1 /*
2 * Copyright (c) 1997, 2015, 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 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/templateInterpreterGenerator.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/arguments.hpp"
40 #include "runtime/deoptimization.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #include "utilities/macros.hpp"
49
50 #ifndef FAST_DISPATCH
51 #define FAST_DISPATCH 1
52 #endif
53 #undef FAST_DISPATCH
54
55
56 // Generation of Interpreter
57 //
58 // The TemplateInterpreterGenerator generates the interpreter into Interpreter::_code.
59
60
61 #define __ _masm->
62
63
64 //----------------------------------------------------------------------------------------------------
65
66
67 void TemplateInterpreterGenerator::save_native_result(void) {
68 // result potentially in O0/O1: save it across calls
69 const Address& l_tmp = InterpreterMacroAssembler::l_tmp;
70
71 // result potentially in F0/F1: save it across calls
72 const Address& d_tmp = InterpreterMacroAssembler::d_tmp;
73
74 // save and restore any potential method result value around the unlocking operation
75 __ stf(FloatRegisterImpl::D, F0, d_tmp);
76 #ifdef _LP64
77 __ stx(O0, l_tmp);
78 #else
79 __ std(O0, l_tmp);
80 #endif
81 }
82
83 void TemplateInterpreterGenerator::restore_native_result(void) {
84 const Address& l_tmp = InterpreterMacroAssembler::l_tmp;
85 const Address& d_tmp = InterpreterMacroAssembler::d_tmp;
86
87 // Restore any method result value
88 __ ldf(FloatRegisterImpl::D, d_tmp, F0);
89 #ifdef _LP64
90 __ ldx(l_tmp, O0);
91 #else
92 __ ldd(l_tmp, O0);
93 #endif
94 }
95
96 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
97 assert(!pass_oop || message == NULL, "either oop or message but not both");
98 address entry = __ pc();
99 // expression stack must be empty before entering the VM if an exception happened
100 __ empty_expression_stack();
101 // load exception object
102 __ set((intptr_t)name, G3_scratch);
103 if (pass_oop) {
104 __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), G3_scratch, Otos_i);
105 } else {
106 __ set((intptr_t)message, G4_scratch);
107 __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), G3_scratch, G4_scratch);
108 }
109 // throw exception
110 assert(Interpreter::throw_exception_entry() != NULL, "generate it first");
111 AddressLiteral thrower(Interpreter::throw_exception_entry());
112 __ jump_to(thrower, G3_scratch);
113 __ delayed()->nop();
114 return entry;
115 }
116
117 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
118 address entry = __ pc();
119 // expression stack must be empty before entering the VM if an exception
120 // happened
121 __ empty_expression_stack();
122 // load exception object
123 __ call_VM(Oexception,
124 CAST_FROM_FN_PTR(address,
125 InterpreterRuntime::throw_ClassCastException),
126 Otos_i);
127 __ should_not_reach_here();
128 return entry;
129 }
130
131
132 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
133 address entry = __ pc();
134 // expression stack must be empty before entering the VM if an exception happened
135 __ empty_expression_stack();
136 // convention: expect aberrant index in register G3_scratch, then shuffle the
137 // index to G4_scratch for the VM call
138 __ mov(G3_scratch, G4_scratch);
139 __ set((intptr_t)name, G3_scratch);
140 __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), G3_scratch, G4_scratch);
141 __ should_not_reach_here();
142 return entry;
143 }
144
145
146 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
147 address entry = __ pc();
148 // expression stack must be empty before entering the VM if an exception happened
149 __ empty_expression_stack();
150 __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
151 __ should_not_reach_here();
152 return entry;
153 }
154
155
156 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
157 address entry = __ pc();
158
159 if (state == atos) {
160 __ profile_return_type(O0, G3_scratch, G1_scratch);
161 }
162
163 #if !defined(_LP64) && defined(COMPILER2)
164 // All return values are where we want them, except for Longs. C2 returns
165 // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1.
166 // Since the interpreter will return longs in G1 and O0/O1 in the 32bit
167 // build even if we are returning from interpreted we just do a little
168 // stupid shuffing.
169 // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to
170 // do this here. Unfortunately if we did a rethrow we'd see an machepilog node
171 // first which would move g1 -> O0/O1 and destroy the exception we were throwing.
172
173 if (state == ltos) {
174 __ srl (G1, 0, O1);
175 __ srlx(G1, 32, O0);
176 }
177 #endif // !_LP64 && COMPILER2
178
179 // The callee returns with the stack possibly adjusted by adapter transition
180 // We remove that possible adjustment here.
181 // All interpreter local registers are untouched. Any result is passed back
182 // in the O0/O1 or float registers. Before continuing, the arguments must be
183 // popped from the java expression stack; i.e., Lesp must be adjusted.
184
185 __ mov(Llast_SP, SP); // Remove any adapter added stack space.
186
187 const Register cache = G3_scratch;
188 const Register index = G1_scratch;
189 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
190
191 const Register flags = cache;
192 __ ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset(), flags);
193 const Register parameter_size = flags;
194 __ and3(flags, ConstantPoolCacheEntry::parameter_size_mask, parameter_size); // argument size in words
195 __ sll(parameter_size, Interpreter::logStackElementSize, parameter_size); // each argument size in bytes
196 __ add(Lesp, parameter_size, Lesp); // pop arguments
197 __ dispatch_next(state, step);
198
199 return entry;
200 }
201
202
203 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
204 address entry = __ pc();
205 __ get_constant_pool_cache(LcpoolCache); // load LcpoolCache
206 #if INCLUDE_JVMCI
207 // Check if we need to take lock at entry of synchronized method.
208 if (UseJVMCICompiler) {
209 Label L;
210 Address pending_monitor_enter_addr(G2_thread, JavaThread::pending_monitorenter_offset());
211 __ ldbool(pending_monitor_enter_addr, Gtemp); // Load if pending monitor enter
212 __ cmp_and_br_short(Gtemp, G0, Assembler::equal, Assembler::pn, L);
213 // Clear flag.
214 __ stbool(G0, pending_monitor_enter_addr);
215 // Take lock.
216 lock_method();
217 __ bind(L);
218 }
219 #endif
220 { Label L;
221 Address exception_addr(G2_thread, Thread::pending_exception_offset());
222 __ ld_ptr(exception_addr, Gtemp); // Load pending exception.
223 __ br_null_short(Gtemp, Assembler::pt, L);
224 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
225 __ should_not_reach_here();
226 __ bind(L);
227 }
228 __ dispatch_next(state, step);
229 return entry;
230 }
231
232 // A result handler converts/unboxes a native call result into
233 // a java interpreter/compiler result. The current frame is an
234 // interpreter frame. The activation frame unwind code must be
235 // consistent with that of TemplateTable::_return(...). In the
236 // case of native methods, the caller's SP was not modified.
237 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
238 address entry = __ pc();
239 Register Itos_i = Otos_i ->after_save();
240 Register Itos_l = Otos_l ->after_save();
241 Register Itos_l1 = Otos_l1->after_save();
242 Register Itos_l2 = Otos_l2->after_save();
243 switch (type) {
244 case T_BOOLEAN: __ subcc(G0, O0, G0); __ addc(G0, 0, Itos_i); break; // !0 => true; 0 => false
245 case T_CHAR : __ sll(O0, 16, O0); __ srl(O0, 16, Itos_i); break; // cannot use and3, 0xFFFF too big as immediate value!
246 case T_BYTE : __ sll(O0, 24, O0); __ sra(O0, 24, Itos_i); break;
247 case T_SHORT : __ sll(O0, 16, O0); __ sra(O0, 16, Itos_i); break;
248 case T_LONG :
249 #ifndef _LP64
250 __ mov(O1, Itos_l2); // move other half of long
251 #endif // ifdef or no ifdef, fall through to the T_INT case
252 case T_INT : __ mov(O0, Itos_i); break;
253 case T_VOID : /* nothing to do */ break;
254 case T_FLOAT : assert(F0 == Ftos_f, "fix this code" ); break;
255 case T_DOUBLE : assert(F0 == Ftos_d, "fix this code" ); break;
256 case T_OBJECT :
257 __ ld_ptr(FP, (frame::interpreter_frame_oop_temp_offset*wordSize) + STACK_BIAS, Itos_i);
258 __ verify_oop(Itos_i);
259 break;
260 default : ShouldNotReachHere();
261 }
262 __ ret(); // return from interpreter activation
263 __ delayed()->restore(I5_savedSP, G0, SP); // remove interpreter frame
264 NOT_PRODUCT(__ emit_int32(0);) // marker for disassembly
265 return entry;
266 }
267
268 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
269 address entry = __ pc();
270 __ push(state);
271 __ call_VM(noreg, runtime_entry);
272 __ dispatch_via(vtos, Interpreter::normal_table(vtos));
273 return entry;
274 }
275
276
277 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
278 address entry = __ pc();
279 __ dispatch_next(state);
280 return entry;
281 }
282
283 //
284 // Helpers for commoning out cases in the various type of method entries.
