templateTable_sparc.cpp revision 6760:22b98ab2a69f
1/*
2 * Copyright (c) 1997, 2014, 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 "interpreter/interpreter.hpp"
27#include "interpreter/interpreterRuntime.hpp"
28#include "interpreter/interp_masm.hpp"
29#include "interpreter/templateTable.hpp"
30#include "memory/universe.inline.hpp"
31#include "oops/methodData.hpp"
32#include "oops/objArrayKlass.hpp"
33#include "oops/oop.inline.hpp"
34#include "prims/methodHandles.hpp"
35#include "runtime/sharedRuntime.hpp"
36#include "runtime/stubRoutines.hpp"
37#include "runtime/synchronizer.hpp"
38#include "utilities/macros.hpp"
39
40#ifndef CC_INTERP
41#define __ _masm->
42
43// Misc helpers
44
45// Do an oop store like *(base + index + offset) = val
46// index can be noreg,
47static void do_oop_store(InterpreterMacroAssembler* _masm,
48                         Register base,
49                         Register index,
50                         int offset,
51                         Register val,
52                         Register tmp,
53                         BarrierSet::Name barrier,
54                         bool precise) {
55  assert(tmp != val && tmp != base && tmp != index, "register collision");
56  assert(index == noreg || offset == 0, "only one offset");
57  switch (barrier) {
58#if INCLUDE_ALL_GCS
59    case BarrierSet::G1SATBCT:
60    case BarrierSet::G1SATBCTLogging:
61      {
62        // Load and record the previous value.
63        __ g1_write_barrier_pre(base, index, offset,
64                                noreg /* pre_val */,
65                                tmp, true /*preserve_o_regs*/);
66
67        // G1 barrier needs uncompressed oop for region cross check.
68        Register new_val = val;
69        if (UseCompressedOops && val != G0) {
70          new_val = tmp;
71          __ mov(val, new_val);
72        }
73
74        if (index == noreg ) {
75          assert(Assembler::is_simm13(offset), "fix this code");
76          __ store_heap_oop(val, base, offset);
77        } else {
78          __ store_heap_oop(val, base, index);
79        }
80
81        // No need for post barrier if storing NULL
82        if (val != G0) {
83          if (precise) {
84            if (index == noreg) {
85              __ add(base, offset, base);
86            } else {
87              __ add(base, index, base);
88            }
89          }
90          __ g1_write_barrier_post(base, new_val, tmp);
91        }
92      }
93      break;
94#endif // INCLUDE_ALL_GCS
95    case BarrierSet::CardTableModRef:
96    case BarrierSet::CardTableExtension:
97      {
98        if (index == noreg ) {
99          assert(Assembler::is_simm13(offset), "fix this code");
100          __ store_heap_oop(val, base, offset);
101        } else {
102          __ store_heap_oop(val, base, index);
103        }
104        // No need for post barrier if storing NULL
105        if (val != G0) {
106          if (precise) {
107            if (index == noreg) {
108              __ add(base, offset, base);
109            } else {
110              __ add(base, index, base);
111            }
112          }
113          __ card_write_barrier_post(base, val, tmp);
114        }
115      }
116      break;
117    case BarrierSet::ModRef:
118    case BarrierSet::Other:
119      ShouldNotReachHere();
120      break;
121    default      :
122      ShouldNotReachHere();
123
124  }
125}
126
127
128//----------------------------------------------------------------------------------------------------
129// Platform-dependent initialization
130
131void TemplateTable::pd_initialize() {
132  // (none)
133}
134
135
136//----------------------------------------------------------------------------------------------------
137// Condition conversion
138Assembler::Condition ccNot(TemplateTable::Condition cc) {
139  switch (cc) {
140    case TemplateTable::equal        : return Assembler::notEqual;
141    case TemplateTable::not_equal    : return Assembler::equal;
142    case TemplateTable::less         : return Assembler::greaterEqual;
143    case TemplateTable::less_equal   : return Assembler::greater;
144    case TemplateTable::greater      : return Assembler::lessEqual;
145    case TemplateTable::greater_equal: return Assembler::less;
146  }
147  ShouldNotReachHere();
148  return Assembler::zero;
149}
150
151//----------------------------------------------------------------------------------------------------
152// Miscelaneous helper routines
153
154
155Address TemplateTable::at_bcp(int offset) {
156  assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
157  return Address(Lbcp, offset);
158}
159
160
161void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
162                                   Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
163                                   int byte_no) {
164  // With sharing on, may need to test Method* flag.
165  if (!RewriteBytecodes)  return;
166  Label L_patch_done;
167
168  switch (bc) {
169  case Bytecodes::_fast_aputfield:
170  case Bytecodes::_fast_bputfield:
171  case Bytecodes::_fast_cputfield:
172  case Bytecodes::_fast_dputfield:
173  case Bytecodes::_fast_fputfield:
174  case Bytecodes::_fast_iputfield:
175  case Bytecodes::_fast_lputfield:
176  case Bytecodes::_fast_sputfield:
177    {
178      // We skip bytecode quickening for putfield instructions when
179      // the put_code written to the constant pool cache is zero.
180      // This is required so that every execution of this instruction
181      // calls out to InterpreterRuntime::resolve_get_put to do
182      // additional, required work.
183      assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
184      assert(load_bc_into_bc_reg, "we use bc_reg as temp");
185      __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1);
186      __ set(bc, bc_reg);
187      __ cmp_and_br_short(temp_reg, 0, Assembler::equal, Assembler::pn, L_patch_done);  // don't patch
188    }
189    break;
190  default:
191    assert(byte_no == -1, "sanity");
192    if (load_bc_into_bc_reg) {
193      __ set(bc, bc_reg);
194    }
195  }
196
197  if (JvmtiExport::can_post_breakpoint()) {
198    Label L_fast_patch;
199    __ ldub(at_bcp(0), temp_reg);
200    __ cmp_and_br_short(temp_reg, Bytecodes::_breakpoint, Assembler::notEqual, Assembler::pt, L_fast_patch);
201    // perform the quickening, slowly, in the bowels of the breakpoint table
202    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), Lmethod, Lbcp, bc_reg);
203    __ ba_short(L_patch_done);
204    __ bind(L_fast_patch);
205  }
206
207#ifdef ASSERT
208  Bytecodes::Code orig_bytecode =  Bytecodes::java_code(bc);
209  Label L_okay;
210  __ ldub(at_bcp(0), temp_reg);
211  __ cmp(temp_reg, orig_bytecode);
212  __ br(Assembler::equal, false, Assembler::pt, L_okay);
213  __ delayed()->cmp(temp_reg, bc_reg);
214  __ br(Assembler::equal, false, Assembler::pt, L_okay);
215  __ delayed()->nop();
216  __ stop("patching the wrong bytecode");
217  __ bind(L_okay);
218#endif
219
220  // patch bytecode
221  __ stb(bc_reg, at_bcp(0));
222  __ bind(L_patch_done);
223}
224
225//----------------------------------------------------------------------------------------------------
226// Individual instructions
227
228void TemplateTable::nop() {
229  transition(vtos, vtos);
230  // nothing to do
231}
232
233void TemplateTable::shouldnotreachhere() {
234  transition(vtos, vtos);
235  __ stop("shouldnotreachhere bytecode");
236}
237
238void TemplateTable::aconst_null() {
239  transition(vtos, atos);
240  __ clr(Otos_i);
241}
242
243
244void TemplateTable::iconst(int value) {
245  transition(vtos, itos);
246  __ set(value, Otos_i);
247}
248
249
250void TemplateTable::lconst(int value) {
251  transition(vtos, ltos);
252  assert(value >= 0, "check this code");
253#ifdef _LP64
254  __ set(value, Otos_l);
255#else
256  __ set(value, Otos_l2);
257  __ clr( Otos_l1);
258#endif
259}
260
261
262void TemplateTable::fconst(int value) {
263  transition(vtos, ftos);
264  static float zero = 0.0, one = 1.0, two = 2.0;
265  float* p;
266  switch( value ) {
267   default: ShouldNotReachHere();
268   case 0:  p = &zero;  break;
269   case 1:  p = &one;   break;
270   case 2:  p = &two;   break;
271  }
272  AddressLiteral a(p);
273  __ sethi(a, G3_scratch);
274  __ ldf(FloatRegisterImpl::S, G3_scratch, a.low10(), Ftos_f);
275}
276
277
278void TemplateTable::dconst(int value) {
279  transition(vtos, dtos);
280  static double zero = 0.0, one = 1.0;
281  double* p;
282  switch( value ) {
283   default: ShouldNotReachHere();
284   case 0:  p = &zero;  break;
285   case 1:  p = &one;   break;
286  }
287  AddressLiteral a(p);
288  __ sethi(a, G3_scratch);
289  __ ldf(FloatRegisterImpl::D, G3_scratch, a.low10(), Ftos_d);
290}
291
292
293// %%%%% Should factore most snippet templates across platforms
294
295void TemplateTable::bipush() {
296  transition(vtos, itos);
297  __ ldsb( at_bcp(1), Otos_i );
298}
299
300void TemplateTable::sipush() {
301  transition(vtos, itos);
302  __ get_2_byte_integer_at_bcp(1, G3_scratch, Otos_i, InterpreterMacroAssembler::Signed);
303}
304
305void TemplateTable::ldc(bool wide) {
306  transition(vtos, vtos);
307  Label call_ldc, notInt, isString, notString, notClass, exit;
308
309  if (wide) {
310    __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
311  } else {
312    __ ldub(Lbcp, 1, O1);
313  }
314  __ get_cpool_and_tags(O0, O2);
315
316  const int base_offset = ConstantPool::header_size() * wordSize;
317  const int tags_offset = Array<u1>::base_offset_in_bytes();
318
319  // get type from tags
320  __ add(O2, tags_offset, O2);
321  __ ldub(O2, O1, O2);
322
323  // unresolved class? If so, must resolve
324  __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClass, Assembler::equal, Assembler::pt, call_ldc);
325
326  // unresolved class in error state
327  __ cmp_and_brx_short(O2, JVM_CONSTANT_UnresolvedClassInError, Assembler::equal, Assembler::pn, call_ldc);
328
329  __ cmp(O2, JVM_CONSTANT_Class);      // need to call vm to get java mirror of the class
330  __ brx(Assembler::notEqual, true, Assembler::pt, notClass);
331  __ delayed()->add(O0, base_offset, O0);
332
333  __ bind(call_ldc);
334  __ set(wide, O1);
335  call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), O1);
336  __ push(atos);
337  __ ba_short(exit);
338
339  __ bind(notClass);
340 // __ add(O0, base_offset, O0);
341  __ sll(O1, LogBytesPerWord, O1);
342  __ cmp(O2, JVM_CONSTANT_Integer);
343  __ brx(Assembler::notEqual, true, Assembler::pt, notInt);
344  __ delayed()->cmp(O2, JVM_CONSTANT_String);
345  __ ld(O0, O1, Otos_i);
346  __ push(itos);
347  __ ba_short(exit);
348
349  __ bind(notInt);
350 // __ cmp(O2, JVM_CONSTANT_String);
351  __ brx(Assembler::notEqual, true, Assembler::pt, notString);
352  __ delayed()->ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
353  __ bind(isString);
354  __ stop("string should be rewritten to fast_aldc");
355  __ ba_short(exit);
356
357  __ bind(notString);
358 // __ ldf(FloatRegisterImpl::S, O0, O1, Ftos_f);
359  __ push(ftos);
360
361  __ bind(exit);
362}
363
364// Fast path for caching oop constants.
365// %%% We should use this to handle Class and String constants also.
366// %%% It will simplify the ldc/primitive path considerably.
367void TemplateTable::fast_aldc(bool wide) {
368  transition(vtos, atos);
369
370  int index_size = wide ? sizeof(u2) : sizeof(u1);
371  Label resolved;
372
373  // We are resolved if the resolved reference cache entry contains a
374  // non-null object (CallSite, etc.)
375  assert_different_registers(Otos_i, G3_scratch);
376  __ get_cache_index_at_bcp(Otos_i, G3_scratch, 1, index_size);  // load index => G3_scratch
377  __ load_resolved_reference_at_index(Otos_i, G3_scratch);
378  __ tst(Otos_i);
379  __ br(Assembler::notEqual, false, Assembler::pt, resolved);
380  __ delayed()->set((int)bytecode(), O1);
381
382  address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
383
384  // first time invocation - must resolve first
385  __ call_VM(Otos_i, entry, O1);
386  __ bind(resolved);
387  __ verify_oop(Otos_i);
388}
389
390
391void TemplateTable::ldc2_w() {
392  transition(vtos, vtos);
393  Label Long, exit;
394
395  __ get_2_byte_integer_at_bcp(1, G3_scratch, O1, InterpreterMacroAssembler::Unsigned);
396  __ get_cpool_and_tags(O0, O2);
397
398  const int base_offset = ConstantPool::header_size() * wordSize;
399  const int tags_offset = Array<u1>::base_offset_in_bytes();
400  // get type from tags
401  __ add(O2, tags_offset, O2);
402  __ ldub(O2, O1, O2);
403
404  __ sll(O1, LogBytesPerWord, O1);
405  __ add(O0, O1, G3_scratch);
406
407  __ cmp_and_brx_short(O2, JVM_CONSTANT_Double, Assembler::notEqual, Assembler::pt, Long);
408  // A double can be placed at word-aligned locations in the constant pool.
409  // Check out Conversions.java for an example.
410  // Also ConstantPool::header_size() is 20, which makes it very difficult
411  // to double-align double on the constant pool.  SG, 11/7/97
412#ifdef _LP64
413  __ ldf(FloatRegisterImpl::D, G3_scratch, base_offset, Ftos_d);
414#else
415  FloatRegister f = Ftos_d;
416  __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset, f);
417  __ ldf(FloatRegisterImpl::S, G3_scratch, base_offset + sizeof(jdouble)/2,
418         f->successor());
419#endif
420  __ push(dtos);
421  __ ba_short(exit);
422
423  __ bind(Long);
424#ifdef _LP64
425  __ ldx(G3_scratch, base_offset, Otos_l);
426#else
427  __ ld(G3_scratch, base_offset, Otos_l);
428  __ ld(G3_scratch, base_offset + sizeof(jlong)/2, Otos_l->successor());
429#endif
430  __ push(ltos);
431
432  __ bind(exit);
433}
434
435
436void TemplateTable::locals_index(Register reg, int offset) {
437  __ ldub( at_bcp(offset), reg );
438}
439
440
441void TemplateTable::locals_index_wide(Register reg) {
442  // offset is 2, not 1, because Lbcp points to wide prefix code
443  __ get_2_byte_integer_at_bcp(2, G4_scratch, reg, InterpreterMacroAssembler::Unsigned);
444}
445
446void TemplateTable::iload() {
447  transition(vtos, itos);
448  // Rewrite iload,iload  pair into fast_iload2
449  //         iload,caload pair into fast_icaload
450  if (RewriteFrequentPairs) {
451    Label rewrite, done;
452
453    // get next byte
454    __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_iload)), G3_scratch);
455
456    // if _iload, wait to rewrite to iload2.  We only want to rewrite the
457    // last two iloads in a pair.  Comparing against fast_iload means that
458    // the next bytecode is neither an iload or a caload, and therefore
459    // an iload pair.
460    __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_iload, Assembler::equal, Assembler::pn, done);
461
462    __ cmp(G3_scratch, (int)Bytecodes::_fast_iload);
463    __ br(Assembler::equal, false, Assembler::pn, rewrite);
464    __ delayed()->set(Bytecodes::_fast_iload2, G4_scratch);
465
466    __ cmp(G3_scratch, (int)Bytecodes::_caload);
467    __ br(Assembler::equal, false, Assembler::pn, rewrite);
468    __ delayed()->set(Bytecodes::_fast_icaload, G4_scratch);
469
470    __ set(Bytecodes::_fast_iload, G4_scratch);  // don't check again
471    // rewrite
472    // G4_scratch: fast bytecode
473    __ bind(rewrite);
474    patch_bytecode(Bytecodes::_iload, G4_scratch, G3_scratch, false);
475    __ bind(done);
476  }
477
478  // Get the local value into tos
479  locals_index(G3_scratch);
480  __ access_local_int( G3_scratch, Otos_i );
481}
482
483void TemplateTable::fast_iload2() {
484  transition(vtos, itos);
485  locals_index(G3_scratch);
486  __ access_local_int( G3_scratch, Otos_i );
487  __ push_i();
488  locals_index(G3_scratch, 3);  // get next bytecode's local index.
