Mercurial > hg > graal-compiler
annotate src/cpu/sparc/vm/interp_masm_sparc.cpp @ 6266:1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
6984705: JSR 292 method handle creation should not go through JNI
Summary: remove assembly code for JDK 7 chained method handles
Reviewed-by: jrose, twisti, kvn, mhaupt
Contributed-by: John Rose <john.r.rose@oracle.com>, Christian Thalinger <christian.thalinger@oracle.com>, Michael Haupt <michael.haupt@oracle.com>
| author | twisti |
|---|---|
| date | Tue, 24 Jul 2012 10:51:00 -0700 |
| parents | 2fe087c3e814 |
| children | da91efe96a93 |
| rev | line source |
|---|---|
| 0 | 1 /* |
|
6123
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
2 * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved. |
| 0 | 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 * | |
|
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1506
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1506
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
|
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1506
diff
changeset
|
21 * questions. |
| 0 | 22 * |
| 23 */ | |
| 24 | |
| 1972 | 25 #include "precompiled.hpp" |
| 26 #include "interp_masm_sparc.hpp" | |
| 27 #include "interpreter/interpreter.hpp" | |
| 28 #include "interpreter/interpreterRuntime.hpp" | |
| 29 #include "oops/arrayOop.hpp" | |
| 30 #include "oops/markOop.hpp" | |
| 31 #include "oops/methodDataOop.hpp" | |
| 32 #include "oops/methodOop.hpp" | |
| 33 #include "prims/jvmtiExport.hpp" | |
| 34 #include "prims/jvmtiRedefineClassesTrace.hpp" | |
| 35 #include "prims/jvmtiThreadState.hpp" | |
| 36 #include "runtime/basicLock.hpp" | |
| 37 #include "runtime/biasedLocking.hpp" | |
| 38 #include "runtime/sharedRuntime.hpp" | |
| 39 #ifdef TARGET_OS_FAMILY_linux | |
| 40 # include "thread_linux.inline.hpp" | |
| 41 #endif | |
| 42 #ifdef TARGET_OS_FAMILY_solaris | |
| 43 # include "thread_solaris.inline.hpp" | |
| 44 #endif | |
| 0 | 45 |
| 46 #ifndef CC_INTERP | |
| 47 #ifndef FAST_DISPATCH | |
| 48 #define FAST_DISPATCH 1 | |
| 49 #endif | |
| 50 #undef FAST_DISPATCH | |
| 51 | |
| 52 // Implementation of InterpreterMacroAssembler | |
| 53 | |
| 54 // This file specializes the assember with interpreter-specific macros | |
| 55 | |
| 727 | 56 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); |
| 57 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); | |
| 0 | 58 |
| 59 #else // CC_INTERP | |
| 60 #ifndef STATE | |
| 61 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name)) | |
| 62 #endif // STATE | |
| 63 | |
| 64 #endif // CC_INTERP | |
| 65 | |
| 66 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { | |
| 67 // Note: this algorithm is also used by C1's OSR entry sequence. | |
| 68 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). | |
| 69 assert_different_registers(args_size, locals_size); | |
| 70 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. | |
| 71 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words | |
| 72 // Use br/mov combination because it works on both V8 and V9 and is | |
| 73 // faster. | |
| 74 Label skip_move; | |
| 75 br(Assembler::negative, true, Assembler::pt, skip_move); | |
| 76 delayed()->mov(G0, delta); | |
| 77 bind(skip_move); | |
| 78 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) | |
| 79 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes | |
| 80 } | |
| 81 | |
| 82 #ifndef CC_INTERP | |
| 83 | |
| 84 // Dispatch code executed in the prolog of a bytecode which does not do it's | |
| 85 // own dispatch. The dispatch address is computed and placed in IdispatchAddress | |
| 86 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { | |
| 87 assert_not_delayed(); | |
| 88 #ifdef FAST_DISPATCH | |
| 89 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since | |
| 90 // they both use I2. | |
| 91 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); | |
| 92 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
| 93 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); | |
| 94 // add offset to correct dispatch table | |
| 95 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
| 96 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr | |
| 97 #else | |
| 727 | 98 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode |
| 0 | 99 // dispatch table to use |
| 727 | 100 AddressLiteral tbl(Interpreter::dispatch_table(state)); |
| 101 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
| 102 set(tbl, G3_scratch); // compute addr of table | |
| 103 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr | |
| 0 | 104 #endif |
| 105 } | |
| 106 | |
| 107 | |
| 108 // Dispatch code executed in the epilog of a bytecode which does not do it's | |
| 109 // own dispatch. The dispatch address in IdispatchAddress is used for the | |
| 110 // dispatch. | |
| 111 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { | |
| 112 assert_not_delayed(); | |
| 113 verify_FPU(1, state); | |
| 114 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 115 jmp( IdispatchAddress, 0 ); | |
| 116 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); | |
| 117 else delayed()->nop(); | |
| 118 } | |
| 119 | |
| 120 | |
| 121 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { | |
| 122 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) | |
| 123 assert_not_delayed(); | |
| 124 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
| 125 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr); | |
| 126 } | |
| 127 | |
| 128 | |
| 129 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { | |
| 130 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) | |
| 131 assert_not_delayed(); | |
| 132 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
| 133 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); | |
| 134 } | |
| 135 | |
| 136 | |
| 137 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { | |
| 138 // load current bytecode | |
| 139 assert_not_delayed(); | |
| 140 ldub( Lbcp, 0, Lbyte_code); // load next bytecode | |
| 141 dispatch_base(state, table); | |
| 142 } | |
| 143 | |
| 144 | |
| 145 void InterpreterMacroAssembler::call_VM_leaf_base( | |
| 146 Register java_thread, | |
| 147 address entry_point, | |
| 148 int number_of_arguments | |
| 149 ) { | |
| 150 if (!java_thread->is_valid()) | |
| 151 java_thread = L7_thread_cache; | |
| 152 // super call | |
| 153 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); | |
| 154 } | |
| 155 | |
| 156 | |
| 157 void InterpreterMacroAssembler::call_VM_base( | |
| 158 Register oop_result, | |
| 159 Register java_thread, | |
| 160 Register last_java_sp, | |
| 161 address entry_point, | |
| 162 int number_of_arguments, | |
| 163 bool check_exception | |
| 164 ) { | |
| 165 if (!java_thread->is_valid()) | |
| 166 java_thread = L7_thread_cache; | |
| 167 // See class ThreadInVMfromInterpreter, which assumes that the interpreter | |
| 168 // takes responsibility for setting its own thread-state on call-out. | |
| 169 // However, ThreadInVMfromInterpreter resets the state to "in_Java". | |
| 170 | |
| 171 //save_bcp(); // save bcp | |
| 172 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); | |
| 173 //restore_bcp(); // restore bcp | |
| 174 //restore_locals(); // restore locals pointer | |
| 175 } | |
| 176 | |
| 177 | |
| 178 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { | |
| 179 if (JvmtiExport::can_pop_frame()) { | |
| 180 Label L; | |
| 181 | |
| 182 // Check the "pending popframe condition" flag in the current thread | |
| 727 | 183 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg); |
| 0 | 184 |
| 185 // Initiate popframe handling only if it is not already being processed. If the flag | |
| 186 // has the popframe_processing bit set, it means that this code is called *during* popframe | |
| 187 // handling - we don't want to reenter. | |
| 188 btst(JavaThread::popframe_pending_bit, scratch_reg); | |
| 189 br(zero, false, pt, L); | |
| 190 delayed()->nop(); | |
| 191 btst(JavaThread::popframe_processing_bit, scratch_reg); | |
| 192 br(notZero, false, pt, L); | |
| 193 delayed()->nop(); | |
| 194 | |
| 195 // Call Interpreter::remove_activation_preserving_args_entry() to get the | |
| 196 // address of the same-named entrypoint in the generated interpreter code. | |
| 197 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); | |
| 198 | |
| 199 // Jump to Interpreter::_remove_activation_preserving_args_entry | |
| 200 jmpl(O0, G0, G0); | |
| 201 delayed()->nop(); | |
| 202 bind(L); | |
| 203 } | |
| 204 } | |
| 205 | |
| 206 | |
| 207 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { | |
| 208 Register thr_state = G4_scratch; | |
| 727 | 209 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); |
| 210 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset()); | |
| 211 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset()); | |
| 212 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset()); | |
| 0 | 213 switch (state) { |
| 214 case ltos: ld_long(val_addr, Otos_l); break; | |
| 215 case atos: ld_ptr(oop_addr, Otos_l); | |
| 216 st_ptr(G0, oop_addr); break; | |
| 217 case btos: // fall through | |
| 218 case ctos: // fall through | |
| 219 case stos: // fall through | |
| 220 case itos: ld(val_addr, Otos_l1); break; | |
| 221 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; | |
| 222 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; | |
| 223 case vtos: /* nothing to do */ break; | |
| 224 default : ShouldNotReachHere(); | |
| 225 } | |
| 226 // Clean up tos value in the jvmti thread state | |
| 227 or3(G0, ilgl, G3_scratch); | |
| 228 stw(G3_scratch, tos_addr); | |
| 229 st_long(G0, val_addr); | |
| 230 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 231 } | |
| 232 | |
| 233 | |
| 234 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { | |
| 235 if (JvmtiExport::can_force_early_return()) { | |
| 236 Label L; | |
