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   interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1144   verify_thread();
1145 
1146   // return tos
1147   assert(Otos_l1 == Otos_i, "adjust code below");
1148   switch (state) {
1149 #ifdef _LP64
1150   case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0
1151 #else
1152   case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through  // O1 -> I1
1153 #endif
1154   case btos:                                      // fall through
1155   case ctos:
1156   case stos:                                      // fall through
1157   case atos:                                      // fall through
1158   case itos: mov(Otos_l1, Otos_l1->after_save());    break;        // O0 -> I0
1159   case ftos:                                      // fall through
1160   case dtos:                                      // fall through
1161   case vtos: /* nothing to do */                     break;
1162   default  : ShouldNotReachHere();
1163   }
1164 
1165 #if defined(COMPILER2) && !defined(_LP64)
1166   if (state == ltos) {
1167     // C2 expects long results in G1 we can't tell if we're returning to interpreted
1168     // or compiled so just be safe use G1 and O0/O1
1169 
1170     // Shift bits into high (msb) of G1
1171     sllx(Otos_l1->after_save(), 32, G1);
1172     // Zero extend low bits
1173     srl (Otos_l2->after_save(), 0, Otos_l2->after_save());
1174     or3 (Otos_l2->after_save(), G1, G1);
1175   }
1176 #endif /* COMPILER2 */
1177 
1178 }
1179 #endif /* CC_INTERP */
1180 
1181 
1182 // Lock object
1183 //
1184 // Argument - lock_reg points to the BasicObjectLock to be used for locking,
1185 //            it must be initialized with the object to lock
1186 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) {
1187   if (UseHeavyMonitors) {
1188     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1189   }
1190   else {
1191     Register obj_reg = Object;
1192     Register mark_reg = G4_scratch;
1193     Register temp_reg = G1_scratch;
1194     Address  lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes());
1195     Address  mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1196     Label    done;
1197 
1198     Label slow_case;
1199 
1200     assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg);
1201 
1202     // load markOop from object into mark_reg
1203     ld_ptr(mark_addr, mark_reg);
1204 
1205     if (UseBiasedLocking) {
1206       biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case);
1207     }
1208 
1209     // get the address of basicLock on stack that will be stored in the object
1210     // we need a temporary register here as we do not want to clobber lock_reg
1211     // (cas clobbers the destination register)
1212     mov(lock_reg, temp_reg);
1213     // set mark reg to be (markOop of object | UNLOCK_VALUE)
1214     or3(mark_reg, markOopDesc::unlocked_value, mark_reg);
1215     // initialize the box  (Must happen before we update the object mark!)
1216     st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1217     // compare and exchange object_addr, markOop | 1, stack address of basicLock
1218     assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1219     cas_ptr(mark_addr.base(), mark_reg, temp_reg);
1220 
1221     // if the compare and exchange succeeded we are done (we saw an unlocked object)
1222     cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done);
1223 
1224     // We did not see an unlocked object so try the fast recursive case
1225 
1226     // Check if owner is self by comparing the value in the markOop of object
1227     // with the stack pointer
1228     sub(temp_reg, SP, temp_reg);
1229 #ifdef _LP64
1230     sub(temp_reg, STACK_BIAS, temp_reg);
1231 #endif
1232     assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1233 
1234     // Composite "andcc" test:
1235     // (a) %sp -vs- markword proximity check, and,
1236     // (b) verify mark word LSBs == 0 (Stack-locked).
1237     //
1238     // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size())
1239     // Note that the page size used for %sp proximity testing is arbitrary and is
1240     // unrelated to the actual MMU page size.  We use a 'logical' page size of
1241     // 4096 bytes.   F..FFF003 is designed to fit conveniently in the SIMM13 immediate
1242     // field of the andcc instruction.
1243     andcc (temp_reg, 0xFFFFF003, G0) ;
1244 
1245     // if condition is true we are done and hence we can store 0 in the displaced
1246     // header indicating it is a recursive lock and be done
1247     brx(Assembler::zero, true, Assembler::pt, done);
1248     delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1249 
1250     // none of the above fast optimizations worked so we have to get into the
1251     // slow case of monitor enter
1252     bind(slow_case);
1253     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1254 
1255     bind(done);
1256   }
1257 }
1258 
1259 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1260 //
1261 // Argument - lock_reg points to the BasicObjectLock for lock
1262 // Throw IllegalMonitorException if object is not locked by current thread
1263 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1264   if (UseHeavyMonitors) {
1265     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1266   } else {
1267     Register obj_reg = G3_scratch;
1268     Register mark_reg = G4_scratch;
1269     Register displaced_header_reg = G1_scratch;
1270     Address  lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes());
1271     Address  mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1272     Label    done;
1273 
1274     if (UseBiasedLocking) {
1275       // load the object out of the BasicObjectLock
1276       ld_ptr(lockobj_addr, obj_reg);
1277       biased_locking_exit(mark_addr, mark_reg, done, true);
1278       st_ptr(G0, lockobj_addr);  // free entry
1279     }
1280 
1281     // Test first if we are in the fast recursive case
1282     Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes());
1283     ld_ptr(lock_addr, displaced_header_reg);
1284     br_null(displaced_header_reg, true, Assembler::pn, done);
1285     delayed()->st_ptr(G0, lockobj_addr);  // free entry
1286 
1287     // See if it is still a light weight lock, if so we just unlock
1288     // the object and we are done
1289 
1290     if (!UseBiasedLocking) {
1291       // load the object out of the BasicObjectLock
1292       ld_ptr(lockobj_addr, obj_reg);
1293     }
1294 
1295     // we have the displaced header in displaced_header_reg
1296     // we expect to see the stack address of the basicLock in case the
1297     // lock is still a light weight lock (lock_reg)
1298     assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1299     cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg);
1300     cmp(lock_reg, displaced_header_reg);
1301     brx(Assembler::equal, true, Assembler::pn, done);
1302     delayed()->st_ptr(G0, lockobj_addr);  // free entry
1303 
1304     // The lock has been converted into a heavy lock and hence
1305     // we need to get into the slow case
1306 
1307     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1308 
1309     bind(done);
1310   }
1311 }
1312 
1313 #ifndef CC_INTERP
1314 
1315 // Get the method data pointer from the Method* and set the
1316 // specified register to its value.
1317 
1318 void InterpreterMacroAssembler::set_method_data_pointer() {
1319   assert(ProfileInterpreter, "must be profiling interpreter");
1320   Label get_continue;
1321 
1322   ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1323   test_method_data_pointer(get_continue);
1324   add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1325   bind(get_continue);
1326 }
1327 
1328 // Set the method data pointer for the current bcp.