285 //
286
287 // increment invocation count & check for overflow
288 //
289 // Note: checking for negative value instead of overflow
290 // so we have a 'sticky' overflow test
291 //
292 // Lmethod: method
293 // ??: invocation counter
294 //
295 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
296 // Note: In tiered we increment either counters in MethodCounters* or in
297 // MDO depending if we're profiling or not.
298 const Register G3_method_counters = G3_scratch;
299 Label done;
300
301 if (TieredCompilation) {
302 const int increment = InvocationCounter::count_increment;
303 Label no_mdo;
304 if (ProfileInterpreter) {
305 // If no method data exists, go to profile_continue.
306 __ ld_ptr(Lmethod, Method::method_data_offset(), G4_scratch);
307 __ br_null_short(G4_scratch, Assembler::pn, no_mdo);
308 // Increment counter
309 Address mdo_invocation_counter(G4_scratch,
310 in_bytes(MethodData::invocation_counter_offset()) +
311 in_bytes(InvocationCounter::counter_offset()));
312 Address mask(G4_scratch, in_bytes(MethodData::invoke_mask_offset()));
313 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask,
314 G3_scratch, Lscratch,
315 Assembler::zero, overflow);
316 __ ba_short(done);
317 }
318
319 // Increment counter in MethodCounters*
320 __ bind(no_mdo);
321 Address invocation_counter(G3_method_counters,
322 in_bytes(MethodCounters::invocation_counter_offset()) +
323 in_bytes(InvocationCounter::counter_offset()));
324 __ get_method_counters(Lmethod, G3_method_counters, done);
325 Address mask(G3_method_counters, in_bytes(MethodCounters::invoke_mask_offset()));
326 __ increment_mask_and_jump(invocation_counter, increment, mask,
327 G4_scratch, Lscratch,
328 Assembler::zero, overflow);
329 __ bind(done);
330 } else { // not TieredCompilation
331 // Update standard invocation counters
332 __ get_method_counters(Lmethod, G3_method_counters, done);
333 __ increment_invocation_counter(G3_method_counters, O0, G4_scratch);
334 if (ProfileInterpreter) {
335 Address interpreter_invocation_counter(G3_method_counters,
336 in_bytes(MethodCounters::interpreter_invocation_counter_offset()));
337 __ ld(interpreter_invocation_counter, G4_scratch);
338 __ inc(G4_scratch);
339 __ st(G4_scratch, interpreter_invocation_counter);
340 }
341
342 if (ProfileInterpreter && profile_method != NULL) {
343 // Test to see if we should create a method data oop
344 Address profile_limit(G3_method_counters, in_bytes(MethodCounters::interpreter_profile_limit_offset()));
345 __ ld(profile_limit, G1_scratch);
346 __ cmp_and_br_short(O0, G1_scratch, Assembler::lessUnsigned, Assembler::pn, *profile_method_continue);
347
348 // if no method data exists, go to profile_method
349 __ test_method_data_pointer(*profile_method);
350 }
351
352 Address invocation_limit(G3_method_counters, in_bytes(MethodCounters::interpreter_invocation_limit_offset()));
353 __ ld(invocation_limit, G3_scratch);
354 __ cmp(O0, G3_scratch);
355 __ br(Assembler::greaterEqualUnsigned, false, Assembler::pn, *overflow); // Far distance
356 __ delayed()->nop();
357 __ bind(done);
358 }
359
360 }
361
362 // Allocate monitor and lock method (asm interpreter)
363 // ebx - Method*
364 //
365 void TemplateInterpreterGenerator::lock_method() {
366 __ ld(Lmethod, in_bytes(Method::access_flags_offset()), O0); // Load access flags.
367
368 #ifdef ASSERT
369 { Label ok;
370 __ btst(JVM_ACC_SYNCHRONIZED, O0);
371 __ br( Assembler::notZero, false, Assembler::pt, ok);
372 __ delayed()->nop();
373 __ stop("method doesn't need synchronization");
374 __ bind(ok);
375 }
376 #endif // ASSERT
377
378 // get synchronization object to O0
379 { Label done;
380 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
381 __ btst(JVM_ACC_STATIC, O0);
382 __ br( Assembler::zero, true, Assembler::pt, done);
383 __ delayed()->ld_ptr(Llocals, Interpreter::local_offset_in_bytes(0), O0); // get receiver for not-static case
384
385 __ ld_ptr( Lmethod, in_bytes(Method::const_offset()), O0);
386 __ ld_ptr( O0, in_bytes(ConstMethod::constants_offset()), O0);
387 __ ld_ptr( O0, ConstantPool::pool_holder_offset_in_bytes(), O0);
388
389 // lock the mirror, not the Klass*
390 __ ld_ptr( O0, mirror_offset, O0);
391
392 #ifdef ASSERT
393 __ tst(O0);
394 __ breakpoint_trap(Assembler::zero, Assembler::ptr_cc);
395 #endif // ASSERT
396
397 __ bind(done);
398 }
399
400 __ add_monitor_to_stack(true, noreg, noreg); // allocate monitor elem
401 __ st_ptr( O0, Lmonitors, BasicObjectLock::obj_offset_in_bytes()); // store object
402 // __ untested("lock_object from method entry");
403 __ lock_object(Lmonitors, O0);
404 }
405
406
407 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rframe_size,
408 Register Rscratch,
409 Register Rscratch2) {
410 const int page_size = os::vm_page_size();
411 Label after_frame_check;
412
413 assert_different_registers(Rframe_size, Rscratch, Rscratch2);
414
415 __ set(page_size, Rscratch);
416 __ cmp_and_br_short(Rframe_size, Rscratch, Assembler::lessEqual, Assembler::pt, after_frame_check);
417
418 // get the stack base, and in debug, verify it is non-zero
419 __ ld_ptr( G2_thread, Thread::stack_base_offset(), Rscratch );
420 #ifdef ASSERT
421 Label base_not_zero;
422 __ br_notnull_short(Rscratch, Assembler::pn, base_not_zero);
423 __ stop("stack base is zero in generate_stack_overflow_check");
424 __ bind(base_not_zero);
425 #endif
426
427 // get the stack size, and in debug, verify it is non-zero
428 assert( sizeof(size_t) == sizeof(intptr_t), "wrong load size" );
429 __ ld_ptr( G2_thread, Thread::stack_size_offset(), Rscratch2 );
430 #ifdef ASSERT
431 Label size_not_zero;
432 __ br_notnull_short(Rscratch2, Assembler::pn, size_not_zero);
433 __ stop("stack size is zero in generate_stack_overflow_check");
434 __ bind(size_not_zero);
435 #endif
436
437 // compute the beginning of the protected zone minus the requested frame size
438 __ sub( Rscratch, Rscratch2, Rscratch );
439 __ set( (StackRedPages+StackYellowPages) * page_size, Rscratch2 );
440 __ add( Rscratch, Rscratch2, Rscratch );
441
442 // Add in the size of the frame (which is the same as subtracting it from the
443 // SP, which would take another register
444 __ add( Rscratch, Rframe_size, Rscratch );
445
446 // the frame is greater than one page in size, so check against
447 // the bottom of the stack
448 __ cmp_and_brx_short(SP, Rscratch, Assembler::greaterUnsigned, Assembler::pt, after_frame_check);
449
450 // the stack will overflow, throw an exception
451
452 // Note that SP is restored to sender's sp (in the delay slot). This
453 // is necessary if the sender's frame is an extended compiled frame
454 // (see gen_c2i_adapter()) and safer anyway in case of JSR292
455 // adaptations.
456
457 // Note also that the restored frame is not necessarily interpreted.
458 // Use the shared runtime version of the StackOverflowError.
459 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
460 AddressLiteral stub(StubRoutines::throw_StackOverflowError_entry());
461 __ jump_to(stub, Rscratch);
462 __ delayed()->mov(O5_savedSP, SP);
463
464 // if you get to here, then there is enough stack space
465 __ bind( after_frame_check );
466 }
467
468
469 //
470 // Generate a fixed interpreter frame. This is identical setup for interpreted
471 // methods and for native methods hence the shared code.