| 237 Register thr_state = G3_scratch; | |
| 727 | 238 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); |
| 3839 | 239 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; |
| 0 | 240 |
| 241 // Initiate earlyret handling only if it is not already being processed. | |
| 242 // If the flag has the earlyret_processing bit set, it means that this code | |
| 243 // is called *during* earlyret handling - we don't want to reenter. | |
| 727 | 244 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch); |
| 3839 | 245 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L); |
| 0 | 246 |
| 247 // Call Interpreter::remove_activation_early_entry() to get the address of the | |
| 248 // same-named entrypoint in the generated interpreter code | |
| 727 | 249 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1); |
| 0 | 250 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); |
| 251 | |
| 252 // Jump to Interpreter::_remove_activation_early_entry | |
| 253 jmpl(O0, G0, G0); | |
| 254 delayed()->nop(); | |
| 255 bind(L); | |
| 256 } | |
| 257 } | |
| 258 | |
| 259 | |
| 1295 | 260 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { |
| 0 | 261 mov(arg_1, O0); |
| 1295 | 262 mov(arg_2, O1); |
| 263 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2); | |
| 0 | 264 } |
| 265 #endif /* CC_INTERP */ | |
| 266 | |
| 267 | |
| 268 #ifndef CC_INTERP | |
| 269 | |
| 270 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { | |
| 271 assert_not_delayed(); | |
| 272 dispatch_Lbyte_code(state, table); | |
| 273 } | |
| 274 | |
| 275 | |
| 276 void InterpreterMacroAssembler::dispatch_normal(TosState state) { | |
| 277 dispatch_base(state, Interpreter::normal_table(state)); | |
| 278 } | |
| 279 | |
| 280 | |
| 281 void InterpreterMacroAssembler::dispatch_only(TosState state) { | |
| 282 dispatch_base(state, Interpreter::dispatch_table(state)); | |
| 283 } | |
| 284 | |
| 285 | |
| 286 // common code to dispatch and dispatch_only | |
| 287 // dispatch value in Lbyte_code and increment Lbcp | |
| 288 | |
| 289 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) { | |
| 290 verify_FPU(1, state); | |
| 291 // %%%%% maybe implement +VerifyActivationFrameSize here | |
| 292 //verify_thread(); //too slow; we will just verify on method entry & exit | |
| 293 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 294 #ifdef FAST_DISPATCH | |
| 295 if (table == Interpreter::dispatch_table(state)) { | |
| 296 // use IdispatchTables | |
| 297 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); | |
| 298 // add offset to correct dispatch table | |
| 299 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
| 300 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr | |
| 301 } else { | |
| 302 #endif | |
| 303 // dispatch table to use | |
| 727 | 304 AddressLiteral tbl(table); |
| 0 | 305 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize |
| 727 | 306 set(tbl, G3_scratch); // compute addr of table |
| 0 | 307 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr |
| 308 #ifdef FAST_DISPATCH | |
| 309 } | |
| 310 #endif | |
| 311 jmp( G3_scratch, 0 ); | |
| 312 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); | |
| 313 else delayed()->nop(); | |
| 314 } | |
| 315 | |
| 316 | |
| 317 // Helpers for expression stack | |
| 318 | |
| 319 // Longs and doubles are Category 2 computational types in the | |
| 320 // JVM specification (section 3.11.1) and take 2 expression stack or | |
| 321 // local slots. | |
| 322 // Aligning them on 32 bit with tagged stacks is hard because the code generated | |
| 323 // for the dup* bytecodes depends on what types are already on the stack. | |
| 324 // If the types are split into the two stack/local slots, that is much easier | |
| 325 // (and we can use 0 for non-reference tags). | |
| 326 | |
| 327 // Known good alignment in _LP64 but unknown otherwise | |
| 328 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { | |
| 329 assert_not_delayed(); | |
| 330 | |
| 331 #ifdef _LP64 | |
| 332 ldf(FloatRegisterImpl::D, r1, offset, d); | |
| 333 #else | |
| 334 ldf(FloatRegisterImpl::S, r1, offset, d); | |
| 1506 | 335 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor()); |
| 0 | 336 #endif |
| 337 } | |
| 338 | |
| 339 // Known good alignment in _LP64 but unknown otherwise | |
| 340 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { | |
| 341 assert_not_delayed(); | |
| 342 | |
| 343 #ifdef _LP64 | |
| 344 stf(FloatRegisterImpl::D, d, r1, offset); | |
| 345 // store something more useful here | |
| 1506 | 346 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) |
| 0 | 347 #else |
| 348 stf(FloatRegisterImpl::S, d, r1, offset); | |
| 1506 | 349 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize); |
| 0 | 350 #endif |
| 351 } | |
| 352 | |
| 353 | |
| 354 // Known good alignment in _LP64 but unknown otherwise | |
| 355 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { | |
| 356 assert_not_delayed(); | |
| 357 #ifdef _LP64 | |
| 358 ldx(r1, offset, rd); | |
| 359 #else | |
| 360 ld(r1, offset, rd); | |
| 1506 | 361 ld(r1, offset + Interpreter::stackElementSize, rd->successor()); |
| 0 | 362 #endif |
| 363 } | |
| 364 | |
| 365 // Known good alignment in _LP64 but unknown otherwise | |
| 366 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { | |
| 367 assert_not_delayed(); | |
| 368 | |
| 369 #ifdef _LP64 | |
| 370 stx(l, r1, offset); | |
| 371 // store something more useful here | |
| 1506 | 372 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);) |
| 0 | 373 #else |
| 374 st(l, r1, offset); | |
| 1506 | 375 st(l->successor(), r1, offset + Interpreter::stackElementSize); |
| 0 | 376 #endif |
| 377 } | |
| 378 | |
| 379 void InterpreterMacroAssembler::pop_i(Register r) { | |
| 380 assert_not_delayed(); | |
| 381 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
| 1506 | 382 inc(Lesp, Interpreter::stackElementSize); |
| 0 | 383 debug_only(verify_esp(Lesp)); |
| 384 } | |
| 385 | |
| 386 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { | |
| 387 assert_not_delayed(); | |
| 388 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
| 1506 | 389 inc(Lesp, Interpreter::stackElementSize); |
| 0 | 390 debug_only(verify_esp(Lesp)); |
| 391 } | |
| 392 | |
| 393 void InterpreterMacroAssembler::pop_l(Register r) { | |
| 394 assert_not_delayed(); | |
| 395 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
| 1506 | 396 inc(Lesp, 2*Interpreter::stackElementSize); |
| 0 | 397 debug_only(verify_esp(Lesp)); |
| 398 } | |
| 399 | |
| 400 | |
| 401 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { | |
| 402 assert_not_delayed(); | |
| 403 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); | |
| 1506 | 404 inc(Lesp, Interpreter::stackElementSize); |
| 0 | 405 debug_only(verify_esp(Lesp)); |
| 406 } | |
| 407 | |
| 408 | |
| 409 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { | |
| 410 assert_not_delayed(); | |
| 411 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); | |
| 1506 | 412 inc(Lesp, 2*Interpreter::stackElementSize); |
| 0 | 413 debug_only(verify_esp(Lesp)); |
| 414 } | |
| 415 | |
| 416 | |
| 417 void InterpreterMacroAssembler::push_i(Register r) { | |
| 418 assert_not_delayed(); | |
| 419 debug_only(verify_esp(Lesp)); | |
| 1506 | 420 st(r, Lesp, 0); |
| 421 dec(Lesp, Interpreter::stackElementSize); | |
| 0 | 422 } |
| 423 | |
| 424 void InterpreterMacroAssembler::push_ptr(Register r) { | |
| 425 assert_not_delayed(); | |
| 1506 | 426 st_ptr(r, Lesp, 0); |
| 427 dec(Lesp, Interpreter::stackElementSize); | |
| 0 | 428 } |
| 429 | |
| 430 // remember: our convention for longs in SPARC is: | |
| 431 // O0 (Otos_l1) has high-order part in first word, | |
| 432 // O1 (Otos_l2) has low-order part in second word | |
| 433 | |
| 434 void InterpreterMacroAssembler::push_l(Register r) { | |
| 435 assert_not_delayed(); | |
| 436 debug_only(verify_esp(Lesp)); | |
| 1506 | 437 // Longs are stored in memory-correct order, even if unaligned. |
| 438 int offset = -Interpreter::stackElementSize; | |
| 0 | 439 store_unaligned_long(r, Lesp, offset); |
| 1506 | 440 dec(Lesp, 2 * Interpreter::stackElementSize); |
| 0 | 441 } |
| 442 | |
| 443 | |
| 444 void InterpreterMacroAssembler::push_f(FloatRegister f) { | |
| 445 assert_not_delayed(); | |
| 446 debug_only(verify_esp(Lesp)); | |
| 1506 | 447 stf(FloatRegisterImpl::S, f, Lesp, 0); |
| 448 dec(Lesp, Interpreter::stackElementSize); | |
| 0 | 449 } |
| 450 | |
| 451 | |
| 452 void InterpreterMacroAssembler::push_d(FloatRegister d) { | |
| 453 assert_not_delayed(); | |
| 454 debug_only(verify_esp(Lesp)); | |
| 1506 | 455 // Longs are stored in memory-correct order, even if unaligned. |
| 456 int offset = -Interpreter::stackElementSize; | |
| 0 | 457 store_unaligned_double(d, Lesp, offset); |
| 1506 | 458 dec(Lesp, 2 * Interpreter::stackElementSize); |
| 0 | 459 } |
| 460 | |
| 461 | |
| 462 void InterpreterMacroAssembler::push(TosState state) { | |
| 463 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 464 switch (state) { | |
| 465 case atos: push_ptr(); break; | |
| 466 case btos: push_i(); break; | |
| 467 case ctos: | |
| 468 case stos: push_i(); break; | |
| 469 case itos: push_i(); break; | |
| 470 case ltos: push_l(); break; | |
| 471 case ftos: push_f(); break; | |
| 472 case dtos: push_d(); break; | |
| 473 case vtos: /* nothing to do */ break; | |
| 474 default : ShouldNotReachHere(); | |
| 475 } | |
| 476 } | |
| 477 | |
| 478 | |
| 479 void InterpreterMacroAssembler::pop(TosState state) { | |
| 480 switch (state) { | |
| 481 case atos: pop_ptr(); break; | |
| 482 case btos: pop_i(); break; | |
| 483 case ctos: | |
| 484 case stos: pop_i(); break; | |
| 485 case itos: pop_i(); break; | |
| 486 case ltos: pop_l(); break; | |
| 487 case ftos: pop_f(); break; | |
| 488 case dtos: pop_d(); break; | |
| 489 case vtos: /* nothing to do */ break; | |