489  __ access_local_int( G3_scratch, Otos_i );
490}
491
492void TemplateTable::fast_iload() {
493  transition(vtos, itos);
494  locals_index(G3_scratch);
495  __ access_local_int( G3_scratch, Otos_i );
496}
497
498void TemplateTable::lload() {
499  transition(vtos, ltos);
500  locals_index(G3_scratch);
501  __ access_local_long( G3_scratch, Otos_l );
502}
503
504
505void TemplateTable::fload() {
506  transition(vtos, ftos);
507  locals_index(G3_scratch);
508  __ access_local_float( G3_scratch, Ftos_f );
509}
510
511
512void TemplateTable::dload() {
513  transition(vtos, dtos);
514  locals_index(G3_scratch);
515  __ access_local_double( G3_scratch, Ftos_d );
516}
517
518
519void TemplateTable::aload() {
520  transition(vtos, atos);
521  locals_index(G3_scratch);
522  __ access_local_ptr( G3_scratch, Otos_i);
523}
524
525
526void TemplateTable::wide_iload() {
527  transition(vtos, itos);
528  locals_index_wide(G3_scratch);
529  __ access_local_int( G3_scratch, Otos_i );
530}
531
532
533void TemplateTable::wide_lload() {
534  transition(vtos, ltos);
535  locals_index_wide(G3_scratch);
536  __ access_local_long( G3_scratch, Otos_l );
537}
538
539
540void TemplateTable::wide_fload() {
541  transition(vtos, ftos);
542  locals_index_wide(G3_scratch);
543  __ access_local_float( G3_scratch, Ftos_f );
544}
545
546
547void TemplateTable::wide_dload() {
548  transition(vtos, dtos);
549  locals_index_wide(G3_scratch);
550  __ access_local_double( G3_scratch, Ftos_d );
551}
552
553
554void TemplateTable::wide_aload() {
555  transition(vtos, atos);
556  locals_index_wide(G3_scratch);
557  __ access_local_ptr( G3_scratch, Otos_i );
558  __ verify_oop(Otos_i);
559}
560
561
562void TemplateTable::iaload() {
563  transition(itos, itos);
564  // Otos_i: index
565  // tos: array
566  __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
567  __ ld(O3, arrayOopDesc::base_offset_in_bytes(T_INT), Otos_i);
568}
569
570
571void TemplateTable::laload() {
572  transition(itos, ltos);
573  // Otos_i: index
574  // O2: array
575  __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
576  __ ld_long(O3, arrayOopDesc::base_offset_in_bytes(T_LONG), Otos_l);
577}
578
579
580void TemplateTable::faload() {
581  transition(itos, ftos);
582  // Otos_i: index
583  // O2: array
584  __ index_check(O2, Otos_i, LogBytesPerInt, G3_scratch, O3);
585  __ ldf(FloatRegisterImpl::S, O3, arrayOopDesc::base_offset_in_bytes(T_FLOAT), Ftos_f);
586}
587
588
589void TemplateTable::daload() {
590  transition(itos, dtos);
591  // Otos_i: index
592  // O2: array
593  __ index_check(O2, Otos_i, LogBytesPerLong, G3_scratch, O3);
594  __ ldf(FloatRegisterImpl::D, O3, arrayOopDesc::base_offset_in_bytes(T_DOUBLE), Ftos_d);
595}
596
597
598void TemplateTable::aaload() {
599  transition(itos, atos);
600  // Otos_i: index
601  // tos: array
602  __ index_check(O2, Otos_i, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O3);
603  __ load_heap_oop(O3, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i);
604  __ verify_oop(Otos_i);
605}
606
607
608void TemplateTable::baload() {
609  transition(itos, itos);
610  // Otos_i: index
611  // tos: array
612  __ index_check(O2, Otos_i, 0, G3_scratch, O3);
613  __ ldsb(O3, arrayOopDesc::base_offset_in_bytes(T_BYTE), Otos_i);
614}
615
616
617void TemplateTable::caload() {
618  transition(itos, itos);
619  // Otos_i: index
620  // tos: array
621  __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
622  __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
623}
624
625void TemplateTable::fast_icaload() {
626  transition(vtos, itos);
627  // Otos_i: index
628  // tos: array
629  locals_index(G3_scratch);
630  __ access_local_int( G3_scratch, Otos_i );
631  __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
632  __ lduh(O3, arrayOopDesc::base_offset_in_bytes(T_CHAR), Otos_i);
633}
634
635
636void TemplateTable::saload() {
637  transition(itos, itos);
638  // Otos_i: index
639  // tos: array
640  __ index_check(O2, Otos_i, LogBytesPerShort, G3_scratch, O3);
641  __ ldsh(O3, arrayOopDesc::base_offset_in_bytes(T_SHORT), Otos_i);
642}
643
644
645void TemplateTable::iload(int n) {
646  transition(vtos, itos);
647  __ ld( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
648}
649
650
651void TemplateTable::lload(int n) {
652  transition(vtos, ltos);
653  assert(n+1 < Argument::n_register_parameters, "would need more code");
654  __ load_unaligned_long(Llocals, Interpreter::local_offset_in_bytes(n+1), Otos_l);
655}
656
657
658void TemplateTable::fload(int n) {
659  transition(vtos, ftos);
660  assert(n < Argument::n_register_parameters, "would need more code");
661  __ ldf( FloatRegisterImpl::S, Llocals, Interpreter::local_offset_in_bytes(n),     Ftos_f );
662}
663
664
665void TemplateTable::dload(int n) {
666  transition(vtos, dtos);
667  FloatRegister dst = Ftos_d;
668  __ load_unaligned_double(Llocals, Interpreter::local_offset_in_bytes(n+1), dst);
669}
670
671
672void TemplateTable::aload(int n) {
673  transition(vtos, atos);
674  __ ld_ptr( Llocals, Interpreter::local_offset_in_bytes(n), Otos_i );
675}
676
677
678void TemplateTable::aload_0() {
679  transition(vtos, atos);
680
681  // According to bytecode histograms, the pairs:
682  //
683  // _aload_0, _fast_igetfield (itos)
684  // _aload_0, _fast_agetfield (atos)
685  // _aload_0, _fast_fgetfield (ftos)
686  //
687  // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
688  // bytecode checks the next bytecode and then rewrites the current
689  // bytecode into a pair bytecode; otherwise it rewrites the current
690  // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
691  //
692  if (RewriteFrequentPairs) {
693    Label rewrite, done;
694
695    // get next byte
696    __ ldub(at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)), G3_scratch);
697
698    // do actual aload_0
699    aload(0);
700
701    // if _getfield then wait with rewrite
702    __ cmp_and_br_short(G3_scratch, (int)Bytecodes::_getfield, Assembler::equal, Assembler::pn, done);
703
704    // if _igetfield then rewrite to _fast_iaccess_0
705    assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
706    __ cmp(G3_scratch, (int)Bytecodes::_fast_igetfield);
707    __ br(Assembler::equal, false, Assembler::pn, rewrite);
708    __ delayed()->set(Bytecodes::_fast_iaccess_0, G4_scratch);
709
710    // if _agetfield then rewrite to _fast_aaccess_0
711    assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
712    __ cmp(G3_scratch, (int)Bytecodes::_fast_agetfield);
713    __ br(Assembler::equal, false, Assembler::pn, rewrite);
714    __ delayed()->set(Bytecodes::_fast_aaccess_0, G4_scratch);
715
716    // if _fgetfield then rewrite to _fast_faccess_0
717    assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
718    __ cmp(G3_scratch, (int)Bytecodes::_fast_fgetfield);
719    __ br(Assembler::equal, false, Assembler::pn, rewrite);
720    __ delayed()->set(Bytecodes::_fast_faccess_0, G4_scratch);
721
722    // else rewrite to _fast_aload0
723    assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "adjust fast bytecode def");
724    __ set(Bytecodes::_fast_aload_0, G4_scratch);
725
726    // rewrite
727    // G4_scratch: fast bytecode
728    __ bind(rewrite);
729    patch_bytecode(Bytecodes::_aload_0, G4_scratch, G3_scratch, false);
730    __ bind(done);
731  } else {
732    aload(0);
733  }
734}
735
736
737void TemplateTable::istore() {
738  transition(itos, vtos);
739  locals_index(G3_scratch);
740  __ store_local_int( G3_scratch, Otos_i );
741}
742
743
744void TemplateTable::lstore() {
745  transition(ltos, vtos);
746  locals_index(G3_scratch);
747  __ store_local_long( G3_scratch, Otos_l );
748}
749
750
751void TemplateTable::fstore() {
752  transition(ftos, vtos);
753  locals_index(G3_scratch);
754  __ store_local_float( G3_scratch, Ftos_f );
755}
756
757
758void TemplateTable::dstore() {
759  transition(dtos, vtos);
760  locals_index(G3_scratch);
761  __ store_local_double( G3_scratch, Ftos_d );
762}
763
764
765void TemplateTable::astore() {
766  transition(vtos, vtos);
767  __ load_ptr(0, Otos_i);
768  __ inc(Lesp, Interpreter::stackElementSize);
769  __ verify_oop_or_return_address(Otos_i, G3_scratch);
770  locals_index(G3_scratch);
771  __ store_local_ptr(G3_scratch, Otos_i);
772}
773
774
775void TemplateTable::wide_istore() {
776  transition(vtos, vtos);
777  __ pop_i();
778  locals_index_wide(G3_scratch);
779  __ store_local_int( G3_scratch, Otos_i );
780}
781
782
783void TemplateTable::wide_lstore() {
784  transition(vtos, vtos);
785  __ pop_l();
786  locals_index_wide(G3_scratch);
787  __ store_local_long( G3_scratch, Otos_l );
788}
789
790
791void TemplateTable::wide_fstore() {
792  transition(vtos, vtos);
793  __ pop_f();
794  locals_index_wide(G3_scratch);
795  __ store_local_float( G3_scratch, Ftos_f );
796}
797
798
799void TemplateTable::wide_dstore() {
800  transition(vtos, vtos);
801  __ pop_d();
802  locals_index_wide(G3_scratch);
803  __ store_local_double( G3_scratch, Ftos_d );
804}
805
806
807void TemplateTable::wide_astore() {
808  transition(vtos, vtos);
809  __ load_ptr(0, Otos_i);
810  __ inc(Lesp, Interpreter::stackElementSize);
811  __ verify_oop_or_return_address(Otos_i, G3_scratch);
812  locals_index_wide(G3_scratch);
813  __ store_local_ptr(G3_scratch, Otos_i);
814}
815
816
817void TemplateTable::iastore() {
818  transition(itos, vtos);
819  __ pop_i(O2); // index
820  // Otos_i: val
821  // O3: array
822  __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
823  __ st(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_INT));
824}
825
826
827void TemplateTable::lastore() {
828  transition(ltos, vtos);
829  __ pop_i(O2); // index
830  // Otos_l: val
831  // O3: array
832  __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
833  __ st_long(Otos_l, O2, arrayOopDesc::base_offset_in_bytes(T_LONG));
834}
835
836
837void TemplateTable::fastore() {
838  transition(ftos, vtos);
839  __ pop_i(O2); // index
840  // Ftos_f: val
841  // O3: array
842  __ index_check(O3, O2, LogBytesPerInt, G3_scratch, O2);
843  __ stf(FloatRegisterImpl::S, Ftos_f, O2, arrayOopDesc::base_offset_in_bytes(T_FLOAT));
844}
845
846
847void TemplateTable::dastore() {
848  transition(dtos, vtos);
849  __ pop_i(O2); // index
850  // Fos_d: val
851  // O3: array
852  __ index_check(O3, O2, LogBytesPerLong, G3_scratch, O2);
853  __ stf(FloatRegisterImpl::D, Ftos_d, O2, arrayOopDesc::base_offset_in_bytes(T_DOUBLE));
854}
855
856
857void TemplateTable::aastore() {
858  Label store_ok, is_null, done;
859  transition(vtos, vtos);
860  __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
861  __ ld(Lesp, Interpreter::expr_offset_in_bytes(1), O2);         // get index
862  __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(2), O3);     // get array
863  // Otos_i: val
864  // O2: index
865  // O3: array
866  __ verify_oop(Otos_i);
867  __ index_check_without_pop(O3, O2, UseCompressedOops ? 2 : LogBytesPerWord, G3_scratch, O1);
868
869  // do array store check - check for NULL value first
870  __ br_null_short( Otos_i, Assembler::pn, is_null );
871
872  __ load_klass(O3, O4); // get array klass
873  __ load_klass(Otos_i, O5); // get value klass
874
875  // do fast instanceof cache test
876
877  __ ld_ptr(O4,     in_bytes(ObjArrayKlass::element_klass_offset()),  O4);
878
879  assert(Otos_i == O0, "just checking");
880
881  // Otos_i:    value
882  // O1:        addr - offset
883  // O2:        index
884  // O3:        array
885  // O4:        array element klass
886  // O5:        value klass
887
888  // Address element(O1, 0, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
889
890  // Generate a fast subtype check.  Branch to store_ok if no
891  // failure.  Throw if failure.
892  __ gen_subtype_check( O5, O4, G3_scratch, G4_scratch, G1_scratch, store_ok );
893
894  // Not a subtype; so must throw exception
895  __ throw_if_not_x( Assembler::never, Interpreter::_throw_ArrayStoreException_entry, G3_scratch );
896
897  // Store is OK.
898  __ bind(store_ok);
899  do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Otos_i, G3_scratch, _bs->kind(), true);
900
901  __ ba(done);
902  __ delayed()->inc(Lesp, 3* Interpreter::stackElementSize); // adj sp (pops array, index and value)
903
904  __ bind(is_null);
905  do_oop_store(_masm, O1, noreg, arrayOopDesc::base_offset_in_bytes(T_OBJECT), G0, G4_scratch, _bs->kind(), true);
906
907  __ profile_null_seen(G3_scratch);
908  __ inc(Lesp, 3* Interpreter::stackElementSize);     // adj sp (pops array, index and value)
909  __ bind(done);
910}
911
912
913void TemplateTable::bastore() {
914  transition(itos, vtos);
915  __ pop_i(O2); // index
916  // Otos_i: val
917  // O3: array
918  __ index_check(O3, O2, 0, G3_scratch, O2);
919  __ stb(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_BYTE));
920}
921
922
923void TemplateTable::castore() {
924  transition(itos, vtos);
925  __ pop_i(O2); // index
926  // Otos_i: val
927  // O3: array
928  __ index_check(O3, O2, LogBytesPerShort, G3_scratch, O2);
929  __ sth(Otos_i, O2, arrayOopDesc::base_offset_in_bytes(T_CHAR));
930}
931
932
933void TemplateTable::sastore() {
934  // %%%%% Factor across platform
935  castore();
936}
937
938
939void TemplateTable::istore(int n) {
940  transition(itos, vtos);
941  __ st(Otos_i, Llocals, Interpreter::local_offset_in_bytes(n));
942}
943
944
945void TemplateTable::lstore(int n) {
946  transition(ltos, vtos);
947  assert(n+1 < Argument::n_register_parameters, "only handle register cases");
948  __ store_unaligned_long(Otos_l, Llocals, Interpreter::local_offset_in_bytes(n+1));
949
950}
951
952
953void TemplateTable::fstore(int n) {
954  transition(ftos, vtos);
955  assert(n < Argument::n_register_parameters, "only handle register cases");
956  __ stf(FloatRegisterImpl::S, Ftos_f, Llocals, Interpreter::local_offset_in_bytes(n));
957}
958
959
960void TemplateTable::dstore(int n) {
961  transition(dtos, vtos);
962  FloatRegister src = Ftos_d;
963  __ store_unaligned_double(src, Llocals, Interpreter::local_offset_in_bytes(n+1));
964}
965
966
967void TemplateTable::astore(int n) {
968  transition(vtos, vtos);
969  __ load_ptr(0, Otos_i);
970  __ inc(Lesp, Interpreter::stackElementSize);
971  __ verify_oop_or_return_address(Otos_i, G3_scratch);
972  __ store_local_ptr(n, Otos_i);
973}
974
975
976void TemplateTable::pop() {
977  transition(vtos, vtos);
978  __ inc(Lesp, Interpreter::stackElementSize);
979}
980
981
982void TemplateTable::pop2() {
983  transition(vtos, vtos);
984  __ inc(Lesp, 2 * Interpreter::stackElementSize);
985}
986
987
988void TemplateTable::dup() {
989  transition(vtos, vtos);
990  // stack: ..., a
991  // load a and tag
992  __ load_ptr(0, Otos_i);
993  __ push_ptr(Otos_i);
994  // stack: ..., a, a
995}
996
997
998void TemplateTable::dup_x1() {
999  transition(vtos, vtos);
1000  // stack: ..., a, b
1001  __ load_ptr( 1, G3_scratch);  // get a
1002  __ load_ptr( 0, Otos_l1);     // get b
1003  __ store_ptr(1, Otos_l1);     // put b
1004  __ store_ptr(0, G3_scratch);  // put a - like swap
1005  __ push_ptr(Otos_l1);         // push b
1006  // stack: ..., b, a, b
1007}
1008
1009
1010void TemplateTable::dup_x2() {
1011  transition(vtos, vtos);
1012  // stack: ..., a, b, c
1013  // get c and push on stack, reuse registers
1014  __ load_ptr( 0, G3_scratch);  // get c
1015  __ push_ptr(G3_scratch);      // push c with tag
1016  // stack: ..., a, b, c, c  (c in reg)  (Lesp - 4)
1017  // (stack offsets n+1 now)
1018  __ load_ptr( 3, Otos_l1);     // get a
1019  __ store_ptr(3, G3_scratch);  // put c at 3
1020  // stack: ..., c, b, c, c  (a in reg)
1021  __ load_ptr( 2, G3_scratch);  // get b
1022  __ store_ptr(2, Otos_l1);     // put a at 2
1023  // stack: ..., c, a, c, c  (b in reg)
1024  __ store_ptr(1, G3_scratch);  // put b at 1
1025  // stack: ..., c, a, b, c
1026}
1027
1028
1029void TemplateTable::dup2() {
1030  transition(vtos, vtos);
1031  __ load_ptr(1, G3_scratch);  // get a
1032  __ load_ptr(0, Otos_l1);     // get b
1033  __ push_ptr(G3_scratch);     // push a
1034  __ push_ptr(Otos_l1);        // push b
1035  // stack: ..., a, b, a, b
1036}
1037
1038
1039void TemplateTable::dup2_x1() {
1040  transition(vtos, vtos);
1041  // stack: ..., a, b, c
1042  __ load_ptr( 1, Lscratch);    // get b
1043  __ load_ptr( 2, Otos_l1);     // get a
1044  __ store_ptr(2, Lscratch);    // put b at a
1045  // stack: ..., b, b, c
1046  __ load_ptr( 0, G3_scratch);  // get c
1047  __ store_ptr(1, G3_scratch);  // put c at b
1048  // stack: ..., b, c, c
1049  __ store_ptr(0, Otos_l1);     // put a at c
1050  // stack: ..., b, c, a
1051  __ push_ptr(Lscratch);        // push b
1052  __ push_ptr(G3_scratch);      // push c
1053  // stack: ..., b, c, a, b, c
1054}
1055
1056
1057// The spec says that these types can be a mixture of category 1 (1 word)
1058// types and/or category 2 types (long and doubles)
1059void TemplateTable::dup2_x2() {
1060  transition(vtos, vtos);
1061  // stack: ..., a, b, c, d
1062  __ load_ptr( 1, Lscratch);    // get c
1063  __ load_ptr( 3, Otos_l1);     // get a
1064  __ store_ptr(3, Lscratch);    // put c at 3
1065  __ store_ptr(1, Otos_l1);     // put a at 1
1066  // stack: ..., c, b, a, d
1067  __ load_ptr( 2, G3_scratch);  // get b
1068  __ load_ptr( 0, Otos_l1);     // get d
1069  __ store_ptr(0, G3_scratch);  // put b at 0
1070  __ store_ptr(2, Otos_l1);     // put d at 2
1071  // stack: ..., c, d, a, b
1072  __ push_ptr(Lscratch);        // push c
1073  __ push_ptr(Otos_l1);         // push d
1074  // stack: ..., c, d, a, b, c, d
1075}
1076
1077
1078void TemplateTable::swap() {
1079  transition(vtos, vtos);
1080  // stack: ..., a, b
1081  __ load_ptr( 1, G3_scratch);  // get a
1082  __ load_ptr( 0, Otos_l1);     // get b
1083  __ store_ptr(0, G3_scratch);  // put b
1084  __ store_ptr(1, Otos_l1);     // put a
1085  // stack: ..., b, a
1086}
1087
1088
1089void TemplateTable::iop2(Operation op) {
1090  transition(itos, itos);
1091  __ pop_i(O1);
1092  switch (op) {
1093   case  add:  __  add(O1, Otos_i, Otos_i);  break;
1094   case  sub:  __  sub(O1, Otos_i, Otos_i);  break;
1095     // %%%%% Mul may not exist: better to call .mul?
