1 /*
   2  * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "interp_masm_sparc.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "oops/arrayOop.hpp"
  30 #include "oops/markOop.hpp"
  31 #include "oops/methodData.hpp"
  32 #include "oops/method.hpp"
  33 #include "oops/methodCounters.hpp"
  34 #include "prims/jvmtiExport.hpp"
  35 #include "prims/jvmtiRedefineClassesTrace.hpp"
  36 #include "prims/jvmtiThreadState.hpp"
  37 #include "runtime/basicLock.hpp"
  38 #include "runtime/biasedLocking.hpp"
  39 #include "runtime/sharedRuntime.hpp"
  40 #include "runtime/thread.inline.hpp"
  41 
  42 #ifndef CC_INTERP
  43 #ifndef FAST_DISPATCH
  44 #define FAST_DISPATCH 1
  45 #endif
  46 #undef FAST_DISPATCH
  47 
  48 // Implementation of InterpreterMacroAssembler
  49 
  50 // This file specializes the assember with interpreter-specific macros
  51 
  52 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS);
  53 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS);
  54 
  55 #else // CC_INTERP
  56 #ifndef STATE
  57 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name))
  58 #endif // STATE
  59 
  60 #endif // CC_INTERP
  61 
  62 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  63   assert(entry, "Entry must have been generated by now");
  64   AddressLiteral al(entry);
  65   jump_to(al, G3_scratch);
  66   delayed()->nop();
  67 }
  68 
  69 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) {
  70   // Note: this algorithm is also used by C1's OSR entry sequence.
  71   // Any changes should also be applied to CodeEmitter::emit_osr_entry().
  72   assert_different_registers(args_size, locals_size);
  73   // max_locals*2 for TAGS.  Assumes that args_size has already been adjusted.
  74   subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words
  75   // Use br/mov combination because it works on both V8 and V9 and is
  76   // faster.
  77   Label skip_move;
  78   br(Assembler::negative, true, Assembler::pt, skip_move);
  79   delayed()->mov(G0, delta);
  80   bind(skip_move);
  81   round_to(delta, WordsPerLong);       // make multiple of 2 (SP must be 2-word aligned)
  82   sll(delta, LogBytesPerWord, delta);  // extra space for locals in bytes
  83 }
  84 
  85 #ifndef CC_INTERP
  86 
  87 // Dispatch code executed in the prolog of a bytecode which does not do it's
  88 // own dispatch. The dispatch address is computed and placed in IdispatchAddress
  89 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
  90   assert_not_delayed();
  91 #ifdef FAST_DISPATCH
  92   // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
  93   // they both use I2.
  94   assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
  95   ldub(Lbcp, bcp_incr, Lbyte_code);                     // load next bytecode
  96   add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
  97                                                         // add offset to correct dispatch table
  98   sll(Lbyte_code, LogBytesPerWord, Lbyte_code);         // multiply by wordSize
  99   ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr
 100 #else
 101   ldub( Lbcp, bcp_incr, Lbyte_code);                    // load next bytecode
 102   // dispatch table to use
 103   AddressLiteral tbl(Interpreter::dispatch_table(state));
 104   sll(Lbyte_code, LogBytesPerWord, Lbyte_code);         // multiply by wordSize
 105   set(tbl, G3_scratch);                                 // compute addr of table
 106   ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress);     // get entry addr
 107 #endif
 108 }
 109 
 110 
 111 // Dispatch code executed in the epilog of a bytecode which does not do it's
 112 // own dispatch. The dispatch address in IdispatchAddress is used for the
 113 // dispatch.
 114 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
 115   assert_not_delayed();
 116   verify_FPU(1, state);
 117   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 118   jmp( IdispatchAddress, 0 );
 119   if (bcp_incr != 0)  delayed()->inc(Lbcp, bcp_incr);
 120   else                delayed()->nop();
 121 }
 122 
 123 
 124 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) {
 125   // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
 126   assert_not_delayed();
 127   ldub( Lbcp, bcp_incr, Lbyte_code);               // load next bytecode
 128   dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr);
 129 }
 130 
 131 
 132 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) {
 133   // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
 134   assert_not_delayed();
 135   ldub( Lbcp, bcp_incr, Lbyte_code);               // load next bytecode
 136   dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false);
 137 }
 138 
 139 
 140 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 141   // load current bytecode
 142   assert_not_delayed();
 143   ldub( Lbcp, 0, Lbyte_code);               // load next bytecode
 144   dispatch_base(state, table);
 145 }
 146 
 147 
 148 void InterpreterMacroAssembler::call_VM_leaf_base(
 149   Register java_thread,
 150   address  entry_point,
 151   int      number_of_arguments
 152 ) {
 153   if (!java_thread->is_valid())
 154     java_thread = L7_thread_cache;
 155   // super call
 156   MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments);
 157 }
 158 
 159 
 160 void InterpreterMacroAssembler::call_VM_base(
 161   Register        oop_result,
 162   Register        java_thread,
 163   Register        last_java_sp,
 164   address         entry_point,
 165   int             number_of_arguments,
 166   bool            check_exception
 167 ) {
 168   if (!java_thread->is_valid())
 169     java_thread = L7_thread_cache;
 170   // See class ThreadInVMfromInterpreter, which assumes that the interpreter
 171   // takes responsibility for setting its own thread-state on call-out.
 172   // However, ThreadInVMfromInterpreter resets the state to "in_Java".
 173 
 174   //save_bcp();                                  // save bcp
 175   MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception);
 176   //restore_bcp();                               // restore bcp
 177   //restore_locals();                            // restore locals pointer
 178 }
 179 
 180 
 181 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
 182   if (JvmtiExport::can_pop_frame()) {
 183     Label L;
 184 
 185     // Check the "pending popframe condition" flag in the current thread
 186     ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg);
 187 
 188     // Initiate popframe handling only if it is not already being processed.  If the flag
 189     // has the popframe_processing bit set, it means that this code is called *during* popframe
 190     // handling - we don't want to reenter.
 191     btst(JavaThread::popframe_pending_bit, scratch_reg);
 192     br(zero, false, pt, L);
 193     delayed()->nop();
 194     btst(JavaThread::popframe_processing_bit, scratch_reg);
 195     br(notZero, false, pt, L);
 196     delayed()->nop();
 197 
 198     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 199     // address of the same-named entrypoint in the generated interpreter code.
 200     call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 201 
 202     // Jump to Interpreter::_remove_activation_preserving_args_entry
 203     jmpl(O0, G0, G0);
 204     delayed()->nop();
 205     bind(L);
 206   }
 207 }
 208 
 209 
 210 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 211   Register thr_state = G4_scratch;
 212   ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
 213   const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset());
 214   const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset());
 215   const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset());
 216   switch (state) {
 217   case ltos: ld_long(val_addr, Otos_l);                   break;
 218   case atos: ld_ptr(oop_addr, Otos_l);
 219              st_ptr(G0, oop_addr);                        break;
 220   case btos:                                           // fall through
 221   case ctos:                                           // fall through
 222   case stos:                                           // fall through
 223   case itos: ld(val_addr, Otos_l1);                       break;
 224   case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break;
 225   case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break;
 226   case vtos: /* nothing to do */                          break;
 227   default  : ShouldNotReachHere();
 228   }
 229   // Clean up tos value in the jvmti thread state
 230   or3(G0, ilgl, G3_scratch);
 231   stw(G3_scratch, tos_addr);
 232   st_long(G0, val_addr);
 233   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 234 }
 235 
 236 
 237 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
 238   if (JvmtiExport::can_force_early_return()) {
 239     Label L;
 240     Register thr_state = G3_scratch;
 241     ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
 242     br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit;
 243 
 244     // Initiate earlyret handling only if it is not already being processed.
 245     // If the flag has the earlyret_processing bit set, it means that this code
 246     // is called *during* earlyret handling - we don't want to reenter.
 247     ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch);
 248     cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L);
 249 
 250     // Call Interpreter::remove_activation_early_entry() to get the address of the
 251     // same-named entrypoint in the generated interpreter code
 252     ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1);
 253     call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1);
 254 
 255     // Jump to Interpreter::_remove_activation_early_entry
 256     jmpl(O0, G0, G0);
 257     delayed()->nop();
 258     bind(L);
 259   }
 260 }
 261 
 262 
 263 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) {
 264   mov(arg_1, O0);
 265   mov(arg_2, O1);
 266   MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2);
 267 }
 268 #endif /* CC_INTERP */
 269 
 270 
 271 #ifndef CC_INTERP
 272 
 273 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) {
 274   assert_not_delayed();
 275   dispatch_Lbyte_code(state, table);
 276 }
 277 
 278 
 279 void InterpreterMacroAssembler::dispatch_normal(TosState state) {
 280   dispatch_base(state, Interpreter::normal_table(state));
 281 }
 282 
 283 
 284 void InterpreterMacroAssembler::dispatch_only(TosState state) {
 285   dispatch_base(state, Interpreter::dispatch_table(state));
 286 }
 287 
 288 
 289 // common code to dispatch and dispatch_only
 290 // dispatch value in Lbyte_code and increment Lbcp
 291 
 292 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) {
 293   verify_FPU(1, state);
 294   // %%%%% maybe implement +VerifyActivationFrameSize here
 295   //verify_thread(); //too slow; we will just verify on method entry & exit
 296   if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 297 #ifdef FAST_DISPATCH
 298   if (table == Interpreter::dispatch_table(state)) {
 299     // use IdispatchTables
 300     add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
 301                                                         // add offset to correct dispatch table
 302     sll(Lbyte_code, LogBytesPerWord, Lbyte_code);       // multiply by wordSize
 303     ld_ptr(IdispatchTables, Lbyte_code, G3_scratch);    // get entry addr
 304   } else {
 305 #endif
 306     // dispatch table to use
 307     AddressLiteral tbl(table);
 308     sll(Lbyte_code, LogBytesPerWord, Lbyte_code);       // multiply by wordSize
 309     set(tbl, G3_scratch);                               // compute addr of table
 310     ld_ptr(G3_scratch, Lbyte_code, G3_scratch);         // get entry addr
 311 #ifdef FAST_DISPATCH
 312   }
 313 #endif
 314   jmp( G3_scratch, 0 );
 315   if (bcp_incr != 0)  delayed()->inc(Lbcp, bcp_incr);
 316   else                delayed()->nop();
 317 }
 318 
 319 
 320 // Helpers for expression stack
 321 
 322 // Longs and doubles are Category 2 computational types in the
 323 // JVM specification (section 3.11.1) and take 2 expression stack or
 324 // local slots.
 325 // Aligning them on 32 bit with tagged stacks is hard because the code generated
 326 // for the dup* bytecodes depends on what types are already on the stack.
 327 // If the types are split into the two stack/local slots, that is much easier
 328 // (and we can use 0 for non-reference tags).