| 490 default : ShouldNotReachHere(); | |
| 491 } | |
| 492 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 493 } | |
| 494 | |
| 495 | |
| 1506 | 496 // Helpers for swap and dup |
| 497 void InterpreterMacroAssembler::load_ptr(int n, Register val) { | |
| 0 | 498 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); |
| 499 } | |
| 1506 | 500 void InterpreterMacroAssembler::store_ptr(int n, Register val) { |
| 0 | 501 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); |
| 502 } | |
| 503 | |
| 504 | |
| 505 void InterpreterMacroAssembler::load_receiver(Register param_count, | |
| 506 Register recv) { | |
| 1506 | 507 sll(param_count, Interpreter::logStackElementSize, param_count); |
|
6266
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
508 ld_ptr(Lesp, param_count, recv); // gets receiver oop |
| 0 | 509 } |
| 510 | |
| 511 void InterpreterMacroAssembler::empty_expression_stack() { | |
| 512 // Reset Lesp. | |
| 513 sub( Lmonitors, wordSize, Lesp ); | |
| 514 | |
| 515 // Reset SP by subtracting more space from Lesp. | |
| 516 Label done; | |
| 517 verify_oop(Lmethod); | |
| 727 | 518 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!"); |
| 0 | 519 |
| 520 // A native does not need to do this, since its callee does not change SP. | |
| 727 | 521 ld(Lmethod, methodOopDesc::access_flags_offset(), Gframe_size); // Load access flags. |
| 0 | 522 btst(JVM_ACC_NATIVE, Gframe_size); |
| 523 br(Assembler::notZero, false, Assembler::pt, done); | |
| 524 delayed()->nop(); | |
| 525 | |
| 526 // Compute max expression stack+register save area | |
| 727 | 527 lduh(Lmethod, in_bytes(methodOopDesc::max_stack_offset()), Gframe_size); // Load max stack. |
| 0 | 528 add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size ); |
| 529 | |
| 530 // | |
| 531 // now set up a stack frame with the size computed above | |
| 532 // | |
| 533 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below | |
| 534 sll( Gframe_size, LogBytesPerWord, Gframe_size ); | |
| 535 sub( Lesp, Gframe_size, Gframe_size ); | |
| 536 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary | |
| 537 debug_only(verify_sp(Gframe_size, G4_scratch)); | |
| 538 #ifdef _LP64 | |
| 539 sub(Gframe_size, STACK_BIAS, Gframe_size ); | |
| 540 #endif | |
| 541 mov(Gframe_size, SP); | |
| 542 | |
| 543 bind(done); | |
| 544 } | |
| 545 | |
| 546 | |
| 547 #ifdef ASSERT | |
| 548 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { | |
| 549 Label Bad, OK; | |
| 550 | |
| 551 // Saved SP must be aligned. | |
| 552 #ifdef _LP64 | |
| 553 btst(2*BytesPerWord-1, Rsp); | |
| 554 #else | |
| 555 btst(LongAlignmentMask, Rsp); | |
| 556 #endif | |
| 557 br(Assembler::notZero, false, Assembler::pn, Bad); | |
| 558 delayed()->nop(); | |
| 559 | |
| 560 // Saved SP, plus register window size, must not be above FP. | |
| 561 add(Rsp, frame::register_save_words * wordSize, Rtemp); | |
| 562 #ifdef _LP64 | |
| 563 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP | |
| 564 #endif | |
| 3839 | 565 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad); |
| 0 | 566 |
| 567 // Saved SP must not be ridiculously below current SP. | |
| 568 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); | |
| 569 set(maxstack, Rtemp); | |
| 570 sub(SP, Rtemp, Rtemp); | |
| 571 #ifdef _LP64 | |
| 572 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp | |
| 573 #endif | |
| 3839 | 574 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad); |
| 575 | |
| 576 ba_short(OK); | |
| 0 | 577 |
| 578 bind(Bad); | |
| 579 stop("on return to interpreted call, restored SP is corrupted"); | |
| 580 | |
| 581 bind(OK); | |
| 582 } | |
| 583 | |
| 584 | |
| 585 void InterpreterMacroAssembler::verify_esp(Register Resp) { | |
| 586 // about to read or write Resp[0] | |
| 587 // make sure it is not in the monitors or the register save area | |
| 588 Label OK1, OK2; | |
| 589 | |
| 590 cmp(Resp, Lmonitors); | |
| 591 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); | |
| 592 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); | |
| 593 stop("too many pops: Lesp points into monitor area"); | |
| 594 bind(OK1); | |
| 595 #ifdef _LP64 | |
| 596 sub(Resp, STACK_BIAS, Resp); | |
| 597 #endif | |
| 598 cmp(Resp, SP); | |
| 599 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); | |
| 600 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); | |
| 601 stop("too many pushes: Lesp points into register window"); | |
| 602 bind(OK2); | |
| 603 } | |
| 604 #endif // ASSERT | |
| 605 | |
| 606 // Load compiled (i2c) or interpreter entry when calling from interpreted and | |
| 607 // do the call. Centralized so that all interpreter calls will do the same actions. | |
| 608 // If jvmti single stepping is on for a thread we must not call compiled code. | |
| 609 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { | |
| 610 | |
| 611 // Assume we want to go compiled if available | |
| 612 | |
| 613 ld_ptr(G5_method, in_bytes(methodOopDesc::from_interpreted_offset()), target); | |
| 614 | |
| 615 if (JvmtiExport::can_post_interpreter_events()) { | |
| 616 // JVMTI events, such as single-stepping, are implemented partly by avoiding running | |
| 617 // compiled code in threads for which the event is enabled. Check here for | |
| 618 // interp_only_mode if these events CAN be enabled. | |
| 619 verify_thread(); | |
| 620 Label skip_compiled_code; | |
| 621 | |
| 727 | 622 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
| 0 | 623 ld(interp_only, scratch); |
| 3839 | 624 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn); |
| 0 | 625 delayed()->ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), target); |
| 626 bind(skip_compiled_code); | |
| 627 } | |
| 628 | |
| 629 // the i2c_adapters need methodOop in G5_method (right? %%%) | |
| 630 // do the call | |
| 631 #ifdef ASSERT | |
| 632 { | |
| 633 Label ok; | |
| 3839 | 634 br_notnull_short(target, Assembler::pt, ok); |
| 0 | 635 stop("null entry point"); |
| 636 bind(ok); | |
| 637 } | |
| 638 #endif // ASSERT | |
| 639 | |
| 640 // Adjust Rret first so Llast_SP can be same as Rret | |
| 641 add(Rret, -frame::pc_return_offset, O7); | |
| 642 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer | |
| 643 // Record SP so we can remove any stack space allocated by adapter transition | |
| 644 jmp(target, 0); | |
| 645 delayed()->mov(SP, Llast_SP); | |
| 646 } | |
| 647 | |
| 648 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { | |
| 649 assert_not_delayed(); | |
| 650 | |
| 651 Label not_taken; | |
| 652 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); | |
| 653 else br (cc, false, Assembler::pn, not_taken); | |
| 654 delayed()->nop(); | |
| 655 | |
| 656 TemplateTable::branch(false,false); | |
| 657 | |
| 658 bind(not_taken); | |
| 659 | |
| 660 profile_not_taken_branch(G3_scratch); | |
| 661 } | |
| 662 | |
| 663 | |
| 664 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( | |
| 665 int bcp_offset, | |
| 666 Register Rtmp, | |
| 667 Register Rdst, | |
| 668 signedOrNot is_signed, | |
| 669 setCCOrNot should_set_CC ) { | |
| 670 assert(Rtmp != Rdst, "need separate temp register"); | |
| 671 assert_not_delayed(); | |
| 672 switch (is_signed) { | |
| 673 default: ShouldNotReachHere(); | |
| 674 | |
| 675 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte | |
| 676 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte | |
| 677 } | |
| 678 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte | |
| 679 sll( Rdst, BitsPerByte, Rdst); | |
| 680 switch (should_set_CC ) { | |
| 681 default: ShouldNotReachHere(); | |
| 682 | |
| 683 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; | |
| 684 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; | |
| 685 } | |
| 686 } | |
| 687 | |
| 688 | |
| 689 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( | |
| 690 int bcp_offset, | |
| 691 Register Rtmp, | |
| 692 Register Rdst, | |
| 693 setCCOrNot should_set_CC ) { | |
| 694 assert(Rtmp != Rdst, "need separate temp register"); | |
| 695 assert_not_delayed(); | |
| 696 add( Lbcp, bcp_offset, Rtmp); | |
| 697 andcc( Rtmp, 3, G0); | |
| 698 Label aligned; | |
| 699 switch (should_set_CC ) { | |
| 700 default: ShouldNotReachHere(); | |
| 701 | |
| 702 case set_CC: break; | |
| 703 case dont_set_CC: break; | |
| 704 } | |
| 705 | |
| 706 br(Assembler::zero, true, Assembler::pn, aligned); | |
| 707 #ifdef _LP64 | |
| 708 delayed()->ldsw(Rtmp, 0, Rdst); | |
| 709 #else | |
| 710 delayed()->ld(Rtmp, 0, Rdst); | |
| 711 #endif | |
| 712 | |
| 713 ldub(Lbcp, bcp_offset + 3, Rdst); | |
| 714 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); | |
| 715 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); | |
| 716 #ifdef _LP64 | |
| 717 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); | |
| 718 #else | |
| 719 // Unsigned load is faster than signed on some implementations | |
| 720 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); | |
| 721 #endif | |
| 722 or3(Rtmp, Rdst, Rdst ); | |
| 723 | |
| 724 bind(aligned); | |
| 725 if (should_set_CC == set_CC) tst(Rdst); | |
| 726 } | |
| 727 | |
| 728 | |
| 1503 | 729 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register cache, Register tmp, |
| 1565 | 730 int bcp_offset, size_t index_size) { |
| 1503 | 731 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); |
| 1565 | 732 if (index_size == sizeof(u2)) { |
| 1503 | 733 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); |
| 1565 | 734 } else if (index_size == sizeof(u4)) { |
|
2416
38fea01eb669
6817525: turn on method handle functionality by default for JSR 292
twisti
parents:
2118
diff
changeset
|
735 assert(EnableInvokeDynamic, "giant index used only for JSR 292"); |
| 1503 | 736 get_4_byte_integer_at_bcp(bcp_offset, cache, tmp); |
| 737 assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line"); | |
| 738 xor3(tmp, -1, tmp); // convert to plain index | |