1096   case  mul:  __ smul(O1, Otos_i, Otos_i);  break;
1097   case _and:  __ and3(O1, Otos_i, Otos_i);  break;
1098   case  _or:  __  or3(O1, Otos_i, Otos_i);  break;
1099   case _xor:  __ xor3(O1, Otos_i, Otos_i);  break;
1100   case  shl:  __  sll(O1, Otos_i, Otos_i);  break;
1101   case  shr:  __  sra(O1, Otos_i, Otos_i);  break;
1102   case ushr:  __  srl(O1, Otos_i, Otos_i);  break;
1103   default: ShouldNotReachHere();
1104  }
1105}
1106
1107
1108void TemplateTable::lop2(Operation op) {
1109  transition(ltos, ltos);
1110  __ pop_l(O2);
1111  switch (op) {
1112#ifdef _LP64
1113   case  add:  __  add(O2, Otos_l, Otos_l);  break;
1114   case  sub:  __  sub(O2, Otos_l, Otos_l);  break;
1115   case _and:  __ and3(O2, Otos_l, Otos_l);  break;
1116   case  _or:  __  or3(O2, Otos_l, Otos_l);  break;
1117   case _xor:  __ xor3(O2, Otos_l, Otos_l);  break;
1118#else
1119   case  add:  __ addcc(O3, Otos_l2, Otos_l2);  __ addc(O2, Otos_l1, Otos_l1);  break;
1120   case  sub:  __ subcc(O3, Otos_l2, Otos_l2);  __ subc(O2, Otos_l1, Otos_l1);  break;
1121   case _and:  __  and3(O3, Otos_l2, Otos_l2);  __ and3(O2, Otos_l1, Otos_l1);  break;
1122   case  _or:  __   or3(O3, Otos_l2, Otos_l2);  __  or3(O2, Otos_l1, Otos_l1);  break;
1123   case _xor:  __  xor3(O3, Otos_l2, Otos_l2);  __ xor3(O2, Otos_l1, Otos_l1);  break;
1124#endif
1125   default: ShouldNotReachHere();
1126  }
1127}
1128
1129
1130void TemplateTable::idiv() {
1131  // %%%%% Later: ForSPARC/V7 call .sdiv library routine,
1132  // %%%%% Use ldsw...sdivx on pure V9 ABI. 64 bit safe.
1133
1134  transition(itos, itos);
1135  __ pop_i(O1); // get 1st op
1136
1137  // Y contains upper 32 bits of result, set it to 0 or all ones
1138  __ wry(G0);
1139  __ mov(~0, G3_scratch);
1140
1141  __ tst(O1);
1142     Label neg;
1143  __ br(Assembler::negative, true, Assembler::pn, neg);
1144  __ delayed()->wry(G3_scratch);
1145  __ bind(neg);
1146
1147     Label ok;
1148  __ tst(Otos_i);
1149  __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch );
1150
1151  const int min_int = 0x80000000;
1152  Label regular;
1153  __ cmp(Otos_i, -1);
1154  __ br(Assembler::notEqual, false, Assembler::pt, regular);
1155#ifdef _LP64
1156  // Don't put set in delay slot
1157  // Set will turn into multiple instructions in 64 bit mode
1158  __ delayed()->nop();
1159  __ set(min_int, G4_scratch);
1160#else
1161  __ delayed()->set(min_int, G4_scratch);
1162#endif
1163  Label done;
1164  __ cmp(O1, G4_scratch);
1165  __ br(Assembler::equal, true, Assembler::pt, done);
1166  __ delayed()->mov(O1, Otos_i);   // (mov only executed if branch taken)
1167
1168  __ bind(regular);
1169  __ sdiv(O1, Otos_i, Otos_i); // note: irem uses O1 after this instruction!
1170  __ bind(done);
1171}
1172
1173
1174void TemplateTable::irem() {
1175  transition(itos, itos);
1176  __ mov(Otos_i, O2); // save divisor
1177  idiv();                               // %%%% Hack: exploits fact that idiv leaves dividend in O1
1178  __ smul(Otos_i, O2, Otos_i);
1179  __ sub(O1, Otos_i, Otos_i);
1180}
1181
1182
1183void TemplateTable::lmul() {
1184  transition(ltos, ltos);
1185  __ pop_l(O2);
1186#ifdef _LP64
1187  __ mulx(Otos_l, O2, Otos_l);
1188#else
1189  __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lmul));
1190#endif
1191
1192}
1193
1194
1195void TemplateTable::ldiv() {
1196  transition(ltos, ltos);
1197
1198  // check for zero
1199  __ pop_l(O2);
1200#ifdef _LP64
1201  __ tst(Otos_l);
1202  __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1203  __ sdivx(O2, Otos_l, Otos_l);
1204#else
1205  __ orcc(Otos_l1, Otos_l2, G0);
1206  __ throw_if_not_icc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1207  __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1208#endif
1209}
1210
1211
1212void TemplateTable::lrem() {
1213  transition(ltos, ltos);
1214
1215  // check for zero
1216  __ pop_l(O2);
1217#ifdef _LP64
1218  __ tst(Otos_l);
1219  __ throw_if_not_xcc( Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1220  __ sdivx(O2, Otos_l, Otos_l2);
1221  __ mulx (Otos_l2, Otos_l, Otos_l2);
1222  __ sub  (O2, Otos_l2, Otos_l);
1223#else
1224  __ orcc(Otos_l1, Otos_l2, G0);
1225  __ throw_if_not_icc(Assembler::notZero, Interpreter::_throw_ArithmeticException_entry, G3_scratch);
1226  __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1227#endif
1228}
1229
1230
1231void TemplateTable::lshl() {
1232  transition(itos, ltos); // %%%% could optimize, fill delay slot or opt for ultra
1233
1234  __ pop_l(O2);                          // shift value in O2, O3
1235#ifdef _LP64
1236  __ sllx(O2, Otos_i, Otos_l);
1237#else
1238  __ lshl(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1239#endif
1240}
1241
1242
1243void TemplateTable::lshr() {
1244  transition(itos, ltos); // %%%% see lshl comment
1245
1246  __ pop_l(O2);                          // shift value in O2, O3
1247#ifdef _LP64
1248  __ srax(O2, Otos_i, Otos_l);
1249#else
1250  __ lshr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1251#endif
1252}
1253
1254
1255
1256void TemplateTable::lushr() {
1257  transition(itos, ltos); // %%%% see lshl comment
1258
1259  __ pop_l(O2);                          // shift value in O2, O3
1260#ifdef _LP64
1261  __ srlx(O2, Otos_i, Otos_l);
1262#else
1263  __ lushr(O2, O3, Otos_i, Otos_l1, Otos_l2, O4);
1264#endif
1265}
1266
1267
1268void TemplateTable::fop2(Operation op) {
1269  transition(ftos, ftos);
1270  switch (op) {
1271   case  add:  __  pop_f(F4); __ fadd(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1272   case  sub:  __  pop_f(F4); __ fsub(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1273   case  mul:  __  pop_f(F4); __ fmul(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1274   case  div:  __  pop_f(F4); __ fdiv(FloatRegisterImpl::S, F4, Ftos_f, Ftos_f);  break;
1275   case  rem:
1276     assert(Ftos_f == F0, "just checking");
1277#ifdef _LP64
1278     // LP64 calling conventions use F1, F3 for passing 2 floats
1279     __ pop_f(F1);
1280     __ fmov(FloatRegisterImpl::S, Ftos_f, F3);
1281#else
1282     __ pop_i(O0);
1283     __ stf(FloatRegisterImpl::S, Ftos_f, __ d_tmp);
1284     __ ld( __ d_tmp, O1 );
1285#endif
1286     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1287     assert( Ftos_f == F0, "fix this code" );
1288     break;
1289
1290   default: ShouldNotReachHere();
1291  }
1292}
1293
1294
1295void TemplateTable::dop2(Operation op) {
1296  transition(dtos, dtos);
1297  switch (op) {
1298   case  add:  __  pop_d(F4); __ fadd(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1299   case  sub:  __  pop_d(F4); __ fsub(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1300   case  mul:  __  pop_d(F4); __ fmul(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1301   case  div:  __  pop_d(F4); __ fdiv(FloatRegisterImpl::D, F4, Ftos_d, Ftos_d);  break;
1302   case  rem:
1303#ifdef _LP64
1304     // Pass arguments in D0, D2
1305     __ fmov(FloatRegisterImpl::D, Ftos_f, F2 );
1306     __ pop_d( F0 );
1307#else
1308     // Pass arguments in O0O1, O2O3
1309     __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
1310     __ ldd( __ d_tmp, O2 );
1311     __ pop_d(Ftos_f);
1312     __ stf(FloatRegisterImpl::D, Ftos_f, __ d_tmp);
1313     __ ldd( __ d_tmp, O0 );
1314#endif
1315     __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1316     assert( Ftos_d == F0, "fix this code" );
1317     break;
1318
1319   default: ShouldNotReachHere();
1320  }
1321}
1322
1323
1324void TemplateTable::ineg() {
1325  transition(itos, itos);
1326  __ neg(Otos_i);
1327}
1328
1329
1330void TemplateTable::lneg() {
1331  transition(ltos, ltos);
1332#ifdef _LP64
1333  __ sub(G0, Otos_l, Otos_l);
1334#else
1335  __ lneg(Otos_l1, Otos_l2);
1336#endif
1337}
1338
1339
1340void TemplateTable::fneg() {
1341  transition(ftos, ftos);
1342  __ fneg(FloatRegisterImpl::S, Ftos_f, Ftos_f);
1343}
1344
1345
1346void TemplateTable::dneg() {
1347  transition(dtos, dtos);
1348  __ fneg(FloatRegisterImpl::D, Ftos_f, Ftos_f);
1349}
1350
1351
1352void TemplateTable::iinc() {
1353  transition(vtos, vtos);
1354  locals_index(G3_scratch);
1355  __ ldsb(Lbcp, 2, O2);  // load constant
1356  __ access_local_int(G3_scratch, Otos_i);
1357  __ add(Otos_i, O2, Otos_i);
1358  __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1359}
1360
1361
1362void TemplateTable::wide_iinc() {
1363  transition(vtos, vtos);
1364  locals_index_wide(G3_scratch);
1365  __ get_2_byte_integer_at_bcp( 4,  O2, O3, InterpreterMacroAssembler::Signed);
1366  __ access_local_int(G3_scratch, Otos_i);
1367  __ add(Otos_i, O3, Otos_i);
1368  __ st(Otos_i, G3_scratch, 0);    // access_local_int puts E.A. in G3_scratch
1369}
1370
1371
1372void TemplateTable::convert() {
1373// %%%%% Factor this first part accross platforms
1374  #ifdef ASSERT
1375    TosState tos_in  = ilgl;
1376    TosState tos_out = ilgl;
1377    switch (bytecode()) {
1378      case Bytecodes::_i2l: // fall through
1379      case Bytecodes::_i2f: // fall through
1380      case Bytecodes::_i2d: // fall through
1381      case Bytecodes::_i2b: // fall through
1382      case Bytecodes::_i2c: // fall through
1383      case Bytecodes::_i2s: tos_in = itos; break;
1384      case Bytecodes::_l2i: // fall through
1385      case Bytecodes::_l2f: // fall through
1386      case Bytecodes::_l2d: tos_in = ltos; break;
1387      case Bytecodes::_f2i: // fall through
1388      case Bytecodes::_f2l: // fall through
1389      case Bytecodes::_f2d: tos_in = ftos; break;
1390      case Bytecodes::_d2i: // fall through
1391      case Bytecodes::_d2l: // fall through
1392      case Bytecodes::_d2f: tos_in = dtos; break;
1393      default             : ShouldNotReachHere();
1394    }
1395    switch (bytecode()) {
1396      case Bytecodes::_l2i: // fall through
1397      case Bytecodes::_f2i: // fall through
1398      case Bytecodes::_d2i: // fall through
1399      case Bytecodes::_i2b: // fall through
1400      case Bytecodes::_i2c: // fall through
1401      case Bytecodes::_i2s: tos_out = itos; break;
1402      case Bytecodes::_i2l: // fall through
1403      case Bytecodes::_f2l: // fall through
1404      case Bytecodes::_d2l: tos_out = ltos; break;
1405      case Bytecodes::_i2f: // fall through
1406      case Bytecodes::_l2f: // fall through
1407      case Bytecodes::_d2f: tos_out = ftos; break;
1408      case Bytecodes::_i2d: // fall through
1409      case Bytecodes::_l2d: // fall through
1410      case Bytecodes::_f2d: tos_out = dtos; break;
1411      default             : ShouldNotReachHere();
1412    }
1413    transition(tos_in, tos_out);
1414  #endif
1415
1416
1417  // Conversion
1418  Label done;
1419  switch (bytecode()) {
1420   case Bytecodes::_i2l:
1421#ifdef _LP64
1422    // Sign extend the 32 bits
1423    __ sra ( Otos_i, 0, Otos_l );
1424#else
1425    __ addcc(Otos_i, 0, Otos_l2);
1426    __ br(Assembler::greaterEqual, true, Assembler::pt, done);
1427    __ delayed()->clr(Otos_l1);
1428    __ set(~0, Otos_l1);
1429#endif
1430    break;
1431
1432   case Bytecodes::_i2f:
1433    __ st(Otos_i, __ d_tmp );
1434    __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1435    __ fitof(FloatRegisterImpl::S, F0, Ftos_f);
1436    break;
1437
1438   case Bytecodes::_i2d:
1439    __ st(Otos_i, __ d_tmp);
1440    __ ldf(FloatRegisterImpl::S,  __ d_tmp, F0);
1441    __ fitof(FloatRegisterImpl::D, F0, Ftos_f);
1442    break;
1443
1444   case Bytecodes::_i2b:
1445    __ sll(Otos_i, 24, Otos_i);
1446    __ sra(Otos_i, 24, Otos_i);
1447    break;
1448
1449   case Bytecodes::_i2c:
1450    __ sll(Otos_i, 16, Otos_i);
1451    __ srl(Otos_i, 16, Otos_i);
1452    break;
1453
1454   case Bytecodes::_i2s:
1455    __ sll(Otos_i, 16, Otos_i);
1456    __ sra(Otos_i, 16, Otos_i);
1457    break;
1458
1459   case Bytecodes::_l2i:
1460#ifndef _LP64
1461    __ mov(Otos_l2, Otos_i);
1462#else
1463    // Sign-extend into the high 32 bits
1464    __ sra(Otos_l, 0, Otos_i);
1465#endif
1466    break;
1467
1468   case Bytecodes::_l2f:
1469   case Bytecodes::_l2d:
1470    __ st_long(Otos_l, __ d_tmp);
1471    __ ldf(FloatRegisterImpl::D, __ d_tmp, Ftos_d);
1472
1473    if (bytecode() == Bytecodes::_l2f) {
1474      __ fxtof(FloatRegisterImpl::S, Ftos_d, Ftos_f);
1475    } else {
1476      __ fxtof(FloatRegisterImpl::D, Ftos_d, Ftos_d);
1477    }
1478    break;
1479
1480  case Bytecodes::_f2i:  {
1481      Label isNaN;
1482      // result must be 0 if value is NaN; test by comparing value to itself
1483      __ fcmp(FloatRegisterImpl::S, Assembler::fcc0, Ftos_f, Ftos_f);
1484      __ fb(Assembler::f_unordered, true, Assembler::pn, isNaN);
1485      __ delayed()->clr(Otos_i);                                     // NaN
1486      __ ftoi(FloatRegisterImpl::S, Ftos_f, F30);
1487      __ stf(FloatRegisterImpl::S, F30, __ d_tmp);
1488      __ ld(__ d_tmp, Otos_i);
1489      __ bind(isNaN);
1490    }
1491    break;
1492
1493   case Bytecodes::_f2l:
1494    // must uncache tos
1495    __ push_f();
1496#ifdef _LP64
1497    __ pop_f(F1);
1498#else
1499    __ pop_i(O0);
1500#endif
1501    __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1502    break;
1503
1504   case Bytecodes::_f2d:
1505    __ ftof( FloatRegisterImpl::S, FloatRegisterImpl::D, Ftos_f, Ftos_f);
1506    break;
1507
1508   case Bytecodes::_d2i:
1509   case Bytecodes::_d2l:
1510    // must uncache tos
1511    __ push_d();
1512#ifdef _LP64
1513    // LP64 calling conventions pass first double arg in D0
1514    __ pop_d( Ftos_d );
1515#else
1516    __ pop_i( O0 );
1517    __ pop_i( O1 );
1518#endif
1519    __ call_VM_leaf(Lscratch,
1520        bytecode() == Bytecodes::_d2i
1521          ? CAST_FROM_FN_PTR(address, SharedRuntime::d2i)
1522          : CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1523    break;
1524
1525    case Bytecodes::_d2f:
1526      __ ftof( FloatRegisterImpl::D, FloatRegisterImpl::S, Ftos_d, Ftos_f);
1527    break;
1528
1529    default: ShouldNotReachHere();
1530  }
1531  __ bind(done);
1532}
1533
1534
1535void TemplateTable::lcmp() {
1536  transition(ltos, itos);
1537
1538#ifdef _LP64
1539  __ pop_l(O1); // pop off value 1, value 2 is in O0
1540  __ lcmp( O1, Otos_l, Otos_i );
1541#else
1542  __ pop_l(O2); // cmp O2,3 to O0,1
1543  __ lcmp( O2, O3, Otos_l1, Otos_l2, Otos_i );
1544#endif
1545}
1546
1547
1548void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1549
1550  if (is_float) __ pop_f(F2);
1551  else          __ pop_d(F2);
1552
1553  assert(Ftos_f == F0  &&  Ftos_d == F0,  "alias checking:");
1554
1555  __ float_cmp( is_float, unordered_result, F2, F0, Otos_i );
1556}
1557
1558void TemplateTable::branch(bool is_jsr, bool is_wide) {
1559  // Note: on SPARC, we use InterpreterMacroAssembler::if_cmp also.