1329 
1330 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1331   assert(ProfileInterpreter, "must be profiling interpreter");
1332   Label zero_continue;
1333 
1334   // Test MDO to avoid the call if it is NULL.
1335   ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1336   test_method_data_pointer(zero_continue);
1337   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
1338   add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1339   add(ImethodDataPtr, O0, ImethodDataPtr);
1340   bind(zero_continue);
1341 }
1342 
1343 // Test ImethodDataPtr.  If it is null, continue at the specified label
1344 
1345 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1346   assert(ProfileInterpreter, "must be profiling interpreter");
1347   br_null_short(ImethodDataPtr, Assembler::pn, zero_continue);
1348 }
1349 
1350 void InterpreterMacroAssembler::verify_method_data_pointer() {
1351   assert(ProfileInterpreter, "must be profiling interpreter");
1352 #ifdef ASSERT
1353   Label verify_continue;
1354   test_method_data_pointer(verify_continue);
1355 
1356   // If the mdp is valid, it will point to a DataLayout header which is
1357   // consistent with the bcp.  The converse is highly probable also.
1358   lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch);
1359   ld_ptr(Lmethod, Method::const_offset(), O5);
1360   add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch);
1361   add(G3_scratch, O5, G3_scratch);
1362   cmp(Lbcp, G3_scratch);
1363   brx(Assembler::equal, false, Assembler::pt, verify_continue);
1364 
1365   Register temp_reg = O5;
1366   delayed()->mov(ImethodDataPtr, temp_reg);
1367   // %%% should use call_VM_leaf here?
1368   //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr);
1369   save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1);
1370   Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
1371   stf(FloatRegisterImpl::D, Ftos_d, d_save);
1372   mov(temp_reg->after_save(), O2);
1373   save_thread(L7_thread_cache);
1374   call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none);
1375   delayed()->nop();
1376   restore_thread(L7_thread_cache);
1377   ldf(FloatRegisterImpl::D, d_save, Ftos_d);
1378   restore();
1379   bind(verify_continue);
1380 #endif // ASSERT
1381 }
1382 
1383 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1384                                                                 Register method_counters,
1385                                                                 Register Rtmp,
1386                                                                 Label &profile_continue) {
1387   assert(ProfileInterpreter, "must be profiling interpreter");
1388   // Control will flow to "profile_continue" if the counter is less than the
1389   // limit or if we call profile_method()
1390 
1391   Label done;
1392 
1393   // if no method data exists, and the counter is high enough, make one
1394   br_notnull_short(ImethodDataPtr, Assembler::pn, done);
1395 
1396   // Test to see if we should create a method data oop
1397   Address profile_limit(method_counters, MethodCounters::interpreter_profile_limit_offset());
1398   ld(profile_limit, Rtmp);
1399   cmp(invocation_count, Rtmp);
1400   // Use long branches because call_VM() code and following code generated by
1401   // test_backedge_count_for_osr() is large in debug VM.
1402   br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue);
1403   delayed()->nop();
1404 
1405   // Build it now.
1406   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1407   set_method_data_pointer_for_bcp();
1408   ba(profile_continue);
1409   delayed()->nop();
1410   bind(done);
1411 }
1412 
1413 // Store a value at some constant offset from the method data pointer.
1414 
1415 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1416   assert(ProfileInterpreter, "must be profiling interpreter");
1417   st_ptr(value, ImethodDataPtr, constant);
1418 }
1419 
1420 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter,
1421                                                       Register bumped_count,
1422                                                       bool decrement) {
1423   assert(ProfileInterpreter, "must be profiling interpreter");
1424 
1425   // Load the counter.
1426   ld_ptr(counter, bumped_count);
1427 
1428   if (decrement) {
1429     // Decrement the register.  Set condition codes.
1430     subcc(bumped_count, DataLayout::counter_increment, bumped_count);
1431 
1432     // If the decrement causes the counter to overflow, stay negative
1433     Label L;
1434     brx(Assembler::negative, true, Assembler::pn, L);
1435 
1436     // Store the decremented counter, if it is still negative.
1437     delayed()->st_ptr(bumped_count, counter);
1438     bind(L);
1439   } else {
1440     // Increment the register.  Set carry flag.
1441     addcc(bumped_count, DataLayout::counter_increment, bumped_count);
1442 
1443     // If the increment causes the counter to overflow, pull back by 1.
1444     assert(DataLayout::counter_increment == 1, "subc works");
1445     subc(bumped_count, G0, bumped_count);
1446 
1447     // Store the incremented counter.
1448     st_ptr(bumped_count, counter);
1449   }
1450 }
1451 
1452 // Increment the value at some constant offset from the method data pointer.
1453 
1454 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1455                                                       Register bumped_count,
1456                                                       bool decrement) {
1457   // Locate the counter at a fixed offset from the mdp:
1458   Address counter(ImethodDataPtr, constant);
1459   increment_mdp_data_at(counter, bumped_count, decrement);
1460 }
1461 
1462 // Increment the value at some non-fixed (reg + constant) offset from
1463 // the method data pointer.
1464 
1465 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1466                                                       int constant,
1467                                                       Register bumped_count,
1468                                                       Register scratch2,
1469                                                       bool decrement) {
1470   // Add the constant to reg to get the offset.
1471   add(ImethodDataPtr, reg, scratch2);
1472   Address counter(scratch2, constant);
1473   increment_mdp_data_at(counter, bumped_count, decrement);
1474 }
1475 
1476 // Set a flag value at the current method data pointer position.
1477 // Updates a single byte of the header, to avoid races with other header bits.
1478 
1479 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1480                                                 Register scratch) {
1481   assert(ProfileInterpreter, "must be profiling interpreter");
1482   // Load the data header
1483   ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch);
1484 
1485   // Set the flag
1486   or3(scratch, flag_constant, scratch);
1487 
1488   // Store the modified header.
1489   stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset()));
1490 }
1491 
1492 // Test the location at some offset from the method data pointer.
1493 // If it is not equal to value, branch to the not_equal_continue Label.
1494 // Set condition codes to match the nullness of the loaded value.
1495 
1496 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1497                                                  Register value,
1498                                                  Label& not_equal_continue,
1499                                                  Register scratch) {
1500   assert(ProfileInterpreter, "must be profiling interpreter");
1501   ld_ptr(ImethodDataPtr, offset, scratch);
1502   cmp(value, scratch);
1503   brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue);
1504   delayed()->tst(scratch);
1505 }
1506 
1507 // Update the method data pointer by the displacement located at some fixed
1508 // offset from the method data pointer.