| 1565 | 739 } else if (index_size == sizeof(u1)) { |
|
2416
38fea01eb669
6817525: turn on method handle functionality by default for JSR 292
twisti
parents:
2118
diff
changeset
|
740 assert(EnableInvokeDynamic, "tiny index used only for JSR 292"); |
| 1565 | 741 ldub(Lbcp, bcp_offset, tmp); |
| 742 } else { | |
| 743 ShouldNotReachHere(); | |
| 1503 | 744 } |
| 745 } | |
| 746 | |
| 747 | |
| 748 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp, | |
| 1565 | 749 int bcp_offset, size_t index_size) { |
| 0 | 750 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); |
| 751 assert_different_registers(cache, tmp); | |
| 752 assert_not_delayed(); | |
| 1565 | 753 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size); |
| 1503 | 754 // convert from field index to ConstantPoolCacheEntry index and from |
| 755 // word index to byte offset | |
| 0 | 756 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); |
| 757 add(LcpoolCache, tmp, cache); | |
| 758 } | |
| 759 | |
| 760 | |
|
3852
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
761 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache, |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
762 Register temp, |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
763 Register bytecode, |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
764 int byte_no, |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
765 int bcp_offset, |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
766 size_t index_size) { |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
767 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size); |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
768 ld_ptr(cache, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode); |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
769 const int shift_count = (1 + byte_no) * BitsPerByte; |
|
6266
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
770 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) || |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
771 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift), |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
772 "correct shift count"); |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
773 srl(bytecode, shift_count, bytecode); |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
774 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask"); |
|
1d7922586cf6
7023639: JSR 292 method handle invocation needs a fast path for compiled code
twisti
parents:
6123
diff
changeset
|
775 and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode); |
|
3852
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
776 } |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
777 |
|
fdb992d83a87
7071653: JSR 292: call site change notification should be pushed not pulled
twisti
parents:
3839
diff
changeset
|
778 |
| 1503 | 779 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, |
| 1565 | 780 int bcp_offset, size_t index_size) { |
| 0 | 781 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); |
| 782 assert_different_registers(cache, tmp); | |
| 783 assert_not_delayed(); | |
| 1565 | 784 if (index_size == sizeof(u2)) { |
| 785 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); | |
| 786 } else { | |
| 787 ShouldNotReachHere(); // other sizes not supported here | |
| 788 } | |
| 0 | 789 // convert from field index to ConstantPoolCacheEntry index |
| 790 // and from word index to byte offset | |
| 791 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); | |
| 792 // skip past the header | |
| 793 add(tmp, in_bytes(constantPoolCacheOopDesc::base_offset()), tmp); | |
| 794 // construct pointer to cache entry | |
| 795 add(LcpoolCache, tmp, cache); | |
| 796 } | |
| 797 | |
| 798 | |
| 799 // Generate a subtype check: branch to ok_is_subtype if sub_klass is | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
800 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. |
| 0 | 801 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, |
| 802 Register Rsuper_klass, | |
| 803 Register Rtmp1, | |
| 804 Register Rtmp2, | |
| 805 Register Rtmp3, | |
| 806 Label &ok_is_subtype ) { | |
|
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
807 Label not_subtype; |
| 0 | 808 |
| 809 // Profile the not-null value's klass. | |
| 810 profile_typecheck(Rsub_klass, Rtmp1); | |
| 811 | |
|
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
812 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
813 Rtmp1, Rtmp2, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
814 &ok_is_subtype, ¬_subtype, NULL); |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
815 |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
816 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
817 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
818 &ok_is_subtype, NULL); |
| 0 | 819 |
| 820 bind(not_subtype); | |
| 821 profile_typecheck_failed(Rtmp1); | |
| 822 } | |
| 823 | |
| 824 // Separate these two to allow for delay slot in middle | |
| 825 // These are used to do a test and full jump to exception-throwing code. | |
| 826 | |
| 827 // %%%%% Could possibly reoptimize this by testing to see if could use | |
| 828 // a single conditional branch (i.e. if span is small enough. | |
| 829 // If you go that route, than get rid of the split and give up | |
| 830 // on the delay-slot hack. | |
| 831 | |
| 832 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, | |
| 833 Label& ok ) { | |
| 834 assert_not_delayed(); | |
| 835 br(ok_condition, true, pt, ok); | |
| 836 // DELAY SLOT | |
| 837 } | |
| 838 | |
| 839 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, | |
| 840 Label& ok ) { | |
| 841 assert_not_delayed(); | |
| 842 bp( ok_condition, true, Assembler::xcc, pt, ok); | |
| 843 // DELAY SLOT | |
| 844 } | |
| 845 | |
| 846 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, | |
| 847 Label& ok ) { | |
| 848 assert_not_delayed(); | |
| 849 brx(ok_condition, true, pt, ok); | |
| 850 // DELAY SLOT | |
| 851 } | |
| 852 | |
| 853 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, | |
| 854 Register Rscratch, | |
| 855 Label& ok ) { | |
| 856 assert(throw_entry_point != NULL, "entry point must be generated by now"); | |
| 727 | 857 AddressLiteral dest(throw_entry_point); |
| 858 jump_to(dest, Rscratch); | |
| 0 | 859 delayed()->nop(); |
| 860 bind(ok); | |
| 861 } | |
| 862 | |
| 863 | |
| 864 // And if you cannot use the delay slot, here is a shorthand: | |
| 865 | |
| 866 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, | |
| 867 address throw_entry_point, | |
| 868 Register Rscratch ) { | |
| 869 Label ok; | |
| 870 if (ok_condition != never) { | |
| 871 throw_if_not_1_icc( ok_condition, ok); | |
| 872 delayed()->nop(); | |
| 873 } | |
| 874 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
| 875 } | |
| 876 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, | |
| 877 address throw_entry_point, | |
| 878 Register Rscratch ) { | |
| 879 Label ok; | |
| 880 if (ok_condition != never) { | |
| 881 throw_if_not_1_xcc( ok_condition, ok); | |
| 882 delayed()->nop(); | |
| 883 } | |
| 884 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
| 885 } | |
| 886 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, | |
| 887 address throw_entry_point, | |
| 888 Register Rscratch ) { | |
| 889 Label ok; | |
| 890 if (ok_condition != never) { | |
| 891 throw_if_not_1_x( ok_condition, ok); | |
| 892 delayed()->nop(); | |
| 893 } | |
| 894 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
| 895 } | |
| 896 | |
| 897 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res | |
| 898 // Note: res is still shy of address by array offset into object. | |
| 899 | |
| 900 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { | |
| 901 assert_not_delayed(); | |
| 902 | |
| 903 verify_oop(array); | |
| 904 #ifdef _LP64 | |
| 905 // sign extend since tos (index) can be a 32bit value | |
| 906 sra(index, G0, index); | |
| 907 #endif // _LP64 | |
| 908 | |
| 909 // check array | |
| 910 Label ptr_ok; | |
| 911 tst(array); | |
| 912 throw_if_not_1_x( notZero, ptr_ok ); | |
| 913 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index | |
| 914 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); | |
| 915 | |
| 916 Label index_ok; | |
| 917 cmp(index, tmp); | |
| 918 throw_if_not_1_icc( lessUnsigned, index_ok ); | |
| 919 if (index_shift > 0) delayed()->sll(index, index_shift, index); | |
| 920 else delayed()->add(array, index, res); // addr - const offset in index | |
| 921 // convention: move aberrant index into G3_scratch for exception message | |
| 922 mov(index, G3_scratch); | |
| 923 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); | |
| 924 | |
| 925 // add offset if didn't do it in delay slot | |
| 926 if (index_shift > 0) add(array, index, res); // addr - const offset in index | |
| 927 } | |
| 928 | |
| 929 | |
| 930 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { | |
| 931 assert_not_delayed(); | |
| 932 | |
| 933 // pop array | |
| 934 pop_ptr(array); | |
| 935 | |
| 936 // check array | |
| 937 index_check_without_pop(array, index, index_shift, tmp, res); | |
| 938 } | |
| 939 | |
| 940 | |
|
6123
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
941 void InterpreterMacroAssembler::get_const(Register Rdst) { |
|
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
942 ld_ptr(Lmethod, in_bytes(methodOopDesc::const_offset()), Rdst); |
|
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
943 } |
|
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
944 |
|
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
945 |
| 0 | 946 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { |
|
6123
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
947 get_const(Rdst); |
|
2fe087c3e814
7172967: Eliminate constMethod's _method backpointer to methodOop.