472
473
474 //----------------------------------------------------------------------------------------------------
475 // Stack frame layout
476 //
477 // When control flow reaches any of the entry types for the interpreter
478 // the following holds ->
479 //
480 // C2 Calling Conventions:
481 //
482 // The entry code below assumes that the following registers are set
483 // when coming in:
484 // G5_method: holds the Method* of the method to call
485 // Lesp: points to the TOS of the callers expression stack
486 // after having pushed all the parameters
487 //
488 // The entry code does the following to setup an interpreter frame
489 // pop parameters from the callers stack by adjusting Lesp
490 // set O0 to Lesp
491 // compute X = (max_locals - num_parameters)
492 // bump SP up by X to accomadate the extra locals
493 // compute X = max_expression_stack
494 // + vm_local_words
495 // + 16 words of register save area
496 // save frame doing a save sp, -X, sp growing towards lower addresses
497 // set Lbcp, Lmethod, LcpoolCache
498 // set Llocals to i0
499 // set Lmonitors to FP - rounded_vm_local_words
500 // set Lesp to Lmonitors - 4
501 //
502 // The frame has now been setup to do the rest of the entry code
503
504 // Try this optimization: Most method entries could live in a
505 // "one size fits all" stack frame without all the dynamic size
506 // calculations. It might be profitable to do all this calculation
507 // statically and approximately for "small enough" methods.
508
509 //-----------------------------------------------------------------------------------------------
510
511 // C1 Calling conventions
512 //
513 // Upon method entry, the following registers are setup:
514 //
515 // g2 G2_thread: current thread
516 // g5 G5_method: method to activate
517 // g4 Gargs : pointer to last argument
518 //
519 //
520 // Stack:
521 //
522 // +---------------+ <--- sp
523 // | |
524 // : reg save area :
525 // | |
526 // +---------------+ <--- sp + 0x40
527 // | |
528 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
529 // | |
530 // +---------------+ <--- sp + 0x5c
531 // | |
532 // : free :
533 // | |
534 // +---------------+ <--- Gargs
535 // | |
536 // : arguments :
537 // | |
538 // +---------------+
539 // | |
540 //
541 //
542 //
543 // AFTER FRAME HAS BEEN SETUP for method interpretation the stack looks like:
544 //
545 // +---------------+ <--- sp
546 // | |
547 // : reg save area :
548 // | |
549 // +---------------+ <--- sp + 0x40
550 // | |
551 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
552 // | |
553 // +---------------+ <--- sp + 0x5c
554 // | |
555 // : :
556 // | | <--- Lesp
557 // +---------------+ <--- Lmonitors (fp - 0x18)
558 // | VM locals |
559 // +---------------+ <--- fp
560 // | |
561 // : reg save area :
562 // | |
563 // +---------------+ <--- fp + 0x40
564 // | |
565 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
566 // | |
567 // +---------------+ <--- fp + 0x5c
568 // | |
569 // : free :
570 // | |
571 // +---------------+
572 // | |
573 // : nonarg locals :
574 // | |
575 // +---------------+
576 // | |
577 // : arguments :
578 // | | <--- Llocals
579 // +---------------+ <--- Gargs
580 // | |
581
582 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
583 //
584 //
585 // The entry code sets up a new interpreter frame in 4 steps:
586 //
587 // 1) Increase caller's SP by for the extra local space needed:
588 // (check for overflow)
589 // Efficient implementation of xload/xstore bytecodes requires
590 // that arguments and non-argument locals are in a contigously
591 // addressable memory block => non-argument locals must be
592 // allocated in the caller's frame.
593 //
594 // 2) Create a new stack frame and register window:
595 // The new stack frame must provide space for the standard
596 // register save area, the maximum java expression stack size,
597 // the monitor slots (0 slots initially), and some frame local
598 // scratch locations.
599 //
600 // 3) The following interpreter activation registers must be setup:
601 // Lesp : expression stack pointer
602 // Lbcp : bytecode pointer
603 // Lmethod : method
604 // Llocals : locals pointer
605 // Lmonitors : monitor pointer
606 // LcpoolCache: constant pool cache
607 //
608 // 4) Initialize the non-argument locals if necessary:
609 // Non-argument locals may need to be initialized to NULL
610 // for GC to work. If the oop-map information is accurate
611 // (in the absence of the JSR problem), no initialization
612 // is necessary.
613 //
614 // (gri - 2/25/2000)
615
616
617 int rounded_vm_local_words = round_to( frame::interpreter_frame_vm_local_words, WordsPerLong );
618
619 const int extra_space =
620 rounded_vm_local_words + // frame local scratch space
621 Method::extra_stack_entries() + // extra stack for jsr 292
622 frame::memory_parameter_word_sp_offset + // register save area
623 (native_call ? frame::interpreter_frame_extra_outgoing_argument_words : 0);
624
625 const Register Glocals_size = G3;
626 const Register RconstMethod = Glocals_size;
627 const Register Otmp1 = O3;
628 const Register Otmp2 = O4;
629 // Lscratch can't be used as a temporary because the call_stub uses
630 // it to assert that the stack frame was setup correctly.
631 const Address constMethod (G5_method, Method::const_offset());
632 const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset());
633
634 __ ld_ptr( constMethod, RconstMethod );
635 __ lduh( size_of_parameters, Glocals_size);
636
637 // Gargs points to first local + BytesPerWord
638 // Set the saved SP after the register window save
639 //
640 assert_different_registers(Gargs, Glocals_size, Gframe_size, O5_savedSP);
641 __ sll(Glocals_size, Interpreter::logStackElementSize, Otmp1);
642 __ add(Gargs, Otmp1, Gargs);
643
644 if (native_call) {
645 __ calc_mem_param_words( Glocals_size, Gframe_size );
646 __ add( Gframe_size, extra_space, Gframe_size);
647 __ round_to( Gframe_size, WordsPerLong );
648 __ sll( Gframe_size, LogBytesPerWord, Gframe_size );
649 } else {
650
651 //
652 // Compute number of locals in method apart from incoming parameters
653 //
654 const Address size_of_locals (Otmp1, ConstMethod::size_of_locals_offset());
655 __ ld_ptr( constMethod, Otmp1 );
656 __ lduh( size_of_locals, Otmp1 );
657 __ sub( Otmp1, Glocals_size, Glocals_size );
658 __ round_to( Glocals_size, WordsPerLong );
659 __ sll( Glocals_size, Interpreter::logStackElementSize, Glocals_size );
660
661 // see if the frame is greater than one page in size. If so,
662 // then we need to verify there is enough stack space remaining
663 // Frame_size = (max_stack + extra_space) * BytesPerWord;
664 __ ld_ptr( constMethod, Gframe_size );
665 __ lduh( Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size );
666 __ add( Gframe_size, extra_space, Gframe_size );
667 __ round_to( Gframe_size, WordsPerLong );
668 __ sll( Gframe_size, Interpreter::logStackElementSize, Gframe_size);
669
670 // Add in java locals size for stack overflow check only
671 __ add( Gframe_size, Glocals_size, Gframe_size );
672
673 const Register Otmp2 = O4;
674 assert_different_registers(Otmp1, Otmp2, O5_savedSP);
675 generate_stack_overflow_check(Gframe_size, Otmp1, Otmp2);
676
677 __ sub( Gframe_size, Glocals_size, Gframe_size);
678
679 //
680 // bump SP to accomodate the extra locals
681 //
682 __ sub( SP, Glocals_size, SP );
683 }
684
685 //
686 // now set up a stack frame with the size computed above
687 //
688 __ neg( Gframe_size );
689 __ save( SP, Gframe_size, SP );
690
691 //
692 // now set up all the local cache registers
693 //
694 // NOTE: At this point, Lbyte_code/Lscratch has been modified. Note
695 // that all present references to Lbyte_code initialize the register
696 // immediately before use
697 if (native_call) {
698 __ mov(G0, Lbcp);
699 } else {
700 __ ld_ptr(G5_method, Method::const_offset(), Lbcp);
701 __ add(Lbcp, in_bytes(ConstMethod::codes_offset()), Lbcp);
702 }
703 __ mov( G5_method, Lmethod); // set Lmethod
704 __ get_constant_pool_cache( LcpoolCache ); // set LcpoolCache
705 __ sub(FP, rounded_vm_local_words * BytesPerWord, Lmonitors ); // set Lmonitors
706 #ifdef _LP64
707 __ add( Lmonitors, STACK_BIAS, Lmonitors ); // Account for 64 bit stack bias
708 #endif
709 __ sub(Lmonitors, BytesPerWord, Lesp); // set Lesp
710
711 // setup interpreter activation registers
712 __ sub(Gargs, BytesPerWord, Llocals); // set Llocals
713
714 if (ProfileInterpreter) {
715 #ifdef FAST_DISPATCH
716 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
717 // they both use I2.