1509 
1510 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1511                                                      Register scratch) {
1512   assert(ProfileInterpreter, "must be profiling interpreter");
1513   ld_ptr(ImethodDataPtr, offset_of_disp, scratch);
1514   add(ImethodDataPtr, scratch, ImethodDataPtr);
1515 }
1516 
1517 // Update the method data pointer by the displacement located at the
1518 // offset (reg + offset_of_disp).
1519 
1520 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1521                                                      int offset_of_disp,
1522                                                      Register scratch) {
1523   assert(ProfileInterpreter, "must be profiling interpreter");
1524   add(reg, offset_of_disp, scratch);
1525   ld_ptr(ImethodDataPtr, scratch, scratch);
1526   add(ImethodDataPtr, scratch, ImethodDataPtr);
1527 }
1528 
1529 // Update the method data pointer by a simple constant displacement.
1530 
1531 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1532   assert(ProfileInterpreter, "must be profiling interpreter");
1533   add(ImethodDataPtr, constant, ImethodDataPtr);
1534 }
1535 
1536 // Update the method data pointer for a _ret bytecode whose target
1537 // was not among our cached targets.
1538 
1539 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1540                                                    Register return_bci) {
1541   assert(ProfileInterpreter, "must be profiling interpreter");
1542   push(state);
1543   st_ptr(return_bci, l_tmp);  // protect return_bci, in case it is volatile
1544   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1545   ld_ptr(l_tmp, return_bci);
1546   pop(state);
1547 }
1548 
1549 // Count a taken branch in the bytecodes.
1550 
1551 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1552   if (ProfileInterpreter) {
1553     Label profile_continue;
1554 
1555     // If no method data exists, go to profile_continue.
1556     test_method_data_pointer(profile_continue);
1557 
1558     // We are taking a branch.  Increment the taken count.
1559     increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count);
1560 
1561     // The method data pointer needs to be updated to reflect the new target.
1562     update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1563     bind (profile_continue);
1564   }
1565 }
1566 
1567 
1568 // Count a not-taken branch in the bytecodes.
1569 
1570 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) {
1571   if (ProfileInterpreter) {
1572     Label profile_continue;
1573 
1574     // If no method data exists, go to profile_continue.
1575     test_method_data_pointer(profile_continue);
1576 
1577     // We are taking a branch.  Increment the not taken count.
1578     increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch);
1579 
1580     // The method data pointer needs to be updated to correspond to the
1581     // next bytecode.
1582     update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1583     bind (profile_continue);
1584   }
1585 }
1586 
1587 
1588 // Count a non-virtual call in the bytecodes.
1589 
1590 void InterpreterMacroAssembler::profile_call(Register scratch) {
1591   if (ProfileInterpreter) {
1592     Label profile_continue;
1593 
1594     // If no method data exists, go to profile_continue.
1595     test_method_data_pointer(profile_continue);
1596 
1597     // We are making a call.  Increment the count.
1598     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1599 
1600     // The method data pointer needs to be updated to reflect the new target.
1601     update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1602     bind (profile_continue);
1603   }
1604 }
1605 
1606 
1607 // Count a final call in the bytecodes.
1608 
1609 void InterpreterMacroAssembler::profile_final_call(Register scratch) {
1610   if (ProfileInterpreter) {
1611     Label profile_continue;
1612 
1613     // If no method data exists, go to profile_continue.
1614     test_method_data_pointer(profile_continue);
1615 
1616     // We are making a call.  Increment the count.
1617     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1618 
1619     // The method data pointer needs to be updated to reflect the new target.
1620     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1621     bind (profile_continue);
1622   }
1623 }
1624 
1625 
1626 // Count a virtual call in the bytecodes.
1627 
1628 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1629                                                      Register scratch,
1630                                                      bool receiver_can_be_null) {
1631   if (ProfileInterpreter) {
1632     Label profile_continue;
1633 
1634     // If no method data exists, go to profile_continue.
1635     test_method_data_pointer(profile_continue);
1636 
1637 
1638     Label skip_receiver_profile;
1639     if (receiver_can_be_null) {
1640       Label not_null;
1641       br_notnull_short(receiver, Assembler::pt, not_null);
1642       // We are making a call.  Increment the count for null receiver.
1643       increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1644       ba_short(skip_receiver_profile);
1645       bind(not_null);
1646     }
1647 
1648     // Record the receiver type.
1649     record_klass_in_profile(receiver, scratch, true);
1650     bind(skip_receiver_profile);
1651 
1652     // The method data pointer needs to be updated to reflect the new target.
1653 #if INCLUDE_JVMCI
1654     if (MethodProfileWidth == 0) {
1655       update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1656     }
1657 #else
1658     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1659 #endif
1660     bind(profile_continue);
1661   }
1662 }
1663 
1664 #if INCLUDE_JVMCI
1665 void InterpreterMacroAssembler::profile_called_method(Register method, Register scratch) {
1666   assert_different_registers(method, scratch);
1667   if (ProfileInterpreter && MethodProfileWidth > 0) {
1668     Label profile_continue;
1669 
1670     // If no method data exists, go to profile_continue.
1671     test_method_data_pointer(profile_continue);
1672 
1673     Label done;
1674     record_item_in_profile_helper(method, scratch, 0, done, MethodProfileWidth,
1675       &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1676     bind(done);
1677 
1678     update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1679     bind(profile_continue);
1680   }
1681 }
1682 #endif // INCLUDE_JVMCI
1683 
1684 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register scratch,
1685                                                                Label& done, bool is_virtual_call) {
1686   if (TypeProfileWidth == 0) {
1687     if (is_virtual_call) {
1688       increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1689     }
1690 #if INCLUDE_JVMCI
1691     else if (EnableJVMCI) {
1692       increment_mdp_data_at(in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()), scratch);
1693     }
1694 #endif
1695   } else {
1696     int non_profiled_offset = -1;
1697     if (is_virtual_call) {
1698       non_profiled_offset = in_bytes(CounterData::count_offset());
1699     }
1700 #if INCLUDE_JVMCI
1701     else if (EnableJVMCI) {
1702       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1703     }
1704 #endif
1705 
1706     record_item_in_profile_helper(receiver, scratch, 0, done, TypeProfileWidth,
1707       &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1708   }
1709 }
1710 
1711 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item,
1712                                           Register scratch, int start_row, Label& done, int total_rows,
1713                                           OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1714                                           int non_profiled_offset) {
1715   int last_row = total_rows - 1;
1716   assert(start_row <= last_row, "must be work left to do");
1717   // Test this row for both the item and for null.
1718   // Take any of three different outcomes:
1719   //   1. found item => increment count and goto done
1720   //   2. found null => keep looking for case 1, maybe allocate this cell
1721   //   3. found something else => keep looking for cases 1 and 2
1722   // Case 3 is handled by a recursive call.
