assembler::notEqual, L);
1227 __ bind(push_double);
1228 __ push(dtos);
1229 __ bind(L);
1230 }
1231 __ push(ltos);
1233 // change thread state
1234 __ get_thread(thread);
1235 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1236 if(os::is_MP()) {
1237 if (UseMembar) {
1238 // Force this write out before the read below
1239 __ membar(Assembler::Membar_mask_bits(
1240 Assembler::LoadLoad | Assembler::LoadStore |
1241 Assembler::StoreLoad | Assembler::StoreStore));
1242 } else {
1243 // Write serialization page so VM thread can do a pseudo remote membar.
1244 // We use the current thread pointer to calculate a thread specific
1245 // offset to write to within the page. This minimizes bus traffic
1246 // due to cache line collision.
1247 __ serialize_memory(thread, rcx);
1248 }
1249 }
1251 if (AlwaysRestoreFPU) {
1252 // Make sure the control word is correct.
1253 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1254 }
1256 // check for safepoint operation in progress and/or pending suspend requests
1257 { Label Continue;
1259 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1260 SafepointSynchronize::_not_synchronized);
1262 Label L;
1263 __ jcc(Assembler::notEqual, L);
1264 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1265 __ jcc(Assembler::equal, Continue);
1266 __ bind(L);
1268 // Don't use call_VM as it will see a possible pending exception and forward it
1269 // and never return here preventing us from clearing _last_native_pc down below.
1270 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1271 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1272 // by hand.
1273 //
1274 __ push(thread);
1275 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1276 JavaThread::check_special_condition_for_native_trans)));
1277 __ increment(rsp, wordSize);
1278 __ get_thread(thread);
1280 __ bind(Continue);
1281 }
1283 // change thread state
1284 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1286 __ reset_last_Java_frame(thread, true, true);
1288 // reset handle block
1289 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1290 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1292 // If result was an oop then unbox and save it in the frame
1293 { Label L;
1294 Label no_oop, store_result;
1295 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1296 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1297 handler.addr());
1298 __ jcc(Assembler::notEqual, no_oop);
1299 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1300 __ pop(ltos);
1301 __ testptr(rax, rax);
1302 __ jcc(Assembler::zero, store_result);
1303 // unbox
1304 __ movptr(rax, Address(rax, 0));
1305 __ bind(store_result);
1306 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1307 // keep stack depth as expected by pushing oop which will eventually be discarded
1308 __ push(ltos);
1309 __ bind(no_oop);
1310 }
1312 {
1313 Label no_reguard;
1314 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1315 __ jcc(Assembler::notEqual, no_reguard);
1317 __ pusha();
1318 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1319 __ popa();
1321 __ bind(no_reguard);
1322 }
1324 // restore rsi to have legal interpreter frame,
1325 // i.e., bci == 0 <=> rsi == code_base()
1326 // Can't call_VM until bcp is within reasonable.
1327 __ get_method(method); // method is junk from thread_in_native to now.
1328 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod*
1329 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
1331 // handle exceptions (exception handling will handle unlocking!)
1332 { Label L;
1333 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1334 __ jcc(Assembler::zero, L);
1335 // Note: At some point we may want to unify this with the code used in call_VM_base();
1336 // i.e., we should use the StubRoutines::forward_exception code. For now this
1337 // doesn't work here because the rsp is not correctly set at this point.
1338 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1339 __ should_not_reach_here();
1340 __ bind(L);
1341 }
1343 // do unlocking if necessary
1344 { Label L;
1345 __ movl(t, Address(method, Method::access_flags_offset()));
1346 __ testl(t, JVM_ACC_SYNCHRONIZED);
1347 __ jcc(Assembler::zero, L);
1348 // the code below should be shared with interpreter macro assembler implementation
1349 { Label unlock;
1350 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1351 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1352 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1354 __ lea(rdx, monitor); // address of first monitor
1356 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1357 __ testptr(t, t);
1358 __ jcc(Assembler::notZero, unlock);
1360 // Entry already unlocked, need to throw exception
1361 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1362 __ should_not_reach_here();
1364 __ bind(unlock);
1365 __ unlock_object(rdx);
1366 }
1367 __ bind(L);
1368 }
1370 // jvmti/dtrace support
1371 // Note: This must happen _after_ handling/throwing any exceptions since
1372 // the exception handler code notifies the runtime of method exits
1373 // too. If this happens before, method entry/exit notifications are
1374 // not properly paired (was bug - gri 11/22/99).