718 assert(0, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
719 #endif // FAST_DISPATCH
720 __ set_method_data_pointer();
721 }
722
723 }
724
725 // Method entry for java.lang.ref.Reference.get.
726 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
727 #if INCLUDE_ALL_GCS
728 // Code: _aload_0, _getfield, _areturn
729 // parameter size = 1
730 //
731 // The code that gets generated by this routine is split into 2 parts:
732 // 1. The "intrinsified" code for G1 (or any SATB based GC),
733 // 2. The slow path - which is an expansion of the regular method entry.
734 //
735 // Notes:-
736 // * In the G1 code we do not check whether we need to block for
737 // a safepoint. If G1 is enabled then we must execute the specialized
738 // code for Reference.get (except when the Reference object is null)
739 // so that we can log the value in the referent field with an SATB
740 // update buffer.
741 // If the code for the getfield template is modified so that the
742 // G1 pre-barrier code is executed when the current method is
743 // Reference.get() then going through the normal method entry
744 // will be fine.
745 // * The G1 code can, however, check the receiver object (the instance
746 // of java.lang.Reference) and jump to the slow path if null. If the
747 // Reference object is null then we obviously cannot fetch the referent
748 // and so we don't need to call the G1 pre-barrier. Thus we can use the
749 // regular method entry code to generate the NPE.
750 //
751 // This code is based on generate_accessor_enty.
752
753 address entry = __ pc();
754
755 const int referent_offset = java_lang_ref_Reference::referent_offset;
756 guarantee(referent_offset > 0, "referent offset not initialized");
757
758 if (UseG1GC) {
759 Label slow_path;
760
761 // In the G1 code we don't check if we need to reach a safepoint. We
762 // continue and the thread will safepoint at the next bytecode dispatch.
763
764 // Check if local 0 != NULL
765 // If the receiver is null then it is OK to jump to the slow path.
766 __ ld_ptr(Gargs, G0, Otos_i ); // get local 0
767 // check if local 0 == NULL and go the slow path
768 __ cmp_and_brx_short(Otos_i, 0, Assembler::equal, Assembler::pn, slow_path);
769
770
771 // Load the value of the referent field.
772 if (Assembler::is_simm13(referent_offset)) {
773 __ load_heap_oop(Otos_i, referent_offset, Otos_i);
774 } else {
775 __ set(referent_offset, G3_scratch);
776 __ load_heap_oop(Otos_i, G3_scratch, Otos_i);
777 }
778
779 // Generate the G1 pre-barrier code to log the value of
780 // the referent field in an SATB buffer. Note with
781 // these parameters the pre-barrier does not generate
782 // the load of the previous value
783
784 __ g1_write_barrier_pre(noreg /* obj */, noreg /* index */, 0 /* offset */,
785 Otos_i /* pre_val */,
786 G3_scratch /* tmp */,
787 true /* preserve_o_regs */);
788
789 // _areturn
790 __ retl(); // return from leaf routine
791 __ delayed()->mov(O5_savedSP, SP);
792
793 // Generate regular method entry
794 __ bind(slow_path);
795 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
796 return entry;
797 }
798 #endif // INCLUDE_ALL_GCS
799
800 // If G1 is not enabled then attempt to go through the accessor entry point
801 // Reference.get is an accessor
802 return NULL;
803 }
804
805 /**
806 * Method entry for static native methods:
807 * int java.util.zip.CRC32.update(int crc, int b)
808 */
809 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
810
811 if (UseCRC32Intrinsics) {
812 address entry = __ pc();
813
814 Label L_slow_path;
815 // If we need a safepoint check, generate full interpreter entry.
816 ExternalAddress state(SafepointSynchronize::address_of_state());
817 __ set(ExternalAddress(SafepointSynchronize::address_of_state()), O2);
818 __ set(SafepointSynchronize::_not_synchronized, O3);
819 __ cmp_and_br_short(O2, O3, Assembler::notEqual, Assembler::pt, L_slow_path);
820
821 // Load parameters
822 const Register crc = O0; // initial crc
823 const Register val = O1; // byte to update with
824 const Register table = O2; // address of 256-entry lookup table
825
826 __ ldub(Gargs, 3, val);
827 __ lduw(Gargs, 8, crc);
828
829 __ set(ExternalAddress(StubRoutines::crc_table_addr()), table);
830
831 __ not1(crc); // ~crc
832 __ clruwu(crc);
833 __ update_byte_crc32(crc, val, table);
834 __ not1(crc); // ~crc
835
836 // result in O0
837 __ retl();
838 __ delayed()->nop();
839
840 // generate a vanilla native entry as the slow path
841 __ bind(L_slow_path);
842 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
843 return entry;
844 }
845 return NULL;
846 }
847
848 /**
849 * Method entry for static native methods:
850 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
851 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
852 */
853 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
854
855 if (UseCRC32Intrinsics) {
856 address entry = __ pc();
857
858 Label L_slow_path;
859 // If we need a safepoint check, generate full interpreter entry.
860 ExternalAddress state(SafepointSynchronize::address_of_state());
861 __ set(ExternalAddress(SafepointSynchronize::address_of_state()), O2);
862 __ set(SafepointSynchronize::_not_synchronized, O3);
863 __ cmp_and_br_short(O2, O3, Assembler::notEqual, Assembler::pt, L_slow_path);
864
865 // Load parameters from the stack
866 const Register crc = O0; // initial crc
867 const Register buf = O1; // source java byte array address
868 const Register len = O2; // len
869 const Register offset = O3; // offset
870
871 // Arguments are reversed on java expression stack
872 // Calculate address of start element
873 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
874 __ lduw(Gargs, 0, len);
875 __ lduw(Gargs, 8, offset);
876 __ ldx( Gargs, 16, buf);
877 __ lduw(Gargs, 32, crc);
878 __ add(buf, offset, buf);
879 } else {
880 __ lduw(Gargs, 0, len);
881 __ lduw(Gargs, 8, offset);
882 __ ldx( Gargs, 16, buf);
883 __ lduw(Gargs, 24, crc);
884 __ add(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE), buf); // account for the header size
885 __ add(buf ,offset, buf);
886 }
887
888 // Call the crc32 kernel
889 __ MacroAssembler::save_thread(L7_thread_cache);
890 __ kernel_crc32(crc, buf, len, O3);
891 __ MacroAssembler::restore_thread(L7_thread_cache);
892
893 // result in O0
894 __ retl();
895 __ delayed()->nop();
896
897 // generate a vanilla native entry as the slow path
898 __ bind(L_slow_path);
899 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
900 return entry;
901 }
902 return NULL;
903 }
904
905 // Not supported
906 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
907 return NULL;
908 }
909
910 // Not supported
911 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
912 return NULL;
913 }
914 //
915 // Interpreter stub for calling a native method. (asm interpreter)
916 // This sets up a somewhat different looking stack for calling the native method
917 // than the typical interpreter frame setup.
918 //
919
920 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
921 address entry = __ pc();
922
923 // the following temporary registers are used during frame creation
924 const Register Gtmp1 = G3_scratch ;
925 const Register Gtmp2 = G1_scratch;
926 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
927
928 // make sure registers are different!
929 assert_different_registers(G2_thread, G5_method, Gargs, Gtmp1, Gtmp2);
930
931 const Address Laccess_flags(Lmethod, Method::access_flags_offset());
932
933 const Register Glocals_size = G3;
934 assert_different_registers(Glocals_size, G4_scratch, Gframe_size);
935
936 // make sure method is native & not abstract
937 // rethink these assertions - they can be simplified and shared (gri 2/25/2000)
938 #ifdef ASSERT
939 __ ld(G5_method, Method::access_flags_offset(), Gtmp1);
940 {
941 Label L;
942 __ btst(JVM_ACC_NATIVE, Gtmp1);
943 __ br(Assembler::notZero, false, Assembler::pt, L);
944 __ delayed()->nop();
945 __ stop("tried to execute non-native method as native");
946 __ bind(L);
947 }
948 { Label L;
949 __ btst(JVM_ACC_ABSTRACT, Gtmp1);
950 __ br(Assembler::zero, false, Assembler::pt, L);
951 __ delayed()->nop();
952 __ stop("tried to execute abstract method as non-abstract");
953 __ bind(L);
954 }
955 #endif // ASSERT
956
957 // generate the code to allocate the interpreter stack frame
958 generate_fixed_frame(true);
959
960 //
961 // No locals to initialize for native method
962 //
963
964 // this slot will be set later, we initialize it to null here just in
965 // case we get a GC before the actual value is stored later
966 __ st_ptr(G0, FP, (frame::interpreter_frame_oop_temp_offset * wordSize) + STACK_BIAS);
967
968 const Address do_not_unlock_if_synchronized(G2_thread,
969 JavaThread::do_not_unlock_if_synchronized_offset());
970 // Since at this point in the method invocation the exception handler
971 // would try to exit the monitor of synchronized methods which hasn't
972 // been entered yet, we set the thread local variable
973 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
974 // runtime, exception handling i.e. unlock_if_synchronized_method will
975 // check this thread local flag.