 329 
 330 // Known good alignment in _LP64 but unknown otherwise
 331 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) {
 332   assert_not_delayed();
 333 
 334 #ifdef _LP64
 335   ldf(FloatRegisterImpl::D, r1, offset, d);
 336 #else
 337   ldf(FloatRegisterImpl::S, r1, offset, d);
 338   ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor());
 339 #endif
 340 }
 341 
 342 // Known good alignment in _LP64 but unknown otherwise
 343 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) {
 344   assert_not_delayed();
 345 
 346 #ifdef _LP64
 347   stf(FloatRegisterImpl::D, d, r1, offset);
 348   // store something more useful here
 349   debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
 350 #else
 351   stf(FloatRegisterImpl::S, d, r1, offset);
 352   stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize);
 353 #endif
 354 }
 355 
 356 
 357 // Known good alignment in _LP64 but unknown otherwise
 358 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) {
 359   assert_not_delayed();
 360 #ifdef _LP64
 361   ldx(r1, offset, rd);
 362 #else
 363   ld(r1, offset, rd);
 364   ld(r1, offset + Interpreter::stackElementSize, rd->successor());
 365 #endif
 366 }
 367 
 368 // Known good alignment in _LP64 but unknown otherwise
 369 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) {
 370   assert_not_delayed();
 371 
 372 #ifdef _LP64
 373   stx(l, r1, offset);
 374   // store something more useful here
 375   debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
 376 #else
 377   st(l, r1, offset);
 378   st(l->successor(), r1, offset + Interpreter::stackElementSize);
 379 #endif
 380 }
 381 
 382 void InterpreterMacroAssembler::pop_i(Register r) {
 383   assert_not_delayed();
 384   ld(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 385   inc(Lesp, Interpreter::stackElementSize);
 386   debug_only(verify_esp(Lesp));
 387 }
 388 
 389 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) {
 390   assert_not_delayed();
 391   ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 392   inc(Lesp, Interpreter::stackElementSize);
 393   debug_only(verify_esp(Lesp));
 394 }
 395 
 396 void InterpreterMacroAssembler::pop_l(Register r) {
 397   assert_not_delayed();
 398   load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r);
 399   inc(Lesp, 2*Interpreter::stackElementSize);
 400   debug_only(verify_esp(Lesp));
 401 }
 402 
 403 
 404 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) {
 405   assert_not_delayed();
 406   ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f);
 407   inc(Lesp, Interpreter::stackElementSize);
 408   debug_only(verify_esp(Lesp));
 409 }
 410 
 411 
 412 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) {
 413   assert_not_delayed();
 414   load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f);
 415   inc(Lesp, 2*Interpreter::stackElementSize);
 416   debug_only(verify_esp(Lesp));
 417 }
 418 
 419 
 420 void InterpreterMacroAssembler::push_i(Register r) {
 421   assert_not_delayed();
 422   debug_only(verify_esp(Lesp));
 423   st(r, Lesp, 0);
 424   dec(Lesp, Interpreter::stackElementSize);
 425 }
 426 
 427 void InterpreterMacroAssembler::push_ptr(Register r) {
 428   assert_not_delayed();
 429   st_ptr(r, Lesp, 0);
 430   dec(Lesp, Interpreter::stackElementSize);
 431 }
 432 
 433 // remember: our convention for longs in SPARC is:
 434 // O0 (Otos_l1) has high-order part in first word,
 435 // O1 (Otos_l2) has low-order part in second word
 436 
 437 void InterpreterMacroAssembler::push_l(Register r) {
 438   assert_not_delayed();
 439   debug_only(verify_esp(Lesp));
 440   // Longs are stored in memory-correct order, even if unaligned.
 441   int offset = -Interpreter::stackElementSize;
 442   store_unaligned_long(r, Lesp, offset);
 443   dec(Lesp, 2 * Interpreter::stackElementSize);
 444 }
 445 
 446 
 447 void InterpreterMacroAssembler::push_f(FloatRegister f) {
 448   assert_not_delayed();
 449   debug_only(verify_esp(Lesp));
 450   stf(FloatRegisterImpl::S, f, Lesp, 0);
 451   dec(Lesp, Interpreter::stackElementSize);
 452 }
 453 
 454 
 455 void InterpreterMacroAssembler::push_d(FloatRegister d)   {
 456   assert_not_delayed();
 457   debug_only(verify_esp(Lesp));
 458   // Longs are stored in memory-correct order, even if unaligned.
 459   int offset = -Interpreter::stackElementSize;
 460   store_unaligned_double(d, Lesp, offset);
 461   dec(Lesp, 2 * Interpreter::stackElementSize);
 462 }
 463 
 464 
 465 void InterpreterMacroAssembler::push(TosState state) {
 466   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 467   switch (state) {
 468     case atos: push_ptr();            break;
 469     case btos: push_i();              break;
 470     case ctos:
 471     case stos: push_i();              break;
 472     case itos: push_i();              break;
 473     case ltos: push_l();              break;
 474     case ftos: push_f();              break;
 475     case dtos: push_d();              break;
 476     case vtos: /* nothing to do */    break;
 477     default  : ShouldNotReachHere();
 478   }
 479 }
 480 
 481 
 482 void InterpreterMacroAssembler::pop(TosState state) {
 483   switch (state) {
 484     case atos: pop_ptr();            break;
 485     case btos: pop_i();              break;
 486     case ctos:
 487     case stos: pop_i();              break;
 488     case itos: pop_i();              break;
 489     case ltos: pop_l();              break;
 490     case ftos: pop_f();              break;
 491     case dtos: pop_d();              break;
 492     case vtos: /* nothing to do */   break;
 493     default  : ShouldNotReachHere();
 494   }
 495   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
 496 }
 497 
 498 
 499 // Helpers for swap and dup
 500 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 501   ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val);
 502 }
 503 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 504   st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n));
 505 }
 506 
 507 
 508 void InterpreterMacroAssembler::load_receiver(Register param_count,
 509                                               Register recv) {
 510   sll(param_count, Interpreter::logStackElementSize, param_count);
 511   ld_ptr(Lesp, param_count, recv);  // gets receiver oop
 512 }
 513 
 514 void InterpreterMacroAssembler::empty_expression_stack() {
 515   // Reset Lesp.
 516   sub( Lmonitors, wordSize, Lesp );
 517 
 518   // Reset SP by subtracting more space from Lesp.
 519   Label done;
 520   assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!");
 521 
 522   // A native does not need to do this, since its callee does not change SP.
 523   ld(Lmethod, Method::access_flags_offset(), Gframe_size);  // Load access flags.
 524   btst(JVM_ACC_NATIVE, Gframe_size);
 525   br(Assembler::notZero, false, Assembler::pt, done);
 526   delayed()->nop();
 527 
 528   // Compute max expression stack+register save area
 529   ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size);
 530   lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size);  // Load max stack.
 531   add(Gframe_size, frame::memory_parameter_word_sp_offset+Method::extra_stack_entries(), Gframe_size );
 532 
 533   //
 534   // now set up a stack frame with the size computed above
 535   //
 536   //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below
 537   sll( Gframe_size, LogBytesPerWord, Gframe_size );
 538   sub( Lesp, Gframe_size, Gframe_size );
 539   and3( Gframe_size, -(2 * wordSize), Gframe_size );          // align SP (downwards) to an 8/16-byte boundary
 540   debug_only(verify_sp(Gframe_size, G4_scratch));
 541 #ifdef _LP64
 542   sub(Gframe_size, STACK_BIAS, Gframe_size );
 543 #endif
 544   mov(Gframe_size, SP);
 545 
 546   bind(done);
 547 }
 548 
 549 
 550 #ifdef ASSERT
 551 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) {
 552   Label Bad, OK;
 553 
 554   // Saved SP must be aligned.
 555 #ifdef _LP64
 556   btst(2*BytesPerWord-1, Rsp);
 557 #else
 558   btst(LongAlignmentMask, Rsp);
 559 #endif
 560   br(Assembler::notZero, false, Assembler::pn, Bad);
 561   delayed()->nop();
 562 
 563   // Saved SP, plus register window size, must not be above FP.
 564   add(Rsp, frame::register_save_words * wordSize, Rtemp);
 565 #ifdef _LP64
 566   sub(Rtemp, STACK_BIAS, Rtemp);  // Bias Rtemp before cmp to FP
 567 #endif
 568   cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad);
 569 
 570   // Saved SP must not be ridiculously below current SP.
 571   size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K);
 572   set(maxstack, Rtemp);
 573   sub(SP, Rtemp, Rtemp);
 574 #ifdef _LP64
 575   add(Rtemp, STACK_BIAS, Rtemp);  // Unbias Rtemp before cmp to Rsp
 576 #endif
 577   cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad);
 578 
 579   ba_short(OK);
 580 
 581   bind(Bad);
 582   stop("on return to interpreted call, restored SP is corrupted");
 583 
 584   bind(OK);
 585 }
 586 
 587 
 588 void InterpreterMacroAssembler::verify_esp(Register Resp) {
 589   // about to read or write Resp[0]
 590   // make sure it is not in the monitors or the register save area
 591   Label OK1, OK2;
 592 
 593   cmp(Resp, Lmonitors);
 594   brx(Assembler::lessUnsigned, true, Assembler::pt, OK1);
 595   delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp);
 596   stop("too many pops:  Lesp points into monitor area");
 597   bind(OK1);
 598 #ifdef _LP64
 599   sub(Resp, STACK_BIAS, Resp);
 600 #endif
 601   cmp(Resp, SP);
 602   brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2);
 603   delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp);
 604   stop("too many pushes:  Lesp points into register window");
 605   bind(OK2);
 606 }
 607 #endif // ASSERT
 608 
 609 // Load compiled (i2c) or interpreter entry when calling from interpreted and
 610 // do the call. Centralized so that all interpreter calls will do the same actions.
 611 // If jvmti single stepping is on for a thread we must not call compiled code.
 612 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) {
 613 
 614   // Assume we want to go compiled if available
 615 
 616   ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target);
 617 
 618   if (JvmtiExport::can_post_interpreter_events()) {
 619     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 620     // compiled code in threads for which the event is enabled.  Check here for
 621     // interp_only_mode if these events CAN be enabled.