1375 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1377 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1378 __ pop(ltos);
1379 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1380 __ call(t);
1382 // remove activation
1383 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1384 __ leave(); // remove frame anchor
1385 __ pop(rdi); // get return address
1386 __ mov(rsp, t); // set sp to sender sp
1387 __ jmp(rdi);
1389 if (inc_counter) {
1390 // Handle overflow of counter and compile method
1391 __ bind(invocation_counter_overflow);
1392 generate_counter_overflow(&continue_after_compile);
1393 }
1395 return entry_point;
1396 }
1398 //
1399 // Generic interpreted method entry to (asm) interpreter
1400 //
1401 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1402 // determine code generation flags
1403 bool inc_counter = UseCompiler || CountCompiledCalls;
1405 // rbx,: Method*
1406 // rsi: sender sp
1407 address entry_point = __ pc();
1409 const Address constMethod (rbx, Method::const_offset());
1410 const Address access_flags (rbx, Method::access_flags_offset());
1411 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1412 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
1414 // get parameter size (always needed)
1415 __ movptr(rdx, constMethod);
1416 __ load_unsigned_short(rcx, size_of_parameters);
1418 // rbx,: Method*
1419 // rcx: size of parameters
1421 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1423 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1424 __ subl(rdx, rcx); // rdx = no. of additional locals
1426 // see if we've got enough room on the stack for locals plus overhead.
1427 generate_stack_overflow_check();
1429 // get return address
1430 __ pop(rax);
1432 // compute beginning of parameters (rdi)
1433 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1435 // rdx - # of additional locals
1436 // allocate space for locals
1437 // explicitly initialize locals
1438 {
1439 Label exit, loop;
1440 __ testl(rdx, rdx);
1441 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1442 __ bind(loop);
1443 __ push((int32_t)NULL_WORD); // initialize local variables
1444 __ decrement(rdx); // until everything initialized
1445 __ jcc(Assembler::greater, loop);
1446 __ bind(exit);
1447 }
1449 // initialize fixed part of activation frame
1450 generate_fixed_frame(false);
1452 // make sure method is not native & not abstract
1453 #ifdef ASSERT
1454 __ movl(rax, access_flags);
1455 {
1456 Label L;
1457 __ testl(rax, JVM_ACC_NATIVE);
1458 __ jcc(Assembler::zero, L);
1459 __ stop("tried to execute native method as non-native");
1460 __ bind(L);
1461 }
1462 { Label L;
1463 __ testl(rax, JVM_ACC_ABSTRACT);
1464 __ jcc(Assembler::zero, L);
1465 __ stop("tried to execute abstract method in interpreter");
1466 __ bind(L);
1467 }
1468 #endif
1470 // Since at this point in the method invocation the exception handler
1471 // would try to exit the monitor of synchronized methods which hasn't
1472 // been entered yet, we set the thread local variable
1473 // _do_not_unlock_if_synchronized to true. The remove_activation will
1474 // check this flag.
1476 __ get_thread(rax);
1477 const Address do_not_unlock_if_synchronized(rax,
1478 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1479 __ movbool(do_not_unlock_if_synchronized, true);
1481 __ profile_parameters_type(rax, rcx, rdx);
1482 // increment invocation count & check for overflow
1483 Label invocation_counter_overflow;
1484 Label profile_method;
1485 Label profile_method_continue;
1486 if (inc_counter) {
1487 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1488 if (ProfileInterpreter) {
1489 __ bind(profile_method_continue);
1490 }
1491 }
1492 Label continue_after_compile;
1493 __ bind(continue_after_compile);
1495 bang_stack_shadow_pages(false);
1497 // reset the _do_not_unlock_if_synchronized flag
1498 __ get_thread(rax);
1499 __ movbool(do_not_unlock_if_synchronized, false);
1501 // check for synchronized methods
1502 // Must happen AFTER invocation_counter check and stack overflow check,
1503 // so method is not locked if overflows.