jiangli
parents:
3852
diff
changeset
|
948 ld_ptr(Rdst, in_bytes(constMethodOopDesc::constants_offset()), Rdst); |
| 0 | 949 } |
| 950 | |
| 951 | |
| 952 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { | |
| 953 get_constant_pool(Rdst); | |
| 954 ld_ptr(Rdst, constantPoolOopDesc::cache_offset_in_bytes(), Rdst); | |
| 955 } | |
| 956 | |
| 957 | |
| 958 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { | |
| 959 get_constant_pool(Rcpool); | |
| 960 ld_ptr(Rcpool, constantPoolOopDesc::tags_offset_in_bytes(), Rtags); | |
| 961 } | |
| 962 | |
| 963 | |
| 964 // unlock if synchronized method | |
| 965 // | |
| 966 // Unlock the receiver if this is a synchronized method. | |
| 967 // Unlock any Java monitors from syncronized blocks. | |
| 968 // | |
| 969 // If there are locked Java monitors | |
| 970 // If throw_monitor_exception | |
| 971 // throws IllegalMonitorStateException | |
| 972 // Else if install_monitor_exception | |
| 973 // installs IllegalMonitorStateException | |
| 974 // Else | |
| 975 // no error processing | |
| 976 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, | |
| 977 bool throw_monitor_exception, | |
| 978 bool install_monitor_exception) { | |
| 979 Label unlocked, unlock, no_unlock; | |
| 980 | |
| 981 // get the value of _do_not_unlock_if_synchronized into G1_scratch | |
| 727 | 982 const Address do_not_unlock_if_synchronized(G2_thread, |
| 983 JavaThread::do_not_unlock_if_synchronized_offset()); | |
| 0 | 984 ldbool(do_not_unlock_if_synchronized, G1_scratch); |
| 985 stbool(G0, do_not_unlock_if_synchronized); // reset the flag | |
| 986 | |
| 987 // check if synchronized method | |
| 727 | 988 const Address access_flags(Lmethod, methodOopDesc::access_flags_offset()); |
| 0 | 989 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); |
| 990 push(state); // save tos | |
| 727 | 991 ld(access_flags, G3_scratch); // Load access flags. |
| 0 | 992 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); |
| 727 | 993 br(zero, false, pt, unlocked); |
| 0 | 994 delayed()->nop(); |
| 995 | |
| 996 // Don't unlock anything if the _do_not_unlock_if_synchronized flag | |
| 997 // is set. | |
| 3839 | 998 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock); |
| 0 | 999 delayed()->nop(); |
| 1000 | |
| 1001 // BasicObjectLock will be first in list, since this is a synchronized method. However, need | |
| 1002 // to check that the object has not been unlocked by an explicit monitorexit bytecode. | |
| 1003 | |
| 1004 //Intel: if (throw_monitor_exception) ... else ... | |
| 1005 // Entry already unlocked, need to throw exception | |
| 1006 //... | |
| 1007 | |
| 1008 // pass top-most monitor elem | |
| 1009 add( top_most_monitor(), O1 ); | |
| 1010 | |
| 1011 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); | |
| 3839 | 1012 br_notnull_short(G3_scratch, pt, unlock); |
| 0 | 1013 |
| 1014 if (throw_monitor_exception) { | |
| 1015 // Entry already unlocked need to throw an exception | |
| 1016 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
| 1017 should_not_reach_here(); | |
| 1018 } else { | |
| 1019 // Monitor already unlocked during a stack unroll. | |
| 1020 // If requested, install an illegal_monitor_state_exception. | |
| 1021 // Continue with stack unrolling. | |
| 1022 if (install_monitor_exception) { | |
| 1023 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); | |
| 1024 } | |
| 3839 | 1025 ba_short(unlocked); |
| 0 | 1026 } |
| 1027 | |
| 1028 bind(unlock); | |
| 1029 | |
| 1030 unlock_object(O1); | |
| 1031 | |
| 1032 bind(unlocked); | |
| 1033 | |
| 1034 // I0, I1: Might contain return value | |
| 1035 | |
| 1036 // Check that all monitors are unlocked | |
| 1037 { Label loop, exception, entry, restart; | |
| 1038 | |
| 1039 Register Rmptr = O0; | |
| 1040 Register Rtemp = O1; | |
| 1041 Register Rlimit = Lmonitors; | |
| 1042 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
| 1043 assert( (delta & LongAlignmentMask) == 0, | |
| 1044 "sizeof BasicObjectLock must be even number of doublewords"); | |
| 1045 | |
| 1046 #ifdef ASSERT | |
| 1047 add(top_most_monitor(), Rmptr, delta); | |
| 1048 { Label L; | |
| 1049 // ensure that Rmptr starts out above (or at) Rlimit | |
| 3839 | 1050 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); |
| 0 | 1051 stop("monitor stack has negative size"); |
| 1052 bind(L); | |
| 1053 } | |
| 1054 #endif | |
| 1055 bind(restart); | |
| 3839 | 1056 ba(entry); |
| 0 | 1057 delayed()-> |
| 1058 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry | |
| 1059 | |
| 1060 // Entry is still locked, need to throw exception | |
| 1061 bind(exception); | |
| 1062 if (throw_monitor_exception) { | |
| 1063 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
| 1064 should_not_reach_here(); | |
| 1065 } else { | |
| 1066 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. | |
| 1067 // Unlock does not block, so don't have to worry about the frame | |
| 1068 unlock_object(Rmptr); | |
| 1069 if (install_monitor_exception) { | |
| 1070 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); | |
| 1071 } | |
| 3839 | 1072 ba_short(restart); |
| 0 | 1073 } |
| 1074 | |
| 1075 bind(loop); | |
| 1076 cmp(Rtemp, G0); // check if current entry is used | |
| 1077 brx(Assembler::notEqual, false, pn, exception); | |
| 1078 delayed()-> | |
| 1079 dec(Rmptr, delta); // otherwise advance to next entry | |
| 1080 #ifdef ASSERT | |
| 1081 { Label L; | |
| 1082 // ensure that Rmptr has not somehow stepped below Rlimit | |
| 3839 | 1083 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L); |
| 0 | 1084 stop("ran off the end of the monitor stack"); |
| 1085 bind(L); | |
| 1086 } | |
| 1087 #endif | |
| 1088 bind(entry); | |
| 1089 cmp(Rmptr, Rlimit); // check if bottom reached | |
| 1090 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry | |
| 1091 delayed()-> | |
| 1092 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); | |
| 1093 } | |
| 1094 | |
| 1095 bind(no_unlock); | |
| 1096 pop(state); | |
| 1097 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 1098 } | |
| 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 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 1124 verify_oop(Lmethod); | |
| 1125 verify_thread(); | |
| 1126 | |
| 1127 // return tos | |
| 1128 assert(Otos_l1 == Otos_i, "adjust code below"); | |
| 1129 switch (state) { | |
| 1130 #ifdef _LP64 | |
| 1131 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 | |
| 1132 #else | |
| 1133 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1 | |
| 1134 #endif | |
| 1135 case btos: // fall through | |
| 1136 case ctos: | |
| 1137 case stos: // fall through | |
| 1138 case atos: // fall through | |
| 1139 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 | |
| 1140 case ftos: // fall through | |
| 1141 case dtos: // fall through | |
| 1142 case vtos: /* nothing to do */ break; | |
| 1143 default : ShouldNotReachHere(); | |
| 1144 } | |
| 1145 | |
| 1146 #if defined(COMPILER2) && !defined(_LP64) | |
| 1147 if (state == ltos) { | |
| 1148 // C2 expects long results in G1 we can't tell if we're returning to interpreted | |
| 1149 // or compiled so just be safe use G1 and O0/O1 | |
| 1150 | |
| 1151 // Shift bits into high (msb) of G1 | |
| 1152 sllx(Otos_l1->after_save(), 32, G1); | |
| 1153 // Zero extend low bits | |
| 1154 srl (Otos_l2->after_save(), 0, Otos_l2->after_save()); | |
| 1155 or3 (Otos_l2->after_save(), G1, G1); | |
| 1156 } | |
| 1157 #endif /* COMPILER2 */ | |
| 1158 | |
| 1159 } | |
| 1160 #endif /* CC_INTERP */ | |
| 1161 | |
| 1162 | |
| 1163 // Lock object | |
| 1164 // | |
| 1165 // Argument - lock_reg points to the BasicObjectLock to be used for locking, | |
| 1166 // it must be initialized with the object to lock | |
| 1167 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { | |
| 1168 if (UseHeavyMonitors) { | |
| 1169 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); | |
| 1170 } | |
| 1171 else { | |
| 1172 Register obj_reg = Object; | |
| 1173 Register mark_reg = G4_scratch; | |
| 1174 Register temp_reg = G1_scratch; | |
| 727 | 1175 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); |
| 1176 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); | |
| 0 | 1177 Label done; |
| 1178 | |
| 1179 Label slow_case; | |
| 1180 | |
| 1181 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); | |
| 1182 | |
| 1183 // load markOop from object into mark_reg | |
| 1184 ld_ptr(mark_addr, mark_reg); | |
| 1185 | |
| 1186 if (UseBiasedLocking) { | |
| 1187 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); | |
| 1188 } | |
| 1189 | |
| 1190 // get the address of basicLock on stack that will be stored in the object | |
| 1191 // we need a temporary register here as we do not want to clobber lock_reg | |
| 1192 // (cas clobbers the destination register) | |
| 1193 mov(lock_reg, temp_reg); | |
| 1194 // set mark reg to be (markOop of object | UNLOCK_VALUE) | |
| 1195 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); | |
| 1196 // initialize the box (Must happen before we update the object mark!) | |
| 1197 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); | |
| 1198 // compare and exchange object_addr, markOop | 1, stack address of basicLock | |
| 1199 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
| 1200 casx_under_lock(mark_addr.base(), mark_reg, temp_reg, | |
| 1201 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
| 1202 | |
| 1203 // if the compare and exchange succeeded we are done (we saw an unlocked object) | |
| 3839 | 1204 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done); |
| 0 | 1205 |
| 1206 // We did not see an unlocked object so try the fast recursive case | |
| 1207 | |
| 1208 // Check if owner is self by comparing the value in the markOop of object | |
| 1209 // with the stack pointer | |
| 1210 sub(temp_reg, SP, temp_reg); | |
| 1211 #ifdef _LP64 | |
| 1212 sub(temp_reg, STACK_BIAS, temp_reg); | |
| 1213 #endif | |
| 1214 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
| 1215 | |
| 1216 // Composite "andcc" test: | |
| 1217 // (a) %sp -vs- markword proximity check, and, | |
| 1218 // (b) verify mark word LSBs == 0 (Stack-locked). | |
| 1219 // | |
| 1220 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) | |
| 1221 // Note that the page size used for %sp proximity testing is arbitrary and is | |
| 1222 // unrelated to the actual MMU page size. We use a 'logical' page size of | |
| 1223 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate | |
| 1224 // field of the andcc instruction. | |
| 1225 andcc (temp_reg, 0xFFFFF003, G0) ; | |
| 1226 | |
| 1227 // if condition is true we are done and hence we can store 0 in the displaced | |
| 1228 // header indicating it is a recursive lock and be done | |
| 1229 brx(Assembler::zero, true, Assembler::pt, done); | |
| 1230 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); | |
| 1231 | |
| 1232 // none of the above fast optimizations worked so we have to get into the | |
| 1233 // slow case of monitor enter | |
| 1234 bind(slow_case); | |
| 1235 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); | |
| 1236 | |
| 1237 bind(done); | |