 622     verify_thread();
 623     Label skip_compiled_code;
 624 
 625     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
 626     ld(interp_only, scratch);
 627     cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn);
 628     delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target);
 629     bind(skip_compiled_code);
 630   }
 631 
 632   // the i2c_adapters need Method* in G5_method (right? %%%)
 633   // do the call
 634 #ifdef ASSERT
 635   {
 636     Label ok;
 637     br_notnull_short(target, Assembler::pt, ok);
 638     stop("null entry point");
 639     bind(ok);
 640   }
 641 #endif // ASSERT
 642 
 643   // Adjust Rret first so Llast_SP can be same as Rret
 644   add(Rret, -frame::pc_return_offset, O7);
 645   add(Lesp, BytesPerWord, Gargs); // setup parameter pointer
 646   // Record SP so we can remove any stack space allocated by adapter transition
 647   jmp(target, 0);
 648   delayed()->mov(SP, Llast_SP);
 649 }
 650 
 651 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) {
 652   assert_not_delayed();
 653 
 654   Label not_taken;
 655   if (ptr_compare) brx(cc, false, Assembler::pn, not_taken);
 656   else             br (cc, false, Assembler::pn, not_taken);
 657   delayed()->nop();
 658 
 659   TemplateTable::branch(false,false);
 660 
 661   bind(not_taken);
 662 
 663   profile_not_taken_branch(G3_scratch);
 664 }
 665 
 666 
 667 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(
 668                                   int         bcp_offset,
 669                                   Register    Rtmp,
 670                                   Register    Rdst,
 671                                   signedOrNot is_signed,
 672                                   setCCOrNot  should_set_CC ) {
 673   assert(Rtmp != Rdst, "need separate temp register");
 674   assert_not_delayed();
 675   switch (is_signed) {
 676    default: ShouldNotReachHere();
 677 
 678    case   Signed:  ldsb( Lbcp, bcp_offset, Rdst  );  break; // high byte
 679    case Unsigned:  ldub( Lbcp, bcp_offset, Rdst  );  break; // high byte
 680   }
 681   ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte
 682   sll( Rdst, BitsPerByte, Rdst);
 683   switch (should_set_CC ) {
 684    default: ShouldNotReachHere();
 685 
 686    case      set_CC:  orcc( Rdst, Rtmp, Rdst ); break;
 687    case dont_set_CC:  or3(  Rdst, Rtmp, Rdst ); break;
 688   }
 689 }
 690 
 691 
 692 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(
 693                                   int        bcp_offset,
 694                                   Register   Rtmp,
 695                                   Register   Rdst,
 696                                   setCCOrNot should_set_CC ) {
 697   assert(Rtmp != Rdst, "need separate temp register");
 698   assert_not_delayed();
 699   add( Lbcp, bcp_offset, Rtmp);
 700   andcc( Rtmp, 3, G0);
 701   Label aligned;
 702   switch (should_set_CC ) {
 703    default: ShouldNotReachHere();
 704 
 705    case      set_CC: break;
 706    case dont_set_CC: break;
 707   }
 708 
 709   br(Assembler::zero, true, Assembler::pn, aligned);
 710 #ifdef _LP64
 711   delayed()->ldsw(Rtmp, 0, Rdst);
 712 #else
 713   delayed()->ld(Rtmp, 0, Rdst);
 714 #endif
 715 
 716   ldub(Lbcp, bcp_offset + 3, Rdst);
 717   ldub(Lbcp, bcp_offset + 2, Rtmp);  sll(Rtmp,  8, Rtmp);  or3(Rtmp, Rdst, Rdst);
 718   ldub(Lbcp, bcp_offset + 1, Rtmp);  sll(Rtmp, 16, Rtmp);  or3(Rtmp, Rdst, Rdst);
 719 #ifdef _LP64
 720   ldsb(Lbcp, bcp_offset + 0, Rtmp);  sll(Rtmp, 24, Rtmp);
 721 #else
 722   // Unsigned load is faster than signed on some implementations
 723   ldub(Lbcp, bcp_offset + 0, Rtmp);  sll(Rtmp, 24, Rtmp);
 724 #endif
 725   or3(Rtmp, Rdst, Rdst );
 726 
 727   bind(aligned);
 728   if (should_set_CC == set_CC) tst(Rdst);
 729 }
 730 
 731 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index,
 732                                                        int bcp_offset, size_t index_size) {
 733   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 734   if (index_size == sizeof(u2)) {
 735     get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned);
 736   } else if (index_size == sizeof(u4)) {
 737     get_4_byte_integer_at_bcp(bcp_offset, temp, index);
 738     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 739     xor3(index, -1, index);  // convert to plain index
 740   } else if (index_size == sizeof(u1)) {
 741     ldub(Lbcp, bcp_offset, index);
 742   } else {
 743     ShouldNotReachHere();
 744   }
 745 }
 746 
 747 
 748 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp,
 749                                                            int bcp_offset, size_t index_size) {
 750   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 751   assert_different_registers(cache, tmp);
 752   assert_not_delayed();
 753   get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size);
 754   // convert from field index to ConstantPoolCacheEntry index and from
 755   // word index to byte offset
 756   sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
 757   add(LcpoolCache, tmp, cache);
 758 }
 759 
 760 
 761 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 762                                                                         Register temp,
 763                                                                         Register bytecode,
 764                                                                         int byte_no,
 765                                                                         int bcp_offset,
 766                                                                         size_t index_size) {
 767   get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size);
 768   ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode);
 769   const int shift_count = (1 + byte_no) * BitsPerByte;
 770   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
 771          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
 772          "correct shift count");
 773   srl(bytecode, shift_count, bytecode);
 774   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
 775   and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode);
 776 }
 777 
 778 
 779 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
 780                                                                int bcp_offset, size_t index_size) {
 781   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 782   assert_different_registers(cache, tmp);
 783   assert_not_delayed();
 784   if (index_size == sizeof(u2)) {
 785     get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned);
 786   } else {
 787     ShouldNotReachHere();  // other sizes not supported here
 788   }
 789               // convert from field index to ConstantPoolCacheEntry index
 790               // and from word index to byte offset
 791   sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
 792               // skip past the header
 793   add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp);
 794               // construct pointer to cache entry
 795   add(LcpoolCache, tmp, cache);
 796 }
 797 
 798 
 799 // Load object from cpool->resolved_references(index)
 800 void InterpreterMacroAssembler::load_resolved_reference_at_index(
 801                                            Register result, Register index) {
 802   assert_different_registers(result, index);
 803   assert_not_delayed();
 804   // convert from field index to resolved_references() index and from
 805   // word index to byte offset. Since this is a java object, it can be compressed
 806   Register tmp = index;  // reuse
 807   sll(index, LogBytesPerHeapOop, tmp);
 808   get_constant_pool(result);
 809   // load pointer for resolved_references[] objArray
 810   ld_ptr(result, ConstantPool::resolved_references_offset_in_bytes(), result);
 811   // JNIHandles::resolve(result)
 812   ld_ptr(result, 0, result);
 813   // Add in the index
 814   add(result, tmp, result);
 815   load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
 816 }
 817 
 818 
 819 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
 820 // a subtype of super_klass.  Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
 821 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 822                                                   Register Rsuper_klass,
 823                                                   Register Rtmp1,
 824                                                   Register Rtmp2,
 825                                                   Register Rtmp3,
 826                                                   Label &ok_is_subtype ) {
 827   Label not_subtype;
 828 
 829   // Profile the not-null value's klass.
 830   profile_typecheck(Rsub_klass, Rtmp1);
 831 
 832   check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass,
 833                                 Rtmp1, Rtmp2,
 834                                 &ok_is_subtype, &not_subtype, NULL);
 835 
 836   check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass,
 837                                 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg,
 838                                 &ok_is_subtype, NULL);
 839 
 840   bind(not_subtype);
 841   profile_typecheck_failed(Rtmp1);
 842 }
 843 
 844 // Separate these two to allow for delay slot in middle
 845 // These are used to do a test and full jump to exception-throwing code.
 846 
 847 // %%%%% Could possibly reoptimize this by testing to see if could use
 848 // a single conditional branch (i.e. if span is small enough.
 849 // If you go that route, than get rid of the split and give up
 850 // on the delay-slot hack.
 851 
 852 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition,
 853                                                     Label&    ok ) {
 854   assert_not_delayed();
 855   br(ok_condition, true, pt, ok);
 856   // DELAY SLOT
 857 }
 858 
 859 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition,
 860                                                     Label&    ok ) {
 861   assert_not_delayed();
 862   bp( ok_condition, true, Assembler::xcc, pt, ok);
 863   // DELAY SLOT
 864 }
 865 
 866 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition,
 867                                                   Label&    ok ) {
 868   assert_not_delayed();
 869   brx(ok_condition, true, pt, ok);
 870   // DELAY SLOT
 871 }
 872 
 873 void InterpreterMacroAssembler::throw_if_not_2( address  throw_entry_point,
 874                                                 Register Rscratch,
 875                                                 Label&   ok ) {
 876   assert(throw_entry_point != NULL, "entry point must be generated by now");
 877   AddressLiteral dest(throw_entry_point);
 878   jump_to(dest, Rscratch);
 879   delayed()->nop();
 880   bind(ok);
 881 }
 882 
 883 
 884 // And if you cannot use the delay slot, here is a shorthand:
 885 
 886 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition,
 887                                                   address   throw_entry_point,
 888                                                   Register  Rscratch ) {
 889   Label ok;
 890   if (ok_condition != never) {
 891     throw_if_not_1_icc( ok_condition, ok);
 892     delayed()->nop();
 893   }
 894   throw_if_not_2( throw_entry_point, Rscratch, ok);
 895 }
 896 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition,
 897                                                   address   throw_entry_point,
 898                                                   Register  Rscratch ) {
 899   Label ok;
 900   if (ok_condition != never) {
 901     throw_if_not_1_xcc( ok_condition, ok);
 902     delayed()->nop();
 903   }
 904   throw_if_not_2( throw_entry_point, Rscratch, ok);
 905 }
 906 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition,
 907                                                 address   throw_entry_point,
 908                                                 Register  Rscratch ) {
 909   Label ok;
 910   if (ok_condition != never) {
 911     throw_if_not_1_x( ok_condition, ok);
 912     delayed()->nop();
 913   }
 914   throw_if_not_2( throw_entry_point, Rscratch, ok);
 915 }
 916 
 917 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res
 918 // Note: res is still shy of address by array offset into object.
 919 
 920 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) {
 921   assert_not_delayed();
 922 
 923   verify_oop(array);
 924 #ifdef _LP64
 925   // sign extend since tos (index) can be a 32bit value
 926   sra(index, G0, index);
 927 #endif // _LP64
 928 
 929   // check array
 930   Label ptr_ok;
 931   tst(array);
 932   throw_if_not_1_x( notZero, ptr_ok );
 933   delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index
 934   throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok);
 935 
 936   Label index_ok;
 937   cmp(index, tmp);
 938   throw_if_not_1_icc( lessUnsigned, index_ok );
 939   if (index_shift > 0)  delayed()->sll(index, index_shift, index);
 940   else                  delayed()->add(array, index, res); // addr - const offset in index
 941   // convention: move aberrant index into G3_scratch for exception message
 942   mov(index, G3_scratch);
 943   throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok);
 944 
 945   // add offset if didn't do it in delay slot
 946   if (index_shift > 0)   add(array, index, res); // addr - const offset in index
 947 }
 948 
 949 
 950 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) {
 951   assert_not_delayed();
 952 
 953   // pop array
 954   pop_ptr(array);
 955 
 956   // check array
 957   index_check_without_pop(array, index, index_shift, tmp, res);
 958 }
 959 
 960 
 961 void InterpreterMacroAssembler::get_const(Register Rdst) {
 962   ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst);
 963 }
 964 
 965 
 966 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
 967   get_const(Rdst);
 968   ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
 969 }
 970 
 971 
 972 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
 973   get_constant_pool(Rdst);
 974   ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
 975 }
 976 
 977 
 978 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
 979   get_constant_pool(Rcpool);
 980   ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags);
 981 }
 982 
 983 
 984 // unlock if synchronized method
 985 //
 986 // Unlock the receiver if this is a synchronized method.
 987 // Unlock any Java monitors from syncronized blocks.
 988 //
 989 // If there are locked Java monitors
 990 //    If throw_monitor_exception
 991 //       throws IllegalMonitorStateException
 992 //    Else if install_monitor_exception
 993 //       installs IllegalMonitorStateException
 994 //    Else
 995 //       no error processing
 996 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
 997                                                               bool throw_monitor_exception,
 998                                                               bool install_monitor_exception) {
 999   Label unlocked, unlock, no_unlock;
1000 
1001   // get the value of _do_not_unlock_if_synchronized into G1_scratch
1002   const Address do_not_unlock_if_synchronized(G2_thread,
1003     JavaThread::do_not_unlock_if_synchronized_offset());
1004   ldbool(do_not_unlock_if_synchronized, G1_scratch);
1005   stbool(G0, do_not_unlock_if_synchronized); // reset the flag
1006 
1007   // check if synchronized method
1008   const Address access_flags(Lmethod, Method::access_flags_offset());
1009   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1010   push(state); // save tos
1011   ld(access_flags, G3_scratch); // Load access flags.