1504 //
1505 if (synchronized) {
1506 // Allocate monitor and lock method
1507 lock_method();
1508 } else {
1509 // no synchronization necessary
1510 #ifdef ASSERT
1511 { Label L;
1512 __ movl(rax, access_flags);
1513 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1514 __ jcc(Assembler::zero, L);
1515 __ stop("method needs synchronization");
1516 __ bind(L);
1517 }
1518 #endif
1519 }
1521 // start execution
1522 #ifdef ASSERT
1523 { Label L;
1524 const Address monitor_block_top (rbp,
1525 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1526 __ movptr(rax, monitor_block_top);
1527 __ cmpptr(rax, rsp);
1528 __ jcc(Assembler::equal, L);
1529 __ stop("broken stack frame setup in interpreter");
1530 __ bind(L);
1531 }
1532 #endif
1534 // jvmti support
1535 __ notify_method_entry();
1537 __ dispatch_next(vtos);
1539 // invocation counter overflow
1540 if (inc_counter) {
1541 if (ProfileInterpreter) {
1542 // We have decided to profile this method in the interpreter
1543 __ bind(profile_method);
1544 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1545 __ set_method_data_pointer_for_bcp();
1546 __ get_method(rbx);
1547 __ jmp(profile_method_continue);
1548 }
1549 // Handle overflow of counter and compile method
1550 __ bind(invocation_counter_overflow);
1551 generate_counter_overflow(&continue_after_compile);
1552 }
1554 return entry_point;
1555 }
1557 //------------------------------------------------------------------------------------------------------------------------
1558 // Entry points
1559 //
1560 // Here we generate the various kind of entries into the interpreter.
1561 // The two main entry type are generic bytecode methods and native call method.
1562 // These both come in synchronized and non-synchronized versions but the
1563 // frame layout they create is very similar. The other method entry
1564 // types are really just special purpose entries that are really entry
1565 // and interpretation all in one. These are for trivial methods like
1566 // accessor, empty, or special math methods.
1567 //
1568 // When control flow reaches any of the entry types for the interpreter
1569 // the following holds ->
1570 //
1571 // Arguments:
1572 //
1573 // rbx,: Method*
1574 // rcx: receiver
1575 //
1576 //
1577 // Stack layout immediately at entry
1578 //
1579 // [ return address ] <--- rsp
1580 // [ parameter n ]
1581 // ...
1582 // [ parameter 1 ]
1583 // [ expression stack ] (caller's java expression stack)
1585 // Assuming that we don't go to one of the trivial specialized
1586 // entries the stack will look like below when we are ready to execute
1587 // the first bytecode (or call the native routine). The register usage
1588 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1589 //
1590 // local variables follow incoming parameters immediately; i.e.
1591 // the return address is moved to the end of the locals).
1592 //
1593 // [ monitor entry ] <--- rsp
1594 // ...
1595 // [ monitor entry ]
1596 // [ expr. stack bottom ]
1597 // [ saved rsi ]
1598 // [ current rdi ]
1599 // [ Method* ]
1600 // [ saved rbp, ] <--- rbp,
1601 // [ return address ]
1602 // [ local variable m ]
1603 // ...
1604 // [ local variable 1 ]
1605 // [ parameter n ]
1606 // ...