976 // This flag has two effects, one is to force an unwind in the topmost
977 // interpreter frame and not perform an unlock while doing so.
978
979 __ movbool(true, G3_scratch);
980 __ stbool(G3_scratch, do_not_unlock_if_synchronized);
981
982 // increment invocation counter and check for overflow
983 //
984 // Note: checking for negative value instead of overflow
985 // so we have a 'sticky' overflow test (may be of
986 // importance as soon as we have true MT/MP)
987 Label invocation_counter_overflow;
988 Label Lcontinue;
989 if (inc_counter) {
990 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
991
992 }
993 __ bind(Lcontinue);
994
995 bang_stack_shadow_pages(true);
996
997 // reset the _do_not_unlock_if_synchronized flag
998 __ stbool(G0, do_not_unlock_if_synchronized);
999
1000 // check for synchronized methods
1001 // Must happen AFTER invocation_counter check and stack overflow check,
1002 // so method is not locked if overflows.
1003
1004 if (synchronized) {
1005 lock_method();
1006 } else {
1007 #ifdef ASSERT
1008 { Label ok;
1009 __ ld(Laccess_flags, O0);
1010 __ btst(JVM_ACC_SYNCHRONIZED, O0);
1011 __ br( Assembler::zero, false, Assembler::pt, ok);
1012 __ delayed()->nop();
1013 __ stop("method needs synchronization");
1014 __ bind(ok);
1015 }
1016 #endif // ASSERT
1017 }
1018
1019
1020 // start execution
1021 __ verify_thread();
1022
1023 // JVMTI support
1024 __ notify_method_entry();
1025
1026 // native call
1027
1028 // (note that O0 is never an oop--at most it is a handle)
1029 // It is important not to smash any handles created by this call,
1030 // until any oop handle in O0 is dereferenced.
1031
1032 // (note that the space for outgoing params is preallocated)
1033
1034 // get signature handler
1035 { Label L;
1036 Address signature_handler(Lmethod, Method::signature_handler_offset());
1037 __ ld_ptr(signature_handler, G3_scratch);
1038 __ br_notnull_short(G3_scratch, Assembler::pt, L);
1039 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), Lmethod);
1040 __ ld_ptr(signature_handler, G3_scratch);
1041 __ bind(L);
1042 }
1043
1044 // Push a new frame so that the args will really be stored in
1045 // Copy a few locals across so the new frame has the variables
1046 // we need but these values will be dead at the jni call and
1047 // therefore not gc volatile like the values in the current
1048 // frame (Lmethod in particular)
1049
1050 // Flush the method pointer to the register save area
1051 __ st_ptr(Lmethod, SP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS);
1052 __ mov(Llocals, O1);
1053
1054 // calculate where the mirror handle body is allocated in the interpreter frame:
1055 __ add(FP, (frame::interpreter_frame_oop_temp_offset * wordSize) + STACK_BIAS, O2);
1056
1057 // Calculate current frame size
1058 __ sub(SP, FP, O3); // Calculate negative of current frame size
1059 __ save(SP, O3, SP); // Allocate an identical sized frame
1060
1061 // Note I7 has leftover trash. Slow signature handler will fill it in
1062 // should we get there. Normal jni call will set reasonable last_Java_pc
1063 // below (and fix I7 so the stack trace doesn't have a meaningless frame
1064 // in it).
1065
1066 // Load interpreter frame's Lmethod into same register here
1067
1068 __ ld_ptr(FP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS, Lmethod);
1069
1070 __ mov(I1, Llocals);
1071 __ mov(I2, Lscratch2); // save the address of the mirror
1072
1073
1074 // ONLY Lmethod and Llocals are valid here!
1075
1076 // call signature handler, It will move the arg properly since Llocals in current frame
1077 // matches that in outer frame
1078
1079 __ callr(G3_scratch, 0);
1080 __ delayed()->nop();
1081
1082 // Result handler is in Lscratch
1083
1084 // Reload interpreter frame's Lmethod since slow signature handler may block
1085 __ ld_ptr(FP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS, Lmethod);
1086
1087 { Label not_static;
1088
1089 __ ld(Laccess_flags, O0);
1090 __ btst(JVM_ACC_STATIC, O0);
1091 __ br( Assembler::zero, false, Assembler::pt, not_static);
1092 // get native function entry point(O0 is a good temp until the very end)
1093 __ delayed()->ld_ptr(Lmethod, in_bytes(Method::native_function_offset()), O0);
1094 // for static methods insert the mirror argument
1095 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1096
1097 __ ld_ptr(Lmethod, Method:: const_offset(), O1);
1098 __ ld_ptr(O1, ConstMethod::constants_offset(), O1);
1099 __ ld_ptr(O1, ConstantPool::pool_holder_offset_in_bytes(), O1);
1100 __ ld_ptr(O1, mirror_offset, O1);
1101 #ifdef ASSERT
1102 if (!PrintSignatureHandlers) // do not dirty the output with this
1103 { Label L;
1104 __ br_notnull_short(O1, Assembler::pt, L);
1105 __ stop("mirror is missing");
1106 __ bind(L);
1107 }
1108 #endif // ASSERT
1109 __ st_ptr(O1, Lscratch2, 0);
1110 __ mov(Lscratch2, O1);
1111 __ bind(not_static);
1112 }
1113
1114 // At this point, arguments have been copied off of stack into
1115 // their JNI positions, which are O1..O5 and SP[68..].
1116 // Oops are boxed in-place on the stack, with handles copied to arguments.
1117 // The result handler is in Lscratch. O0 will shortly hold the JNIEnv*.
1118
1119 #ifdef ASSERT
1120 { Label L;
1121 __ br_notnull_short(O0, Assembler::pt, L);
1122 __ stop("native entry point is missing");
1123 __ bind(L);
1124 }
1125 #endif // ASSERT
1126
1127 //
1128 // setup the frame anchor
1129 //
1130 // The scavenge function only needs to know that the PC of this frame is
1131 // in the interpreter method entry code, it doesn't need to know the exact
1132 // PC and hence we can use O7 which points to the return address from the
1133 // previous call in the code stream (signature handler function)
1134 //
1135 // The other trick is we set last_Java_sp to FP instead of the usual SP because
1136 // we have pushed the extra frame in order to protect the volatile register(s)
1137 // in that frame when we return from the jni call
1138 //
1139
1140 __ set_last_Java_frame(FP, O7);
1141 __ mov(O7, I7); // make dummy interpreter frame look like one above,
1142 // not meaningless information that'll confuse me.
1143
1144 // flush the windows now. We don't care about the current (protection) frame
1145 // only the outer frames
1146
1147 __ flushw();
1148
1149 // mark windows as flushed
1150 Address flags(G2_thread, JavaThread::frame_anchor_offset() + JavaFrameAnchor::flags_offset());
1151 __ set(JavaFrameAnchor::flushed, G3_scratch);
1152 __ st(G3_scratch, flags);
1153
1154 // Transition from _thread_in_Java to _thread_in_native. We are already safepoint ready.
1155
1156 Address thread_state(G2_thread, JavaThread::thread_state_offset());
1157 #ifdef ASSERT
1158 { Label L;
1159 __ ld(thread_state, G3_scratch);
1160 __ cmp_and_br_short(G3_scratch, _thread_in_Java, Assembler::equal, Assembler::pt, L);
1161 __ stop("Wrong thread state in native stub");
1162 __ bind(L);
1163 }
1164 #endif // ASSERT
1165 __ set(_thread_in_native, G3_scratch);
1166 __ st(G3_scratch, thread_state);
1167
1168 // Call the jni method, using the delay slot to set the JNIEnv* argument.
1169 __ save_thread(L7_thread_cache); // save Gthread
1170 __ callr(O0, 0);
1171 __ delayed()->
1172 add(L7_thread_cache, in_bytes(JavaThread::jni_environment_offset()), O0);
1173
1174 // Back from jni method Lmethod in this frame is DEAD, DEAD, DEAD
1175
1176 __ restore_thread(L7_thread_cache); // restore G2_thread
1177 __ reinit_heapbase();
1178
1179 // must we block?