1012   btst(JVM_ACC_SYNCHRONIZED, G3_scratch);
1013   br(zero, false, pt, unlocked);
1014   delayed()->nop();
1015 
1016   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1017   // is set.
1018   cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock);
1019   delayed()->nop();
1020 
1021   // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1022   // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1023 
1024   //Intel: if (throw_monitor_exception) ... else ...
1025   // Entry already unlocked, need to throw exception
1026   //...
1027 
1028   // pass top-most monitor elem
1029   add( top_most_monitor(), O1 );
1030 
1031   ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch);
1032   br_notnull_short(G3_scratch, pt, unlock);
1033 
1034   if (throw_monitor_exception) {
1035     // Entry already unlocked need to throw an exception
1036     MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1037     should_not_reach_here();
1038   } else {
1039     // Monitor already unlocked during a stack unroll.
1040     // If requested, install an illegal_monitor_state_exception.
1041     // Continue with stack unrolling.
1042     if (install_monitor_exception) {
1043       MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1044     }
1045     ba_short(unlocked);
1046   }
1047 
1048   bind(unlock);
1049 
1050   unlock_object(O1);
1051 
1052   bind(unlocked);
1053 
1054   // I0, I1: Might contain return value
1055 
1056   // Check that all monitors are unlocked
1057   { Label loop, exception, entry, restart;
1058 
1059     Register Rmptr   = O0;
1060     Register Rtemp   = O1;
1061     Register Rlimit  = Lmonitors;
1062     const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1063     assert( (delta & LongAlignmentMask) == 0,
1064             "sizeof BasicObjectLock must be even number of doublewords");
1065 
1066     #ifdef ASSERT
1067     add(top_most_monitor(), Rmptr, delta);
1068     { Label L;
1069       // ensure that Rmptr starts out above (or at) Rlimit
1070       cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1071       stop("monitor stack has negative size");
1072       bind(L);
1073     }
1074     #endif
1075     bind(restart);
1076     ba(entry);
1077     delayed()->
1078     add(top_most_monitor(), Rmptr, delta);      // points to current entry, starting with bottom-most entry
1079 
1080     // Entry is still locked, need to throw exception
1081     bind(exception);
1082     if (throw_monitor_exception) {
1083       MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1084       should_not_reach_here();
1085     } else {
1086       // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
1087       // Unlock does not block, so don't have to worry about the frame
1088       unlock_object(Rmptr);
1089       if (install_monitor_exception) {
1090         MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1091       }
1092       ba_short(restart);
1093     }
1094 
1095     bind(loop);
1096     cmp(Rtemp, G0);                             // check if current entry is used
1097     brx(Assembler::notEqual, false, pn, exception);
1098     delayed()->
1099     dec(Rmptr, delta);                          // otherwise advance to next entry
1100     #ifdef ASSERT
1101     { Label L;
1102       // ensure that Rmptr has not somehow stepped below Rlimit
1103       cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1104       stop("ran off the end of the monitor stack");
1105       bind(L);
1106     }
1107     #endif
1108     bind(entry);
1109     cmp(Rmptr, Rlimit);                         // check if bottom reached
1110     brx(Assembler::notEqual, true, pn, loop);   // if not at bottom then check this entry
1111     delayed()->
1112     ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp);
1113   }
1114 
1115   bind(no_unlock);
1116   pop(state);
1117   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1118 }
1119 
1120 
1121 // remove activation
1122 //
1123 // Unlock the receiver if this is a synchronized method.
1124 // Unlock any Java monitors from syncronized blocks.
1125 // Remove the activation from the stack.
1126 //
1127 // If there are locked Java monitors
1128 //    If throw_monitor_exception
1129 //       throws IllegalMonitorStateException
1130 //    Else if install_monitor_exception
1131 //       installs IllegalMonitorStateException
1132 //    Else
1133 //       no error processing
1134 void InterpreterMacroAssembler::remove_activation(TosState state,
1135                                                   bool throw_monitor_exception,
1136                                                   bool install_monitor_exception) {
1137 
1138   unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
1139 
1140   // save result (push state before jvmti call and pop it afterwards) and notify jvmti
1141   notify_method_exit(false, state, NotifyJVMTI);
1142 
1143   if (StackReservedPages > 0) {
1144     // testing if Stack Reserved Area needs to be re-enabled
1145     Label no_reserved_zone_enabling;
1146     ld_ptr(G2_thread, JavaThread::reserved_stack_activation_offset(), G3_scratch);
1147     cmp_and_brx_short(SP, G3_scratch, Assembler::lessUnsigned, Assembler::pt, no_reserved_zone_enabling);
1148 
1149     call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), G2_thread);
1150     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError), G2_thread);
1151     should_not_reach_here();
1152     
1153     bind(no_reserved_zone_enabling);
1154   }
1155   
1156   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1157   verify_thread();
1158 
1159   // return tos
1160   assert(Otos_l1 == Otos_i, "adjust code below");
1161   switch (state) {
1162 #ifdef _LP64
1163   case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0
1164 #else
1165   case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through  // O1 -> I1
1166 #endif
1167   case btos:                                      // fall through
1168   case ctos:
1169   case stos:                                      // fall through
1170   case atos:                                      // fall through
1171   case itos: mov(Otos_l1, Otos_l1->after_save());    break;        // O0 -> I0
1172   case ftos:                                      // fall through
1173   case dtos:                                      // fall through
1174   case vtos: /* nothing to do */                     break;
1175   default  : ShouldNotReachHere();
1176   }
1177 
1178 #if defined(COMPILER2) && !defined(_LP64)
1179   if (state == ltos) {
1180     // C2 expects long results in G1 we can't tell if we're returning to interpreted
1181     // or compiled so just be safe use G1 and O0/O1
1182 
1183     // Shift bits into high (msb) of G1
1184     sllx(Otos_l1->after_save(), 32, G1);
1185     // Zero extend low bits
1186     srl (Otos_l2->after_save(), 0, Otos_l2->after_save());
1187     or3 (Otos_l2->after_save(), G1, G1);
1188   }
1189 #endif /* COMPILER2 */
1190 
1191 }
1192 #endif /* CC_INTERP */
1193 
1194 
1195 // Lock object
1196 //
1197 // Argument - lock_reg points to the BasicObjectLock to be used for locking,
1198 //            it must be initialized with the object to lock
1199 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) {
1200   if (UseHeavyMonitors) {
1201     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1202   }
1203   else {
1204     Register obj_reg = Object;
1205     Register mark_reg = G4_scratch;
1206     Register temp_reg = G1_scratch;
1207     Address  lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes());
1208     Address  mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1209     Label    done;
1210 
1211     Label slow_case;
1212 
1213     assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg);
1214 
1215     // load markOop from object into mark_reg
1216     ld_ptr(mark_addr, mark_reg);
1217 
1218     if (UseBiasedLocking) {
1219       biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case);
1220     }
1221 
1222     // get the address of basicLock on stack that will be stored in the object
1223     // we need a temporary register here as we do not want to clobber lock_reg
1224     // (cas clobbers the destination register)
1225     mov(lock_reg, temp_reg);
1226     // set mark reg to be (markOop of object | UNLOCK_VALUE)
1227     or3(mark_reg, markOopDesc::unlocked_value, mark_reg);
1228     // initialize the box  (Must happen before we update the object mark!)
1229     st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1230     // compare and exchange object_addr, markOop | 1, stack address of basicLock
1231     assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1232     cas_ptr(mark_addr.base(), mark_reg, temp_reg);
1233 
1234     // if the compare and exchange succeeded we are done (we saw an unlocked object)
1235     cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done);
1236 
1237     // We did not see an unlocked object so try the fast recursive case
1238 
1239     // Check if owner is self by comparing the value in the markOop of object
1240     // with the stack pointer
1241     sub(temp_reg, SP, temp_reg);
1242 #ifdef _LP64
1243     sub(temp_reg, STACK_BIAS, temp_reg);
1244 #endif
1245     assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1246 
1247     // Composite "andcc" test:
1248     // (a) %sp -vs- markword proximity check, and,
1249     // (b) verify mark word LSBs == 0 (Stack-locked).
1250     //
1251     // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size())
1252     // Note that the page size used for %sp proximity testing is arbitrary and is
1253     // unrelated to the actual MMU page size.  We use a 'logical' page size of
1254     // 4096 bytes.   F..FFF003 is designed to fit conveniently in the SIMM13 immediate
1255     // field of the andcc instruction.
1256     andcc (temp_reg, 0xFFFFF003, G0) ;
1257 
1258     // if condition is true we are done and hence we can store 0 in the displaced
1259     // header indicating it is a recursive lock and be done
1260     brx(Assembler::zero, true, Assembler::pt, done);
1261     delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1262 
1263     // none of the above fast optimizations worked so we have to get into the
1264     // slow case of monitor enter
1265     bind(slow_case);
1266     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1267 
1268     bind(done);
1269   }
1270 }
1271 
1272 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1273 //
1274 // Argument - lock_reg points to the BasicObjectLock for lock
1275 // Throw IllegalMonitorException if object is not locked by current thread
1276 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1277   if (UseHeavyMonitors) {
1278     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1279   } else {
1280     Register obj_reg = G3_scratch;
1281     Register mark_reg = G4_scratch;
1282     Register displaced_header_reg = G1_scratch;
1283     Address  lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes());
1284     Address  mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1285     Label    done;
1286 
1287     if (UseBiasedLocking) {
1288       // load the object out of the BasicObjectLock
1289       ld_ptr(lockobj_addr, obj_reg);
1290       biased_locking_exit(mark_addr, mark_reg, done, true);
1291       st_ptr(G0, lockobj_addr);  // free entry
1292     }
1293 
1294     // Test first if we are in the fast recursive case
1295     Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes());
1296     ld_ptr(lock_addr, displaced_header_reg);
1297     br_null(displaced_header_reg, true, Assembler::pn, done);
1298     delayed()->st_ptr(G0, lockobj_addr);  // free entry
1299 
1300     // See if it is still a light weight lock, if so we just unlock
1301     // the object and we are done
1302 
1303     if (!UseBiasedLocking) {
1304       // load the object out of the BasicObjectLock
1305       ld_ptr(lockobj_addr, obj_reg);
1306     }
1307 
1308     // we have the displaced header in displaced_header_reg
1309     // we expect to see the stack address of the basicLock in case the
1310     // lock is still a light weight lock (lock_reg)
1311     assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1312     cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg);
1313     cmp(lock_reg, displaced_header_reg);
1314     brx(Assembler::equal, true, Assembler::pn, done);
1315     delayed()->st_ptr(G0, lockobj_addr);  // free entry
1316 
1317     // The lock has been converted into a heavy lock and hence
1318     // we need to get into the slow case
1319 
1320     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1321 
1322     bind(done);
1323   }
1324 }
1325 
1326 #ifndef CC_INTERP
1327 
1328 // Get the method data pointer from the Method* and set the
1329 // specified register to its value.
1330 
1331 void InterpreterMacroAssembler::set_method_data_pointer() {
1332   assert(ProfileInterpreter, "must be profiling interpreter");
1333   Label get_continue;
1334 
1335   ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1336   test_method_data_pointer(get_continue);
1337   add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1338   bind(get_continue);
1339 }
1340 
1341 // Set the method data pointer for the current bcp.