1607 // [ parameter 1 ] <--- rdi
1609 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1610 // determine code generation flags
1611 bool synchronized = false;
1612 address entry_point = NULL;
1613 InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1615 switch (kind) {
1616 case Interpreter::zerolocals : break;
1617 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1618 case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break;
1619 case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break;
1620 case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break;
1621 case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break;
1622 case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break;
1624 case Interpreter::java_lang_math_sin : // fall thru
1625 case Interpreter::java_lang_math_cos : // fall thru
1626 case Interpreter::java_lang_math_tan : // fall thru
1627 case Interpreter::java_lang_math_abs : // fall thru
1628 case Interpreter::java_lang_math_log : // fall thru
1629 case Interpreter::java_lang_math_log10 : // fall thru
1630 case Interpreter::java_lang_math_sqrt : // fall thru
1631 case Interpreter::java_lang_math_pow : // fall thru
1632 case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break;
1633 case Interpreter::java_lang_ref_reference_get
1634 : entry_point = ig_this->generate_Reference_get_entry(); break;
1635 case Interpreter::java_util_zip_CRC32_update
1636 : entry_point = ig_this->generate_CRC32_update_entry(); break;
1637 case Interpreter::java_util_zip_CRC32_updateBytes
1638 : // fall thru
1639 case Interpreter::java_util_zip_CRC32_updateByteBuffer
1640 : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1641 default:
1642 fatal(err_msg("unexpected method kind: %d", kind));
1643 break;
1644 }
1646 if (entry_point) return entry_point;
1648 return ig_this->generate_normal_entry(synchronized);
1650 }
1652 // These should never be compiled since the interpreter will prefer
1653 // the compiled version to the intrinsic version.
1654 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1655 switch (method_kind(m)) {
1656 case Interpreter::java_lang_math_sin : // fall thru
1657 case Interpreter::java_lang_math_cos : // fall thru
1658 case Interpreter::java_lang_math_tan : // fall thru
1659 case Interpreter::java_lang_math_abs : // fall thru
1660 case Interpreter::java_lang_math_log : // fall thru
1661 case Interpreter::java_lang_math_log10 : // fall thru
1662 case Interpreter::java_lang_math_sqrt : // fall thru
1663 case Interpreter::java_lang_math_pow : // fall thru
1664 case Interpreter::java_lang_math_exp :
1665 return false;
1666 default:
1667 return true;
1668 }
1669 }
1671 // How much stack a method activation needs in words.
1672 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1674 const int stub_code = 4; // see generate_call_stub
1675 // Save space for one monitor to get into the interpreted method in case
1676 // the method is synchronized
1677 int monitor_size = method->is_synchronized() ?
1678 1*frame::interpreter_frame_monitor_size() : 0;
1680 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1681 // be sure to change this if you add/subtract anything to/from the overhead area
1682 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1684 const int method_stack = (method->max_locals() + method->max_stack()) *
1685 Interpreter::stackElementWords;
1686 return overhead_size + method_stack + stub_code;
1687 }
1689 //------------------------------------------------------------------------------------------------------------------------
1690 // Exceptions
1692 void TemplateInterpreterGenerator::generate_throw_exception() {
1693 // Entry point in previous activation (i.e., if the caller was interpreted)
1694 Interpreter::_rethrow_exception_entry = __ pc();
1695 const Register thread = rcx;
1697 // Restore sp to interpreter_frame_last_sp even though we are going
1698 // to empty the expression stack for the exception processing.
1699 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1700 // rax,: exception
1701 // rdx: return address/pc that threw exception
1702 __ restore_bcp(); // rsi points to call/send
1703 __ restore_locals();
1705 // Entry point for exceptions thrown within interpreter code
1706 Interpreter::_throw_exception_entry = __ pc();
1707 // expression stack is undefined here
1708 // rax,: exception
1709 // rsi: exception bcp
1710 __ verify_oop(rax);
1712 // expression stack must be empty before entering the VM in case of an exception
1713 __ empty_expression_stack();
1714 __ empty_FPU_stack();
1715 // find exception handler address and preserve exception oop
1716 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1717 // rax,: exception handler entry point
1718 // rdx: preserved exception oop
1719 // rsi: bcp for exception handler
1720 __ push_ptr(rdx); // push exception which is now the only value on the stack
1721 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1723 // If the exception is not handled in the current frame the frame is removed and
1724 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1725 //
1726 // Note: At this point the bci is still the bxi for the instruction which caused
1727 // the exception and the expression stack is empty. Thus, for any VM calls
1728 // at this point, GC will find a legal oop map (with empty expression stack).
1730 // In current activation
1731 // tos: exception
1732 // rsi: exception bcp
1734 //
1735 // JVMTI PopFrame support
1736 //
1738 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1739 __ empty_expression_stack();
1740 __ empty_FPU_stack();
1741 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1742 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1743 // popframe handling cycles.