1180
1181 // Block, if necessary, before resuming in _thread_in_Java state.
1182 // In order for GC to work, don't clear the last_Java_sp until after blocking.
1183 { Label no_block;
1184 AddressLiteral sync_state(SafepointSynchronize::address_of_state());
1185
1186 // Switch thread to "native transition" state before reading the synchronization state.
1187 // This additional state is necessary because reading and testing the synchronization
1188 // state is not atomic w.r.t. GC, as this scenario demonstrates:
1189 // Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1190 // VM thread changes sync state to synchronizing and suspends threads for GC.
1191 // Thread A is resumed to finish this native method, but doesn't block here since it
1192 // didn't see any synchronization is progress, and escapes.
1193 __ set(_thread_in_native_trans, G3_scratch);
1194 __ st(G3_scratch, thread_state);
1195 if(os::is_MP()) {
1196 if (UseMembar) {
1197 // Force this write out before the read below
1198 __ membar(Assembler::StoreLoad);
1199 } else {
1200 // Write serialization page so VM thread can do a pseudo remote membar.
1201 // We use the current thread pointer to calculate a thread specific
1202 // offset to write to within the page. This minimizes bus traffic
1203 // due to cache line collision.
1204 __ serialize_memory(G2_thread, G1_scratch, G3_scratch);
1205 }
1206 }
1207 __ load_contents(sync_state, G3_scratch);
1208 __ cmp(G3_scratch, SafepointSynchronize::_not_synchronized);
1209
1210 Label L;
1211 __ br(Assembler::notEqual, false, Assembler::pn, L);
1212 __ delayed()->ld(G2_thread, JavaThread::suspend_flags_offset(), G3_scratch);
1213 __ cmp_and_br_short(G3_scratch, 0, Assembler::equal, Assembler::pt, no_block);
1214 __ bind(L);
1215
1216 // Block. Save any potential method result value before the operation and
1217 // use a leaf call to leave the last_Java_frame setup undisturbed.
1218 save_native_result();
1219 __ call_VM_leaf(L7_thread_cache,
1220 CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1221 G2_thread);
1222
1223 // Restore any method result value
1224 restore_native_result();
1225 __ bind(no_block);
1226 }
1227
1228 // Clear the frame anchor now
1229
1230 __ reset_last_Java_frame();
1231
1232 // Move the result handler address
1233 __ mov(Lscratch, G3_scratch);
1234 // return possible result to the outer frame
1235 #ifndef __LP64
1236 __ mov(O0, I0);
1237 __ restore(O1, G0, O1);
1238 #else
1239 __ restore(O0, G0, O0);
1240 #endif /* __LP64 */
1241
1242 // Move result handler to expected register
1243 __ mov(G3_scratch, Lscratch);
1244
1245 // Back in normal (native) interpreter frame. State is thread_in_native_trans
1246 // switch to thread_in_Java.
1247
1248 __ set(_thread_in_Java, G3_scratch);
1249 __ st(G3_scratch, thread_state);
1250
1251 // reset handle block
1252 __ ld_ptr(G2_thread, JavaThread::active_handles_offset(), G3_scratch);
1253 __ st(G0, G3_scratch, JNIHandleBlock::top_offset_in_bytes());
1254
1255 // If we have an oop result store it where it will be safe for any further gc
1256 // until we return now that we've released the handle it might be protected by
1257
1258 {
1259 Label no_oop, store_result;
1260
1261 __ set((intptr_t)AbstractInterpreter::result_handler(T_OBJECT), G3_scratch);
1262 __ cmp_and_brx_short(G3_scratch, Lscratch, Assembler::notEqual, Assembler::pt, no_oop);
1263 __ addcc(G0, O0, O0);
1264 __ brx(Assembler::notZero, true, Assembler::pt, store_result); // if result is not NULL:
1265 __ delayed()->ld_ptr(O0, 0, O0); // unbox it
1266 __ mov(G0, O0);
1267
1268 __ bind(store_result);
1269 // Store it where gc will look for it and result handler expects it.
1270 __ st_ptr(O0, FP, (frame::interpreter_frame_oop_temp_offset*wordSize) + STACK_BIAS);
1271
1272 __ bind(no_oop);
1273
1274 }
1275
1276
1277 // handle exceptions (exception handling will handle unlocking!)
1278 { Label L;
1279 Address exception_addr(G2_thread, Thread::pending_exception_offset());
1280 __ ld_ptr(exception_addr, Gtemp);
1281 __ br_null_short(Gtemp, Assembler::pt, L);
1282 // Note: This could be handled more efficiently since we know that the native
1283 // method doesn't have an exception handler. We could directly return
1284 // to the exception handler for the caller.
1285 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1286 __ should_not_reach_here();
1287 __ bind(L);
1288 }
1289
1290 // JVMTI support (preserves thread register)
1291 __ notify_method_exit(true, ilgl, InterpreterMacroAssembler::NotifyJVMTI);
1292
1293 if (synchronized) {
1294 // save and restore any potential method result value around the unlocking operation
1295 save_native_result();
1296
1297 __ add( __ top_most_monitor(), O1);
1298 __ unlock_object(O1);
1299
1300 restore_native_result();
1301 }
1302
1303 #if defined(COMPILER2) && !defined(_LP64)
1304
1305 // C2 expects long results in G1 we can't tell if we're returning to interpreted
1306 // or compiled so just be safe.
1307
1308 __ sllx(O0, 32, G1); // Shift bits into high G1
1309 __ srl (O1, 0, O1); // Zero extend O1
1310 __ or3 (O1, G1, G1); // OR 64 bits into G1
1311
1312 #endif /* COMPILER2 && !_LP64 */
1313
1314 // dispose of return address and remove activation
1315 #ifdef ASSERT
1316 {
1317 Label ok;
1318 __ cmp_and_brx_short(I5_savedSP, FP, Assembler::greaterEqualUnsigned, Assembler::pt, ok);
1319 __ stop("bad I5_savedSP value");
1320 __ should_not_reach_here();
1321 __ bind(ok);
1322 }
1323 #endif
1324 if (TraceJumps) {
1325 // Move target to register that is recordable
1326 __ mov(Lscratch, G3_scratch);
1327 __ JMP(G3_scratch, 0);
1328 } else {
1329 __ jmp(Lscratch, 0);
1330 }
1331 __ delayed()->nop();
1332
1333
1334 if (inc_counter) {
1335 // handle invocation counter overflow
1336 __ bind(invocation_counter_overflow);
1337 generate_counter_overflow(Lcontinue);
1338 }
1339
1340
1341
1342 return entry;
1343 }
1344
1345
1346 // Generic method entry to (asm) interpreter
1347 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1348 address entry = __ pc();
1349
1350 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1351
1352 // the following temporary registers are used during frame creation
1353 const Register Gtmp1 = G3_scratch ;
1354 const Register Gtmp2 = G1_scratch;
1355
1356 // make sure registers are different!
1357 assert_different_registers(G2_thread, G5_method, Gargs, Gtmp1, Gtmp2);
1358
1359 const Address constMethod (G5_method, Method::const_offset());
1360 // Seems like G5_method is live at the point this is used. So we could make this look consistent
1361 // and use in the asserts.
1362 const Address access_flags (Lmethod, Method::access_flags_offset());
1363
1364 const Register Glocals_size = G3;
1365 assert_different_registers(Glocals_size, G4_scratch, Gframe_size);
1366
1367 // make sure method is not native & not abstract
1368 // rethink these assertions - they can be simplified and shared (gri 2/25/2000)
1369 #ifdef ASSERT
1370 __ ld(G5_method, Method::access_flags_offset(), Gtmp1);
1371 {
1372 Label L;
1373 __ btst(JVM_ACC_NATIVE, Gtmp1);
1374 __ br(Assembler::zero, false, Assembler::pt, L);
1375 __ delayed()->nop();
1376 __ stop("tried to execute native method as non-native");
1377 __ bind(L);
1378 }
1379 { Label L;
1380 __ btst(JVM_ACC_ABSTRACT, Gtmp1);
1381 __ br(Assembler::zero, false, Assembler::pt, L);
1382 __ delayed()->nop();
1383 __ stop("tried to execute abstract method as non-abstract");
1384 __ bind(L);
1385 }
1386 #endif // ASSERT
1387
1388 // generate the code to allocate the interpreter stack frame
1389
1390 generate_fixed_frame(false);
1391
1392 #ifdef FAST_DISPATCH
1393 __ set((intptr_t)Interpreter::dispatch_table(), IdispatchTables);
1394 // set bytecode dispatch table base
1395 #endif
1396
1397 //
1398 // Code to initialize the extra (i.e. non-parm) locals
1399 //
1400 Register init_value = noreg; // will be G0 if we must clear locals
1401 // The way the code was setup before zerolocals was always true for vanilla java entries.