1342 
1343 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1344   assert(ProfileInterpreter, "must be profiling interpreter");
1345   Label zero_continue;
1346 
1347   // Test MDO to avoid the call if it is NULL.
1348   ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1349   test_method_data_pointer(zero_continue);
1350   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
1351   add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1352   add(ImethodDataPtr, O0, ImethodDataPtr);
1353   bind(zero_continue);
1354 }
1355 
1356 // Test ImethodDataPtr.  If it is null, continue at the specified label
1357 
1358 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1359   assert(ProfileInterpreter, "must be profiling interpreter");
1360   br_null_short(ImethodDataPtr, Assembler::pn, zero_continue);
1361 }
1362 
1363 void InterpreterMacroAssembler::verify_method_data_pointer() {
1364   assert(ProfileInterpreter, "must be profiling interpreter");
1365 #ifdef ASSERT
1366   Label verify_continue;
1367   test_method_data_pointer(verify_continue);
1368 
1369   // If the mdp is valid, it will point to a DataLayout header which is
1370   // consistent with the bcp.  The converse is highly probable also.
1371   lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch);
1372   ld_ptr(Lmethod, Method::const_offset(), O5);
1373   add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch);
1374   add(G3_scratch, O5, G3_scratch);
1375   cmp(Lbcp, G3_scratch);
1376   brx(Assembler::equal, false, Assembler::pt, verify_continue);
1377 
1378   Register temp_reg = O5;
1379   delayed()->mov(ImethodDataPtr, temp_reg);
1380   // %%% should use call_VM_leaf here?
1381   //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr);
1382   save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1);
1383   Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
1384   stf(FloatRegisterImpl::D, Ftos_d, d_save);
1385   mov(temp_reg->after_save(), O2);
1386   save_thread(L7_thread_cache);
1387   call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none);
1388   delayed()->nop();
1389   restore_thread(L7_thread_cache);
1390   ldf(FloatRegisterImpl::D, d_save, Ftos_d);
1391   restore();
1392   bind(verify_continue);
1393 #endif // ASSERT
1394 }
1395 
1396 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1397                                                                 Register method_counters,
1398                                                                 Register Rtmp,
1399                                                                 Label &profile_continue) {
1400   assert(ProfileInterpreter, "must be profiling interpreter");
1401   // Control will flow to "profile_continue" if the counter is less than the
1402   // limit or if we call profile_method()
1403 
1404   Label done;
1405 
1406   // if no method data exists, and the counter is high enough, make one
1407   br_notnull_short(ImethodDataPtr, Assembler::pn, done);
1408 
1409   // Test to see if we should create a method data oop
1410   Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset());
1411   ld(profile_limit, Rtmp);
1412   cmp(invocation_count, Rtmp);
1413   // Use long branches because call_VM() code and following code generated by
1414   // test_backedge_count_for_osr() is large in debug VM.
1415   br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue);
1416   delayed()->nop();
1417 
1418   // Build it now.
1419   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1420   set_method_data_pointer_for_bcp();
1421   ba(profile_continue);
1422   delayed()->nop();
1423   bind(done);
1424 }
1425 
1426 // Store a value at some constant offset from the method data pointer.
1427 
1428 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1429   assert(ProfileInterpreter, "must be profiling interpreter");
1430   st_ptr(value, ImethodDataPtr, constant);
1431 }
1432 
1433 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter,
1434                                                       Register bumped_count,
1435                                                       bool decrement) {
1436   assert(ProfileInterpreter, "must be profiling interpreter");
1437 
1438   // Load the counter.
1439   ld_ptr(counter, bumped_count);
1440 
1441   if (decrement) {
1442     // Decrement the register.  Set condition codes.
1443     subcc(bumped_count, DataLayout::counter_increment, bumped_count);
1444 
1445     // If the decrement causes the counter to overflow, stay negative
1446     Label L;
1447     brx(Assembler::negative, true, Assembler::pn, L);
1448 
1449     // Store the decremented counter, if it is still negative.
1450     delayed()->st_ptr(bumped_count, counter);
1451     bind(L);
1452   } else {
1453     // Increment the register.  Set carry flag.
1454     addcc(bumped_count, DataLayout::counter_increment, bumped_count);
1455 
1456     // If the increment causes the counter to overflow, pull back by 1.
1457     assert(DataLayout::counter_increment == 1, "subc works");
1458     subc(bumped_count, G0, bumped_count);
1459 
1460     // Store the incremented counter.
1461     st_ptr(bumped_count, counter);
1462   }
1463 }
1464 
1465 // Increment the value at some constant offset from the method data pointer.
1466 
1467 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1468                                                       Register bumped_count,
1469                                                       bool decrement) {
1470   // Locate the counter at a fixed offset from the mdp:
1471   Address counter(ImethodDataPtr, constant);
1472   increment_mdp_data_at(counter, bumped_count, decrement);
1473 }
1474 
1475 // Increment the value at some non-fixed (reg + constant) offset from
1476 // the method data pointer.
1477 
1478 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1479                                                       int constant,
1480                                                       Register bumped_count,
1481                                                       Register scratch2,
1482                                                       bool decrement) {
1483   // Add the constant to reg to get the offset.
1484   add(ImethodDataPtr, reg, scratch2);
1485   Address counter(scratch2, constant);
1486   increment_mdp_data_at(counter, bumped_count, decrement);
1487 }
1488 
1489 // Set a flag value at the current method data pointer position.
1490 // Updates a single byte of the header, to avoid races with other header bits.
1491 
1492 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1493                                                 Register scratch) {
1494   assert(ProfileInterpreter, "must be profiling interpreter");
1495   // Load the data header
1496   ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch);
1497 
1498   // Set the flag
1499   or3(scratch, flag_constant, scratch);
1500 
1501   // Store the modified header.
1502   stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset()));
1503 }
1504 
1505 // Test the location at some offset from the method data pointer.
1506 // If it is not equal to value, branch to the not_equal_continue Label.
1507 // Set condition codes to match the nullness of the loaded value.
1508 
1509 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1510                                                  Register value,
1511                                                  Label& not_equal_continue,
1512                                                  Register scratch) {
1513   assert(ProfileInterpreter, "must be profiling interpreter");
1514   ld_ptr(ImethodDataPtr, offset, scratch);
1515   cmp(value, scratch);
1516   brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue);
1517   delayed()->tst(scratch);
1518 }
1519 
1520 // Update the method data pointer by the displacement located at some fixed
1521 // offset from the method data pointer.
1522 
1523 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1524                                                      Register scratch) {
1525   assert(ProfileInterpreter, "must be profiling interpreter");
1526   ld_ptr(ImethodDataPtr, offset_of_disp, scratch);
1527   add(ImethodDataPtr, scratch, ImethodDataPtr);
1528 }
1529 
1530 // Update the method data pointer by the displacement located at the
1531 // offset (reg + offset_of_disp).
1532 
1533 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1534                                                      int offset_of_disp,
1535                                                      Register scratch) {
1536   assert(ProfileInterpreter, "must be profiling interpreter");
1537   add(reg, offset_of_disp, scratch);
1538   ld_ptr(ImethodDataPtr, scratch, scratch);
1539   add(ImethodDataPtr, scratch, ImethodDataPtr);
1540 }
1541 
1542 // Update the method data pointer by a simple constant displacement.
1543 
1544 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1545   assert(ProfileInterpreter, "must be profiling interpreter");
1546   add(ImethodDataPtr, constant, ImethodDataPtr);
1547 }
1548 
1549 // Update the method data pointer for a _ret bytecode whose target
1550 // was not among our cached targets.
1551 
1552 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1553                                                    Register return_bci) {
1554   assert(ProfileInterpreter, "must be profiling interpreter");
1555   push(state);
1556   st_ptr(return_bci, l_tmp);  // protect return_bci, in case it is volatile
1557   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1558   ld_ptr(l_tmp, return_bci);
1559   pop(state);
1560 }
1561 
1562 // Count a taken branch in the bytecodes.
1563 
1564 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1565   if (ProfileInterpreter) {
1566     Label profile_continue;
1567 
1568     // If no method data exists, go to profile_continue.
1569     test_method_data_pointer(profile_continue);
1570 
1571     // We are taking a branch.  Increment the taken count.
1572     increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count);
1573 
1574     // The method data pointer needs to be updated to reflect the new target.
1575     update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1576     bind (profile_continue);
1577   }
1578 }
1579 
1580 
1581 // Count a not-taken branch in the bytecodes.
1582 
1583 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) {
1584   if (ProfileInterpreter) {
1585     Label profile_continue;
1586 
1587     // If no method data exists, go to profile_continue.
1588     test_method_data_pointer(profile_continue);
1589 
1590     // We are taking a branch.  Increment the not taken count.
1591     increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch);
1592 
1593     // The method data pointer needs to be updated to correspond to the
1594     // next bytecode.
1595     update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1596     bind (profile_continue);
1597   }
1598 }
1599 
1600 
1601 // Count a non-virtual call in the bytecodes.
1602 
1603 void InterpreterMacroAssembler::profile_call(Register scratch) {
1604   if (ProfileInterpreter) {
1605     Label profile_continue;
1606 
1607     // If no method data exists, go to profile_continue.
1608     test_method_data_pointer(profile_continue);
1609 
1610     // We are making a call.  Increment the count.
1611     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1612 
1613     // The method data pointer needs to be updated to reflect the new target.
1614     update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1615     bind (profile_continue);
1616   }
1617 }
1618 
1619 
1620 // Count a final call in the bytecodes.
1621 
1622 void InterpreterMacroAssembler::profile_final_call(Register scratch) {
1623   if (ProfileInterpreter) {
1624     Label profile_continue;
1625 
1626     // If no method data exists, go to profile_continue.
1627     test_method_data_pointer(profile_continue);
1628 
1629     // We are making a call.  Increment the count.
1630     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1631 
1632     // The method data pointer needs to be updated to reflect the new target.
1633     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1634     bind (profile_continue);
1635   }
1636 }
1637 
1638 
1639 // Count a virtual call in the bytecodes.
1640 
1641 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1642                                                      Register scratch,
1643                                                      bool receiver_can_be_null) {
1644   if (ProfileInterpreter) {
1645     Label profile_continue;
1646 
1647     // If no method data exists, go to profile_continue.
1648     test_method_data_pointer(profile_continue);
1649 
1650 
1651     Label skip_receiver_profile;
1652     if (receiver_can_be_null) {
1653       Label not_null;
1654       br_notnull_short(receiver, Assembler::pt, not_null);
1655       // We are making a call.  Increment the count for null receiver.
1656       increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1657       ba_short(skip_receiver_profile);
1658       bind(not_null);
1659     }
1660 
1661     // Record the receiver type.
1662     record_klass_in_profile(receiver, scratch, true);
1663     bind(skip_receiver_profile);
1664 
1665     // The method data pointer needs to be updated to reflect the new target.
1666 #if INCLUDE_JVMCI
1667     if (MethodProfileWidth == 0) {
1668       update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1669     }
1670 #else
1671     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1672 #endif
1673     bind(profile_continue);
1674   }
1675 }
1676 
1677 #if INCLUDE_JVMCI
1678 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) {
1679   assert_different_registers(method, scratch);
1680   if (ProfileInterpreter && MethodProfileWidth > 0) {
1681     Label profile_continue;
1682 
1683     // If no method data exists, go to profile_continue.