1744 __ get_thread(thread);
1745 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1746 __ orl(rdx, JavaThread::popframe_processing_bit);
1747 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1749 {
1750 // Check to see whether we are returning to a deoptimized frame.
1751 // (The PopFrame call ensures that the caller of the popped frame is
1752 // either interpreted or compiled and deoptimizes it if compiled.)
1753 // In this case, we can't call dispatch_next() after the frame is
1754 // popped, but instead must save the incoming arguments and restore
1755 // them after deoptimization has occurred.
1756 //
1757 // Note that we don't compare the return PC against the
1758 // deoptimization blob's unpack entry because of the presence of
1759 // adapter frames in C2.
1760 Label caller_not_deoptimized;
1761 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1762 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1763 __ testl(rax, rax);
1764 __ jcc(Assembler::notZero, caller_not_deoptimized);
1766 // Compute size of arguments for saving when returning to deoptimized caller
1767 __ get_method(rax);
1768 __ movptr(rax, Address(rax, Method::const_offset()));
1769 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1770 __ shlptr(rax, Interpreter::logStackElementSize);
1771 __ restore_locals();
1772 __ subptr(rdi, rax);
1773 __ addptr(rdi, wordSize);
1774 // Save these arguments
1775 __ get_thread(thread);
1776 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1778 __ remove_activation(vtos, rdx,
1779 /* throw_monitor_exception */ false,
1780 /* install_monitor_exception */ false,
1781 /* notify_jvmdi */ false);
1783 // Inform deoptimization that it is responsible for restoring these arguments
1784 __ get_thread(thread);
1785 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1787 // Continue in deoptimization handler
1788 __ jmp(rdx);
1790 __ bind(caller_not_deoptimized);
1791 }
1793 __ remove_activation(vtos, rdx,
1794 /* throw_monitor_exception */ false,
1795 /* install_monitor_exception */ false,
1796 /* notify_jvmdi */ false);
1798 // Finish with popframe handling
1799 // A previous I2C followed by a deoptimization might have moved the
1800 // outgoing arguments further up the stack. PopFrame expects the
1801 // mutations to those outgoing arguments to be preserved and other
1802 // constraints basically require this frame to look exactly as
1803 // though it had previously invoked an interpreted activation with
1804 // no space between the top of the expression stack (current
1805 // last_sp) and the top of stack. Rather than force deopt to
1806 // maintain this kind of invariant all the time we call a small
1807 // fixup routine to move the mutated arguments onto the top of our
1808 // expression stack if necessary.
1809 __ mov(rax, rsp);
1810 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1811 __ get_thread(thread);
1812 // PC must point into interpreter here
1813 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1814 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1815 __ get_thread(thread);
1816 __ reset_last_Java_frame(thread, true, true);
1817 // Restore the last_sp and null it out
1818 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1819 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1821 __ restore_bcp();
1822 __ restore_locals();
1823 // The method data pointer was incremented already during
1824 // call profiling. We have to restore the mdp for the current bcp.