1723   for (int row = start_row; row <= last_row; row++) {
1724     Label next_test;
1725     bool test_for_null_also = (row == start_row);
1726 
1727     // See if the item is item[n].
1728     int item_offset = in_bytes(item_offset_fn(row));
1729     test_mdp_data_at(item_offset, item, next_test, scratch);
1730     // delayed()->tst(scratch);
1731 
1732     // The receiver is item[n].  Increment count[n].
1733     int count_offset = in_bytes(item_count_offset_fn(row));
1734     increment_mdp_data_at(count_offset, scratch);
1735     ba_short(done);
1736     bind(next_test);
1737 
1738     if (test_for_null_also) {
1739       Label found_null;
1740       // Failed the equality check on item[n]...  Test for null.
1741       if (start_row == last_row) {
1742         // The only thing left to do is handle the null case.
1743         if (non_profiled_offset >= 0) {
1744           brx(Assembler::zero, false, Assembler::pn, found_null);
1745           delayed()->nop();
1746           // Item did not match any saved item and there is no empty row for it.
1747           // Increment total counter to indicate polymorphic case.
1748           increment_mdp_data_at(non_profiled_offset, scratch);
1749           ba_short(done);
1750           bind(found_null);
1751         } else {
1752           brx(Assembler::notZero, false, Assembler::pt, done);
1753           delayed()->nop();
1754         }
1755         break;
1756       }
1757       // Since null is rare, make it be the branch-taken case.
1758       brx(Assembler::zero, false, Assembler::pn, found_null);
1759       delayed()->nop();
1760 
1761       // Put all the "Case 3" tests here.
1762       record_item_in_profile_helper(item, scratch, start_row + 1, done, total_rows,
1763         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1764 
1765       // Found a null.  Keep searching for a matching item,
1766       // but remember that this is an empty (unused) slot.
1767       bind(found_null);
1768     }
1769   }
1770 
1771   // In the fall-through case, we found no matching item, but we
1772   // observed the item[start_row] is NULL.
1773 
1774   // Fill in the item field and increment the count.
1775   int item_offset = in_bytes(item_offset_fn(start_row));
1776   set_mdp_data_at(item_offset, item);
1777   int count_offset = in_bytes(item_count_offset_fn(start_row));
1778   mov(DataLayout::counter_increment, scratch);
1779   set_mdp_data_at(count_offset, scratch);
1780   if (start_row > 0) {
1781     ba_short(done);
1782   }
1783 }
1784 
1785 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1786                                                         Register scratch, bool is_virtual_call) {
1787   assert(ProfileInterpreter, "must be profiling");
1788   Label done;
1789 
1790   record_klass_in_profile_helper(receiver, scratch, done, is_virtual_call);
1791 
1792   bind (done);
1793 }
1794 
1795 
1796 // Count a ret in the bytecodes.
1797 
1798 void InterpreterMacroAssembler::profile_ret(TosState state,
1799                                             Register return_bci,
1800                                             Register scratch) {
1801   if (ProfileInterpreter) {
1802     Label profile_continue;
1803     uint row;
1804 
1805     // If no method data exists, go to profile_continue.
1806     test_method_data_pointer(profile_continue);
1807 
1808     // Update the total ret count.
1809     increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1810 
1811     for (row = 0; row < RetData::row_limit(); row++) {
1812       Label next_test;
1813 
1814       // See if return_bci is equal to bci[n]:
1815       test_mdp_data_at(in_bytes(RetData::bci_offset(row)),
1816                        return_bci, next_test, scratch);
1817 
1818       // return_bci is equal to bci[n].  Increment the count.
1819       increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch);
1820 
1821       // The method data pointer needs to be updated to reflect the new target.
1822       update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch);
1823       ba_short(profile_continue);
1824       bind(next_test);
1825     }
1826 
1827     update_mdp_for_ret(state, return_bci);
1828 
1829     bind (profile_continue);
1830   }
1831 }
1832 
1833 // Profile an unexpected null in the bytecodes.
1834 void InterpreterMacroAssembler::profile_null_seen(Register scratch) {
1835   if (ProfileInterpreter) {
1836     Label profile_continue;
1837 
1838     // If no method data exists, go to profile_continue.
1839     test_method_data_pointer(profile_continue);
1840 
1841     set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch);
1842 
1843     // The method data pointer needs to be updated.
1844     int mdp_delta = in_bytes(BitData::bit_data_size());
1845     if (TypeProfileCasts) {
1846       mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1847     }
1848     update_mdp_by_constant(mdp_delta);
1849 
1850     bind (profile_continue);
1851   }
1852 }
1853 
1854 void InterpreterMacroAssembler::profile_typecheck(Register klass,
1855                                                   Register scratch) {
1856   if (ProfileInterpreter) {
1857     Label profile_continue;
1858 
1859     // If no method data exists, go to profile_continue.
1860     test_method_data_pointer(profile_continue);
1861 
1862     int mdp_delta = in_bytes(BitData::bit_data_size());
1863     if (TypeProfileCasts) {
1864       mdp_delta = in_bytes(ReceiverTypeData::receiver_type_data_size());
1865 
1866       // Record the object type.
1867       record_klass_in_profile(klass, scratch, false);
1868     }
1869 
1870     // The method data pointer needs to be updated.
1871     update_mdp_by_constant(mdp_delta);
1872 
1873     bind (profile_continue);
1874   }
1875 }
1876 
1877 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) {
1878   if (ProfileInterpreter && TypeProfileCasts) {
1879     Label profile_continue;
1880 
1881     // If no method data exists, go to profile_continue.
1882     test_method_data_pointer(profile_continue);
1883 
1884     int count_offset = in_bytes(CounterData::count_offset());
1885     // Back up the address, since we have already bumped the mdp.
1886     count_offset -= in_bytes(ReceiverTypeData::receiver_type_data_size());
1887 
1888     // *Decrement* the counter.  We expect to see zero or small negatives.
1889     increment_mdp_data_at(count_offset, scratch, true);
1890 
1891     bind (profile_continue);
1892   }
1893 }
1894 
1895 // Count the default case of a switch construct.
1896 
1897 void InterpreterMacroAssembler::profile_switch_default(Register scratch) {
1898   if (ProfileInterpreter) {
1899     Label profile_continue;
1900 
1901     // If no method data exists, go to profile_continue.
1902     test_method_data_pointer(profile_continue);
1903 
1904     // Update the default case count
1905     increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1906                           scratch);
1907 
1908     // The method data pointer needs to be updated.
1909     update_mdp_by_offset(
1910                     in_bytes(MultiBranchData::default_displacement_offset()),
1911                     scratch);
1912 
1913     bind (profile_continue);
1914   }
1915 }
1916 
1917 // Count the index'th case of a switch construct.