1684     test_method_data_pointer(profile_continue);
1685 
1686     Label done;
1687     record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth,
1688       &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1689     bind(done);
1690 
1691     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1692     bind(profile_continue);
1693   }
1694 }
1695 #endif // INCLUDE_JVMCI
1696 
1697 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch,
1698                                                                Label& done, bool is_virtual_call) {
1699   if (TypeProfileWidth == 0) {
1700     if (is_virtual_call) {
1701       increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1702     }
1703 #if INCLUDE_JVMCI
1704     else if (EnableJVMCI) {
1705       increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch);
1706     }
1707 #endif
1708   } else {
1709     int non_profiled_offset = -1;
1710     if (is_virtual_call) {
1711       non_profiled_offset = in_bytes(CounterData::count_offset());
1712     }
1713 #if INCLUDE_JVMCI
1714     else if (EnableJVMCI) {
1715       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1716     }
1717 #endif
1718 
1719     record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth,
1720       &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1721   }
1722 }
1723 
1724 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item,
1725                                           Register scratch, int start_row, Label& done, int total_rows,
1726                                           OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1727                                           int non_profiled_offset) {
1728   int last_row = total_rows - 1;
1729   assert(start_row <= last_row, "must be work left to do");
1730   // Test this row for both the item and for null.
1731   // Take any of three different outcomes:
1732   //   1. found item => increment count and goto done
1733   //   2. found null => keep looking for case 1, maybe allocate this cell
1734   //   3. found something else => keep looking for cases 1 and 2
1735   // Case 3 is handled by a recursive call.
1736   for (int row = start_row; row <= last_row; row++) {
1737     Label next_test;
1738     bool test_for_null_also = (row == start_row);
1739 
1740     // See if the item is item[n].
1741     int item_offset = in_bytes(item_offset_fn(row));
1742     test_mdp_data_at(item_offset, item, next_test, scratch);
1743     // delayed()->tst(scratch);
1744 
1745     // The receiver is item[n].  Increment count[n].
1746     int count_offset = in_bytes(item_count_offset_fn(row));
1747     increment_mdp_data_at(count_offset, scratch);
1748     ba_short(done);
1749     bind(next_test);
1750 
1751     if (test_for_null_also) {
1752       Label found_null;
1753       // Failed the equality check on item[n]...  Test for null.
1754       if (start_row == last_row) {
1755         // The only thing left to do is handle the null case.
1756         if (non_profiled_offset >= 0) {
1757           brx(Assembler::zero, false, Assembler::pn, found_null);
1758           delayed()->nop();
1759           // Item did not match any saved item and there is no empty row for it.
1760           // Increment total counter to indicate polymorphic case.
1761           increment_mdp_data_at(non_profiled_offset, scratch);
1762           ba_short(done);
1763           bind(found_null);
1764         } else {
1765           brx(Assembler::notZero, false, Assembler::pt, done);
1766           delayed()->nop();
1767         }
1768         break;
1769       }
1770       // Since null is rare, make it be the branch-taken case.
1771       brx(Assembler::zero, false, Assembler::pn, found_null);
1772       delayed()->nop();
1773 
1774       // Put all the "Case 3" tests here.
1775       record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows,
1776         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1777 
1778       // Found a null.  Keep searching for a matching item,
1779       // but remember that this is an empty (unused) slot.
1780       bind(found_null);
1781     }
1782   }
1783 
1784   // In the fall-through case, we found no matching item, but we
1785   // observed the item[start_row] is NULL.
1786 
1787   // Fill in the item field and increment the count.
1788   int item_offset = in_bytes(item_offset_fn(start_row));
1789   set_mdp_data_at(item_offset, item);
1790   int count_offset = in_bytes(item_count_offset_fn(start_row));
1791   mov(DataLayout::counter_increment, scratch);
1792   set_mdp_data_at(count_offset, scratch);
1793   if (start_row > 0) {
1794     ba_short(done);
1795   }
1796 }
1797 
1798 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1799                                                         Register scratch, bool is_virtual_call) {
1800   assert(ProfileInterpreter, "must be profiling");
1801   Label done;
1802 
1803   record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call);
1804 
1805   bind (done);
1806 }
1807 
1808 
1809 // Count a ret in the bytecodes.
1810 
1811 void InterpreterMacroAssembler::profile_ret(TosState state,
1812                                             Register return_bci,
1813                                             Register scratch) {
1814   if (ProfileInterpreter) {
1815     Label profile_continue;
1816     uint row;
1817 
1818     // If no method data exists, go to profile_continue.
1819     test_method_data_pointer(profile_continue);
1820 
1821     // Update the total ret count.
1822     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1823 
1824     for (row = 0; row < RetData::row_limit(); row++) {
1825       Label next_test;
1826 
1827       // See if return_bci is equal to bci[n]:
1828       test_mdp_data_at(in_bytes(RetData::bci_offset(row)),
1829                        return_bci, next_test, scratch);
1830 
1831       // return_bci is equal to bci[n].  Increment the count.
1832       increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch);
1833 
1834       // The method data pointer needs to be updated to reflect the new target.
1835       update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch);
1836       ba_short(profile_continue);
1837       bind(next_test);
1838     }
1839 
1840     update_mdp_for_ret(state, return_bci);
1841 
1842     bind (profile_continue);
1843   }
1844 }
1845 
1846 // Profile an unexpected null in the bytecodes.
1847 void InterpreterMacroAssembler::profile_null_seen(Register scratch) {
1848   if (ProfileInterpreter) {
1849     Label profile_continue;
1850 
1851     // If no method data exists, go to profile_continue.
1852     test_method_data_pointer(profile_continue);
1853 
1854     set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch);
1855 
1856     // The method data pointer needs to be updated.
1857     int mdp_delta = in_bytes(BitData::bit_data_size());
1858     if (TypeProfileCasts) {
1859       mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1860     }
1861     update_mdp_by_constant(mdp_delta);
1862 
1863     bind (profile_continue);
1864   }
1865 }
1866 
1867 void InterpreterMacroAssembler::profile_typecheck(Register klass,
1868                                                   Register scratch) {
1869   if (ProfileInterpreter) {
1870     Label profile_continue;
1871 
1872     // If no method data exists, go to profile_continue.
1873     test_method_data_pointer(profile_continue);
1874 
1875     int mdp_delta = in_bytes(BitData::bit_data_size());
1876     if (TypeProfileCasts) {
1877       mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1878 
1879       // Record the object type.
1880       record_klass_in_profile(klass, scratch, false);
1881     }
1882 
1883     // The method data pointer needs to be updated.
1884     update_mdp_by_constant(mdp_delta);
1885 
1886     bind (profile_continue);
1887   }
1888 }
1889 
1890 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) {
1891   if (ProfileInterpreter && TypeProfileCasts) {
1892     Label profile_continue;
1893 
1894     // If no method data exists, go to profile_continue.
1895     test_method_data_pointer(profile_continue);
1896 
1897     int count_offset = in_bytes(CounterData::count_offset());
1898     // Back up the address, since we have already bumped the mdp.
1899     count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size());
1900 
1901     // *Decrement* the counter.  We expect to see zero or small negatives.
1902     increment_mdp_data_at(count_offset, scratch, true);
1903 
1904     bind (profile_continue);
1905   }
1906 }
1907 
1908 // Count the default case of a switch construct.
1909 
1910 void InterpreterMacroAssembler::profile_switch_default(Register scratch) {
1911   if (ProfileInterpreter) {
1912     Label profile_continue;
1913 
1914     // If no method data exists, go to profile_continue.
1915     test_method_data_pointer(profile_continue);
1916 
1917     // Update the default case count
1918     increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1919                           scratch);
1920 
1921     // The method data pointer needs to be updated.
1922     update_mdp_by_offset(
1923                     in_bytes(MultiBranchData::default_displacement_offset()),
1924                     scratch);
1925 
1926     bind (profile_continue);
1927   }
1928 }
1929 
1930 // Count the index'th case of a switch construct.
1931 
1932 void InterpreterMacroAssembler::profile_switch_case(Register index,
1933                                                     Register scratch,
1934                                                     Register scratch2,
1935                                                     Register scratch3) {
1936   if (ProfileInterpreter) {
1937     Label profile_continue;
1938 
1939     // If no method data exists, go to profile_continue.
1940     test_method_data_pointer(profile_continue);
1941 
1942     // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1943     set(in_bytes(MultiBranchData::per_case_size()), scratch);
1944     smul(index, scratch, scratch);
1945     add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch);
1946 
1947     // Update the case count
1948     increment_mdp_data_at(scratch,
1949                           in_bytes(MultiBranchData::relative_count_offset()),
1950                           scratch2,
1951                           scratch3);
1952 
1953     // The method data pointer needs to be updated.
1954     update_mdp_by_offset(scratch,
1955                      in_bytes(MultiBranchData::relative_displacement_offset()),
1956                      scratch2);
1957 
1958     bind (profile_continue);
1959   }
1960 }
1961 
1962 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
1963   Label not_null, do_nothing, do_update;
1964 
1965   assert_different_registers(obj, mdo_addr.base(), tmp);
1966 
1967   verify_oop(obj);
1968 
1969   ld_ptr(mdo_addr, tmp);
1970 
1971   br_notnull_short(obj, pt, not_null);
1972   or3(tmp, TypeEntries::null_seen, tmp);
1973   ba_short(do_update);
1974 
1975   bind(not_null);
1976   load_klass(obj, obj);
1977 
1978   xor3(obj, tmp, obj);
1979   btst(TypeEntries::type_klass_mask, obj);
1980   // klass seen before, nothing to do. The unknown bit may have been
1981   // set already but no need to check.
1982   brx(zero, false, pt, do_nothing);
1983   delayed()->
1984 
1985   btst(TypeEntries::type_unknown, obj);
1986   // already unknown. Nothing to do anymore.
1987   brx(notZero, false, pt, do_nothing);
1988   delayed()->
1989 
1990   btst(TypeEntries::type_mask, tmp);
1991   brx(zero, true, pt, do_update);
1992   // first time here. Set profile type.
1993   delayed()->or3(tmp, obj, tmp);
1994 
1995   // different than before. Cannot keep accurate profile.
1996   or3(tmp, TypeEntries::type_unknown, tmp);
1997 
1998   bind(do_update);
1999   // update profile
2000   st_ptr(tmp, mdo_addr);
2001 
2002   bind(do_nothing);
2003 }
2004 
2005 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
2006   if (!ProfileInterpreter) {
2007     return;
2008   }
2009 
2010   assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr);
2011 
2012   if (MethodData::profile_arguments() || MethodData::profile_return()) {
2013     Label profile_continue;
2014 
2015     test_method_data_pointer(profile_continue);
2016 
2017     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
2018 
2019     ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1);
2020     cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue);
2021 
2022     if (MethodData::profile_arguments()) {
2023       Label done;
2024       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
2025       add(ImethodDataPtr, off_to_args, ImethodDataPtr);
2026 
2027       for (int i = 0; i < TypeProfileArgsLimit; i++) {
2028         if (i > 0 || MethodData::profile_return()) {
2029           // If return value type is profiled we may have no argument to profile
2030           ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2031           sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1);
2032           cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done);
2033         }
2034         ld_ptr(Address(callee, Method::const_offset()), tmp1);
2035         lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1);
2036         // stack offset o (zero based) from the start of the argument
2037         // list, for n arguments translates into offset n - o - 1 from
2038         // the end of the argument list. But there's an extra slot at
2039         // the stop of the stack. So the offset is n - o from Lesp.