1825 if (ProfileInterpreter) {
1826 __ set_method_data_pointer_for_bcp();
1827 }
1829 // Clear the popframe condition flag
1830 __ get_thread(thread);
1831 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1833 #if INCLUDE_JVMTI
1834 if (EnableInvokeDynamic) {
1835 Label L_done;
1836 const Register local0 = rdi;
1838 __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
1839 __ jcc(Assembler::notEqual, L_done);
1841 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1842 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1844 __ get_method(rdx);
1845 __ movptr(rax, Address(local0, 0));
1846 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
1848 __ testptr(rax, rax);
1849 __ jcc(Assembler::zero, L_done);
1851 __ movptr(Address(rbx, 0), rax);
1852 __ bind(L_done);
1853 }
1854 #endif // INCLUDE_JVMTI
1856 __ dispatch_next(vtos);
1857 // end of PopFrame support
1859 Interpreter::_remove_activation_entry = __ pc();
1861 // preserve exception over this code sequence
1862 __ pop_ptr(rax);
1863 __ get_thread(thread);
1864 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1865 // remove the activation (without doing throws on illegalMonitorExceptions)
1866 __ remove_activation(vtos, rdx, false, true, false);
1867 // restore exception
1868 __ get_thread(thread);
1869 __ get_vm_result(rax, thread);
1871 // Inbetween activations - previous activation type unknown yet
1872 // compute continuation point - the continuation point expects
1873 // the following registers set up:
1874 //
1875 // rax: exception
1876 // rdx: return address/pc that threw exception
1877 // rsp: expression stack of caller
1878 // rbp: rbp, of caller
1879 __ push(rax); // save exception
1880 __ push(rdx); // save return address
1881 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1882 __ mov(rbx, rax); // save exception handler
1883 __ pop(rdx); // restore return address
1884 __ pop(rax); // restore exception
1885 // Note that an "issuing PC" is actually the next PC after the call
1886 __ jmp(rbx); // jump to exception handler of caller
1887 }
1890 //
1891 // JVMTI ForceEarlyReturn support
1892 //
1893 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1894 address entry = __ pc();
1895 const Register thread = rcx;
1897 __ restore_bcp();
1898 __ restore_locals();
1899 __ empty_expression_stack();
1900 __ empty_FPU_stack();
1901 __ load_earlyret_value(state);
1903 __ get_thread(thread);
1904 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1905 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1907 // Clear the earlyret state
1908 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1910 __ remove_activation(state, rsi,
1911 false, /* throw_monitor_exception */
1912 false, /* install_monitor_exception */
1913 true); /* notify_jvmdi */
1914 __ jmp(rsi);
1915 return entry;
1916 } // end of ForceEarlyReturn support
1919 //------------------------------------------------------------------------------------------------------------------------
1920 // Helper for vtos entry point generation
1922 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) {
1923 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1924 Label L;
1925 fep = __ pc(); __ push(ftos); __ jmp(L);
1926 dep = __ pc(); __ push(dtos); __ jmp(L);
1927 lep = __ pc(); __ push(ltos); __ jmp(L);
1928 aep = __ pc(); __ push(atos); __ jmp(L);
1929 bep = cep = sep = // fall through
1930 iep = __ pc(); __ push(itos); // fall through
1931 vep = __ pc(); __ bind(L); // fall through
1932 generate_and_dispatch(t);
1933 }
1935 //------------------------------------------------------------------------------------------------------------------------
1936 // Generation of individual instructions
1938 // helpers for generate_and_dispatch
1942 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1943 : TemplateInterpreterGenerator(code) {
1944 generate_all(); // down here so it can be "virtual"
1945 }
1947 //------------------------------------------------------------------------------------------------------------------------
1949 // Non-product code
1950 #ifndef PRODUCT
1951 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1952 address entry = __ pc();
1954 // prepare expression stack
1955 __ pop(rcx); // pop return address so expression stack is 'pure'
1956 __ push(state); // save tosca
1958 // pass tosca registers as arguments & call tracer
1959 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1960 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1961 __ pop(state); // restore tosca
1963 // return
1964 __ jmp(rcx);
1966 return entry;
1967 }
1970 void TemplateInterpreterGenerator::count_bytecode() {
1971 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1972 }
1975 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1976 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1977 }
1980 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1981 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1982 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1983 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1984 ExternalAddress table((address) BytecodePairHistogram::_counters);
1985 Address index(noreg, rbx, Address::times_4);
1986 __ incrementl(ArrayAddress(table, index));
1987 }
1990 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1991 // Call a little run-time stub to avoid blow-up for each bytecode.
1992 // The run-time runtime saves the right registers, depending on
1993 // the tosca in-state for the given template.
1994 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1995 "entry must have been generated");
1996 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1997 }
2000 void TemplateInterpreterGenerator::stop_interpreter_at() {
2001 Label L;
2002 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
2003 StopInterpreterAt);
2004 __ jcc(Assembler::notEqual, L);
2005 __ int3();
2006 __ bind(L);
2007 }
2008 #endif // !PRODUCT
2009 #endif // CC_INTERP