| 1238 } | |
| 1239 } | |
| 1240 | |
| 1241 // Unlocks an object. Used in monitorexit bytecode and remove_activation. | |
| 1242 // | |
| 1243 // Argument - lock_reg points to the BasicObjectLock for lock | |
| 1244 // Throw IllegalMonitorException if object is not locked by current thread | |
| 1245 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { | |
| 1246 if (UseHeavyMonitors) { | |
| 1247 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); | |
| 1248 } else { | |
| 1249 Register obj_reg = G3_scratch; | |
| 1250 Register mark_reg = G4_scratch; | |
| 1251 Register displaced_header_reg = G1_scratch; | |
| 727 | 1252 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); |
| 1253 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); | |
| 0 | 1254 Label done; |
| 1255 | |
| 1256 if (UseBiasedLocking) { | |
| 1257 // load the object out of the BasicObjectLock | |
| 1258 ld_ptr(lockobj_addr, obj_reg); | |
| 1259 biased_locking_exit(mark_addr, mark_reg, done, true); | |
| 1260 st_ptr(G0, lockobj_addr); // free entry | |
| 1261 } | |
| 1262 | |
| 1263 // Test first if we are in the fast recursive case | |
| 727 | 1264 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); |
| 1265 ld_ptr(lock_addr, displaced_header_reg); | |
| 0 | 1266 br_null(displaced_header_reg, true, Assembler::pn, done); |
| 1267 delayed()->st_ptr(G0, lockobj_addr); // free entry | |
| 1268 | |
| 1269 // See if it is still a light weight lock, if so we just unlock | |
| 1270 // the object and we are done | |
| 1271 | |
| 1272 if (!UseBiasedLocking) { | |
| 1273 // load the object out of the BasicObjectLock | |
| 1274 ld_ptr(lockobj_addr, obj_reg); | |
| 1275 } | |
| 1276 | |
| 1277 // we have the displaced header in displaced_header_reg | |
| 1278 // we expect to see the stack address of the basicLock in case the | |
| 1279 // lock is still a light weight lock (lock_reg) | |
| 1280 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
| 1281 casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg, | |
| 1282 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
| 1283 cmp(lock_reg, displaced_header_reg); | |
| 1284 brx(Assembler::equal, true, Assembler::pn, done); | |
| 1285 delayed()->st_ptr(G0, lockobj_addr); // free entry | |
| 1286 | |
| 1287 // The lock has been converted into a heavy lock and hence | |
| 1288 // we need to get into the slow case | |
| 1289 | |
| 1290 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); | |
| 1291 | |
| 1292 bind(done); | |
| 1293 } | |
| 1294 } | |
| 1295 | |
| 1296 #ifndef CC_INTERP | |
| 1297 | |
| 1298 // Get the method data pointer from the methodOop and set the | |
| 1299 // specified register to its value. | |
| 1300 | |
|
2118
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1301 void InterpreterMacroAssembler::set_method_data_pointer() { |
| 0 | 1302 assert(ProfileInterpreter, "must be profiling interpreter"); |
| 1303 Label get_continue; | |
| 1304 | |
| 1305 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr); | |
| 1306 test_method_data_pointer(get_continue); | |
| 1307 add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr); | |
| 1308 bind(get_continue); | |
| 1309 } | |
| 1310 | |
| 1311 // Set the method data pointer for the current bcp. | |
| 1312 | |
| 1313 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { | |
| 1314 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1315 Label zero_continue; | |
| 1316 | |
| 1317 // Test MDO to avoid the call if it is NULL. | |
|
2118
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1318 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr); |
| 0 | 1319 test_method_data_pointer(zero_continue); |
| 1320 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); | |
|
2118
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1321 add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr); |
|
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1322 add(ImethodDataPtr, O0, ImethodDataPtr); |
| 0 | 1323 bind(zero_continue); |
| 1324 } | |
| 1325 | |
| 1326 // Test ImethodDataPtr. If it is null, continue at the specified label | |
| 1327 | |
| 1328 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { | |
| 1329 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 3839 | 1330 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue); |
| 0 | 1331 } |
| 1332 | |
| 1333 void InterpreterMacroAssembler::verify_method_data_pointer() { | |
| 1334 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1335 #ifdef ASSERT | |
| 1336 Label verify_continue; | |
| 1337 test_method_data_pointer(verify_continue); | |
| 1338 | |
| 1339 // If the mdp is valid, it will point to a DataLayout header which is | |
| 1340 // consistent with the bcp. The converse is highly probable also. | |
| 1341 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); | |
| 727 | 1342 ld_ptr(Lmethod, methodOopDesc::const_offset(), O5); |
| 0 | 1343 add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), G3_scratch); |
| 1344 add(G3_scratch, O5, G3_scratch); | |
| 1345 cmp(Lbcp, G3_scratch); | |
| 1346 brx(Assembler::equal, false, Assembler::pt, verify_continue); | |
| 1347 | |
| 1348 Register temp_reg = O5; | |
| 1349 delayed()->mov(ImethodDataPtr, temp_reg); | |
| 1350 // %%% should use call_VM_leaf here? | |
| 1351 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); | |
| 1352 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); | |
| 727 | 1353 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); |
| 0 | 1354 stf(FloatRegisterImpl::D, Ftos_d, d_save); |
| 1355 mov(temp_reg->after_save(), O2); | |
| 1356 save_thread(L7_thread_cache); | |
| 1357 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); | |
| 1358 delayed()->nop(); | |
| 1359 restore_thread(L7_thread_cache); | |
| 1360 ldf(FloatRegisterImpl::D, d_save, Ftos_d); | |
| 1361 restore(); | |
| 1362 bind(verify_continue); | |
| 1363 #endif // ASSERT | |
| 1364 } | |
| 1365 | |
| 1366 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, | |
| 1367 Register Rtmp, | |
| 1368 Label &profile_continue) { | |
| 1369 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1370 // Control will flow to "profile_continue" if the counter is less than the | |
| 1371 // limit or if we call profile_method() | |
| 1372 | |
| 1373 Label done; | |
| 1374 | |
| 1375 // if no method data exists, and the counter is high enough, make one | |
| 3839 | 1376 br_notnull_short(ImethodDataPtr, Assembler::pn, done); |
| 0 | 1377 |
| 1378 // Test to see if we should create a method data oop | |
| 727 | 1379 AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit); |
| 1380 sethi(profile_limit, Rtmp); | |
| 1381 ld(Rtmp, profile_limit.low10(), Rtmp); | |
| 3839 | 1382 cmp_and_br_short(invocation_count, Rtmp, Assembler::lessUnsigned, Assembler::pn, profile_continue); |
| 0 | 1383 |
| 1384 // Build it now. | |
|
2118
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1385 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); |
|
dd031b2226de
4930919: race condition in MDO creation at back branch locations
iveresov
parents:
1972
diff
changeset
|
1386 set_method_data_pointer_for_bcp(); |
| 3839 | 1387 ba_short(profile_continue); |
| 0 | 1388 bind(done); |
| 1389 } | |
| 1390 | |
| 1391 // Store a value at some constant offset from the method data pointer. | |
| 1392 | |
| 1393 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { | |
| 1394 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1395 st_ptr(value, ImethodDataPtr, constant); | |
| 1396 } | |
| 1397 | |
| 1398 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, | |
| 1399 Register bumped_count, | |
| 1400 bool decrement) { | |
| 1401 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1402 | |
| 1403 // Load the counter. | |
| 1404 ld_ptr(counter, bumped_count); | |
| 1405 | |
| 1406 if (decrement) { | |
| 1407 // Decrement the register. Set condition codes. | |
| 1408 subcc(bumped_count, DataLayout::counter_increment, bumped_count); | |
| 1409 | |
| 1410 // If the decrement causes the counter to overflow, stay negative | |
| 1411 Label L; | |
| 1412 brx(Assembler::negative, true, Assembler::pn, L); | |
| 1413 | |
| 1414 // Store the decremented counter, if it is still negative. | |
| 1415 delayed()->st_ptr(bumped_count, counter); | |
| 1416 bind(L); | |
| 1417 } else { | |
| 1418 // Increment the register. Set carry flag. | |
| 1419 addcc(bumped_count, DataLayout::counter_increment, bumped_count); | |
| 1420 | |
| 1421 // If the increment causes the counter to overflow, pull back by 1. | |
| 1422 assert(DataLayout::counter_increment == 1, "subc works"); | |
| 1423 subc(bumped_count, G0, bumped_count); | |
| 1424 | |
| 1425 // Store the incremented counter. | |
| 1426 st_ptr(bumped_count, counter); | |
| 1427 } | |
| 1428 } | |
| 1429 | |
| 1430 // Increment the value at some constant offset from the method data pointer. | |
| 1431 | |
| 1432 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, | |
| 1433 Register bumped_count, | |
| 1434 bool decrement) { | |
| 1435 // Locate the counter at a fixed offset from the mdp: | |
| 727 | 1436 Address counter(ImethodDataPtr, constant); |
| 0 | 1437 increment_mdp_data_at(counter, bumped_count, decrement); |
| 1438 } | |
| 1439 | |
| 1440 // Increment the value at some non-fixed (reg + constant) offset from | |
| 1441 // the method data pointer. | |
| 1442 | |
| 1443 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, | |
| 1444 int constant, | |
| 1445 Register bumped_count, | |
| 1446 Register scratch2, | |
| 1447 bool decrement) { | |
| 1448 // Add the constant to reg to get the offset. | |
| 1449 add(ImethodDataPtr, reg, scratch2); | |
| 727 | 1450 Address counter(scratch2, constant); |
| 0 | 1451 increment_mdp_data_at(counter, bumped_count, decrement); |
| 1452 } | |
| 1453 | |
| 1454 // Set a flag value at the current method data pointer position. | |
| 1455 // Updates a single byte of the header, to avoid races with other header bits. | |
| 1456 | |
| 1457 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, | |
| 1458 Register scratch) { | |
| 1459 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1460 // Load the data header | |
| 1461 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); | |
| 1462 | |
| 1463 // Set the flag | |
| 1464 or3(scratch, flag_constant, scratch); | |
| 1465 | |
| 1466 // Store the modified header. | |
| 1467 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); | |
| 1468 } | |
| 1469 | |
| 1470 // Test the location at some offset from the method data pointer. | |
| 1471 // If it is not equal to value, branch to the not_equal_continue Label. | |
| 1472 // Set condition codes to match the nullness of the loaded value. | |
| 1473 | |
| 1474 void InterpreterMacroAssembler::test_mdp_data_at(int offset, | |
| 1475 Register value, | |
| 1476 Label& not_equal_continue, | |
| 1477 Register scratch) { | |
| 1478 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1479 ld_ptr(ImethodDataPtr, offset, scratch); | |
| 1480 cmp(value, scratch); | |
| 1481 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); | |
| 1482 delayed()->tst(scratch); | |
| 1483 } | |
| 1484 | |