1402 // It could only be false for the specialized entries like accessor or empty which have
1403 // no extra locals so the testing was a waste of time and the extra locals were always
1404 // initialized. We removed this extra complication to already over complicated code.
1405
1406 init_value = G0;
1407 Label clear_loop;
1408
1409 const Register RconstMethod = O1;
1410 const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset());
1411 const Address size_of_locals (RconstMethod, ConstMethod::size_of_locals_offset());
1412
1413 // NOTE: If you change the frame layout, this code will need to
1414 // be updated!
1415 __ ld_ptr( constMethod, RconstMethod );
1416 __ lduh( size_of_locals, O2 );
1417 __ lduh( size_of_parameters, O1 );
1418 __ sll( O2, Interpreter::logStackElementSize, O2);
1419 __ sll( O1, Interpreter::logStackElementSize, O1 );
1420 __ sub( Llocals, O2, O2 );
1421 __ sub( Llocals, O1, O1 );
1422
1423 __ bind( clear_loop );
1424 __ inc( O2, wordSize );
1425
1426 __ cmp( O2, O1 );
1427 __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, clear_loop );
1428 __ delayed()->st_ptr( init_value, O2, 0 );
1429
1430 const Address do_not_unlock_if_synchronized(G2_thread,
1431 JavaThread::do_not_unlock_if_synchronized_offset());
1432 // Since at this point in the method invocation the exception handler
1433 // would try to exit the monitor of synchronized methods which hasn't
1434 // been entered yet, we set the thread local variable
1435 // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1436 // runtime, exception handling i.e. unlock_if_synchronized_method will
1437 // check this thread local flag.
1438 __ movbool(true, G3_scratch);
1439 __ stbool(G3_scratch, do_not_unlock_if_synchronized);
1440
1441 __ profile_parameters_type(G1_scratch, G3_scratch, G4_scratch, Lscratch);
1442 // increment invocation counter and check for overflow
1443 //
1444 // Note: checking for negative value instead of overflow
1445 // so we have a 'sticky' overflow test (may be of
1446 // importance as soon as we have true MT/MP)
1447 Label invocation_counter_overflow;
1448 Label profile_method;
1449 Label profile_method_continue;
1450 Label Lcontinue;
1451 if (inc_counter) {
1452 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1453 if (ProfileInterpreter) {
1454 __ bind(profile_method_continue);
1455 }
1456 }
1457 __ bind(Lcontinue);
1458
1459 bang_stack_shadow_pages(false);
1460
1461 // reset the _do_not_unlock_if_synchronized flag
1462 __ stbool(G0, do_not_unlock_if_synchronized);
1463
1464 // check for synchronized methods
1465 // Must happen AFTER invocation_counter check and stack overflow check,
1466 // so method is not locked if overflows.
1467
1468 if (synchronized) {
1469 lock_method();
1470 } else {
1471 #ifdef ASSERT
1472 { Label ok;
1473 __ ld(access_flags, O0);
1474 __ btst(JVM_ACC_SYNCHRONIZED, O0);
1475 __ br( Assembler::zero, false, Assembler::pt, ok);
1476 __ delayed()->nop();
1477 __ stop("method needs synchronization");
1478 __ bind(ok);
1479 }
1480 #endif // ASSERT
1481 }
1482
1483 // start execution
1484
1485 __ verify_thread();
1486
1487 // jvmti support
1488 __ notify_method_entry();
1489
1490 // start executing instructions
1491 __ dispatch_next(vtos);
1492
1493
1494 if (inc_counter) {
1495 if (ProfileInterpreter) {
1496 // We have decided to profile this method in the interpreter
1497 __ bind(profile_method);
1498
1499 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1500 __ set_method_data_pointer_for_bcp();
1501 __ ba_short(profile_method_continue);
1502 }
1503
1504 // handle invocation counter overflow
1505 __ bind(invocation_counter_overflow);
1506 generate_counter_overflow(Lcontinue);
1507 }
1508
1509
1510 return entry;
1511 }
1512
1513 //----------------------------------------------------------------------------------------------------
1514 // Exceptions
1515 void TemplateInterpreterGenerator::generate_throw_exception() {
1516
1517 // Entry point in previous activation (i.e., if the caller was interpreted)
1518 Interpreter::_rethrow_exception_entry = __ pc();
1519 // O0: exception
1520
1521 // entry point for exceptions thrown within interpreter code
1522 Interpreter::_throw_exception_entry = __ pc();
1523 __ verify_thread();
1524 // expression stack is undefined here
1525 // O0: exception, i.e. Oexception
1526 // Lbcp: exception bcp
1527 __ verify_oop(Oexception);
1528
1529
1530 // expression stack must be empty before entering the VM in case of an exception
1531 __ empty_expression_stack();
1532 // find exception handler address and preserve exception oop
1533 // call C routine to find handler and jump to it
1534 __ call_VM(O1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Oexception);
1535 __ push_ptr(O1); // push exception for exception handler bytecodes
1536
1537 __ JMP(O0, 0); // jump to exception handler (may be remove activation entry!)
1538 __ delayed()->nop();
1539
1540
1541 // if the exception is not handled in the current frame
1542 // the frame is removed and the exception is rethrown
1543 // (i.e. exception continuation is _rethrow_exception)
1544 //
1545 // Note: At this point the bci is still the bxi for the instruction which caused
1546 // the exception and the expression stack is empty. Thus, for any VM calls
1547 // at this point, GC will find a legal oop map (with empty expression stack).
1548
1549 // in current activation
1550 // tos: exception
1551 // Lbcp: exception bcp
1552
1553 //
1554 // JVMTI PopFrame support
1555 //
1556
1557 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1558 Address popframe_condition_addr(G2_thread, JavaThread::popframe_condition_offset());
1559 // Set the popframe_processing bit in popframe_condition indicating that we are
1560 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1561 // popframe handling cycles.
1562
1563 __ ld(popframe_condition_addr, G3_scratch);
1564 __ or3(G3_scratch, JavaThread::popframe_processing_bit, G3_scratch);
1565 __ stw(G3_scratch, popframe_condition_addr);
1566
1567 // Empty the expression stack, as in normal exception handling
1568 __ empty_expression_stack();
1569 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
1570
1571 {
1572 // Check to see whether we are returning to a deoptimized frame.
1573 // (The PopFrame call ensures that the caller of the popped frame is
1574 // either interpreted or compiled and deoptimizes it if compiled.)
1575 // In this case, we can't call dispatch_next() after the frame is
1576 // popped, but instead must save the incoming arguments and restore
1577 // them after deoptimization has occurred.
1578 //
1579 // Note that we don't compare the return PC against the
1580 // deoptimization blob's unpack entry because of the presence of
1581 // adapter frames in C2.
1582 Label caller_not_deoptimized;
1583 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), I7);
1584 __ br_notnull_short(O0, Assembler::pt, caller_not_deoptimized);
1585
1586 const Register Gtmp1 = G3_scratch;
1587 const Register Gtmp2 = G1_scratch;
1588 const Register RconstMethod = Gtmp1;
1589 const Address constMethod(Lmethod, Method::const_offset());
1590 const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset());
1591
1592 // Compute size of arguments for saving when returning to deoptimized caller
1593 __ ld_ptr(constMethod, RconstMethod);
1594 __ lduh(size_of_parameters, Gtmp1);
1595 __ sll(Gtmp1, Interpreter::logStackElementSize, Gtmp1);
1596 __ sub(Llocals, Gtmp1, Gtmp2);
1597 __ add(Gtmp2, wordSize, Gtmp2);
1598 // Save these arguments
1599 __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), G2_thread, Gtmp1, Gtmp2);
1600 // Inform deoptimization that it is responsible for restoring these arguments
1601 __ set(JavaThread::popframe_force_deopt_reexecution_bit, Gtmp1);
1602 Address popframe_condition_addr(G2_thread, JavaThread::popframe_condition_offset());
1603 __ st(Gtmp1, popframe_condition_addr);
1604
1605 // Return from the current method
1606 // The caller's SP was adjusted upon method entry to accomodate
1607 // the callee's non-argument locals. Undo that adjustment.
1608 __ ret();
1609 __ delayed()->restore(I5_savedSP, G0, SP);
1610
1611 __ bind(caller_not_deoptimized);
1612 }
1613
1614 // Clear the popframe condition flag
1615 __ stw(G0 /* popframe_inactive */, popframe_condition_addr);
1616
1617 // Get out of the current method (how this is done depends on the particular compiler calling
1618 // convention that the interpreter currently follows)
1619 // The caller's SP was adjusted upon method entry to accomodate
1620 // the callee's non-argument locals. Undo that adjustment.