1560  __ verify_thread();
1561
1562  const Register O2_bumped_count = O2;
1563  __ profile_taken_branch(G3_scratch, O2_bumped_count);
1564
1565  // get (wide) offset to O1_disp
1566  const Register O1_disp = O1;
1567  if (is_wide)  __ get_4_byte_integer_at_bcp( 1,  G4_scratch, O1_disp,                                    InterpreterMacroAssembler::set_CC);
1568  else          __ get_2_byte_integer_at_bcp( 1,  G4_scratch, O1_disp, InterpreterMacroAssembler::Signed, InterpreterMacroAssembler::set_CC);
1569
1570  // Handle all the JSR stuff here, then exit.
1571  // It's much shorter and cleaner than intermingling with the
1572  // non-JSR normal-branch stuff occurring below.
1573  if( is_jsr ) {
1574    // compute return address as bci in Otos_i
1575    __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1576    __ sub(Lbcp, G3_scratch, G3_scratch);
1577    __ sub(G3_scratch, in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3), Otos_i);
1578
1579    // Bump Lbcp to target of JSR
1580    __ add(Lbcp, O1_disp, Lbcp);
1581    // Push returnAddress for "ret" on stack
1582    __ push_ptr(Otos_i);
1583    // And away we go!
1584    __ dispatch_next(vtos);
1585    return;
1586  }
1587
1588  // Normal (non-jsr) branch handling
1589
1590  // Save the current Lbcp
1591  const Register l_cur_bcp = Lscratch;
1592  __ mov( Lbcp, l_cur_bcp );
1593
1594  bool increment_invocation_counter_for_backward_branches = UseCompiler && UseLoopCounter;
1595  if ( increment_invocation_counter_for_backward_branches ) {
1596    Label Lforward;
1597    // check branch direction
1598    __ br( Assembler::positive, false,  Assembler::pn, Lforward );
1599    // Bump bytecode pointer by displacement (take the branch)
1600    __ delayed()->add( O1_disp, Lbcp, Lbcp );     // add to bc addr
1601
1602    const Register Rcounters = G3_scratch;
1603    __ get_method_counters(Lmethod, Rcounters, Lforward);
1604
1605    if (TieredCompilation) {
1606      Label Lno_mdo, Loverflow;
1607      int increment = InvocationCounter::count_increment;
1608      int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1609      if (ProfileInterpreter) {
1610        // If no method data exists, go to profile_continue.
1611        __ ld_ptr(Lmethod, Method::method_data_offset(), G4_scratch);
1612        __ br_null_short(G4_scratch, Assembler::pn, Lno_mdo);
1613
1614        // Increment backedge counter in the MDO
1615        Address mdo_backedge_counter(G4_scratch, in_bytes(MethodData::backedge_counter_offset()) +
1616                                                 in_bytes(InvocationCounter::counter_offset()));
1617        __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, G3_scratch, O0,
1618                                   Assembler::notZero, &Lforward);
1619        __ ba_short(Loverflow);
1620      }
1621
1622      // If there's no MDO, increment counter in MethodCounters*
1623      __ bind(Lno_mdo);
1624      Address backedge_counter(Rcounters,
1625              in_bytes(MethodCounters::backedge_counter_offset()) +
1626              in_bytes(InvocationCounter::counter_offset()));
1627      __ increment_mask_and_jump(backedge_counter, increment, mask, G4_scratch, O0,
1628                                 Assembler::notZero, &Lforward);
1629      __ bind(Loverflow);
1630
1631      // notify point for loop, pass branch bytecode
1632      __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), l_cur_bcp);
1633
1634      // Was an OSR adapter generated?
1635      // O0 = osr nmethod
1636      __ br_null_short(O0, Assembler::pn, Lforward);
1637
1638      // Has the nmethod been invalidated already?
1639      __ ld(O0, nmethod::entry_bci_offset(), O2);
1640      __ cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, Lforward);
1641
1642      // migrate the interpreter frame off of the stack
1643
1644      __ mov(G2_thread, L7);
1645      // save nmethod
1646      __ mov(O0, L6);
1647      __ set_last_Java_frame(SP, noreg);
1648      __ call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
1649      __ reset_last_Java_frame();
1650      __ mov(L7, G2_thread);
1651
1652      // move OSR nmethod to I1
1653      __ mov(L6, I1);
1654
1655      // OSR buffer to I0
1656      __ mov(O0, I0);
1657
1658      // remove the interpreter frame
1659      __ restore(I5_savedSP, 0, SP);
1660
1661      // Jump to the osr code.
1662      __ ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
1663      __ jmp(O2, G0);
1664      __ delayed()->nop();
1665
1666    } else {
1667      // Update Backedge branch separately from invocations
1668      const Register G4_invoke_ctr = G4;
1669      __ increment_backedge_counter(Rcounters, G4_invoke_ctr, G1_scratch);
1670      if (ProfileInterpreter) {
1671        __ test_invocation_counter_for_mdp(G4_invoke_ctr, G3_scratch, Lforward);
1672        if (UseOnStackReplacement) {
1673          __ test_backedge_count_for_osr(O2_bumped_count, l_cur_bcp, G3_scratch);
1674        }
1675      } else {
1676        if (UseOnStackReplacement) {
1677          __ test_backedge_count_for_osr(G4_invoke_ctr, l_cur_bcp, G3_scratch);
1678        }
1679      }
1680    }
1681
1682    __ bind(Lforward);
1683  } else
1684    // Bump bytecode pointer by displacement (take the branch)
1685    __ add( O1_disp, Lbcp, Lbcp );// add to bc addr
1686
1687  // continue with bytecode @ target
1688  // %%%%% Like Intel, could speed things up by moving bytecode fetch to code above,
1689  // %%%%% and changing dispatch_next to dispatch_only
1690  __ dispatch_next(vtos);
1691}
1692
1693
1694// Note Condition in argument is TemplateTable::Condition
1695// arg scope is within class scope
1696
1697void TemplateTable::if_0cmp(Condition cc) {
1698  // no pointers, integer only!
1699  transition(itos, vtos);
1700  // assume branch is more often taken than not (loops use backward branches)
1701  __ cmp( Otos_i, 0);
1702  __ if_cmp(ccNot(cc), false);
1703}
1704
1705
1706void TemplateTable::if_icmp(Condition cc) {
1707  transition(itos, vtos);
1708  __ pop_i(O1);
1709  __ cmp(O1, Otos_i);
1710  __ if_cmp(ccNot(cc), false);
1711}
1712
1713
1714void TemplateTable::if_nullcmp(Condition cc) {
1715  transition(atos, vtos);
1716  __ tst(Otos_i);
1717  __ if_cmp(ccNot(cc), true);
1718}
1719
1720
1721void TemplateTable::if_acmp(Condition cc) {
1722  transition(atos, vtos);
1723  __ pop_ptr(O1);
1724  __ verify_oop(O1);
1725  __ verify_oop(Otos_i);
1726  __ cmp(O1, Otos_i);
1727  __ if_cmp(ccNot(cc), true);
1728}
1729
1730
1731
1732void TemplateTable::ret() {
1733  transition(vtos, vtos);
1734  locals_index(G3_scratch);
1735  __ access_local_returnAddress(G3_scratch, Otos_i);
1736  // Otos_i contains the bci, compute the bcp from that
1737
1738#ifdef _LP64
1739#ifdef ASSERT
1740  // jsr result was labeled as an 'itos' not an 'atos' because we cannot GC
1741  // the result.  The return address (really a BCI) was stored with an
1742  // 'astore' because JVM specs claim it's a pointer-sized thing.  Hence in
1743  // the 64-bit build the 32-bit BCI is actually in the low bits of a 64-bit
1744  // loaded value.
1745  { Label zzz ;
1746     __ set (65536, G3_scratch) ;
1747     __ cmp (Otos_i, G3_scratch) ;
1748     __ bp( Assembler::lessEqualUnsigned, false, Assembler::xcc, Assembler::pn, zzz);
1749     __ delayed()->nop();
1750     __ stop("BCI is in the wrong register half?");
1751     __ bind (zzz) ;
1752  }
1753#endif
1754#endif
1755
1756  __ profile_ret(vtos, Otos_i, G4_scratch);
1757
1758  __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1759  __ add(G3_scratch, Otos_i, G3_scratch);
1760  __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
1761  __ dispatch_next(vtos);
1762}
1763
1764
1765void TemplateTable::wide_ret() {
1766  transition(vtos, vtos);
1767  locals_index_wide(G3_scratch);
1768  __ access_local_returnAddress(G3_scratch, Otos_i);
1769  // Otos_i contains the bci, compute the bcp from that
1770
1771  __ profile_ret(vtos, Otos_i, G4_scratch);
1772
1773  __ ld_ptr(Lmethod, Method::const_offset(), G3_scratch);
1774  __ add(G3_scratch, Otos_i, G3_scratch);
1775  __ add(G3_scratch, in_bytes(ConstMethod::codes_offset()), Lbcp);
1776  __ dispatch_next(vtos);
1777}
1778
1779
1780void TemplateTable::tableswitch() {
1781  transition(itos, vtos);
1782  Label default_case, continue_execution;
1783
1784  // align bcp
1785  __ add(Lbcp, BytesPerInt, O1);
1786  __ and3(O1, -BytesPerInt, O1);
1787  // load lo, hi
1788  __ ld(O1, 1 * BytesPerInt, O2);       // Low Byte
1789  __ ld(O1, 2 * BytesPerInt, O3);       // High Byte
1790#ifdef _LP64
1791  // Sign extend the 32 bits
1792  __ sra ( Otos_i, 0, Otos_i );
1793#endif /* _LP64 */
1794
1795  // check against lo & hi
1796  __ cmp( Otos_i, O2);
1797  __ br( Assembler::less, false, Assembler::pn, default_case);
1798  __ delayed()->cmp( Otos_i, O3 );
1799  __ br( Assembler::greater, false, Assembler::pn, default_case);
1800  // lookup dispatch offset
1801  __ delayed()->sub(Otos_i, O2, O2);
1802  __ profile_switch_case(O2, O3, G3_scratch, G4_scratch);
1803  __ sll(O2, LogBytesPerInt, O2);
1804  __ add(O2, 3 * BytesPerInt, O2);
1805  __ ba(continue_execution);
1806  __ delayed()->ld(O1, O2, O2);
1807  // handle default
1808  __ bind(default_case);
1809  __ profile_switch_default(O3);
1810  __ ld(O1, 0, O2); // get default offset
1811  // continue execution
1812  __ bind(continue_execution);
1813  __ add(Lbcp, O2, Lbcp);
1814  __ dispatch_next(vtos);
1815}
1816
1817
1818void TemplateTable::lookupswitch() {
1819  transition(itos, itos);
1820  __ stop("lookupswitch bytecode should have been rewritten");
1821}
1822
1823void TemplateTable::fast_linearswitch() {
1824  transition(itos, vtos);
1825    Label loop_entry, loop, found, continue_execution;
1826  // align bcp
1827  __ add(Lbcp, BytesPerInt, O1);
1828  __ and3(O1, -BytesPerInt, O1);
1829 // set counter
1830  __ ld(O1, BytesPerInt, O2);
1831  __ sll(O2, LogBytesPerInt + 1, O2); // in word-pairs
1832  __ add(O1, 2 * BytesPerInt, O3); // set first pair addr
1833  __ ba(loop_entry);
1834  __ delayed()->add(O3, O2, O2); // counter now points past last pair
1835
1836  // table search
1837  __ bind(loop);
1838  __ cmp(O4, Otos_i);
1839  __ br(Assembler::equal, true, Assembler::pn, found);
1840  __ delayed()->ld(O3, BytesPerInt, O4); // offset -> O4
1841  __ inc(O3, 2 * BytesPerInt);
1842
1843  __ bind(loop_entry);
1844  __ cmp(O2, O3);
1845  __ brx(Assembler::greaterUnsigned, true, Assembler::pt, loop);
1846  __ delayed()->ld(O3, 0, O4);
1847
1848  // default case
1849  __ ld(O1, 0, O4); // get default offset
1850  if (ProfileInterpreter) {
1851    __ profile_switch_default(O3);
1852    __ ba_short(continue_execution);
1853  }
1854
1855  // entry found -> get offset
1856  __ bind(found);
1857  if (ProfileInterpreter) {
1858    __ sub(O3, O1, O3);
1859    __ sub(O3, 2*BytesPerInt, O3);
1860    __ srl(O3, LogBytesPerInt + 1, O3); // in word-pairs
1861    __ profile_switch_case(O3, O1, O2, G3_scratch);
1862
1863    __ bind(continue_execution);
1864  }
1865  __ add(Lbcp, O4, Lbcp);
1866  __ dispatch_next(vtos);
1867}
1868
1869
1870void TemplateTable::fast_binaryswitch() {
1871  transition(itos, vtos);
1872  // Implementation using the following core algorithm: (copied from Intel)
1873  //
1874  // int binary_search(int key, LookupswitchPair* array, int n) {
1875  //   // Binary search according to "Methodik des Programmierens" by
1876  //   // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1877  //   int i = 0;
1878  //   int j = n;
1879  //   while (i+1 < j) {
1880  //     // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1881  //     // with      Q: for all i: 0 <= i < n: key < a[i]
1882  //     // where a stands for the array and assuming that the (inexisting)
1883  //     // element a[n] is infinitely big.
1884  //     int h = (i + j) >> 1;
1885  //     // i < h < j
1886  //     if (key < array[h].fast_match()) {
1887  //       j = h;
1888  //     } else {
1889  //       i = h;
1890  //     }
1891  //   }
1892  //   // R: a[i] <= key < a[i+1] or Q
1893  //   // (i.e., if key is within array, i is the correct index)
1894  //   return i;
1895  // }
1896
1897  // register allocation
1898  assert(Otos_i == O0, "alias checking");
1899  const Register Rkey     = Otos_i;                    // already set (tosca)
1900  const Register Rarray   = O1;
1901  const Register Ri       = O2;
1902  const Register Rj       = O3;
1903  const Register Rh       = O4;
1904  const Register Rscratch = O5;
1905
1906  const int log_entry_size = 3;
1907  const int entry_size = 1 << log_entry_size;
1908
1909  Label found;
1910  // Find Array start
1911  __ add(Lbcp, 3 * BytesPerInt, Rarray);
1912  __ and3(Rarray, -BytesPerInt, Rarray);
1913  // initialize i & j (in delay slot)
1914  __ clr( Ri );
1915
1916  // and start
1917  Label entry;
1918  __ ba(entry);
1919  __ delayed()->ld( Rarray, -BytesPerInt, Rj);
1920  // (Rj is already in the native byte-ordering.)
1921
1922  // binary search loop
1923  { Label loop;
1924    __ bind( loop );
1925    // int h = (i + j) >> 1;
1926    __ sra( Rh, 1, Rh );
1927    // if (key < array[h].fast_match()) {
1928    //   j = h;
1929    // } else {
1930    //   i = h;
1931    // }
1932    __ sll( Rh, log_entry_size, Rscratch );
1933    __ ld( Rarray, Rscratch, Rscratch );
1934    // (Rscratch is already in the native byte-ordering.)
1935    __ cmp( Rkey, Rscratch );
1936    __ movcc( Assembler::less,         false, Assembler::icc, Rh, Rj );  // j = h if (key <  array[h].fast_match())
1937    __ movcc( Assembler::greaterEqual, false, Assembler::icc, Rh, Ri );  // i = h if (key >= array[h].fast_match())
1938
1939    // while (i+1 < j)
1940    __ bind( entry );
1941    __ add( Ri, 1, Rscratch );
1942    __ cmp(Rscratch, Rj);
1943    __ br( Assembler::less, true, Assembler::pt, loop );
1944    __ delayed()->add( Ri, Rj, Rh ); // start h = i + j  >> 1;
1945  }
1946
1947  // end of binary search, result index is i (must check again!)
1948  Label default_case;
1949  Label continue_execution;
1950  if (ProfileInterpreter) {
1951    __ mov( Ri, Rh );              // Save index in i for profiling
1952  }
1953  __ sll( Ri, log_entry_size, Ri );
1954  __ ld( Rarray, Ri, Rscratch );
1955  // (Rscratch is already in the native byte-ordering.)
1956  __ cmp( Rkey, Rscratch );
1957  __ br( Assembler::notEqual, true, Assembler::pn, default_case );
1958  __ delayed()->ld( Rarray, -2 * BytesPerInt, Rj ); // load default offset -> j
1959
1960  // entry found -> j = offset
1961  __ inc( Ri, BytesPerInt );
1962  __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
1963  __ ld( Rarray, Ri, Rj );
1964  // (Rj is already in the native byte-ordering.)