1918 
1919 void InterpreterMacroAssembler::profile_switch_case(Register index,
1920                                                     Register scratch,
1921                                                     Register scratch2,
1922                                                     Register scratch3) {
1923   if (ProfileInterpreter) {
1924     Label profile_continue;
1925 
1926     // If no method data exists, go to profile_continue.
1927     test_method_data_pointer(profile_continue);
1928 
1929     // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1930     set(in_bytes(MultiBranchData::per_case_size()), scratch);
1931     smul(index, scratch, scratch);
1932     add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch);
1933 
1934     // Update the case count
1935     increment_mdp_data_at(scratch,
1936                           in_bytes(MultiBranchData::relative_count_offset()),
1937                           scratch2,
1938                           scratch3);
1939 
1940     // The method data pointer needs to be updated.
1941     update_mdp_by_offset(scratch,
1942                      in_bytes(MultiBranchData::relative_displacement_offset()),
1943                      scratch2);
1944 
1945     bind (profile_continue);
1946   }
1947 }
1948 
1949 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
1950   Label not_null, do_nothing, do_update;
1951 
1952   assert_different_registers(obj, mdo_addr.base(), tmp);
1953 
1954   verify_oop(obj);
1955 
1956   ld_ptr(mdo_addr, tmp);
1957 
1958   br_notnull_short(obj, pt, not_null);
1959   or3(tmp, TypeEntries::null_seen, tmp);
1960   ba_short(do_update);
1961 
1962   bind(not_null);
1963   load_klass(obj, obj);
1964 
1965   xor3(obj, tmp, obj);
1966   btst(TypeEntries::type_klass_mask, obj);
1967   // klass seen before, nothing to do. The unknown bit may have been
1968   // set already but no need to check.
1969   brx(zero, false, pt, do_nothing);
1970   delayed()->
1971 
1972   btst(TypeEntries::type_unknown, obj);
1973   // already unknown. Nothing to do anymore.
1974   brx(notZero, false, pt, do_nothing);
1975   delayed()->
1976 
1977   btst(TypeEntries::type_mask, tmp);
1978   brx(zero, true, pt, do_update);
1979   // first time here. Set profile type.
1980   delayed()->or3(tmp, obj, tmp);
1981 
1982   // different than before. Cannot keep accurate profile.
1983   or3(tmp, TypeEntries::type_unknown, tmp);
1984 
1985   bind(do_update);
1986   // update profile
1987   st_ptr(tmp, mdo_addr);
1988 
1989   bind(do_nothing);
1990 }
1991 
1992 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
1993   if (!ProfileInterpreter) {
1994     return;
1995   }
1996 
1997   assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr);
1998 
1999   if (MethodData::profile_arguments() || MethodData::profile_return()) {
2000     Label profile_continue;
2001 
2002     test_method_data_pointer(profile_continue);
2003 
2004     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
2005 
2006     ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1);
2007     cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue);
2008 
2009     if (MethodData::profile_arguments()) {
2010       Label done;
2011       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
2012       add(ImethodDataPtr, off_to_args, ImethodDataPtr);
2013 
2014       for (int i = 0; i < TypeProfileArgsLimit; i++) {
2015         if (i > 0 || MethodData::profile_return()) {
2016           // If return value type is profiled we may have no argument to profile
2017           ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2018           sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1);
2019           cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done);
2020         }
2021         ld_ptr(Address(callee, Method::const_offset()), tmp1);
2022         lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1);
2023         // stack offset o (zero based) from the start of the argument
2024         // list, for n arguments translates into offset n - o - 1 from
2025         // the end of the argument list. But there's an extra slot at
2026         // the stop of the stack. So the offset is n - o from Lesp.
2027         ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2);
2028         sub(tmp1, tmp2, tmp1);
2029 
2030         // Can't use MacroAssembler::argument_address() which needs Gargs to be set up
2031         sll(tmp1, Interpreter::logStackElementSize, tmp1);
2032         ld_ptr(Lesp, tmp1, tmp1);
2033 
2034         Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
2035         profile_obj_type(tmp1, mdo_arg_addr, tmp2);
2036 
2037         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
2038         add(ImethodDataPtr, to_add, ImethodDataPtr);
2039         off_to_args += to_add;
2040       }
2041 
2042       if (MethodData::profile_return()) {
2043         ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
2044         sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1);
2045       }
2046 
2047       bind(done);
2048 
2049       if (MethodData::profile_return()) {
2050         // We're right after the type profile for the last
2051         // argument. tmp1 is the number of cells left in the
2052         // CallTypeData/VirtualCallTypeData to reach its end. Non null
2053         // if there's a return to profile.
2054         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
2055         sll(tmp1, exact_log2(DataLayout::cell_size), tmp1);
2056         add(ImethodDataPtr, tmp1, ImethodDataPtr);
2057       }
2058     } else {
2059       assert(MethodData::profile_return(), "either profile call args or call ret");
2060       update_mdp_by_constant(in_bytes(TypeEntriesAtCall::return_only_size()));
2061     }
2062 
2063     // mdp points right after the end of the
2064     // CallTypeData/VirtualCallTypeData, right after the cells for the
2065     // return value type if there's one.
2066 
2067     bind(profile_continue);
2068   }
2069 }
2070 
2071 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
2072   assert_different_registers(ret, tmp1, tmp2);
2073   if (ProfileInterpreter && MethodData::profile_return()) {
2074     Label profile_continue, done;
2075 
2076     test_method_data_pointer(profile_continue);
2077 
2078     if (MethodData::profile_return_jsr292_only()) {
2079       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
2080 
2081       // If we don't profile all invoke bytecodes we must make sure
2082       // it's a bytecode we indeed profile. We can't go back to the
2083       // begining of the ProfileData we intend to update to check its
2084       // type because we're right after it and we don't known its
2085       // length.
2086       Label do_profile;
2087       ldub(Lbcp, 0, tmp1);
2088       cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile);
2089       cmp(tmp1, Bytecodes::_invokehandle);
2090       br(equal, false, pn, do_profile);
2091       delayed()->lduh(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1);
2092       cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue);
2093 
2094       bind(do_profile);
2095     }
2096 
2097     Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size()));
2098     mov(ret, tmp1);
2099     profile_obj_type(tmp1, mdo_ret_addr, tmp2);
2100 
2101     bind(profile_continue);
2102   }
2103 }
2104 
2105 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
2106   if (ProfileInterpreter && MethodData::profile_parameters()) {
2107     Label profile_continue, done;
2108 
2109     test_method_data_pointer(profile_continue);
2110 
2111     // Load the offset of the area within the MDO used for
2112     // parameters. If it's negative we're not profiling any parameters.