| 1485 // Update the method data pointer by the displacement located at some fixed | |
| 1486 // offset from the method data pointer. | |
| 1487 | |
| 1488 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, | |
| 1489 Register scratch) { | |
| 1490 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1491 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); | |
| 1492 add(ImethodDataPtr, scratch, ImethodDataPtr); | |
| 1493 } | |
| 1494 | |
| 1495 // Update the method data pointer by the displacement located at the | |
| 1496 // offset (reg + offset_of_disp). | |
| 1497 | |
| 1498 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, | |
| 1499 int offset_of_disp, | |
| 1500 Register scratch) { | |
| 1501 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1502 add(reg, offset_of_disp, scratch); | |
| 1503 ld_ptr(ImethodDataPtr, scratch, scratch); | |
| 1504 add(ImethodDataPtr, scratch, ImethodDataPtr); | |
| 1505 } | |
| 1506 | |
| 1507 // Update the method data pointer by a simple constant displacement. | |
| 1508 | |
| 1509 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { | |
| 1510 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1511 add(ImethodDataPtr, constant, ImethodDataPtr); | |
| 1512 } | |
| 1513 | |
| 1514 // Update the method data pointer for a _ret bytecode whose target | |
| 1515 // was not among our cached targets. | |
| 1516 | |
| 1517 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, | |
| 1518 Register return_bci) { | |
| 1519 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1520 push(state); | |
| 1521 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile | |
| 1522 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); | |
| 1523 ld_ptr(l_tmp, return_bci); | |
| 1524 pop(state); | |
| 1525 } | |
| 1526 | |
| 1527 // Count a taken branch in the bytecodes. | |
| 1528 | |
| 1529 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { | |
| 1530 if (ProfileInterpreter) { | |
| 1531 Label profile_continue; | |
| 1532 | |
| 1533 // If no method data exists, go to profile_continue. | |
| 1534 test_method_data_pointer(profile_continue); | |
| 1535 | |
| 1536 // We are taking a branch. Increment the taken count. | |
| 1537 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); | |
| 1538 | |
| 1539 // The method data pointer needs to be updated to reflect the new target. | |
| 1540 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); | |
| 1541 bind (profile_continue); | |
| 1542 } | |
| 1543 } | |
| 1544 | |
| 1545 | |
| 1546 // Count a not-taken branch in the bytecodes. | |
| 1547 | |
| 1548 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { | |
| 1549 if (ProfileInterpreter) { | |
| 1550 Label profile_continue; | |
| 1551 | |
| 1552 // If no method data exists, go to profile_continue. | |
| 1553 test_method_data_pointer(profile_continue); | |
| 1554 | |
| 1555 // We are taking a branch. Increment the not taken count. | |
| 1556 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); | |
| 1557 | |
| 1558 // The method data pointer needs to be updated to correspond to the | |
| 1559 // next bytecode. | |
| 1560 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); | |
| 1561 bind (profile_continue); | |
| 1562 } | |
| 1563 } | |
| 1564 | |
| 1565 | |
| 1566 // Count a non-virtual call in the bytecodes. | |
| 1567 | |
| 1568 void InterpreterMacroAssembler::profile_call(Register scratch) { | |
| 1569 if (ProfileInterpreter) { | |
| 1570 Label profile_continue; | |
| 1571 | |
| 1572 // If no method data exists, go to profile_continue. | |
| 1573 test_method_data_pointer(profile_continue); | |
| 1574 | |
| 1575 // We are making a call. Increment the count. | |
| 1576 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
| 1577 | |
| 1578 // The method data pointer needs to be updated to reflect the new target. | |
| 1579 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); | |
| 1580 bind (profile_continue); | |
| 1581 } | |
| 1582 } | |
| 1583 | |
| 1584 | |
| 1585 // Count a final call in the bytecodes. | |
| 1586 | |
| 1587 void InterpreterMacroAssembler::profile_final_call(Register scratch) { | |
| 1588 if (ProfileInterpreter) { | |
| 1589 Label profile_continue; | |
| 1590 | |
| 1591 // If no method data exists, go to profile_continue. | |
| 1592 test_method_data_pointer(profile_continue); | |
| 1593 | |
| 1594 // We are making a call. Increment the count. | |
| 1595 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
| 1596 | |
| 1597 // The method data pointer needs to be updated to reflect the new target. | |
| 1598 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); | |
| 1599 bind (profile_continue); | |
| 1600 } | |
| 1601 } | |
| 1602 | |
| 1603 | |
| 1604 // Count a virtual call in the bytecodes. | |
| 1605 | |
| 1606 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, | |
| 1503 | 1607 Register scratch, |
| 1608 bool receiver_can_be_null) { | |
| 0 | 1609 if (ProfileInterpreter) { |
| 1610 Label profile_continue; | |
| 1611 | |
| 1612 // If no method data exists, go to profile_continue. | |
| 1613 test_method_data_pointer(profile_continue); | |
| 1614 | |
| 1503 | 1615 |
| 1616 Label skip_receiver_profile; | |
| 1617 if (receiver_can_be_null) { | |
| 1618 Label not_null; | |
| 3839 | 1619 br_notnull_short(receiver, Assembler::pt, not_null); |
| 1503 | 1620 // We are making a call. Increment the count for null receiver. |
| 1621 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
| 3839 | 1622 ba_short(skip_receiver_profile); |
| 1503 | 1623 bind(not_null); |
| 1624 } | |
| 1625 | |
| 0 | 1626 // Record the receiver type. |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1627 record_klass_in_profile(receiver, scratch, true); |
| 1503 | 1628 bind(skip_receiver_profile); |
| 0 | 1629 |
| 1630 // The method data pointer needs to be updated to reflect the new target. | |
| 1631 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); | |
| 1632 bind (profile_continue); | |
| 1633 } | |
| 1634 } | |
| 1635 | |
| 1636 void InterpreterMacroAssembler::record_klass_in_profile_helper( | |
| 1637 Register receiver, Register scratch, | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1638 int start_row, Label& done, bool is_virtual_call) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1639 if (TypeProfileWidth == 0) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1640 if (is_virtual_call) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1641 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1642 } |
|
967
6918603297f7
6858208: jvm crash when specifying TypeProfileWidth=0 on jdk 6.0
poonam
parents:
727
diff
changeset
|
1643 return; |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1644 } |
|
967
6918603297f7
6858208: jvm crash when specifying TypeProfileWidth=0 on jdk 6.0
poonam
parents:
727
diff
changeset
|
1645 |
| 0 | 1646 int last_row = VirtualCallData::row_limit() - 1; |
| 1647 assert(start_row <= last_row, "must be work left to do"); | |
| 1648 // Test this row for both the receiver and for null. | |
| 1649 // Take any of three different outcomes: | |
| 1650 // 1. found receiver => increment count and goto done | |
| 1651 // 2. found null => keep looking for case 1, maybe allocate this cell | |
| 1652 // 3. found something else => keep looking for cases 1 and 2 | |
| 1653 // Case 3 is handled by a recursive call. | |
| 1654 for (int row = start_row; row <= last_row; row++) { | |
| 1655 Label next_test; | |
| 1656 bool test_for_null_also = (row == start_row); | |
| 1657 | |
| 1658 // See if the receiver is receiver[n]. | |
| 1659 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); | |
| 1660 test_mdp_data_at(recvr_offset, receiver, next_test, scratch); | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1661 // delayed()->tst(scratch); |
| 0 | 1662 |
| 1663 // The receiver is receiver[n]. Increment count[n]. | |
| 1664 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); | |
| 1665 increment_mdp_data_at(count_offset, scratch); | |
| 3839 | 1666 ba_short(done); |
| 0 | 1667 bind(next_test); |
| 1668 | |
| 1669 if (test_for_null_also) { | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1670 Label found_null; |
| 0 | 1671 // Failed the equality check on receiver[n]... Test for null. |
| 1672 if (start_row == last_row) { | |
| 1673 // The only thing left to do is handle the null case. | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1674 if (is_virtual_call) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1675 brx(Assembler::zero, false, Assembler::pn, found_null); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1676 delayed()->nop(); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1677 // Receiver did not match any saved receiver and there is no empty row for it. |
|
1251
576e77447e3c
6923002: assert(false,"this call site should not be polymorphic")
kvn
parents:
1206
diff
changeset
|
1678 // Increment total counter to indicate polymorphic case. |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1679 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); |
| 3839 | 1680 ba_short(done); |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1681 bind(found_null); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1682 } else { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1683 brx(Assembler::notZero, false, Assembler::pt, done); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1684 delayed()->nop(); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1685 } |
| 0 | 1686 break; |
| 1687 } | |
| 1688 // Since null is rare, make it be the branch-taken case. | |
| 1689 brx(Assembler::zero, false, Assembler::pn, found_null); | |
| 1690 delayed()->nop(); | |
| 1691 | |
| 1692 // Put all the "Case 3" tests here. | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1693 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call); |
| 0 | 1694 |
| 1695 // Found a null. Keep searching for a matching receiver, | |
| 1696 // but remember that this is an empty (unused) slot. | |
| 1697 bind(found_null); | |
| 1698 } | |
| 1699 } | |
| 1700 | |
| 1701 // In the fall-through case, we found no matching receiver, but we | |
| 1702 // observed the receiver[start_row] is NULL. | |
| 1703 | |
| 1704 // Fill in the receiver field and increment the count. | |
| 1705 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); | |
| 1706 set_mdp_data_at(recvr_offset, receiver); | |
| 1707 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); | |
| 1708 mov(DataLayout::counter_increment, scratch); | |
| 1709 set_mdp_data_at(count_offset, scratch); | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1710 if (start_row > 0) { |
| 3839 | 1711 ba_short(done); |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1712 } |
| 0 | 1713 } |
| 1714 | |
| 1715 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1716 Register scratch, bool is_virtual_call) { |
| 0 | 1717 assert(ProfileInterpreter, "must be profiling"); |
| 1718 Label done; | |
| 1719 | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1720 record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call); |
| 0 | 1721 |
| 1722 bind (done); | |
| 1723 } | |
| 1724 | |
| 1725 | |