2040         ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2);
2041         sub(tmp1, tmp2, tmp1);
2042 
2043         // Can't use MacroAssembler::argument_address() which needs Gargs to be set up
2044         sll(tmp1, Interpreter::logStackElementSize, tmp1);
2045         ld_ptr(Lesp, tmp1, tmp1);
2046 
2047         Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
2048         profile_obj_type(tmp1, mdo_arg_addr, tmp2);
2049 
2050         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
2051         add(ImethodDataPtr, to_add, ImethodDataPtr);
2052         off_to_args += to_add;
2053       }
2054 
2055       if (MethodData::profile_return()) {
2056         ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2057         sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1);
2058       }
2059 
2060       bind(done);
2061 
2062       if (MethodData::profile_return()) {
2063         // We're right after the type profile for the last
2064         // argument. tmp1 is the number of cells left in the
2065         // CallTypeData/VirtualCallTypeData to reach its end. Non null
2066         // if there's a return to profile.
2067         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
2068         sll(tmp1, exact_log2(DataLayout::cell_size), tmp1);
2069         add(ImethodDataPtr, tmp1, ImethodDataPtr);
2070       }
2071     } else {
2072       assert(MethodData::profile_return(), "either profile call args or call ret");
2073       update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
2074     }
2075 
2076     // mdp points right after the end of the
2077     // CallTypeData/VirtualCallTypeData, right after the cells for the
2078     // return value type if there's one.
2079 
2080     bind(profile_continue);
2081   }
2082 }
2083 
2084 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
2085   assert_different_registers(ret, tmp1, tmp2);
2086   if (ProfileInterpreter && MethodData::profile_return()) {
2087     Label profile_continue, done;
2088 
2089     test_method_data_pointer(profile_continue);
2090 
2091     if (MethodData::profile_return_jsr292_only()) {
2092       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
2093 
2094       // If we don't profile all invoke bytecodes we must make sure
2095       // it's a bytecode we indeed profile. We can't go back to the
2096       // begining of the ProfileData we intend to update to check its
2097       // type because we're right after it and we don't known its
2098       // length.
2099       Label do_profile;
2100       ldub(Lbcp, 0, tmp1);
2101       cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile);
2102       cmp(tmp1, Bytecodes::_invokehandle);
2103       br(equal, false, pn, do_profile);
2104       delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1);
2105       cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue);
2106 
2107       bind(do_profile);
2108     }
2109 
2110     Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size()));
2111     mov(ret, tmp1);
2112     profile_obj_type(tmp1, mdo_ret_addr, tmp2);
2113 
2114     bind(profile_continue);
2115   }
2116 }
2117 
2118 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
2119   if (ProfileInterpreter && MethodData::profile_parameters()) {
2120     Label profile_continue, done;
2121 
2122     test_method_data_pointer(profile_continue);
2123 
2124     // Load the offset of the area within the MDO used for
2125     // parameters. If it's negative we're not profiling any parameters.
2126     lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1);
2127     cmp_and_br_short(tmp1, 0, less, pn, profile_continue);
2128 
2129     // Compute a pointer to the area for parameters from the offset
2130     // and move the pointer to the slot for the last
2131     // parameters. Collect profiling from last parameter down.
2132     // mdo start + parameters offset + array length - 1
2133 
2134     // Pointer to the parameter area in the MDO
2135     Register mdp = tmp1;
2136     add(ImethodDataPtr, tmp1, mdp);
2137 
2138     // offset of the current profile entry to update
2139     Register entry_offset = tmp2;
2140     // entry_offset = array len in number of cells
2141     ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset);
2142 
2143     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
2144     assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
2145 
2146     // entry_offset (number of cells)  = array len - size of 1 entry + offset of the stack slot field
2147     sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset);
2148     // entry_offset in bytes
2149     sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset);
2150 
2151     Label loop;
2152     bind(loop);
2153 
2154     // load offset on the stack from the slot for this parameter
2155     ld_ptr(mdp, entry_offset, tmp3);
2156     sll(tmp3,Interpreter::logStackElementSize, tmp3);
2157     neg(tmp3);
2158     // read the parameter from the local area
2159     ld_ptr(Llocals, tmp3, tmp3);
2160 
2161     // make entry_offset now point to the type field for this parameter
2162     int type_base = in_bytes(ParametersTypeData::type_offset(0));
2163     assert(type_base > off_base, "unexpected");
2164     add(entry_offset, type_base - off_base, entry_offset);
2165 
2166     // profile the parameter
2167     Address arg_type(mdp, entry_offset);
2168     profile_obj_type(tmp3, arg_type, tmp4);
2169 
2170     // go to next parameter
2171     sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset);
2172     cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop);
2173 
2174     bind(profile_continue);
2175   }
2176 }
2177 
2178 // add a InterpMonitorElem to stack (see frame_sparc.hpp)
2179 
2180 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty,
2181                                                       Register Rtemp,
2182                                                       Register Rtemp2 ) {
2183 
2184   Register Rlimit = Lmonitors;
2185   const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2186   assert( (delta & LongAlignmentMask) == 0,
2187           "sizeof BasicObjectLock must be even number of doublewords");
2188 
2189   sub( SP,        delta, SP);
2190   sub( Lesp,      delta, Lesp);
2191   sub( Lmonitors, delta, Lmonitors);
2192 
2193   if (!stack_is_empty) {
2194 
2195     // must copy stack contents down
2196 
2197     Label start_copying, next;
2198 
2199     // untested("monitor stack expansion");
2200     compute_stack_base(Rtemp);
2201     ba(start_copying);
2202     delayed()->cmp(Rtemp, Rlimit); // done? duplicated below
2203 
2204     // note: must copy from low memory upwards
2205     // On entry to loop,
2206     // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS)
2207     // Loop mutates Rtemp
2208 
2209     bind( next);
2210 
2211     st_ptr(Rtemp2, Rtemp, 0);
2212     inc(Rtemp, wordSize);
2213     cmp(Rtemp, Rlimit); // are we done? (duplicated above)
2214 
2215     bind( start_copying );
2216 
2217     brx( notEqual, true, pn, next );
2218     delayed()->ld_ptr( Rtemp, delta, Rtemp2 );
2219 
2220     // done copying stack
2221   }
2222 }
2223 
2224 // Locals
2225 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) {
2226   assert_not_delayed();
2227   sll(index, Interpreter::logStackElementSize, index);
2228   sub(Llocals, index, index);
2229   ld_ptr(index, 0, dst);
2230   // Note:  index must hold the effective address--the iinc template uses it
2231 }
2232 
2233 // Just like access_local_ptr but the tag is a returnAddress
2234 void InterpreterMacroAssembler::access_local_returnAddress(Register index,
2235                                                            Register dst ) {
2236   assert_not_delayed();
2237   sll(index, Interpreter::logStackElementSize, index);
2238   sub(Llocals, index, index);
2239   ld_ptr(index, 0, dst);
2240 }
2241 
2242 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) {
2243   assert_not_delayed();
2244   sll(index, Interpreter::logStackElementSize, index);
2245   sub(Llocals, index, index);
2246   ld(index, 0, dst);
2247   // Note:  index must hold the effective address--the iinc template uses it
2248 }
2249 
2250 
2251 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) {
2252   assert_not_delayed();
2253   sll(index, Interpreter::logStackElementSize, index);
2254   sub(Llocals, index, index);
2255   // First half stored at index n+1 (which grows down from Llocals[n])
2256   load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst);
2257 }
2258 
2259 
2260 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) {
2261   assert_not_delayed();
2262   sll(index, Interpreter::logStackElementSize, index);
2263   sub(Llocals, index, index);
2264   ldf(FloatRegisterImpl::S, index, 0, dst);
2265 }
2266 
2267 
2268 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) {
2269   assert_not_delayed();
2270   sll(index, Interpreter::logStackElementSize, index);
2271   sub(Llocals, index, index);
2272   load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst);
2273 }
2274 
2275 
2276 #ifdef ASSERT
2277 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) {
2278   Label L;
2279 
2280   assert(Rindex != Rscratch, "Registers cannot be same");
2281   assert(Rindex != Rscratch1, "Registers cannot be same");
2282   assert(Rlimit != Rscratch, "Registers cannot be same");
2283   assert(Rlimit != Rscratch1, "Registers cannot be same");
2284   assert(Rscratch1 != Rscratch, "Registers cannot be same");
2285 
2286   // untested("reg area corruption");
2287   add(Rindex, offset, Rscratch);
2288   add(Rlimit, 64 + STACK_BIAS, Rscratch1);
2289   cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L);
2290   stop("regsave area is being clobbered");
2291   bind(L);
2292 }
2293 #endif // ASSERT
2294 
2295 
2296 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) {
2297   assert_not_delayed();
2298   sll(index, Interpreter::logStackElementSize, index);
2299   sub(Llocals, index, index);
2300   debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);)
2301   st(src, index, 0);
2302 }
2303 
2304 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) {
2305   assert_not_delayed();
2306   sll(index, Interpreter::logStackElementSize, index);
2307   sub(Llocals, index, index);
2308 #ifdef ASSERT
2309   check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2310 #endif
2311   st_ptr(src, index, 0);
2312 }
2313 
2314 
2315 
2316 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) {
2317   st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n));
2318 }
2319 
2320 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) {
2321   assert_not_delayed();
2322   sll(index, Interpreter::logStackElementSize, index);
2323   sub(Llocals, index, index);
2324 #ifdef ASSERT
2325   check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2326 #endif
2327   store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1
2328 }
2329 
2330 
2331 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) {
2332   assert_not_delayed();
2333   sll(index, Interpreter::logStackElementSize, index);
2334   sub(Llocals, index, index);
2335 #ifdef ASSERT
2336   check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2337 #endif
2338   stf(FloatRegisterImpl::S, src, index, 0);
2339 }
2340 
2341 
2342 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) {
2343   assert_not_delayed();
2344   sll(index, Interpreter::logStackElementSize, index);
2345   sub(Llocals, index, index);
2346 #ifdef ASSERT
2347   check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2348 #endif
2349   store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1));
2350 }
2351 
2352 
2353 int InterpreterMacroAssembler::top_most_monitor_byte_offset() {
2354   const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2355   int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong);
2356   return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS;
2357 }
2358 
2359 
2360 Address InterpreterMacroAssembler::top_most_monitor() {
2361   return Address(FP, top_most_monitor_byte_offset());
2362 }
2363 
2364 
2365 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) {
2366   add( Lesp,      wordSize,                                    Rdest );
2367 }
2368 
2369 #endif /* CC_INTERP */
2370 
2371 void InterpreterMacroAssembler::get_method_counters(Register method,
2372                                                     Register Rcounters,
2373                                                     Label& skip) {
2374   Label has_counters;
2375   Address method_counters(method, in_bytes(Method::method_counters_offset()));
2376   ld_ptr(method_counters, Rcounters);
2377   br_notnull_short(Rcounters, Assembler::pt, has_counters);
2378   call_VM(noreg, CAST_FROM_FN_PTR(address,
2379           InterpreterRuntime::build_method_counters), method);
2380   ld_ptr(method_counters, Rcounters);
2381   br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory
2382   delayed()->nop();
2383   bind(has_counters);
2384 }
2385 
2386 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2387   assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
2388   assert_different_registers(Rcounters, Rtmp, Rtmp2);
2389 
2390   Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2391                                  InvocationCounter::counter_offset());
2392   Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2393                                  InvocationCounter::counter_offset());
2394   int delta = InvocationCounter::count_increment;
2395 
2396   // Load each counter in a register
2397   ld( inv_counter, Rtmp );
2398   ld( be_counter, Rtmp2 );
2399 
2400   assert( is_simm13( delta ), " delta too large.");
2401 
2402   // Add the delta to the invocation counter and store the result
2403   add( Rtmp, delta, Rtmp );
2404 
2405   // Mask the backedge counter
2406   and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2407 
2408   // Store value
2409   st( Rtmp, inv_counter);
2410 
2411   // Add invocation counter + backedge counter
2412   add( Rtmp, Rtmp2, Rtmp);
2413 
2414   // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2415 }
2416 
2417 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2418   assert(UseCompiler, "incrementing must be useful");
2419   assert_different_registers(Rcounters, Rtmp, Rtmp2);
2420 
2421   Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2422                                  InvocationCounter::counter_offset());
2423   Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2424                                  InvocationCounter::counter_offset());
2425 
2426   int delta = InvocationCounter::count_increment;
2427   // Load each counter in a register
2428   ld( be_counter, Rtmp );
2429   ld( inv_counter, Rtmp2 );
2430 
2431   // Add the delta to the backedge counter
2432   add( Rtmp, delta, Rtmp );
2433 
2434   // Mask the invocation counter, add to backedge counter
2435   and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2436 
2437   // and store the result to memory
2438   st( Rtmp, be_counter );
2439 
2440   // Add backedge + invocation counter
2441   add( Rtmp, Rtmp2, Rtmp );
2442 
2443   // Note that this macro must leave backedge_count + invocation_count in Rtmp!