2113     lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1);
2114     cmp_and_br_short(tmp1, 0, less, pn, profile_continue);
2115 
2116     // Compute a pointer to the area for parameters from the offset
2117     // and move the pointer to the slot for the last
2118     // parameters. Collect profiling from last parameter down.
2119     // mdo start + parameters offset + array length - 1
2120 
2121     // Pointer to the parameter area in the MDO
2122     Register mdp = tmp1;
2123     add(ImethodDataPtr, tmp1, mdp);
2124 
2125     // offset of the current profile entry to update
2126     Register entry_offset = tmp2;
2127     // entry_offset = array len in number of cells
2128     ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset);
2129 
2130     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
2131     assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
2132 
2133     // entry_offset (number of cells)  = array len - size of 1 entry + offset of the stack slot field
2134     sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset);
2135     // entry_offset in bytes
2136     sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset);
2137 
2138     Label loop;
2139     bind(loop);
2140 
2141     // load offset on the stack from the slot for this parameter
2142     ld_ptr(mdp, entry_offset, tmp3);
2143     sll(tmp3,Interpreter::logStackElementSize, tmp3);
2144     neg(tmp3);
2145     // read the parameter from the local area
2146     ld_ptr(Llocals, tmp3, tmp3);
2147 
2148     // make entry_offset now point to the type field for this parameter
2149     int type_base = in_bytes(ParametersTypeData::type_offset(0));
2150     assert(type_base > off_base, "unexpected");
2151     add(entry_offset, type_base - off_base, entry_offset);
2152 
2153     // profile the parameter
2154     Address arg_type(mdp, entry_offset);
2155     profile_obj_type(tmp3, arg_type, tmp4);
2156 
2157     // go to next parameter
2158     sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset);
2159     cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop);
2160 
2161     bind(profile_continue);
2162   }
2163 }
2164 
2165 // add a InterpMonitorElem to stack (see frame_sparc.hpp)
2166 
2167 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty,
2168                                                       Register Rtemp,
2169                                                       Register Rtemp2 ) {
2170 
2171   Register Rlimit = Lmonitors;
2172   const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2173   assert( (delta & LongAlignmentMask) == 0,
2174           "sizeof BasicObjectLock must be even number of doublewords");
2175 
2176   sub( SP,        delta, SP);
2177   sub( Lesp,      delta, Lesp);
2178   sub( Lmonitors, delta, Lmonitors);
2179 
2180   if (!stack_is_empty) {
2181 
2182     // must copy stack contents down
2183 
2184     Label start_copying, next;
2185 
2186     // untested("monitor stack expansion");
2187     compute_stack_base(Rtemp);
2188     ba(start_copying);
2189     delayed()->cmp(Rtemp, Rlimit); // done? duplicated below
2190 
2191     // note: must copy from low memory upwards
2192     // On entry to loop,
2193     // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS)
2194     // Loop mutates Rtemp
2195 
2196     bind( next);
2197 
2198     st_ptr(Rtemp2, Rtemp, 0);
2199     inc(Rtemp, wordSize);
2200     cmp(Rtemp, Rlimit); // are we done? (duplicated above)
2201 
2202     bind( start_copying );
2203 
2204     brx( notEqual, true, pn, next );
2205     delayed()->ld_ptr( Rtemp, delta, Rtemp2 );
2206 
2207     // done copying stack
2208   }
2209 }
2210 
2211 // Locals
2212 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) {
2213   assert_not_delayed();
2214   sll(index, Interpreter::logStackElementSize, index);
2215   sub(Llocals, index, index);
2216   ld_ptr(index, 0, dst);
2217   // Note:  index must hold the effective address--the iinc template uses it
2218 }
2219 
2220 // Just like access_local_ptr but the tag is a returnAddress
2221 void InterpreterMacroAssembler::access_local_returnAddress(Register index,
2222                                                            Register dst ) {
2223   assert_not_delayed();
2224   sll(index, Interpreter::logStackElementSize, index);
2225   sub(Llocals, index, index);
2226   ld_ptr(index, 0, dst);
2227 }
2228 
2229 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) {
2230   assert_not_delayed();
2231   sll(index, Interpreter::logStackElementSize, index);
2232   sub(Llocals, index, index);
2233   ld(index, 0, dst);
2234   // Note:  index must hold the effective address--the iinc template uses it
2235 }
2236 
2237 
2238 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) {
2239   assert_not_delayed();
2240   sll(index, Interpreter::logStackElementSize, index);
2241   sub(Llocals, index, index);
2242   // First half stored at index n+1 (which grows down from Llocals[n])
2243   load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst);
2244 }
2245 
2246 
2247 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) {
2248   assert_not_delayed();
2249   sll(index, Interpreter::logStackElementSize, index);
2250   sub(Llocals, index, index);
2251   ldf(FloatRegisterImpl::S, index, 0, dst);
2252 }
2253 
2254 
2255 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) {
2256   assert_not_delayed();
2257   sll(index, Interpreter::logStackElementSize, index);
2258   sub(Llocals, index, index);
2259   load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst);
2260 }
2261 
2262 
2263 #ifdef ASSERT
2264 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) {
2265   Label L;
2266 
2267   assert(Rindex != Rscratch, "Registers cannot be same");
2268   assert(Rindex != Rscratch1, "Registers cannot be same");
2269   assert(Rlimit != Rscratch, "Registers cannot be same");
2270   assert(Rlimit != Rscratch1, "Registers cannot be same");
2271   assert(Rscratch1 != Rscratch, "Registers cannot be same");
2272 
2273   // untested("reg area corruption");
2274   add(Rindex, offset, Rscratch);
2275   add(Rlimit, 64 + STACK_BIAS, Rscratch1);
2276   cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L);
2277   stop("regsave area is being clobbered");
2278   bind(L);
2279 }
2280 #endif // ASSERT
2281 
2282 
2283 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) {
2284   assert_not_delayed();
2285   sll(index, Interpreter::logStackElementSize, index);
2286   sub(Llocals, index, index);
2287   debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);)
2288   st(src, index, 0);
2289 }
2290 
2291 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) {
2292   assert_not_delayed();
2293   sll(index, Interpreter::logStackElementSize, index);
2294   sub(Llocals, index, index);
2295 #ifdef ASSERT
2296   check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2297 #endif
2298   st_ptr(src, index, 0);
2299 }
2300 
2301 
2302 
2303 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) {
2304   st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n));
2305 }
2306 