1965
1966  if (ProfileInterpreter) {
1967    __ ba_short(continue_execution);
1968  }
1969
1970  __ bind(default_case); // fall through (if not profiling)
1971  __ profile_switch_default(Ri);
1972
1973  __ bind(continue_execution);
1974  __ add( Lbcp, Rj, Lbcp );
1975  __ dispatch_next( vtos );
1976}
1977
1978
1979void TemplateTable::_return(TosState state) {
1980  transition(state, state);
1981  assert(_desc->calls_vm(), "inconsistent calls_vm information");
1982
1983  if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
1984    assert(state == vtos, "only valid state");
1985    __ mov(G0, G3_scratch);
1986    __ access_local_ptr(G3_scratch, Otos_i);
1987    __ load_klass(Otos_i, O2);
1988    __ set(JVM_ACC_HAS_FINALIZER, G3);
1989    __ ld(O2, in_bytes(Klass::access_flags_offset()), O2);
1990    __ andcc(G3, O2, G0);
1991    Label skip_register_finalizer;
1992    __ br(Assembler::zero, false, Assembler::pn, skip_register_finalizer);
1993    __ delayed()->nop();
1994
1995    // Call out to do finalizer registration
1996    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), Otos_i);
1997
1998    __ bind(skip_register_finalizer);
1999  }
2000
2001  __ remove_activation(state, /* throw_monitor_exception */ true);
2002
2003  // The caller's SP was adjusted upon method entry to accomodate
2004  // the callee's non-argument locals. Undo that adjustment.
2005  __ ret();                             // return to caller
2006  __ delayed()->restore(I5_savedSP, G0, SP);
2007}
2008
2009
2010// ----------------------------------------------------------------------------
2011// Volatile variables demand their effects be made known to all CPU's in
2012// order.  Store buffers on most chips allow reads & writes to reorder; the
2013// JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2014// memory barrier (i.e., it's not sufficient that the interpreter does not
2015// reorder volatile references, the hardware also must not reorder them).
2016//
2017// According to the new Java Memory Model (JMM):
2018// (1) All volatiles are serialized wrt to each other.
2019// ALSO reads & writes act as aquire & release, so:
2020// (2) A read cannot let unrelated NON-volatile memory refs that happen after
2021// the read float up to before the read.  It's OK for non-volatile memory refs
2022// that happen before the volatile read to float down below it.
2023// (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2024// that happen BEFORE the write float down to after the write.  It's OK for
2025// non-volatile memory refs that happen after the volatile write to float up
2026// before it.
2027//
2028// We only put in barriers around volatile refs (they are expensive), not
2029// _between_ memory refs (that would require us to track the flavor of the
2030// previous memory refs).  Requirements (2) and (3) require some barriers
2031// before volatile stores and after volatile loads.  These nearly cover
2032// requirement (1) but miss the volatile-store-volatile-load case.  This final
2033// case is placed after volatile-stores although it could just as well go
2034// before volatile-loads.
2035void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint) {
2036  // Helper function to insert a is-volatile test and memory barrier
2037  // All current sparc implementations run in TSO, needing only StoreLoad
2038  if ((order_constraint & Assembler::StoreLoad) == 0) return;
2039  __ membar( order_constraint );
2040}
2041
2042// ----------------------------------------------------------------------------
2043void TemplateTable::resolve_cache_and_index(int byte_no,
2044                                            Register Rcache,
2045                                            Register index,
2046                                            size_t index_size) {
2047  // Depends on cpCacheOop layout!
2048  Label resolved;
2049
2050    assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2051    __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, Lbyte_code, byte_no, 1, index_size);
2052    __ cmp(Lbyte_code, (int) bytecode());  // have we resolved this bytecode?
2053    __ br(Assembler::equal, false, Assembler::pt, resolved);
2054    __ delayed()->set((int)bytecode(), O1);
2055
2056  address entry;
2057  switch (bytecode()) {
2058    case Bytecodes::_getstatic      : // fall through
2059    case Bytecodes::_putstatic      : // fall through
2060    case Bytecodes::_getfield       : // fall through
2061    case Bytecodes::_putfield       : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2062    case Bytecodes::_invokevirtual  : // fall through
2063    case Bytecodes::_invokespecial  : // fall through
2064    case Bytecodes::_invokestatic   : // fall through
2065    case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);  break;
2066    case Bytecodes::_invokehandle   : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);  break;
2067    case Bytecodes::_invokedynamic  : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);  break;
2068    default:
2069      fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2070      break;
2071  }
2072  // first time invocation - must resolve first
2073  __ call_VM(noreg, entry, O1);
2074  // Update registers with resolved info
2075  __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2076  __ bind(resolved);
2077}
2078
2079void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2080                                               Register method,
2081                                               Register itable_index,
2082                                               Register flags,
2083                                               bool is_invokevirtual,
2084                                               bool is_invokevfinal,
2085                                               bool is_invokedynamic) {
2086  // Uses both G3_scratch and G4_scratch
2087  Register cache = G3_scratch;
2088  Register index = G4_scratch;
2089  assert_different_registers(cache, method, itable_index);
2090
2091  // determine constant pool cache field offsets
2092  assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2093  const int method_offset = in_bytes(
2094      ConstantPoolCache::base_offset() +
2095      ((byte_no == f2_byte)
2096       ? ConstantPoolCacheEntry::f2_offset()
2097       : ConstantPoolCacheEntry::f1_offset()
2098      )
2099    );
2100  const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2101                                    ConstantPoolCacheEntry::flags_offset());
2102  // access constant pool cache fields
2103  const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2104                                    ConstantPoolCacheEntry::f2_offset());
2105
2106  if (is_invokevfinal) {
2107    __ get_cache_and_index_at_bcp(cache, index, 1);
2108    __ ld_ptr(Address(cache, method_offset), method);
2109  } else {
2110    size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2111    resolve_cache_and_index(byte_no, cache, index, index_size);
2112    __ ld_ptr(Address(cache, method_offset), method);
2113  }
2114
2115  if (itable_index != noreg) {
2116    // pick up itable or appendix index from f2 also:
2117    __ ld_ptr(Address(cache, index_offset), itable_index);
2118  }
2119  __ ld_ptr(Address(cache, flags_offset), flags);
2120}
2121
2122// The Rcache register must be set before call
2123void TemplateTable::load_field_cp_cache_entry(Register Robj,
2124                                              Register Rcache,
2125                                              Register index,
2126                                              Register Roffset,
2127                                              Register Rflags,
2128                                              bool is_static) {
2129  assert_different_registers(Rcache, Rflags, Roffset);
2130
2131  ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2132
2133  __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2134  __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2135  if (is_static) {
2136    __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f1_offset(), Robj);
2137    const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2138    __ ld_ptr( Robj, mirror_offset, Robj);
2139  }
2140}
2141
2142// The registers Rcache and index expected to be set before call.
2143// Correct values of the Rcache and index registers are preserved.
2144void TemplateTable::jvmti_post_field_access(Register Rcache,
2145                                            Register index,
2146                                            bool is_static,
2147                                            bool has_tos) {
2148  ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2149
2150  if (JvmtiExport::can_post_field_access()) {
2151    // Check to see if a field access watch has been set before we take
2152    // the time to call into the VM.
2153    Label Label1;
2154    assert_different_registers(Rcache, index, G1_scratch);
2155    AddressLiteral get_field_access_count_addr(JvmtiExport::get_field_access_count_addr());
2156    __ load_contents(get_field_access_count_addr, G1_scratch);
2157    __ cmp_and_br_short(G1_scratch, 0, Assembler::equal, Assembler::pt, Label1);
2158
2159    __ add(Rcache, in_bytes(cp_base_offset), Rcache);
2160
2161    if (is_static) {
2162      __ clr(Otos_i);
2163    } else {
2164      if (has_tos) {
2165      // save object pointer before call_VM() clobbers it
2166        __ push_ptr(Otos_i);  // put object on tos where GC wants it.
2167      } else {
2168        // Load top of stack (do not pop the value off the stack);
2169        __ ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), Otos_i);
2170      }
2171      __ verify_oop(Otos_i);
2172    }
2173    // Otos_i: object pointer or NULL if static
2174    // Rcache: cache entry pointer
2175    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2176               Otos_i, Rcache);
2177    if (!is_static && has_tos) {
2178      __ pop_ptr(Otos_i);  // restore object pointer
2179      __ verify_oop(Otos_i);
2180    }
2181    __ get_cache_and_index_at_bcp(Rcache, index, 1);
2182    __ bind(Label1);
2183  }
2184}
2185
2186void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2187  transition(vtos, vtos);
2188
2189  Register Rcache = G3_scratch;
2190  Register index  = G4_scratch;
2191  Register Rclass = Rcache;
2192  Register Roffset= G4_scratch;
2193  Register Rflags = G1_scratch;
2194  ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2195
2196  resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
2197  jvmti_post_field_access(Rcache, index, is_static, false);
2198  load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2199
2200  if (!is_static) {
2201    pop_and_check_object(Rclass);
2202  } else {
2203    __ verify_oop(Rclass);
2204  }
2205
2206  Label exit;
2207
2208  Assembler::Membar_mask_bits membar_bits =
2209    Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2210
2211  if (__ membar_has_effect(membar_bits)) {
2212    // Get volatile flag
2213    __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2214    __ and3(Rflags, Lscratch, Lscratch);
2215  }
2216
2217  Label checkVolatile;
2218
2219  // compute field type
2220  Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj;
2221  __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2222  // Make sure we don't need to mask Rflags after the above shift
2223  ConstantPoolCacheEntry::verify_tos_state_shift();
2224
2225  // Check atos before itos for getstatic, more likely (in Queens at least)
2226  __ cmp(Rflags, atos);
2227  __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2228  __ delayed() ->cmp(Rflags, itos);
2229
2230  // atos
2231  __ load_heap_oop(Rclass, Roffset, Otos_i);
2232  __ verify_oop(Otos_i);
2233  __ push(atos);
2234  if (!is_static) {
2235    patch_bytecode(Bytecodes::_fast_agetfield, G3_scratch, G4_scratch);
2236  }
2237  __ ba(checkVolatile);
2238  __ delayed()->tst(Lscratch);
2239
2240  __ bind(notObj);
2241
2242  // cmp(Rflags, itos);
2243  __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2244  __ delayed() ->cmp(Rflags, ltos);
2245
2246  // itos
2247  __ ld(Rclass, Roffset, Otos_i);
2248  __ push(itos);
2249  if (!is_static) {
2250    patch_bytecode(Bytecodes::_fast_igetfield, G3_scratch, G4_scratch);
2251  }
2252  __ ba(checkVolatile);
2253  __ delayed()->tst(Lscratch);
2254
2255  __ bind(notInt);
2256
2257  // cmp(Rflags, ltos);
2258  __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2259  __ delayed() ->cmp(Rflags, btos);
2260
2261  // ltos
2262  // load must be atomic
2263  __ ld_long(Rclass, Roffset, Otos_l);
2264  __ push(ltos);
2265  if (!is_static) {
2266    patch_bytecode(Bytecodes::_fast_lgetfield, G3_scratch, G4_scratch);
2267  }
2268  __ ba(checkVolatile);
2269  __ delayed()->tst(Lscratch);
2270
2271  __ bind(notLong);
2272
2273  // cmp(Rflags, btos);
2274  __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2275  __ delayed() ->cmp(Rflags, ctos);
2276
2277  // btos
2278  __ ldsb(Rclass, Roffset, Otos_i);
2279  __ push(itos);
2280  if (!is_static) {
2281    patch_bytecode(Bytecodes::_fast_bgetfield, G3_scratch, G4_scratch);
2282  }
2283  __ ba(checkVolatile);
2284  __ delayed()->tst(Lscratch);
2285
2286  __ bind(notByte);
2287
2288  // cmp(Rflags, ctos);
2289  __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2290  __ delayed() ->cmp(Rflags, stos);
2291
2292  // ctos
2293  __ lduh(Rclass, Roffset, Otos_i);
2294  __ push(itos);
2295  if (!is_static) {
2296    patch_bytecode(Bytecodes::_fast_cgetfield, G3_scratch, G4_scratch);
2297  }
2298  __ ba(checkVolatile);
2299  __ delayed()->tst(Lscratch);
2300
2301  __ bind(notChar);
2302
2303  // cmp(Rflags, stos);
2304  __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2305  __ delayed() ->cmp(Rflags, ftos);
2306
2307  // stos
2308  __ ldsh(Rclass, Roffset, Otos_i);
2309  __ push(itos);
2310  if (!is_static) {
2311    patch_bytecode(Bytecodes::_fast_sgetfield, G3_scratch, G4_scratch);
2312  }
2313  __ ba(checkVolatile);
2314  __ delayed()->tst(Lscratch);
2315
2316  __ bind(notShort);
2317
2318
2319  // cmp(Rflags, ftos);
2320  __ br(Assembler::notEqual, false, Assembler::pt, notFloat);
2321  __ delayed() ->tst(Lscratch);
2322
2323  // ftos
2324  __ ldf(FloatRegisterImpl::S, Rclass, Roffset, Ftos_f);
2325  __ push(ftos);
2326  if (!is_static) {
2327    patch_bytecode(Bytecodes::_fast_fgetfield, G3_scratch, G4_scratch);
2328  }
2329  __ ba(checkVolatile);
2330  __ delayed()->tst(Lscratch);
2331
2332  __ bind(notFloat);
2333
2334
2335  // dtos
2336  __ ldf(FloatRegisterImpl::D, Rclass, Roffset, Ftos_d);
2337  __ push(dtos);
2338  if (!is_static) {
2339    patch_bytecode(Bytecodes::_fast_dgetfield, G3_scratch, G4_scratch);
2340  }
2341
2342  __ bind(checkVolatile);
2343  if (__ membar_has_effect(membar_bits)) {
2344    // __ tst(Lscratch); executed in delay slot
2345    __ br(Assembler::zero, false, Assembler::pt, exit);
2346    __ delayed()->nop();
2347    volatile_barrier(membar_bits);
2348  }
2349
2350  __ bind(exit);
2351}
2352
2353
2354void TemplateTable::getfield(int byte_no) {
2355  getfield_or_static(byte_no, false);
2356}
2357
2358void TemplateTable::getstatic(int byte_no) {
2359  getfield_or_static(byte_no, true);
2360}
2361
2362
2363void TemplateTable::fast_accessfield(TosState state) {
2364  transition(atos, state);
2365  Register Rcache  = G3_scratch;
2366  Register index   = G4_scratch;
2367  Register Roffset = G4_scratch;
2368  Register Rflags  = Rcache;
2369  ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2370
2371  __ get_cache_and_index_at_bcp(Rcache, index, 1);
2372  jvmti_post_field_access(Rcache, index, /*is_static*/false, /*has_tos*/true);
2373
2374  __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2375
2376  __ null_check(Otos_i);
2377  __ verify_oop(Otos_i);
2378
2379  Label exit;
2380
2381  Assembler::Membar_mask_bits membar_bits =
2382    Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2383  if (__ membar_has_effect(membar_bits)) {
2384    // Get volatile flag
2385    __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Rflags);
2386    __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2387  }
2388
2389  switch (bytecode()) {
2390    case Bytecodes::_fast_bgetfield:
2391      __ ldsb(Otos_i, Roffset, Otos_i);
2392      break;
2393    case Bytecodes::_fast_cgetfield:
2394      __ lduh(Otos_i, Roffset, Otos_i);
2395      break;
2396    case Bytecodes::_fast_sgetfield:
2397      __ ldsh(Otos_i, Roffset, Otos_i);
2398      break;
2399    case Bytecodes::_fast_igetfield:
2400      __ ld(Otos_i, Roffset, Otos_i);
2401      break;
2402    case Bytecodes::_fast_lgetfield:
2403      __ ld_long(Otos_i, Roffset, Otos_l);
2404      break;
2405    case Bytecodes::_fast_fgetfield:
2406      __ ldf(FloatRegisterImpl::S, Otos_i, Roffset, Ftos_f);
2407      break;
2408    case Bytecodes::_fast_dgetfield:
2409      __ ldf(FloatRegisterImpl::D, Otos_i, Roffset, Ftos_d);
2410      break;
2411    case Bytecodes::_fast_agetfield:
2412      __ load_heap_oop(Otos_i, Roffset, Otos_i);
2413      break;
2414    default:
2415      ShouldNotReachHere();
2416  }
2417
2418  if (__ membar_has_effect(membar_bits)) {
2419    __ btst(Lscratch, Rflags);
2420    __ br(Assembler::zero, false, Assembler::pt, exit);
2421    __ delayed()->nop();
2422    volatile_barrier(membar_bits);
2423    __ bind(exit);
2424  }
2425
2426  if (state == atos) {
2427    __ verify_oop(Otos_i);    // does not blow flags!
2428  }
2429}
2430
2431void TemplateTable::jvmti_post_fast_field_mod() {
2432  if (JvmtiExport::can_post_field_modification()) {
2433    // Check to see if a field modification watch has been set before we take
2434    // the time to call into the VM.
2435    Label done;
2436    AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2437    __ load_contents(get_field_modification_count_addr, G4_scratch);
2438    __ cmp_and_br_short(G4_scratch, 0, Assembler::equal, Assembler::pt, done);
2439    __ pop_ptr(G4_scratch);     // copy the object pointer from tos
2440    __ verify_oop(G4_scratch);
2441    __ push_ptr(G4_scratch);    // put the object pointer back on tos
2442    __ get_cache_entry_pointer_at_bcp(G1_scratch, G3_scratch, 1);
2443    // Save tos values before call_VM() clobbers them. Since we have
2444    // to do it for every data type, we use the saved values as the
2445    // jvalue object.
