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