2307 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) {
2308   assert_not_delayed();
2309   sll(index, Interpreter::logStackElementSize, index);
2310   sub(Llocals, index, index);
2311 #ifdef ASSERT
2312   check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2313 #endif
2314   store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1
2315 }
2316 
2317 
2318 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) {
2319   assert_not_delayed();
2320   sll(index, Interpreter::logStackElementSize, index);
2321   sub(Llocals, index, index);
2322 #ifdef ASSERT
2323   check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2324 #endif
2325   stf(FloatRegisterImpl::S, src, index, 0);
2326 }
2327 
2328 
2329 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) {
2330   assert_not_delayed();
2331   sll(index, Interpreter::logStackElementSize, index);
2332   sub(Llocals, index, index);
2333 #ifdef ASSERT
2334   check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2335 #endif
2336   store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1));
2337 }
2338 
2339 
2340 int InterpreterMacroAssembler::top_most_monitor_byte_offset() {
2341   const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2342   int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong);
2343   return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS;
2344 }
2345 
2346 
2347 Address InterpreterMacroAssembler::top_most_monitor() {
2348   return Address(FP, top_most_monitor_byte_offset());
2349 }
2350 
2351 
2352 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) {
2353   add( Lesp,      wordSize,                                    Rdest );
2354 }
2355 
2356 #endif /* CC_INTERP */
2357 
2358 void InterpreterMacroAssembler::get_method_counters(Register method,
2359                                                     Register Rcounters,
2360                                                     Label& skip) {
2361   Label has_counters;
2362   Address method_counters(method, in_bytes(Method::method_counters_offset()));
2363   ld_ptr(method_counters, Rcounters);
2364   br_notnull_short(Rcounters, Assembler::pt, has_counters);
2365   call_VM(noreg, CAST_FROM_FN_PTR(address,
2366           InterpreterRuntime::build_method_counters), method);
2367   ld_ptr(method_counters, Rcounters);
2368   br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory
2369   delayed()->nop();
2370   bind(has_counters);
2371 }
2372 
2373 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2374   assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
2375   assert_different_registers(Rcounters, Rtmp, Rtmp2);
2376 
2377   Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2378                                  InvocationCounter::counter_offset());
2379   Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2380                                  InvocationCounter::counter_offset());
2381   int delta = InvocationCounter::count_increment;
2382 
2383   // Load each counter in a register
2384   ld( inv_counter, Rtmp );
2385   ld( be_counter, Rtmp2 );
2386 
2387   assert( is_simm13( delta ), " delta too large.");
2388 
2389   // Add the delta to the invocation counter and store the result
2390   add( Rtmp, delta, Rtmp );
2391 
2392   // Mask the backedge counter
2393   and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2394 
2395   // Store value
2396   st( Rtmp, inv_counter);
2397 
2398   // Add invocation counter + backedge counter
2399   add( Rtmp, Rtmp2, Rtmp);
2400 
2401   // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2402 }
2403 
2404 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2405   assert(UseCompiler, "incrementing must be useful");
2406   assert_different_registers(Rcounters, Rtmp, Rtmp2);
2407 
2408   Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2409                                  InvocationCounter::counter_offset());
2410   Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2411                                  InvocationCounter::counter_offset());
2412 
2413   int delta = InvocationCounter::count_increment;
2414   // Load each counter in a register
2415   ld( be_counter, Rtmp );
2416   ld( inv_counter, Rtmp2 );
2417 
2418   // Add the delta to the backedge counter
2419   add( Rtmp, delta, Rtmp );
2420 
2421   // Mask the invocation counter, add to backedge counter
2422   and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2423 
2424   // and store the result to memory
2425   st( Rtmp, be_counter );
2426 
2427   // Add backedge + invocation counter
2428   add( Rtmp, Rtmp2, Rtmp );
2429 
2430   // Note that this macro must leave backedge_count + invocation_count in Rtmp!
2431 }
2432 
2433 #ifndef CC_INTERP
2434 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count,
2435                                                              Register method_counters,
2436                                                              Register branch_bcp,
2437                                                              Register Rtmp ) {
2438   Label did_not_overflow;
2439   Label overflow_with_error;
2440   assert_different_registers(backedge_count, Rtmp, branch_bcp);
2441   assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
2442 
2443   Address limit(method_counters, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset()));
2444   ld(limit, Rtmp);
2445   cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow);
2446 
2447   // When ProfileInterpreter is on, the backedge_count comes from the
2448   // MethodData*, which value does not get reset on the call to
2449   // frequency_counter_overflow().  To avoid excessive calls to the overflow
2450   // routine while the method is being compiled, add a second test to make sure
2451   // the overflow function is called only once every overflow_frequency.
2452   if (ProfileInterpreter) {
2453     const int overflow_frequency = 1024;
2454     andcc(backedge_count, overflow_frequency-1, Rtmp);
2455     brx(Assembler::notZero, false, Assembler::pt, did_not_overflow);
2456     delayed()->nop();
2457   }
2458 
2459   // overflow in loop, pass branch bytecode
2460   set(6,Rtmp);
2461   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp);
2462 
2463   // Was an OSR adapter generated?
2464   // O0 = osr nmethod
2465   br_null_short(O0, Assembler::pn, overflow_with_error);
2466 
2467   // Has the nmethod been invalidated already?
2468   ldub(O0, nmethod::state_offset(), O2);
2469   cmp_and_br_short(O2, nmethod::in_use, Assembler::notEqual, Assembler::pn, overflow_with_error);
2470 
2471   // migrate the interpreter frame off of the stack
2472 
2473   mov(G2_thread, L7);
2474   // save nmethod
2475   mov(O0, L6);
2476   set_last_Java_frame(SP, noreg);
2477   call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
2478   reset_last_Java_frame();
2479   mov(L7, G2_thread);
2480 
2481   // move OSR nmethod to I1
2482   mov(L6, I1);
2483 
2484   // OSR buffer to I0
2485   mov(O0, I0);
2486 
2487   // remove the interpreter frame
2488   restore(I5_savedSP, 0, SP);
2489 
2490   // Jump to the osr code.