2446    switch (bytecode()) {  // save tos values before call_VM() clobbers them
2447    case Bytecodes::_fast_aputfield: __ push_ptr(Otos_i); break;
2448    case Bytecodes::_fast_bputfield: // fall through
2449    case Bytecodes::_fast_sputfield: // fall through
2450    case Bytecodes::_fast_cputfield: // fall through
2451    case Bytecodes::_fast_iputfield: __ push_i(Otos_i); break;
2452    case Bytecodes::_fast_dputfield: __ push_d(Ftos_d); break;
2453    case Bytecodes::_fast_fputfield: __ push_f(Ftos_f); break;
2454    // get words in right order for use as jvalue object
2455    case Bytecodes::_fast_lputfield: __ push_l(Otos_l); break;
2456    }
2457    // setup pointer to jvalue object
2458    __ mov(Lesp, G3_scratch);  __ inc(G3_scratch, wordSize);
2459    // G4_scratch:  object pointer
2460    // G1_scratch: cache entry pointer
2461    // G3_scratch: jvalue object on the stack
2462    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), G4_scratch, G1_scratch, G3_scratch);
2463    switch (bytecode()) {             // restore tos values
2464    case Bytecodes::_fast_aputfield: __ pop_ptr(Otos_i); break;
2465    case Bytecodes::_fast_bputfield: // fall through
2466    case Bytecodes::_fast_sputfield: // fall through
2467    case Bytecodes::_fast_cputfield: // fall through
2468    case Bytecodes::_fast_iputfield: __ pop_i(Otos_i); break;
2469    case Bytecodes::_fast_dputfield: __ pop_d(Ftos_d); break;
2470    case Bytecodes::_fast_fputfield: __ pop_f(Ftos_f); break;
2471    case Bytecodes::_fast_lputfield: __ pop_l(Otos_l); break;
2472    }
2473    __ bind(done);
2474  }
2475}
2476
2477// The registers Rcache and index expected to be set before call.
2478// The function may destroy various registers, just not the Rcache and index registers.
2479void TemplateTable::jvmti_post_field_mod(Register Rcache, Register index, bool is_static) {
2480  ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2481
2482  if (JvmtiExport::can_post_field_modification()) {
2483    // Check to see if a field modification watch has been set before we take
2484    // the time to call into the VM.
2485    Label Label1;
2486    assert_different_registers(Rcache, index, G1_scratch);
2487    AddressLiteral get_field_modification_count_addr(JvmtiExport::get_field_modification_count_addr());
2488    __ load_contents(get_field_modification_count_addr, G1_scratch);
2489    __ cmp_and_br_short(G1_scratch, 0, Assembler::zero, Assembler::pt, Label1);
2490
2491    // The Rcache and index registers have been already set.
2492    // This allows to eliminate this call but the Rcache and index
2493    // registers must be correspondingly used after this line.
2494    __ get_cache_and_index_at_bcp(G1_scratch, G4_scratch, 1);
2495
2496    __ add(G1_scratch, in_bytes(cp_base_offset), G3_scratch);
2497    if (is_static) {
2498      // Life is simple.  Null out the object pointer.
2499      __ clr(G4_scratch);
2500    } else {
2501      Register Rflags = G1_scratch;
2502      // Life is harder. The stack holds the value on top, followed by the
2503      // object.  We don't know the size of the value, though; it could be
2504      // one or two words depending on its type. As a result, we must find
2505      // the type to determine where the object is.
2506
2507      Label two_word, valsizeknown;
2508      __ ld_ptr(G1_scratch, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2509      __ mov(Lesp, G4_scratch);
2510      __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2511      // Make sure we don't need to mask Rflags after the above shift
2512      ConstantPoolCacheEntry::verify_tos_state_shift();
2513      __ cmp(Rflags, ltos);
2514      __ br(Assembler::equal, false, Assembler::pt, two_word);
2515      __ delayed()->cmp(Rflags, dtos);
2516      __ br(Assembler::equal, false, Assembler::pt, two_word);
2517      __ delayed()->nop();
2518      __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(1));
2519      __ ba_short(valsizeknown);
2520      __ bind(two_word);
2521
2522      __ inc(G4_scratch, Interpreter::expr_offset_in_bytes(2));
2523
2524      __ bind(valsizeknown);
2525      // setup object pointer
2526      __ ld_ptr(G4_scratch, 0, G4_scratch);
2527      __ verify_oop(G4_scratch);
2528    }
2529    // setup pointer to jvalue object
2530    __ mov(Lesp, G1_scratch);  __ inc(G1_scratch, wordSize);
2531    // G4_scratch:  object pointer or NULL if static
2532    // G3_scratch: cache entry pointer
2533    // G1_scratch: jvalue object on the stack
2534    __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2535               G4_scratch, G3_scratch, G1_scratch);
2536    __ get_cache_and_index_at_bcp(Rcache, index, 1);
2537    __ bind(Label1);
2538  }
2539}
2540
2541void TemplateTable::pop_and_check_object(Register r) {
2542  __ pop_ptr(r);
2543  __ null_check(r);  // for field access must check obj.
2544  __ verify_oop(r);
2545}
2546
2547void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2548  transition(vtos, vtos);
2549  Register Rcache = G3_scratch;
2550  Register index  = G4_scratch;
2551  Register Rclass = Rcache;
2552  Register Roffset= G4_scratch;
2553  Register Rflags = G1_scratch;
2554  ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2555
2556  resolve_cache_and_index(byte_no, Rcache, index, sizeof(u2));
2557  jvmti_post_field_mod(Rcache, index, is_static);
2558  load_field_cp_cache_entry(Rclass, Rcache, index, Roffset, Rflags, is_static);
2559
2560  Assembler::Membar_mask_bits read_bits =
2561    Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2562  Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2563
2564  Label notVolatile, checkVolatile, exit;
2565  if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2566    __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2567    __ and3(Rflags, Lscratch, Lscratch);
2568
2569    if (__ membar_has_effect(read_bits)) {
2570      __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2571      volatile_barrier(read_bits);
2572      __ bind(notVolatile);
2573    }
2574  }
2575
2576  __ srl(Rflags, ConstantPoolCacheEntry::tos_state_shift, Rflags);
2577  // Make sure we don't need to mask Rflags after the above shift
2578  ConstantPoolCacheEntry::verify_tos_state_shift();
2579
2580  // compute field type
2581  Label notInt, notShort, notChar, notObj, notByte, notLong, notFloat;
2582
2583  if (is_static) {
2584    // putstatic with object type most likely, check that first
2585    __ cmp(Rflags, atos);
2586    __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2587    __ delayed()->cmp(Rflags, itos);
2588
2589    // atos
2590    {
2591      __ pop_ptr();
2592      __ verify_oop(Otos_i);
2593      do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2594      __ ba(checkVolatile);
2595      __ delayed()->tst(Lscratch);
2596    }
2597
2598    __ bind(notObj);
2599    // cmp(Rflags, itos);
2600    __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2601    __ delayed()->cmp(Rflags, btos);
2602
2603    // itos
2604    {
2605      __ pop_i();
2606      __ st(Otos_i, Rclass, Roffset);
2607      __ ba(checkVolatile);
2608      __ delayed()->tst(Lscratch);
2609    }
2610
2611    __ bind(notInt);
2612  } else {
2613    // putfield with int type most likely, check that first
2614    __ cmp(Rflags, itos);
2615    __ br(Assembler::notEqual, false, Assembler::pt, notInt);
2616    __ delayed()->cmp(Rflags, atos);
2617
2618    // itos
2619    {
2620      __ pop_i();
2621      pop_and_check_object(Rclass);
2622      __ st(Otos_i, Rclass, Roffset);
2623      patch_bytecode(Bytecodes::_fast_iputfield, G3_scratch, G4_scratch, true, byte_no);
2624      __ ba(checkVolatile);
2625      __ delayed()->tst(Lscratch);
2626    }
2627
2628    __ bind(notInt);
2629    // cmp(Rflags, atos);
2630    __ br(Assembler::notEqual, false, Assembler::pt, notObj);
2631    __ delayed()->cmp(Rflags, btos);
2632
2633    // atos
2634    {
2635      __ pop_ptr();
2636      pop_and_check_object(Rclass);
2637      __ verify_oop(Otos_i);
2638      do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2639      patch_bytecode(Bytecodes::_fast_aputfield, G3_scratch, G4_scratch, true, byte_no);
2640      __ ba(checkVolatile);
2641      __ delayed()->tst(Lscratch);
2642    }
2643
2644    __ bind(notObj);
2645  }
2646
2647  // cmp(Rflags, btos);
2648  __ br(Assembler::notEqual, false, Assembler::pt, notByte);
2649  __ delayed()->cmp(Rflags, ltos);
2650
2651  // btos
2652  {
2653    __ pop_i();
2654    if (!is_static) pop_and_check_object(Rclass);
2655    __ stb(Otos_i, Rclass, Roffset);
2656    if (!is_static) {
2657      patch_bytecode(Bytecodes::_fast_bputfield, G3_scratch, G4_scratch, true, byte_no);
2658    }
2659    __ ba(checkVolatile);
2660    __ delayed()->tst(Lscratch);
2661  }
2662
2663  __ bind(notByte);
2664  // cmp(Rflags, ltos);
2665  __ br(Assembler::notEqual, false, Assembler::pt, notLong);
2666  __ delayed()->cmp(Rflags, ctos);
2667
2668  // ltos
2669  {
2670    __ pop_l();
2671    if (!is_static) pop_and_check_object(Rclass);
2672    __ st_long(Otos_l, Rclass, Roffset);
2673    if (!is_static) {
2674      patch_bytecode(Bytecodes::_fast_lputfield, G3_scratch, G4_scratch, true, byte_no);
2675    }
2676    __ ba(checkVolatile);
2677    __ delayed()->tst(Lscratch);
2678  }
2679
2680  __ bind(notLong);
2681  // cmp(Rflags, ctos);
2682  __ br(Assembler::notEqual, false, Assembler::pt, notChar);
2683  __ delayed()->cmp(Rflags, stos);
2684
2685  // ctos (char)
2686  {
2687    __ pop_i();
2688    if (!is_static) pop_and_check_object(Rclass);
2689    __ sth(Otos_i, Rclass, Roffset);
2690    if (!is_static) {
2691      patch_bytecode(Bytecodes::_fast_cputfield, G3_scratch, G4_scratch, true, byte_no);
2692    }
2693    __ ba(checkVolatile);
2694    __ delayed()->tst(Lscratch);
2695  }
2696
2697  __ bind(notChar);
2698  // cmp(Rflags, stos);
2699  __ br(Assembler::notEqual, false, Assembler::pt, notShort);
2700  __ delayed()->cmp(Rflags, ftos);
2701
2702  // stos (short)
2703  {
2704    __ pop_i();
2705    if (!is_static) pop_and_check_object(Rclass);
2706    __ sth(Otos_i, Rclass, Roffset);
2707    if (!is_static) {
2708      patch_bytecode(Bytecodes::_fast_sputfield, G3_scratch, G4_scratch, true, byte_no);
2709    }
2710    __ ba(checkVolatile);
2711    __ delayed()->tst(Lscratch);
2712  }
2713
2714  __ bind(notShort);
2715  // cmp(Rflags, ftos);
2716  __ br(Assembler::notZero, false, Assembler::pt, notFloat);
2717  __ delayed()->nop();
2718
2719  // ftos
2720  {
2721    __ pop_f();
2722    if (!is_static) pop_and_check_object(Rclass);
2723    __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2724    if (!is_static) {
2725      patch_bytecode(Bytecodes::_fast_fputfield, G3_scratch, G4_scratch, true, byte_no);
2726    }
2727    __ ba(checkVolatile);
2728    __ delayed()->tst(Lscratch);
2729  }
2730
2731  __ bind(notFloat);
2732
2733  // dtos
2734  {
2735    __ pop_d();
2736    if (!is_static) pop_and_check_object(Rclass);
2737    __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2738    if (!is_static) {
2739      patch_bytecode(Bytecodes::_fast_dputfield, G3_scratch, G4_scratch, true, byte_no);
2740    }
2741  }
2742
2743  __ bind(checkVolatile);
2744  __ tst(Lscratch);
2745
2746  if (__ membar_has_effect(write_bits)) {
2747    // __ tst(Lscratch); in delay slot
2748    __ br(Assembler::zero, false, Assembler::pt, exit);
2749    __ delayed()->nop();
2750    volatile_barrier(Assembler::StoreLoad);
2751    __ bind(exit);
2752  }
2753}
2754
2755void TemplateTable::fast_storefield(TosState state) {
2756  transition(state, vtos);
2757  Register Rcache = G3_scratch;
2758  Register Rclass = Rcache;
2759  Register Roffset= G4_scratch;
2760  Register Rflags = G1_scratch;
2761  ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2762
2763  jvmti_post_fast_field_mod();
2764
2765  __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 1);
2766
2767  Assembler::Membar_mask_bits read_bits =
2768    Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore);
2769  Assembler::Membar_mask_bits write_bits = Assembler::StoreLoad;
2770
2771  Label notVolatile, checkVolatile, exit;
2772  if (__ membar_has_effect(read_bits) || __ membar_has_effect(write_bits)) {
2773    __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::flags_offset(), Rflags);
2774    __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2775    __ and3(Rflags, Lscratch, Lscratch);
2776    if (__ membar_has_effect(read_bits)) {
2777      __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, notVolatile);
2778      volatile_barrier(read_bits);
2779      __ bind(notVolatile);
2780    }
2781  }
2782
2783  __ ld_ptr(Rcache, cp_base_offset + ConstantPoolCacheEntry::f2_offset(), Roffset);
2784  pop_and_check_object(Rclass);
2785
2786  switch (bytecode()) {
2787    case Bytecodes::_fast_bputfield: __ stb(Otos_i, Rclass, Roffset); break;
2788    case Bytecodes::_fast_cputfield: /* fall through */
2789    case Bytecodes::_fast_sputfield: __ sth(Otos_i, Rclass, Roffset); break;
2790    case Bytecodes::_fast_iputfield: __ st(Otos_i, Rclass, Roffset);  break;
2791    case Bytecodes::_fast_lputfield: __ st_long(Otos_l, Rclass, Roffset); break;
2792    case Bytecodes::_fast_fputfield:
2793      __ stf(FloatRegisterImpl::S, Ftos_f, Rclass, Roffset);
2794      break;
2795    case Bytecodes::_fast_dputfield:
2796      __ stf(FloatRegisterImpl::D, Ftos_d, Rclass, Roffset);
2797      break;
2798    case Bytecodes::_fast_aputfield:
2799      do_oop_store(_masm, Rclass, Roffset, 0, Otos_i, G1_scratch, _bs->kind(), false);
2800      break;
2801    default:
2802      ShouldNotReachHere();
2803  }
2804
2805  if (__ membar_has_effect(write_bits)) {
2806    __ cmp_and_br_short(Lscratch, 0, Assembler::equal, Assembler::pt, exit);
2807    volatile_barrier(Assembler::StoreLoad);
2808    __ bind(exit);
2809  }
2810}
2811
2812
2813void TemplateTable::putfield(int byte_no) {
2814  putfield_or_static(byte_no, false);
2815}
2816
2817void TemplateTable::putstatic(int byte_no) {
2818  putfield_or_static(byte_no, true);
2819}
2820
2821
2822void TemplateTable::fast_xaccess(TosState state) {
2823  transition(vtos, state);
2824  Register Rcache = G3_scratch;
2825  Register Roffset = G4_scratch;
2826  Register Rflags  = G4_scratch;
2827  Register Rreceiver = Lscratch;
2828
2829  __ ld_ptr(Llocals, 0, Rreceiver);
2830
2831  // access constant pool cache  (is resolved)
2832  __ get_cache_and_index_at_bcp(Rcache, G4_scratch, 2);
2833  __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset(), Roffset);
2834  __ add(Lbcp, 1, Lbcp);       // needed to report exception at the correct bcp
2835
2836  __ verify_oop(Rreceiver);
2837  __ null_check(Rreceiver);
2838  if (state == atos) {
2839    __ load_heap_oop(Rreceiver, Roffset, Otos_i);
2840  } else if (state == itos) {
2841    __ ld (Rreceiver, Roffset, Otos_i) ;
2842  } else if (state == ftos) {
2843    __ ldf(FloatRegisterImpl::S, Rreceiver, Roffset, Ftos_f);
2844  } else {
2845    ShouldNotReachHere();
2846  }
2847
2848  Assembler::Membar_mask_bits membar_bits =
2849    Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore);
2850  if (__ membar_has_effect(membar_bits)) {
2851
2852    // Get is_volatile value in Rflags and check if membar is needed
2853    __ ld_ptr(Rcache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset(), Rflags);
2854
2855    // Test volatile
2856    Label notVolatile;
2857    __ set((1 << ConstantPoolCacheEntry::is_volatile_shift), Lscratch);
2858    __ btst(Rflags, Lscratch);
2859    __ br(Assembler::zero, false, Assembler::pt, notVolatile);
2860    __ delayed()->nop();
2861    volatile_barrier(membar_bits);
2862    __ bind(notVolatile);
2863  }
2864
2865  __ interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
2866  __ sub(Lbcp, 1, Lbcp);
2867}
2868
2869//----------------------------------------------------------------------------------------------------
2870// Calls
2871
2872void TemplateTable::count_calls(Register method, Register temp) {
2873  // implemented elsewhere
2874  ShouldNotReachHere();
2875}
2876
2877void TemplateTable::prepare_invoke(int byte_no,
2878                                   Register method,  // linked method (or i-klass)
2879                                   Register ra,      // return address
2880                                   Register index,   // itable index, MethodType, etc.
2881                                   Register recv,    // if caller wants to see it
2882                                   Register flags    // if caller wants to test it
2883                                   ) {
2884  // determine flags
2885  const Bytecodes::Code code = bytecode();
2886  const bool is_invokeinterface  = code == Bytecodes::_invokeinterface;
2887  const bool is_invokedynamic    = code == Bytecodes::_invokedynamic;
2888  const bool is_invokehandle     = code == Bytecodes::_invokehandle;
2889  const bool is_invokevirtual    = code == Bytecodes::_invokevirtual;
2890  const bool is_invokespecial    = code == Bytecodes::_invokespecial;
2891  const bool load_receiver       = (recv != noreg);
2892  assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2893  assert(recv  == noreg || recv  == O0, "");
2894  assert(flags == noreg || flags == O1, "");
2895
2896  // setup registers & access constant pool cache
2897  if (recv  == noreg)  recv  = O0;
2898  if (flags == noreg)  flags = O1;
2899  const Register temp = O2;
2900  assert_different_registers(method, ra, index, recv, flags, temp);
2901
2902  load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2903
2904  __ mov(SP, O5_savedSP);  // record SP that we wanted the callee to restore
2905
2906  // maybe push appendix to arguments
2907  if (is_invokedynamic || is_invokehandle) {
2908    Label L_no_push;
2909    __ set((1 << ConstantPoolCacheEntry::has_appendix_shift), temp);
2910    __ btst(flags, temp);
2911    __ br(Assembler::zero, false, Assembler::pt, L_no_push);
2912    __ delayed()->nop();
2913    // Push the appendix as a trailing parameter.