1621 __ restore(I5_savedSP, G0, SP);
1622 // The method data pointer was incremented already during
1623 // call profiling. We have to restore the mdp for the current bcp.
1624 if (ProfileInterpreter) {
1625 __ set_method_data_pointer_for_bcp();
1626 }
1627
1628 #if INCLUDE_JVMTI
1629 {
1630 Label L_done;
1631
1632 __ ldub(Address(Lbcp, 0), G1_scratch); // Load current bytecode
1633 __ cmp_and_br_short(G1_scratch, Bytecodes::_invokestatic, Assembler::notEqual, Assembler::pn, L_done);
1634
1635 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1636 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1637
1638 __ call_VM(G1_scratch, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), I0, Lmethod, Lbcp);
1639
1640 __ br_null(G1_scratch, false, Assembler::pn, L_done);
1641 __ delayed()->nop();
1642
1643 __ st_ptr(G1_scratch, Lesp, wordSize);
1644 __ bind(L_done);
1645 }
1646 #endif // INCLUDE_JVMTI
1647
1648 // Resume bytecode interpretation at the current bcp
1649 __ dispatch_next(vtos);
1650 // end of JVMTI PopFrame support
1651
1652 Interpreter::_remove_activation_entry = __ pc();
1653
1654 // preserve exception over this code sequence (remove activation calls the vm, but oopmaps are not correct here)
1655 __ pop_ptr(Oexception); // get exception
1656
1657 // Intel has the following comment:
1658 //// remove the activation (without doing throws on illegalMonitorExceptions)
1659 // They remove the activation without checking for bad monitor state.
1660 // %%% We should make sure this is the right semantics before implementing.
1661
1662 __ set_vm_result(Oexception);
1663 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false);
1664
1665 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI);
1666
1667 __ get_vm_result(Oexception);
1668 __ verify_oop(Oexception);
1669
1670 const int return_reg_adjustment = frame::pc_return_offset;
1671 Address issuing_pc_addr(I7, return_reg_adjustment);
1672
1673 // We are done with this activation frame; find out where to go next.
1674 // The continuation point will be an exception handler, which expects
1675 // the following registers set up:
1676 //
1677 // Oexception: exception
1678 // Oissuing_pc: the local call that threw exception
1679 // Other On: garbage
1680 // In/Ln: the contents of the caller's register window
1681 //
1682 // We do the required restore at the last possible moment, because we
1683 // need to preserve some state across a runtime call.
1684 // (Remember that the caller activation is unknown--it might not be
1685 // interpreted, so things like Lscratch are useless in the caller.)
1686
1687 // Although the Intel version uses call_C, we can use the more
1688 // compact call_VM. (The only real difference on SPARC is a
1689 // harmlessly ignored [re]set_last_Java_frame, compared with
1690 // the Intel code which lacks this.)
1691 __ mov(Oexception, Oexception ->after_save()); // get exception in I0 so it will be on O0 after restore
1692 __ add(issuing_pc_addr, Oissuing_pc->after_save()); // likewise set I1 to a value local to the caller
1693 __ super_call_VM_leaf(L7_thread_cache,
1694 CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1695 G2_thread, Oissuing_pc->after_save());
1696
1697 // The caller's SP was adjusted upon method entry to accomodate
1698 // the callee's non-argument locals. Undo that adjustment.
1699 __ JMP(O0, 0); // return exception handler in caller
1700 __ delayed()->restore(I5_savedSP, G0, SP);
1701
1702 // (same old exception object is already in Oexception; see above)
1703 // Note that an "issuing PC" is actually the next PC after the call
1704 }
1705
1706
1707 //
1708 // JVMTI ForceEarlyReturn support
1709 //
1710
1711 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1712 address entry = __ pc();
1713
1714 __ empty_expression_stack();
1715 __ load_earlyret_value(state);
1716
1717 __ ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), G3_scratch);
1718 Address cond_addr(G3_scratch, JvmtiThreadState::earlyret_state_offset());
1719
1720 // Clear the earlyret state
1721 __ stw(G0 /* JvmtiThreadState::earlyret_inactive */, cond_addr);
1722
1723 __ remove_activation(state,
1724 /* throw_monitor_exception */ false,
1725 /* install_monitor_exception */ false);
1726
1727 // The caller's SP was adjusted upon method entry to accomodate
1728 // the callee's non-argument locals. Undo that adjustment.
1729 __ ret(); // return to caller
1730 __ delayed()->restore(I5_savedSP, G0, SP);
1731
1732 return entry;
1733 } // end of JVMTI ForceEarlyReturn support
1734
1735
1736 //------------------------------------------------------------------------------------------------------------------------
1737 // Helper for vtos entry point generation
1738
1739 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) {
1740 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1741 Label L;
1742 aep = __ pc(); __ push_ptr(); __ ba_short(L);
1743 fep = __ pc(); __ push_f(); __ ba_short(L);
1744 dep = __ pc(); __ push_d(); __ ba_short(L);
1745 lep = __ pc(); __ push_l(); __ ba_short(L);
1746 iep = __ pc(); __ push_i();
1747 bep = cep = sep = iep; // there aren't any
1748 vep = __ pc(); __ bind(L); // fall through
1749 generate_and_dispatch(t);
1750 }
1751
1752 // --------------------------------------------------------------------------------
1753
1754 // Non-product code
1755 #ifndef PRODUCT
1756 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1757 address entry = __ pc();
1758
1759 __ push(state);
1760 __ mov(O7, Lscratch); // protect return address within interpreter
1761
1762 // Pass a 0 (not used in sparc) and the top of stack to the bytecode tracer
1763 __ mov( Otos_l2, G3_scratch );
1764 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), G0, Otos_l1, G3_scratch);
1765 __ mov(Lscratch, O7); // restore return address
1766 __ pop(state);
1767 __ retl();
1768 __ delayed()->nop();
1769
1770 return entry;
1771 }
1772
1773
1774 // helpers for generate_and_dispatch
1775
1776 void TemplateInterpreterGenerator::count_bytecode() {
1777 __ inc_counter(&BytecodeCounter::_counter_value, G3_scratch, G4_scratch);
1778 }
1779
1780
1781 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1782 __ inc_counter(&BytecodeHistogram::_counters[t->bytecode()], G3_scratch, G4_scratch);
1783 }
1784
1785
1786 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1787 AddressLiteral index (&BytecodePairHistogram::_index);
1788 AddressLiteral counters((address) &BytecodePairHistogram::_counters);
1789
1790 // get index, shift out old bytecode, bring in new bytecode, and store it
1791 // _index = (_index >> log2_number_of_codes) |
1792 // (bytecode << log2_number_of_codes);
1793
1794 __ load_contents(index, G4_scratch);
1795 __ srl( G4_scratch, BytecodePairHistogram::log2_number_of_codes, G4_scratch );
1796 __ set( ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes, G3_scratch );
1797 __ or3( G3_scratch, G4_scratch, G4_scratch );
1798 __ store_contents(G4_scratch, index, G3_scratch);
1799
1800 // bump bucket contents
1801 // _counters[_index] ++;
1802
1803 __ set(counters, G3_scratch); // loads into G3_scratch
1804 __ sll( G4_scratch, LogBytesPerWord, G4_scratch ); // Index is word address
1805 __ add (G3_scratch, G4_scratch, G3_scratch); // Add in index
1806 __ ld (G3_scratch, 0, G4_scratch);
1807 __ inc (G4_scratch);
1808 __ st (G4_scratch, 0, G3_scratch);
1809 }
1810
1811
1812 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1813 // Call a little run-time stub to avoid blow-up for each bytecode.
1814 // The run-time runtime saves the right registers, depending on
1815 // the tosca in-state for the given template.
1816 address entry = Interpreter::trace_code(t->tos_in());
1817 guarantee(entry != NULL, "entry must have been generated");
1818 __ call(entry, relocInfo::none);
1819 __ delayed()->nop();
1820 }
1821
1822
1823 void TemplateInterpreterGenerator::stop_interpreter_at() {
1824 AddressLiteral counter(&BytecodeCounter::_counter_value);
1825 __ load_contents(counter, G3_scratch);
1826 AddressLiteral stop_at(&StopInterpreterAt);
1827 __ load_ptr_contents(stop_at, G4_scratch);
1828 __ cmp(G3_scratch, G4_scratch);
1829 __ breakpoint_trap(Assembler::equal, Assembler::icc);
1830 }
1831 #endif // not PRODUCT