2491   ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
2492   jmp(O2, G0);
2493   delayed()->nop();
2494 
2495   bind(overflow_with_error);
2496 
2497   bind(did_not_overflow);
2498 }
2499 
2500 
2501 
2502 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) {
2503   if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); }
2504 }
2505 
2506 
2507 // local helper function for the verify_oop_or_return_address macro
2508 static bool verify_return_address(Method* m, int bci) {
2509 #ifndef PRODUCT
2510   address pc = (address)(m->constMethod())
2511              + in_bytes(ConstMethod::codes_offset()) + bci;
2512   // assume it is a valid return address if it is inside m and is preceded by a jsr
2513   if (!m->contains(pc))                                          return false;
2514   address jsr_pc;
2515   jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2516   if (*jsr_pc == Bytecodes::_jsr   && jsr_pc >= m->code_base())    return true;
2517   jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2518   if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base())    return true;
2519 #endif // PRODUCT
2520   return false;
2521 }
2522 
2523 
2524 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2525   if (!VerifyOops)  return;
2526   // the VM documentation for the astore[_wide] bytecode allows
2527   // the TOS to be not only an oop but also a return address
2528   Label test;
2529   Label skip;
2530   // See if it is an address (in the current method):
2531 
2532   mov(reg, Rtmp);
2533   const int log2_bytecode_size_limit = 16;
2534   srl(Rtmp, log2_bytecode_size_limit, Rtmp);
2535   br_notnull_short( Rtmp, pt, test );
2536 
2537   // %%% should use call_VM_leaf here?
2538   save_frame_and_mov(0, Lmethod, O0, reg, O1);
2539   save_thread(L7_thread_cache);
2540   call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none);
2541   delayed()->nop();
2542   restore_thread(L7_thread_cache);
2543   br_notnull( O0, false, pt, skip );
2544   delayed()->restore();
2545 
2546   // Perform a more elaborate out-of-line call
2547   // Not an address; verify it:
2548   bind(test);
2549   verify_oop(reg);
2550   bind(skip);
2551 }
2552 
2553 
2554 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2555   if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
2556 }
2557 
2558 
2559 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2560 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2561                                                         int increment, Address mask_addr,
2562                                                         Register scratch1, Register scratch2,
2563                                                         Condition cond, Label *where) {
2564   ld(counter_addr, scratch1);
2565   add(scratch1, increment, scratch1);
2566   ld(mask_addr, scratch2);
2567   andcc(scratch1, scratch2,  G0);
2568   br(cond, false, Assembler::pn, *where);
2569   delayed()->st(scratch1, counter_addr);
2570 }
2571 #endif /* CC_INTERP */
2572 
2573 // Inline assembly for:
2574 //
2575 // if (thread is in interp_only_mode) {
2576 //   InterpreterRuntime::post_method_entry();
2577 // }
2578 // if (DTraceMethodProbes) {
2579 //   SharedRuntime::dtrace_method_entry(method, receiver);
2580 // }
2581 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2582 //   SharedRuntime::rc_trace_method_entry(method, receiver);
2583 // }
2584 
2585 void InterpreterMacroAssembler::notify_method_entry() {
2586 
2587   // C++ interpreter only uses this for native methods.
2588 
2589   // Whenever JVMTI puts a thread in interp_only_mode, method
2590   // entry/exit events are sent for that thread to track stack
2591   // depth.  If it is possible to enter interp_only_mode we add
2592   // the code to check if the event should be sent.
2593   if (JvmtiExport::can_post_interpreter_events()) {
2594     Label L;
2595     Register temp_reg = O5;
2596     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2597     ld(interp_only, temp_reg);
2598     cmp_and_br_short(temp_reg, 0, equal, pt, L);
2599     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2600     bind(L);
2601   }
2602 
2603   {
2604     Register temp_reg = O5;
2605     SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2606     call_VM_leaf(noreg,
2607       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2608       G2_thread, Lmethod);
2609   }
2610 
2611   // RedefineClasses() tracing support for obsolete method entry
2612   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2613     call_VM_leaf(noreg,
2614       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2615       G2_thread, Lmethod);
2616   }
2617 }
2618 
2619 
2620 // Inline assembly for:
2621 //
2622 // if (thread is in interp_only_mode) {
2623 //   // save result
2624 //   InterpreterRuntime::post_method_exit();
2625 //   // restore result
2626 // }
2627 // if (DTraceMethodProbes) {
2628 //   SharedRuntime::dtrace_method_exit(thread, method);
2629 // }
2630 //
2631 // Native methods have their result stored in d_tmp and l_tmp
2632 // Java methods have their result stored in the expression stack
2633 
2634 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method,
2635                                                    TosState state,
2636                                                    NotifyMethodExitMode mode) {
2637   // C++ interpreter only uses this for native methods.
2638 
2639   // Whenever JVMTI puts a thread in interp_only_mode, method
2640   // entry/exit events are sent for that thread to track stack
2641   // depth.  If it is possible to enter interp_only_mode we add
2642   // the code to check if the event should be sent.
2643   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2644     Label L;
2645     Register temp_reg = O5;
2646     const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2647     ld(interp_only, temp_reg);
2648     cmp_and_br_short(temp_reg, 0, equal, pt, L);
2649 
2650     // Note: frame::interpreter_frame_result has a dependency on how the
2651     // method result is saved across the call to post_method_exit. For
2652     // native methods it assumes the result registers are saved to
2653     // l_scratch and d_scratch. If this changes then the interpreter_frame_result
2654     // implementation will need to be updated too.
2655 
2656     save_return_value(state, is_native_method);
2657     call_VM(noreg,
2658             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2659     restore_return_value(state, is_native_method);
2660     bind(L);
2661   }
2662 
2663   {
2664     Register temp_reg = O5;
2665     // Dtrace notification
2666     SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2667     save_return_value(state, is_native_method);
2668     call_VM_leaf(
2669       noreg,
2670       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2671       G2_thread, Lmethod);
2672     restore_return_value(state, is_native_method);
2673   }
2674 }
2675 
2676 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) {
2677 #ifdef CC_INTERP
2678   // result potentially in O0/O1: save it across calls
2679   stf(FloatRegisterImpl::D, F0, STATE(_native_fresult));
2680 #ifdef _LP64
2681   stx(O0, STATE(_native_lresult));
2682 #else
2683   std(O0, STATE(_native_lresult));
2684 #endif
2685 #else // CC_INTERP
2686   if (is_native_call) {
2687     stf(FloatRegisterImpl::D, F0, d_tmp);
2688 #ifdef _LP64
2689     stx(O0, l_tmp);
2690 #else
2691     std(O0, l_tmp);
2692 #endif
2693   } else {
2694     push(state);
2695   }
2696 #endif // CC_INTERP
2697 }
2698 
2699 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) {
2700 #ifdef CC_INTERP
2701   ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0);
2702 #ifdef _LP64
2703   ldx(STATE(_native_lresult), O0);
2704 #else
2705   ldd(STATE(_native_lresult), O0);
2706 #endif
2707 #else // CC_INTERP
2708   if (is_native_call) {
2709     ldf(FloatRegisterImpl::D, d_tmp, F0);
2710 #ifdef _LP64
2711     ldx(l_tmp, O0);
2712 #else
2713     ldd(l_tmp, O0);
2714 #endif
2715   } else {
2716     pop(state);
2717   }
2718 #endif // CC_INTERP
2719 }
--- EOF ---