2914    // This must be done before we get the receiver,
2915    // since the parameter_size includes it.
2916    assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2917    __ load_resolved_reference_at_index(temp, index);
2918    __ verify_oop(temp);
2919    __ push_ptr(temp);  // push appendix (MethodType, CallSite, etc.)
2920    __ bind(L_no_push);
2921  }
2922
2923  // load receiver if needed (after appendix is pushed so parameter size is correct)
2924  if (load_receiver) {
2925    __ and3(flags, ConstantPoolCacheEntry::parameter_size_mask, temp);  // get parameter size
2926    __ load_receiver(temp, recv);  //  __ argument_address uses Gargs but we need Lesp
2927    __ verify_oop(recv);
2928  }
2929
2930  // compute return type
2931  __ srl(flags, ConstantPoolCacheEntry::tos_state_shift, ra);
2932  // Make sure we don't need to mask flags after the above shift
2933  ConstantPoolCacheEntry::verify_tos_state_shift();
2934  // load return address
2935  {
2936    const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2937    AddressLiteral table(table_addr);
2938    __ set(table, temp);
2939    __ sll(ra, LogBytesPerWord, ra);
2940    __ ld_ptr(Address(temp, ra), ra);
2941  }
2942}
2943
2944
2945void TemplateTable::generate_vtable_call(Register Rrecv, Register Rindex, Register Rret) {
2946  Register Rcall = Rindex;
2947  assert_different_registers(Rcall, G5_method, Gargs, Rret);
2948
2949  // get target Method* & entry point
2950  __ lookup_virtual_method(Rrecv, Rindex, G5_method);
2951  __ profile_arguments_type(G5_method, Rcall, Gargs, true);
2952  __ call_from_interpreter(Rcall, Gargs, Rret);
2953}
2954
2955void TemplateTable::invokevirtual(int byte_no) {
2956  transition(vtos, vtos);
2957  assert(byte_no == f2_byte, "use this argument");
2958
2959  Register Rscratch = G3_scratch;
2960  Register Rtemp    = G4_scratch;
2961  Register Rret     = Lscratch;
2962  Register O0_recv  = O0;
2963  Label notFinal;
2964
2965  load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Rret, true, false, false);
2966  __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
2967
2968  // Check for vfinal
2969  __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), G4_scratch);
2970  __ btst(Rret, G4_scratch);
2971  __ br(Assembler::zero, false, Assembler::pt, notFinal);
2972  __ delayed()->and3(Rret, 0xFF, G4_scratch);      // gets number of parameters
2973
2974  patch_bytecode(Bytecodes::_fast_invokevfinal, Rscratch, Rtemp);
2975
2976  invokevfinal_helper(Rscratch, Rret);
2977
2978  __ bind(notFinal);
2979
2980  __ mov(G5_method, Rscratch);  // better scratch register
2981  __ load_receiver(G4_scratch, O0_recv);  // gets receiverOop
2982  // receiver is in O0_recv
2983  __ verify_oop(O0_recv);
2984
2985  // get return address
2986  AddressLiteral table(Interpreter::invoke_return_entry_table());
2987  __ set(table, Rtemp);
2988  __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
2989  // Make sure we don't need to mask Rret after the above shift
2990  ConstantPoolCacheEntry::verify_tos_state_shift();
2991  __ sll(Rret,  LogBytesPerWord, Rret);
2992  __ ld_ptr(Rtemp, Rret, Rret);         // get return address
2993
2994  // get receiver klass
2995  __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
2996  __ load_klass(O0_recv, O0_recv);
2997  __ verify_klass_ptr(O0_recv);
2998
2999  __ profile_virtual_call(O0_recv, O4);
3000
3001  generate_vtable_call(O0_recv, Rscratch, Rret);
3002}
3003
3004void TemplateTable::fast_invokevfinal(int byte_no) {
3005  transition(vtos, vtos);
3006  assert(byte_no == f2_byte, "use this argument");
3007
3008  load_invoke_cp_cache_entry(byte_no, G5_method, noreg, Lscratch, true,
3009                             /*is_invokevfinal*/true, false);
3010  __ mov(SP, O5_savedSP); // record SP that we wanted the callee to restore
3011  invokevfinal_helper(G3_scratch, Lscratch);
3012}
3013
3014void TemplateTable::invokevfinal_helper(Register Rscratch, Register Rret) {
3015  Register Rtemp = G4_scratch;
3016
3017  // Load receiver from stack slot
3018  __ ld_ptr(G5_method, in_bytes(Method::const_offset()), G4_scratch);
3019  __ lduh(G4_scratch, in_bytes(ConstMethod::size_of_parameters_offset()), G4_scratch);
3020  __ load_receiver(G4_scratch, O0);
3021
3022  // receiver NULL check
3023  __ null_check(O0);
3024
3025  __ profile_final_call(O4);
3026  __ profile_arguments_type(G5_method, Rscratch, Gargs, true);
3027
3028  // get return address
3029  AddressLiteral table(Interpreter::invoke_return_entry_table());
3030  __ set(table, Rtemp);
3031  __ srl(Rret, ConstantPoolCacheEntry::tos_state_shift, Rret);          // get return type
3032  // Make sure we don't need to mask Rret after the above shift
3033  ConstantPoolCacheEntry::verify_tos_state_shift();
3034  __ sll(Rret,  LogBytesPerWord, Rret);
3035  __ ld_ptr(Rtemp, Rret, Rret);         // get return address
3036
3037
3038  // do the call
3039  __ call_from_interpreter(Rscratch, Gargs, Rret);
3040}
3041
3042
3043void TemplateTable::invokespecial(int byte_no) {
3044  transition(vtos, vtos);
3045  assert(byte_no == f1_byte, "use this argument");
3046
3047  const Register Rret     = Lscratch;
3048  const Register O0_recv  = O0;
3049  const Register Rscratch = G3_scratch;
3050
3051  prepare_invoke(byte_no, G5_method, Rret, noreg, O0_recv);  // get receiver also for null check
3052  __ null_check(O0_recv);
3053
3054  // do the call
3055  __ profile_call(O4);
3056  __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3057  __ call_from_interpreter(Rscratch, Gargs, Rret);
3058}
3059
3060
3061void TemplateTable::invokestatic(int byte_no) {
3062  transition(vtos, vtos);
3063  assert(byte_no == f1_byte, "use this argument");
3064
3065  const Register Rret     = Lscratch;
3066  const Register Rscratch = G3_scratch;
3067
3068  prepare_invoke(byte_no, G5_method, Rret);  // get f1 Method*
3069
3070  // do the call
3071  __ profile_call(O4);
3072  __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3073  __ call_from_interpreter(Rscratch, Gargs, Rret);
3074}
3075
3076void TemplateTable::invokeinterface_object_method(Register RKlass,
3077                                                  Register Rcall,
3078                                                  Register Rret,
3079                                                  Register Rflags) {
3080  Register Rscratch = G4_scratch;
3081  Register Rindex = Lscratch;
3082
3083  assert_different_registers(Rscratch, Rindex, Rret);
3084
3085  Label notFinal;
3086
3087  // Check for vfinal
3088  __ set((1 << ConstantPoolCacheEntry::is_vfinal_shift), Rscratch);
3089  __ btst(Rflags, Rscratch);
3090  __ br(Assembler::zero, false, Assembler::pt, notFinal);
3091  __ delayed()->nop();
3092
3093  __ profile_final_call(O4);
3094
3095  // do the call - the index (f2) contains the Method*
3096  assert_different_registers(G5_method, Gargs, Rcall);
3097  __ mov(Rindex, G5_method);
3098  __ profile_arguments_type(G5_method, Rcall, Gargs, true);
3099  __ call_from_interpreter(Rcall, Gargs, Rret);
3100  __ bind(notFinal);
3101
3102  __ profile_virtual_call(RKlass, O4);
3103  generate_vtable_call(RKlass, Rindex, Rret);
3104}
3105
3106
3107void TemplateTable::invokeinterface(int byte_no) {
3108  transition(vtos, vtos);
3109  assert(byte_no == f1_byte, "use this argument");
3110
3111  const Register Rinterface  = G1_scratch;
3112  const Register Rret        = G3_scratch;
3113  const Register Rindex      = Lscratch;
3114  const Register O0_recv     = O0;
3115  const Register O1_flags    = O1;
3116  const Register O2_Klass    = O2;
3117  const Register Rscratch    = G4_scratch;
3118  assert_different_registers(Rscratch, G5_method);
3119
3120  prepare_invoke(byte_no, Rinterface, Rret, Rindex, O0_recv, O1_flags);
3121
3122  // get receiver klass
3123  __ null_check(O0_recv, oopDesc::klass_offset_in_bytes());
3124  __ load_klass(O0_recv, O2_Klass);
3125
3126  // Special case of invokeinterface called for virtual method of
3127  // java.lang.Object.  See cpCacheOop.cpp for details.
3128  // This code isn't produced by javac, but could be produced by
3129  // another compliant java compiler.
3130  Label notMethod;
3131  __ set((1 << ConstantPoolCacheEntry::is_forced_virtual_shift), Rscratch);
3132  __ btst(O1_flags, Rscratch);
3133  __ br(Assembler::zero, false, Assembler::pt, notMethod);
3134  __ delayed()->nop();
3135
3136  invokeinterface_object_method(O2_Klass, Rinterface, Rret, O1_flags);
3137
3138  __ bind(notMethod);
3139
3140  __ profile_virtual_call(O2_Klass, O4);
3141
3142  //
3143  // find entry point to call
3144  //
3145
3146  // compute start of first itableOffsetEntry (which is at end of vtable)
3147  const int base = InstanceKlass::vtable_start_offset() * wordSize;
3148  Label search;
3149  Register Rtemp = O1_flags;
3150
3151  __ ld(O2_Klass, InstanceKlass::vtable_length_offset() * wordSize, Rtemp);
3152  if (align_object_offset(1) > 1) {
3153    __ round_to(Rtemp, align_object_offset(1));
3154  }
3155  __ sll(Rtemp, LogBytesPerWord, Rtemp);   // Rscratch *= 4;
3156  if (Assembler::is_simm13(base)) {
3157    __ add(Rtemp, base, Rtemp);
3158  } else {
3159    __ set(base, Rscratch);
3160    __ add(Rscratch, Rtemp, Rtemp);
3161  }
3162  __ add(O2_Klass, Rtemp, Rscratch);
3163
3164  __ bind(search);
3165
3166  __ ld_ptr(Rscratch, itableOffsetEntry::interface_offset_in_bytes(), Rtemp);
3167  {
3168    Label ok;
3169
3170    // Check that entry is non-null.  Null entries are probably a bytecode
3171    // problem.  If the interface isn't implemented by the receiver class,
3172    // the VM should throw IncompatibleClassChangeError.  linkResolver checks
3173    // this too but that's only if the entry isn't already resolved, so we
3174    // need to check again.
3175    __ br_notnull_short( Rtemp, Assembler::pt, ok);
3176    call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeError));
3177    __ should_not_reach_here();
3178    __ bind(ok);
3179  }
3180
3181  __ cmp(Rinterface, Rtemp);
3182  __ brx(Assembler::notEqual, true, Assembler::pn, search);
3183  __ delayed()->add(Rscratch, itableOffsetEntry::size() * wordSize, Rscratch);
3184
3185  // entry found and Rscratch points to it
3186  __ ld(Rscratch, itableOffsetEntry::offset_offset_in_bytes(), Rscratch);
3187
3188  assert(itableMethodEntry::method_offset_in_bytes() == 0, "adjust instruction below");
3189  __ sll(Rindex, exact_log2(itableMethodEntry::size() * wordSize), Rindex);       // Rindex *= 8;
3190  __ add(Rscratch, Rindex, Rscratch);
3191  __ ld_ptr(O2_Klass, Rscratch, G5_method);
3192
3193  // Check for abstract method error.
3194  {
3195    Label ok;
3196    __ br_notnull_short(G5_method, Assembler::pt, ok);
3197    call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3198    __ should_not_reach_here();
3199    __ bind(ok);
3200  }
3201
3202  Register Rcall = Rinterface;
3203  assert_different_registers(Rcall, G5_method, Gargs, Rret);
3204
3205  __ profile_arguments_type(G5_method, Rcall, Gargs, true);
3206  __ call_from_interpreter(Rcall, Gargs, Rret);
3207}
3208
3209void TemplateTable::invokehandle(int byte_no) {
3210  transition(vtos, vtos);
3211  assert(byte_no == f1_byte, "use this argument");
3212
3213  const Register Rret       = Lscratch;
3214  const Register G4_mtype   = G4_scratch;
3215  const Register O0_recv    = O0;
3216  const Register Rscratch   = G3_scratch;
3217
3218  prepare_invoke(byte_no, G5_method, Rret, G4_mtype, O0_recv);
3219  __ null_check(O0_recv);
3220
3221  // G4: MethodType object (from cpool->resolved_references[f1], if necessary)
3222  // G5: MH.invokeExact_MT method (from f2)
3223
3224  // Note:  G4_mtype is already pushed (if necessary) by prepare_invoke
3225
3226  // do the call
3227  __ verify_oop(G4_mtype);
3228  __ profile_final_call(O4);  // FIXME: profile the LambdaForm also
3229  __ profile_arguments_type(G5_method, Rscratch, Gargs, true);
3230  __ call_from_interpreter(Rscratch, Gargs, Rret);
3231}
3232
3233
3234void TemplateTable::invokedynamic(int byte_no) {
3235  transition(vtos, vtos);
3236  assert(byte_no == f1_byte, "use this argument");
3237
3238  const Register Rret        = Lscratch;
3239  const Register G4_callsite = G4_scratch;
3240  const Register Rscratch    = G3_scratch;
3241
3242  prepare_invoke(byte_no, G5_method, Rret, G4_callsite);
3243
3244  // G4: CallSite object (from cpool->resolved_references[f1])
3245  // G5: MH.linkToCallSite method (from f2)
3246
3247  // Note:  G4_callsite is already pushed by prepare_invoke
3248
3249  // %%% should make a type profile for any invokedynamic that takes a ref argument
3250  // profile this call
3251  __ profile_call(O4);
3252
3253  // do the call
3254  __ verify_oop(G4_callsite);
3255  __ profile_arguments_type(G5_method, Rscratch, Gargs, false);
3256  __ call_from_interpreter(Rscratch, Gargs, Rret);
3257}
3258
3259
3260//----------------------------------------------------------------------------------------------------
3261// Allocation
3262
3263void TemplateTable::_new() {
3264  transition(vtos, atos);
3265
3266  Label slow_case;
3267  Label done;
3268  Label initialize_header;
3269  Label initialize_object;  // including clearing the fields
3270
3271  Register RallocatedObject = Otos_i;
3272  Register RinstanceKlass = O1;
3273  Register Roffset = O3;
3274  Register Rscratch = O4;
3275
3276  __ get_2_byte_integer_at_bcp(1, Rscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3277  __ get_cpool_and_tags(Rscratch, G3_scratch);
3278  // make sure the class we're about to instantiate has been resolved
3279  // This is done before loading InstanceKlass to be consistent with the order
3280  // how Constant Pool is updated (see ConstantPool::klass_at_put)
3281  __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3282  __ ldub(G3_scratch, Roffset, G3_scratch);
3283  __ cmp(G3_scratch, JVM_CONSTANT_Class);
3284  __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3285  __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3286  // get InstanceKlass
3287  //__ sll(Roffset, LogBytesPerWord, Roffset);        // executed in delay slot
3288  __ add(Roffset, sizeof(ConstantPool), Roffset);
3289  __ ld_ptr(Rscratch, Roffset, RinstanceKlass);
3290
3291  // make sure klass is fully initialized:
3292  __ ldub(RinstanceKlass, in_bytes(InstanceKlass::init_state_offset()), G3_scratch);
3293  __ cmp(G3_scratch, InstanceKlass::fully_initialized);
3294  __ br(Assembler::notEqual, false, Assembler::pn, slow_case);
3295  __ delayed()->ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
3296
3297  // get instance_size in InstanceKlass (already aligned)
3298  //__ ld(RinstanceKlass, in_bytes(Klass::layout_helper_offset()), Roffset);
3299
3300  // make sure klass does not have has_finalizer, or is abstract, or interface or java/lang/Class
3301  __ btst(Klass::_lh_instance_slow_path_bit, Roffset);
3302  __ br(Assembler::notZero, false, Assembler::pn, slow_case);
3303  __ delayed()->nop();
3304
3305  // allocate the instance
3306  // 1) Try to allocate in the TLAB
3307  // 2) if fail, and the TLAB is not full enough to discard, allocate in the shared Eden
3308  // 3) if the above fails (or is not applicable), go to a slow case
3309  // (creates a new TLAB, etc.)