2444 }
2445 
2446 #ifndef CC_INTERP
2447 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count,
2448                                                              Register method_counters,
2449                                                              Register branch_bcp,
2450                                                              Register Rtmp ) {
2451   Label did_not_overflow;
2452   Label overflow_with_error;
2453   assert_different_registers(backedge_count, Rtmp, branch_bcp);
2454   assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
2455 
2456   Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()));
2457   ld(limit, Rtmp);
2458   cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow);
2459 
2460   // When ProfileInterpreter is on, the backedge_count comes from the
2461   // MethodData*, which value does not get reset on the call to
2462   // frequency_counter_overflow().  To avoid excessive calls to the overflow
2463   // routine while the method is being compiled, add a second test to make sure
2464   // the overflow function is called only once every overflow_frequency.
2465   if (ProfileInterpreter) {
2466     const int overflow_frequency = 1024;
2467     andcc(backedge_count, overflow_frequency-1, Rtmp);
2468     brx(Assembler::notZero, false, Assembler::pt, did_not_overflow);
2469     delayed()->nop();
2470   }
2471 
2472   // overflow in loop, pass branch bytecode
2473   set(6,Rtmp);
2474   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp);
2475 
2476   // Was an OSR adapter generated?
2477   // O0 = osr nmethod
2478   br_null_short(O0, Assembler::pn, overflow_with_error);
2479 
2480   // Has the nmethod been invalidated already?
2481   ldub(O0, nmethod::state_offset(), O2);
2482   cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error);
2483 
2484   // migrate the interpreter frame off of the stack
2485 
2486   mov(G2_thread, L7);
2487   // save nmethod
2488   mov(O0, L6);
2489   set_last_Java_frame(SP, noreg);
2490   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
2491   reset_last_Java_frame();
2492   mov(L7, G2_thread);
2493 
2494   // move OSR nmethod to I1
2495   mov(L6, I1);
2496 
2497   // OSR buffer to I0
2498   mov(O0, I0);
2499 
2500   // remove the interpreter frame
2501   restore(I5_savedSP, 0, SP);
2502 
2503   // Jump to the osr code.
2504   ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
2505   jmp(O2, G0);
2506   delayed()->nop();
2507 
2508   bind(overflow_with_error);
2509 
2510   bind(did_not_overflow);
2511 }
2512 
2513 
2514 
2515 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) {
2516   if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); }
2517 }
2518 
2519 
2520 // local helper function for the verify_oop_or_return_address macro
2521 static bool verify_return_address(Method* m, int bci) {
2522 #ifndef PRODUCT
2523   address pc = (address)(m->constMethod())
2524              + in_bytes(ConstMethod::codes_offset()) + bci;
2525   // assume it is a valid return address if it is inside m and is preceded by a jsr
2526   if (!m->contains(pc))                                          return false;
2527   address jsr_pc;
2528   jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2529   if (*jsr_pc == Bytecodes::_jsr   && jsr_pc >= m->code_base())    return true;
2530   jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2531   if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base())    return true;
2532 #endif // PRODUCT
2533   return false;
2534 }
2535 
2536 
2537 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2538   if (!VerifyOops)  return;
2539   // the VM documentation for the astore[_wide] bytecode allows
2540   // the TOS to be not only an oop but also a return address
2541   Label test;
2542   Label skip;
2543   // See if it is an address (in the current method):
2544 
2545   mov(reg, Rtmp);
2546   const int log2_bytecode_size_limit = 16;
2547   srl(Rtmp, log2_bytecode_size_limit, Rtmp);
2548   br_notnull_short( Rtmp, pt, test );
2549 
2550   // %%% should use call_VM_leaf here?
2551   save_frame_and_mov(0, Lmethod, O0, reg, O1);
2552   save_thread(L7_thread_cache);
2553   call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none);
2554   delayed()->nop();
2555   restore_thread(L7_thread_cache);
2556   br_notnull( O0, false, pt, skip );
2557   delayed()->restore();
2558 
2559   // Perform a more elaborate out-of-line call
2560   // Not an address; verify it:
2561   bind(test);
2562   verify_oop(reg);
2563   bind(skip);
2564 }
2565 
2566 
2567 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2568   if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
2569 }
2570 
2571 
2572 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2573 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2574                                                         int increment, Address mask_addr,
2575                                                         Register scratch1, Register scratch2,
2576                                                         Condition cond, Label *where) {
2577   ld(counter_addr, scratch1);
2578   add(scratch1, increment, scratch1);
2579   ld(mask_addr, scratch2);
2580   andcc(scratch1, scratch2,  G0);
2581   br(cond, false, Assembler::pn, *where);
2582   delayed()->st(scratch1, counter_addr);
2583 }
2584 #endif /* CC_INTERP */
2585 
2586 // Inline assembly for:
2587 //
2588 // if (thread is in interp_only_mode) {
2589 //   InterpreterRuntime::post_method_entry();
2590 // }
2591 // if (DTraceMethodProbes) {
2592 //   SharedRuntime::dtrace_method_entry(method, receiver);
2593 // }
2594 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2595 //   SharedRuntime::rc_trace_method_entry(method, receiver);
2596 // }
2597 
2598 void InterpreterMacroAssembler::notify_method_entry() {
2599 
2600   // C++ interpreter only uses this for native methods.
2601 
2602   // Whenever JVMTI puts a thread in interp_only_mode, method
2603   // entry/exit events are sent for that thread to track stack
2604   // depth.  If it is possible to enter interp_only_mode we add
2605   // the code to check if the event should be sent.
2606   if (JvmtiExport::can_post_interpreter_events()) {
2607     Label L;
2608     Register temp_reg = O5;
2609     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2610     ld(interp_only, temp_reg);
2611     cmp_and_br_short(temp_reg, 0, equal, pt, L);
2612     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2613     bind(L);
2614   }
2615 
2616   {
2617     Register temp_reg = O5;
2618     SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2619     call_VM_leaf(noreg,
2620       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2621       G2_thread, Lmethod);
2622   }
2623 
2624   // RedefineClasses() tracing support for obsolete method entry
2625   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2626     call_VM_leaf(noreg,
2627       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2628       G2_thread, Lmethod);
2629   }
2630 }
2631 
2632 
2633 // Inline assembly for:
2634 //
2635 // if (thread is in interp_only_mode) {
2636 //   // save result
2637 //   InterpreterRuntime::post_method_exit();
2638 //   // restore result
2639 // }
2640 // if (DTraceMethodProbes) {
2641 //   SharedRuntime::dtrace_method_exit(thread, method);
2642 // }
2643 //
2644 // Native methods have their result stored in d_tmp and l_tmp
2645 // Java methods have their result stored in the expression stack
2646 
2647 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method,
2648                                                    TosState state,
2649                                                    NotifyMethodExitMode mode) {
2650   // C++ interpreter only uses this for native methods.
2651 
2652   // Whenever JVMTI puts a thread in interp_only_mode, method
2653   // entry/exit events are sent for that thread to track stack
2654   // depth.  If it is possible to enter interp_only_mode we add
2655   // the code to check if the event should be sent.
2656   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2657     Label L;
2658     Register temp_reg = O5;
2659     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2660     ld(interp_only, temp_reg);
2661     cmp_and_br_short(temp_reg, 0, equal, pt, L);
2662 
2663     // Note: frame::interpreter_frame_result has a dependency on how the
2664     // method result is saved across the call to post_method_exit. For
2665     // native methods it assumes the result registers are saved to
2666     // l_scratch and d_scratch. If this changes then the interpreter_frame_result
2667     // implementation will need to be updated too.
2668 
2669     save_return_value(state, is_native_method);
2670     call_VM(noreg,
2671             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2672     restore_return_value(state, is_native_method);
2673     bind(L);
2674   }
2675 
2676   {
2677     Register temp_reg = O5;
2678     // Dtrace notification
2679     SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2680     save_return_value(state, is_native_method);
2681     call_VM_leaf(
2682       noreg,
2683       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2684       G2_thread, Lmethod);
2685     restore_return_value(state, is_native_method);
2686   }
2687 }
2688 
2689 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) {
2690 #ifdef CC_INTERP
2691   // result potentially in O0/O1: save it across calls
2692   stf(FloatRegisterImpl::D, F0, STATE(_native_fresult));
2693 #ifdef _LP64
2694   stx(O0, STATE(_native_lresult));
2695 #else
2696   std(O0, STATE(_native_lresult));
2697 #endif
2698 #else // CC_INTERP
2699   if (is_native_call) {
2700     stf(FloatRegisterImpl::D, F0, d_tmp);
2701 #ifdef _LP64
2702     stx(O0, l_tmp);
2703 #else
2704     std(O0, l_tmp);
2705 #endif
2706   } else {
2707     push(state);
2708   }
2709 #endif // CC_INTERP
2710 }
2711 
2712 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) {
2713 #ifdef CC_INTERP
2714   ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0);
2715 #ifdef _LP64
2716   ldx(STATE(_native_lresult), O0);
2717 #else
2718   ldd(STATE(_native_lresult), O0);
2719 #endif
2720 #else // CC_INTERP
2721   if (is_native_call) {
2722     ldf(FloatRegisterImpl::D, d_tmp, F0);
2723 #ifdef _LP64
2724     ldx(l_tmp, O0);
2725 #else
2726     ldd(l_tmp, O0);
2727 #endif
2728   } else {
2729     pop(state);
2730   }
2731 #endif // CC_INTERP
2732 }