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