3310
3311  const bool allow_shared_alloc =
3312    Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3313
3314  if(UseTLAB) {
3315    Register RoldTopValue = RallocatedObject;
3316    Register RtlabWasteLimitValue = G3_scratch;
3317    Register RnewTopValue = G1_scratch;
3318    Register RendValue = Rscratch;
3319    Register RfreeValue = RnewTopValue;
3320
3321    // check if we can allocate in the TLAB
3322    __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), RoldTopValue); // sets up RalocatedObject
3323    __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), RendValue);
3324    __ add(RoldTopValue, Roffset, RnewTopValue);
3325
3326    // if there is enough space, we do not CAS and do not clear
3327    __ cmp(RnewTopValue, RendValue);
3328    if(ZeroTLAB) {
3329      // the fields have already been cleared
3330      __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_header);
3331    } else {
3332      // initialize both the header and fields
3333      __ brx(Assembler::lessEqualUnsigned, true, Assembler::pt, initialize_object);
3334    }
3335    __ delayed()->st_ptr(RnewTopValue, G2_thread, in_bytes(JavaThread::tlab_top_offset()));
3336
3337    if (allow_shared_alloc) {
3338      // Check if tlab should be discarded (refill_waste_limit >= free)
3339      __ ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), RtlabWasteLimitValue);
3340      __ sub(RendValue, RoldTopValue, RfreeValue);
3341#ifdef _LP64
3342      __ srlx(RfreeValue, LogHeapWordSize, RfreeValue);
3343#else
3344      __ srl(RfreeValue, LogHeapWordSize, RfreeValue);
3345#endif
3346      __ cmp_and_brx_short(RtlabWasteLimitValue, RfreeValue, Assembler::greaterEqualUnsigned, Assembler::pt, slow_case); // tlab waste is small
3347
3348      // increment waste limit to prevent getting stuck on this slow path
3349      __ add(RtlabWasteLimitValue, ThreadLocalAllocBuffer::refill_waste_limit_increment(), RtlabWasteLimitValue);
3350      __ st_ptr(RtlabWasteLimitValue, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()));
3351    } else {
3352      // No allocation in the shared eden.
3353      __ ba_short(slow_case);
3354    }
3355  }
3356
3357  // Allocation in the shared Eden
3358  if (allow_shared_alloc) {
3359    Register RoldTopValue = G1_scratch;
3360    Register RtopAddr = G3_scratch;
3361    Register RnewTopValue = RallocatedObject;
3362    Register RendValue = Rscratch;
3363
3364    __ set((intptr_t)Universe::heap()->top_addr(), RtopAddr);
3365
3366    Label retry;
3367    __ bind(retry);
3368    __ set((intptr_t)Universe::heap()->end_addr(), RendValue);
3369    __ ld_ptr(RendValue, 0, RendValue);
3370    __ ld_ptr(RtopAddr, 0, RoldTopValue);
3371    __ add(RoldTopValue, Roffset, RnewTopValue);
3372
3373    // RnewTopValue contains the top address after the new object
3374    // has been allocated.
3375    __ cmp_and_brx_short(RnewTopValue, RendValue, Assembler::greaterUnsigned, Assembler::pn, slow_case);
3376
3377    __ cas_ptr(RtopAddr, RoldTopValue, RnewTopValue);
3378
3379    // if someone beat us on the allocation, try again, otherwise continue
3380    __ cmp_and_brx_short(RoldTopValue, RnewTopValue, Assembler::notEqual, Assembler::pn, retry);
3381
3382    // bump total bytes allocated by this thread
3383    // RoldTopValue and RtopAddr are dead, so can use G1 and G3
3384    __ incr_allocated_bytes(Roffset, G1_scratch, G3_scratch);
3385  }
3386
3387  if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3388    // clear object fields
3389    __ bind(initialize_object);
3390    __ deccc(Roffset, sizeof(oopDesc));
3391    __ br(Assembler::zero, false, Assembler::pt, initialize_header);
3392    __ delayed()->add(RallocatedObject, sizeof(oopDesc), G3_scratch);
3393
3394    // initialize remaining object fields
3395    if (UseBlockZeroing) {
3396      // Use BIS for zeroing
3397      __ bis_zeroing(G3_scratch, Roffset, G1_scratch, initialize_header);
3398    } else {
3399      Label loop;
3400      __ subcc(Roffset, wordSize, Roffset);
3401      __ bind(loop);
3402      //__ subcc(Roffset, wordSize, Roffset);      // executed above loop or in delay slot
3403      __ st_ptr(G0, G3_scratch, Roffset);
3404      __ br(Assembler::notEqual, false, Assembler::pt, loop);
3405      __ delayed()->subcc(Roffset, wordSize, Roffset);
3406    }
3407    __ ba_short(initialize_header);
3408  }
3409
3410  // slow case
3411  __ bind(slow_case);
3412  __ get_2_byte_integer_at_bcp(1, G3_scratch, O2, InterpreterMacroAssembler::Unsigned);
3413  __ get_constant_pool(O1);
3414
3415  call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), O1, O2);
3416
3417  __ ba_short(done);
3418
3419  // Initialize the header: mark, klass
3420  __ bind(initialize_header);
3421
3422  if (UseBiasedLocking) {
3423    __ ld_ptr(RinstanceKlass, in_bytes(Klass::prototype_header_offset()), G4_scratch);
3424  } else {
3425    __ set((intptr_t)markOopDesc::prototype(), G4_scratch);
3426  }
3427  __ st_ptr(G4_scratch, RallocatedObject, oopDesc::mark_offset_in_bytes());       // mark
3428  __ store_klass_gap(G0, RallocatedObject);         // klass gap if compressed
3429  __ store_klass(RinstanceKlass, RallocatedObject); // klass (last for cms)
3430
3431  {
3432    SkipIfEqual skip_if(
3433      _masm, G4_scratch, &DTraceAllocProbes, Assembler::zero);
3434    // Trigger dtrace event
3435    __ push(atos);
3436    __ call_VM_leaf(noreg,
3437       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), O0);
3438    __ pop(atos);
3439  }
3440
3441  // continue
3442  __ bind(done);
3443}
3444
3445
3446
3447void TemplateTable::newarray() {
3448  transition(itos, atos);
3449  __ ldub(Lbcp, 1, O1);
3450     call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), O1, Otos_i);
3451}
3452
3453
3454void TemplateTable::anewarray() {
3455  transition(itos, atos);
3456  __ get_constant_pool(O1);
3457  __ get_2_byte_integer_at_bcp(1, G4_scratch, O2, InterpreterMacroAssembler::Unsigned);
3458     call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), O1, O2, Otos_i);
3459}
3460
3461
3462void TemplateTable::arraylength() {
3463  transition(atos, itos);
3464  Label ok;
3465  __ verify_oop(Otos_i);
3466  __ tst(Otos_i);
3467  __ throw_if_not_1_x( Assembler::notZero, ok );
3468  __ delayed()->ld(Otos_i, arrayOopDesc::length_offset_in_bytes(), Otos_i);
3469  __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3470}
3471
3472
3473void TemplateTable::checkcast() {
3474  transition(atos, atos);
3475  Label done, is_null, quicked, cast_ok, resolved;
3476  Register Roffset = G1_scratch;
3477  Register RobjKlass = O5;
3478  Register RspecifiedKlass = O4;
3479
3480  // Check for casting a NULL
3481  __ br_null_short(Otos_i, Assembler::pn, is_null);
3482
3483  // Get value klass in RobjKlass
3484  __ load_klass(Otos_i, RobjKlass); // get value klass
3485
3486  // Get constant pool tag
3487  __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3488
3489  // See if the checkcast has been quickened
3490  __ get_cpool_and_tags(Lscratch, G3_scratch);
3491  __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3492  __ ldub(G3_scratch, Roffset, G3_scratch);
3493  __ cmp(G3_scratch, JVM_CONSTANT_Class);
3494  __ br(Assembler::equal, true, Assembler::pt, quicked);
3495  __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3496
3497  __ push_ptr(); // save receiver for result, and for GC
3498  call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3499  __ get_vm_result_2(RspecifiedKlass);
3500  __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3501
3502  __ ba_short(resolved);
3503
3504  // Extract target class from constant pool
3505  __ bind(quicked);
3506  __ add(Roffset, sizeof(ConstantPool), Roffset);
3507  __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
3508  __ bind(resolved);
3509  __ load_klass(Otos_i, RobjKlass); // get value klass
3510
3511  // Generate a fast subtype check.  Branch to cast_ok if no
3512  // failure.  Throw exception if failure.
3513  __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, cast_ok );
3514
3515  // Not a subtype; so must throw exception
3516  __ throw_if_not_x( Assembler::never, Interpreter::_throw_ClassCastException_entry, G3_scratch );
3517
3518  __ bind(cast_ok);
3519
3520  if (ProfileInterpreter) {
3521    __ ba_short(done);
3522  }
3523  __ bind(is_null);
3524  __ profile_null_seen(G3_scratch);
3525  __ bind(done);
3526}
3527
3528
3529void TemplateTable::instanceof() {
3530  Label done, is_null, quicked, resolved;
3531  transition(atos, itos);
3532  Register Roffset = G1_scratch;
3533  Register RobjKlass = O5;
3534  Register RspecifiedKlass = O4;
3535
3536  // Check for casting a NULL
3537  __ br_null_short(Otos_i, Assembler::pt, is_null);
3538
3539  // Get value klass in RobjKlass
3540  __ load_klass(Otos_i, RobjKlass); // get value klass
3541
3542  // Get constant pool tag
3543  __ get_2_byte_integer_at_bcp(1, Lscratch, Roffset, InterpreterMacroAssembler::Unsigned);
3544
3545  // See if the checkcast has been quickened
3546  __ get_cpool_and_tags(Lscratch, G3_scratch);
3547  __ add(G3_scratch, Array<u1>::base_offset_in_bytes(), G3_scratch);
3548  __ ldub(G3_scratch, Roffset, G3_scratch);
3549  __ cmp(G3_scratch, JVM_CONSTANT_Class);
3550  __ br(Assembler::equal, true, Assembler::pt, quicked);
3551  __ delayed()->sll(Roffset, LogBytesPerWord, Roffset);
3552
3553  __ push_ptr(); // save receiver for result, and for GC
3554  call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3555  __ get_vm_result_2(RspecifiedKlass);
3556  __ pop_ptr(Otos_i, G3_scratch); // restore receiver
3557
3558  __ ba_short(resolved);
3559
3560  // Extract target class from constant pool
3561  __ bind(quicked);
3562  __ add(Roffset, sizeof(ConstantPool), Roffset);
3563  __ get_constant_pool(Lscratch);
3564  __ ld_ptr(Lscratch, Roffset, RspecifiedKlass);
3565  __ bind(resolved);
3566  __ load_klass(Otos_i, RobjKlass); // get value klass
3567
3568  // Generate a fast subtype check.  Branch to cast_ok if no
3569  // failure.  Return 0 if failure.
3570  __ or3(G0, 1, Otos_i);      // set result assuming quick tests succeed
3571  __ gen_subtype_check( RobjKlass, RspecifiedKlass, G3_scratch, G4_scratch, G1_scratch, done );
3572  // Not a subtype; return 0;
3573  __ clr( Otos_i );
3574
3575  if (ProfileInterpreter) {
3576    __ ba_short(done);
3577  }
3578  __ bind(is_null);
3579  __ profile_null_seen(G3_scratch);
3580  __ bind(done);
3581}
3582
3583void TemplateTable::_breakpoint() {
3584
3585   // Note: We get here even if we are single stepping..
3586   // jbug inists on setting breakpoints at every bytecode
3587   // even if we are in single step mode.
3588
3589   transition(vtos, vtos);
3590   // get the unpatched byte code
3591   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), Lmethod, Lbcp);
3592   __ mov(O0, Lbyte_code);
3593
3594   // post the breakpoint event
3595   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), Lmethod, Lbcp);
3596
3597   // complete the execution of original bytecode
3598   __ dispatch_normal(vtos);
3599}
3600
3601
3602//----------------------------------------------------------------------------------------------------
3603// Exceptions
3604
3605void TemplateTable::athrow() {
3606  transition(atos, vtos);
3607
3608  // This works because exception is cached in Otos_i which is same as O0,
3609  // which is same as what throw_exception_entry_expects
3610  assert(Otos_i == Oexception, "see explanation above");
3611
3612  __ verify_oop(Otos_i);
3613  __ null_check(Otos_i);
3614  __ throw_if_not_x(Assembler::never, Interpreter::throw_exception_entry(), G3_scratch);
3615}
3616
3617
3618//----------------------------------------------------------------------------------------------------
3619// Synchronization
3620
3621
3622// See frame_sparc.hpp for monitor block layout.
3623// Monitor elements are dynamically allocated by growing stack as needed.
3624
3625void TemplateTable::monitorenter() {
3626  transition(atos, vtos);
3627  __ verify_oop(Otos_i);
3628  // Try to acquire a lock on the object
3629  // Repeat until succeeded (i.e., until
3630  // monitorenter returns true).
3631
3632  {   Label ok;
3633    __ tst(Otos_i);
3634    __ throw_if_not_1_x( Assembler::notZero,  ok);
3635    __ delayed()->mov(Otos_i, Lscratch); // save obj
3636    __ throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ok);
3637  }
3638
3639  assert(O0 == Otos_i, "Be sure where the object to lock is");
3640
3641  // find a free slot in the monitor block
3642
3643
3644  // initialize entry pointer
3645  __ clr(O1); // points to free slot or NULL
3646
3647  {
3648    Label entry, loop, exit;
3649    __ add( __ top_most_monitor(), O2 ); // last one to check
3650    __ ba( entry );
3651    __ delayed()->mov( Lmonitors, O3 ); // first one to check
3652
3653
3654    __ bind( loop );
3655
3656    __ verify_oop(O4);          // verify each monitor's oop
3657    __ tst(O4); // is this entry unused?
3658    __ movcc( Assembler::zero, false, Assembler::ptr_cc, O3, O1);
3659
3660    __ cmp(O4, O0); // check if current entry is for same object
3661    __ brx( Assembler::equal, false, Assembler::pn, exit );
3662    __ delayed()->inc( O3, frame::interpreter_frame_monitor_size() * wordSize ); // check next one
3663
3664    __ bind( entry );
3665
3666    __ cmp( O3, O2 );
3667    __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3668    __ delayed()->ld_ptr(O3, BasicObjectLock::obj_offset_in_bytes(), O4);
3669
3670    __ bind( exit );
3671  }
3672
3673  { Label allocated;
3674
3675    // found free slot?
3676    __ br_notnull_short(O1, Assembler::pn, allocated);
3677
3678    __ add_monitor_to_stack( false, O2, O3 );
3679    __ mov(Lmonitors, O1);
3680
3681    __ bind(allocated);
3682  }
3683
3684  // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3685  // The object has already been poped from the stack, so the expression stack looks correct.
3686  __ inc(Lbcp);
3687
3688  __ st_ptr(O0, O1, BasicObjectLock::obj_offset_in_bytes()); // store object
3689  __ lock_object(O1, O0);
3690
3691  // check if there's enough space on the stack for the monitors after locking
3692  __ generate_stack_overflow_check(0);
3693
3694  // The bcp has already been incremented. Just need to dispatch to next instruction.
3695  __ dispatch_next(vtos);
3696}
3697
3698
3699void TemplateTable::monitorexit() {
3700  transition(atos, vtos);
3701  __ verify_oop(Otos_i);
3702  __ tst(Otos_i);
3703  __ throw_if_not_x( Assembler::notZero, Interpreter::_throw_NullPointerException_entry, G3_scratch );
3704
3705  assert(O0 == Otos_i, "just checking");
3706
3707  { Label entry, loop, found;
3708    __ add( __ top_most_monitor(), O2 ); // last one to check
3709    __ ba(entry);
3710    // use Lscratch to hold monitor elem to check, start with most recent monitor,
3711    // By using a local it survives the call to the C routine.
3712    __ delayed()->mov( Lmonitors, Lscratch );
3713
3714    __ bind( loop );
3715
3716    __ verify_oop(O4);          // verify each monitor's oop
3717    __ cmp(O4, O0); // check if current entry is for desired object
3718    __ brx( Assembler::equal, true, Assembler::pt, found );
3719    __ delayed()->mov(Lscratch, O1); // pass found entry as argument to monitorexit
3720
3721    __ inc( Lscratch, frame::interpreter_frame_monitor_size() * wordSize ); // advance to next
3722
3723    __ bind( entry );
3724
3725    __ cmp( Lscratch, O2 );
3726    __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, loop );
3727    __ delayed()->ld_ptr(Lscratch, BasicObjectLock::obj_offset_in_bytes(), O4);
3728
3729    call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3730    __ should_not_reach_here();
3731
3732    __ bind(found);
3733  }
3734  __ unlock_object(O1);
3735}
3736
3737
3738//----------------------------------------------------------------------------------------------------
3739// Wide instructions
3740
3741void TemplateTable::wide() {
3742  transition(vtos, vtos);
3743  __ ldub(Lbcp, 1, G3_scratch);// get next bc
3744  __ sll(G3_scratch, LogBytesPerWord, G3_scratch);
3745  AddressLiteral ep(Interpreter::_wentry_point);
3746  __ set(ep, G4_scratch);
3747  __ ld_ptr(G4_scratch, G3_scratch, G3_scratch);
3748  __ jmp(G3_scratch, G0);
3749  __ delayed()->nop();
3750  // Note: the Lbcp increment step is part of the individual wide bytecode implementations
3751}
3752
3753
3754//----------------------------------------------------------------------------------------------------
3755// Multi arrays
3756
3757void TemplateTable::multianewarray() {
3758  transition(vtos, atos);
3759     // put ndims * wordSize into Lscratch
3760  __ ldub( Lbcp,     3,               Lscratch);
3761  __ sll(  Lscratch, Interpreter::logStackElementSize, Lscratch);
3762     // Lesp points past last_dim, so set to O1 to first_dim address
3763  __ add(  Lesp,     Lscratch,        O1);
3764     call_VM(Otos_i, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), O1);
3765  __ add(  Lesp,     Lscratch,        Lesp); // pop all dimensions off the stack
3766}
3767#endif /* !CC_INTERP */
3768