| 1726 // Count a ret in the bytecodes. | |
| 1727 | |
| 1728 void InterpreterMacroAssembler::profile_ret(TosState state, | |
| 1729 Register return_bci, | |
| 1730 Register scratch) { | |
| 1731 if (ProfileInterpreter) { | |
| 1732 Label profile_continue; | |
| 1733 uint row; | |
| 1734 | |
| 1735 // If no method data exists, go to profile_continue. | |
| 1736 test_method_data_pointer(profile_continue); | |
| 1737 | |
| 1738 // Update the total ret count. | |
| 1739 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
| 1740 | |
| 1741 for (row = 0; row < RetData::row_limit(); row++) { | |
| 1742 Label next_test; | |
| 1743 | |
| 1744 // See if return_bci is equal to bci[n]: | |
| 1745 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), | |
| 1746 return_bci, next_test, scratch); | |
| 1747 | |
| 1748 // return_bci is equal to bci[n]. Increment the count. | |
| 1749 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); | |
| 1750 | |
| 1751 // The method data pointer needs to be updated to reflect the new target. | |
| 1752 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); | |
| 3839 | 1753 ba_short(profile_continue); |
| 0 | 1754 bind(next_test); |
| 1755 } | |
| 1756 | |
| 1757 update_mdp_for_ret(state, return_bci); | |
| 1758 | |
| 1759 bind (profile_continue); | |
| 1760 } | |
| 1761 } | |
| 1762 | |
| 1763 // Profile an unexpected null in the bytecodes. | |
| 1764 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { | |
| 1765 if (ProfileInterpreter) { | |
| 1766 Label profile_continue; | |
| 1767 | |
| 1768 // If no method data exists, go to profile_continue. | |
| 1769 test_method_data_pointer(profile_continue); | |
| 1770 | |
| 1771 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); | |
| 1772 | |
| 1773 // The method data pointer needs to be updated. | |
| 1774 int mdp_delta = in_bytes(BitData::bit_data_size()); | |
| 1775 if (TypeProfileCasts) { | |
| 1776 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); | |
| 1777 } | |
| 1778 update_mdp_by_constant(mdp_delta); | |
| 1779 | |
| 1780 bind (profile_continue); | |
| 1781 } | |
| 1782 } | |
| 1783 | |
| 1784 void InterpreterMacroAssembler::profile_typecheck(Register klass, | |
| 1785 Register scratch) { | |
| 1786 if (ProfileInterpreter) { | |
| 1787 Label profile_continue; | |
| 1788 | |
| 1789 // If no method data exists, go to profile_continue. | |
| 1790 test_method_data_pointer(profile_continue); | |
| 1791 | |
| 1792 int mdp_delta = in_bytes(BitData::bit_data_size()); | |
| 1793 if (TypeProfileCasts) { | |
| 1794 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); | |
| 1795 | |
| 1796 // Record the object type. | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1797 record_klass_in_profile(klass, scratch, false); |
| 0 | 1798 } |
| 1799 | |
| 1800 // The method data pointer needs to be updated. | |
| 1801 update_mdp_by_constant(mdp_delta); | |
| 1802 | |
| 1803 bind (profile_continue); | |
| 1804 } | |
| 1805 } | |
| 1806 | |
| 1807 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { | |
| 1808 if (ProfileInterpreter && TypeProfileCasts) { | |
| 1809 Label profile_continue; | |
| 1810 | |
| 1811 // If no method data exists, go to profile_continue. | |
| 1812 test_method_data_pointer(profile_continue); | |
| 1813 | |
| 1814 int count_offset = in_bytes(CounterData::count_offset()); | |
| 1815 // Back up the address, since we have already bumped the mdp. | |
| 1816 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); | |
| 1817 | |
| 1818 // *Decrement* the counter. We expect to see zero or small negatives. | |
| 1819 increment_mdp_data_at(count_offset, scratch, true); | |
| 1820 | |
| 1821 bind (profile_continue); | |
| 1822 } | |
| 1823 } | |
| 1824 | |
| 1825 // Count the default case of a switch construct. | |
| 1826 | |
| 1827 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { | |
| 1828 if (ProfileInterpreter) { | |
| 1829 Label profile_continue; | |
| 1830 | |
| 1831 // If no method data exists, go to profile_continue. | |
| 1832 test_method_data_pointer(profile_continue); | |
| 1833 | |
| 1834 // Update the default case count | |
| 1835 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), | |
| 1836 scratch); | |
| 1837 | |
| 1838 // The method data pointer needs to be updated. | |
| 1839 update_mdp_by_offset( | |
| 1840 in_bytes(MultiBranchData::default_displacement_offset()), | |
| 1841 scratch); | |
| 1842 | |
| 1843 bind (profile_continue); | |
| 1844 } | |
| 1845 } | |
| 1846 | |
| 1847 // Count the index'th case of a switch construct. | |
| 1848 | |
| 1849 void InterpreterMacroAssembler::profile_switch_case(Register index, | |
| 1850 Register scratch, | |
| 1851 Register scratch2, | |
| 1852 Register scratch3) { | |
| 1853 if (ProfileInterpreter) { | |
| 1854 Label profile_continue; | |
| 1855 | |
| 1856 // If no method data exists, go to profile_continue. | |
| 1857 test_method_data_pointer(profile_continue); | |
| 1858 | |
| 1859 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() | |
| 1860 set(in_bytes(MultiBranchData::per_case_size()), scratch); | |
| 1861 smul(index, scratch, scratch); | |
| 1862 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); | |
| 1863 | |
| 1864 // Update the case count | |
| 1865 increment_mdp_data_at(scratch, | |
| 1866 in_bytes(MultiBranchData::relative_count_offset()), | |
| 1867 scratch2, | |
| 1868 scratch3); | |
| 1869 | |
| 1870 // The method data pointer needs to be updated. | |
| 1871 update_mdp_by_offset(scratch, | |
| 1872 in_bytes(MultiBranchData::relative_displacement_offset()), | |
| 1873 scratch2); | |
| 1874 | |
| 1875 bind (profile_continue); | |
| 1876 } | |
| 1877 } | |
| 1878 | |
| 1879 // add a InterpMonitorElem to stack (see frame_sparc.hpp) | |
| 1880 | |
| 1881 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, | |
| 1882 Register Rtemp, | |
| 1883 Register Rtemp2 ) { | |
| 1884 | |
| 1885 Register Rlimit = Lmonitors; | |
| 1886 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
| 1887 assert( (delta & LongAlignmentMask) == 0, | |
| 1888 "sizeof BasicObjectLock must be even number of doublewords"); | |
| 1889 | |
| 1890 sub( SP, delta, SP); | |
| 1891 sub( Lesp, delta, Lesp); | |
| 1892 sub( Lmonitors, delta, Lmonitors); | |
| 1893 | |
| 1894 if (!stack_is_empty) { | |
| 1895 | |
| 1896 // must copy stack contents down | |
| 1897 | |
| 1898 Label start_copying, next; | |
| 1899 | |
| 1900 // untested("monitor stack expansion"); | |
| 1901 compute_stack_base(Rtemp); | |
| 3839 | 1902 ba(start_copying); |
| 1903 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below | |
| 0 | 1904 |
| 1905 // note: must copy from low memory upwards | |
| 1906 // On entry to loop, | |
| 1907 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) | |
| 1908 // Loop mutates Rtemp | |
| 1909 | |
| 1910 bind( next); | |
| 1911 | |
| 1912 st_ptr(Rtemp2, Rtemp, 0); | |
| 1913 inc(Rtemp, wordSize); | |
| 1914 cmp(Rtemp, Rlimit); // are we done? (duplicated above) | |
| 1915 | |
| 1916 bind( start_copying ); | |
| 1917 | |
| 1918 brx( notEqual, true, pn, next ); | |
| 1919 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); | |
| 1920 | |
| 1921 // done copying stack | |
| 1922 } | |
| 1923 } | |
| 1924 | |
| 1925 // Locals | |
| 1926 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { | |
| 1927 assert_not_delayed(); | |
| 1506 | 1928 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 1929 sub(Llocals, index, index); |
| 1506 | 1930 ld_ptr(index, 0, dst); |
| 0 | 1931 // Note: index must hold the effective address--the iinc template uses it |
| 1932 } | |
| 1933 | |
| 1934 // Just like access_local_ptr but the tag is a returnAddress | |
| 1935 void InterpreterMacroAssembler::access_local_returnAddress(Register index, | |
| 1936 Register dst ) { | |
| 1937 assert_not_delayed(); | |
| 1506 | 1938 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 1939 sub(Llocals, index, index); |
| 1506 | 1940 ld_ptr(index, 0, dst); |
| 0 | 1941 } |
| 1942 | |
| 1943 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { | |
| 1944 assert_not_delayed(); | |
| 1506 | 1945 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 1946 sub(Llocals, index, index); |
| 1506 | 1947 ld(index, 0, dst); |
| 0 | 1948 // Note: index must hold the effective address--the iinc template uses it |
| 1949 } | |
| 1950 | |
| 1951 | |
| 1952 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { | |
| 1953 assert_not_delayed(); | |
| 1506 | 1954 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 1955 sub(Llocals, index, index); |
| 1956 // First half stored at index n+1 (which grows down from Llocals[n]) | |
| 1957 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); | |
| 1958 } | |
| 1959 | |
| 1960 | |
| 1961 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { | |
| 1962 assert_not_delayed(); | |
| 1506 | 1963 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 1964 sub(Llocals, index, index); |
| 1506 | 1965 ldf(FloatRegisterImpl::S, index, 0, dst); |
| 0 | 1966 } |
| 1967 | |
| 1968 | |
| 1969 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { | |
| 1970 assert_not_delayed(); | |
| 1506 | 1971 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 1972 sub(Llocals, index, index); |
| 1973 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); | |
| 1974 } | |
| 1975 | |
| 1976 | |
| 1977 #ifdef ASSERT | |
| 1978 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { | |
| 1979 Label L; | |
| 1980 | |
| 1981 assert(Rindex != Rscratch, "Registers cannot be same"); | |
| 1982 assert(Rindex != Rscratch1, "Registers cannot be same"); | |
| 1983 assert(Rlimit != Rscratch, "Registers cannot be same"); | |
| 1984 assert(Rlimit != Rscratch1, "Registers cannot be same"); | |
| 1985 assert(Rscratch1 != Rscratch, "Registers cannot be same"); | |
| 1986 | |
| 1987 // untested("reg area corruption"); | |
| 1988 add(Rindex, offset, Rscratch); | |
| 1989 add(Rlimit, 64 + STACK_BIAS, Rscratch1); | |
| 3839 | 1990 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L); |
| 0 | 1991 stop("regsave area is being clobbered"); |
| 1992 bind(L); | |
| 1993 } | |
| 1994 #endif // ASSERT | |
| 1995 | |
| 1996 | |
| 1997 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { | |
| 1998 assert_not_delayed(); | |
| 1506 | 1999 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 2000 sub(Llocals, index, index); |
| 1506 | 2001 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);) |
| 2002 st(src, index, 0); | |
| 0 | 2003 } |
| 2004 | |
| 1506 | 2005 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) { |
| 0 | 2006 assert_not_delayed(); |
| 1506 | 2007 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 2008 sub(Llocals, index, index); |
| 1506 | 2009 #ifdef ASSERT |
| 2010 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); | |
| 2011 #endif | |
| 2012 st_ptr(src, index, 0); | |
| 0 | 2013 } |
| 2014 | |
| 2015 | |
| 2016 | |
| 1506 | 2017 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) { |
| 2018 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); | |
| 0 | 2019 } |
| 2020 | |
| 2021 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { | |
| 2022 assert_not_delayed(); | |
| 1506 | 2023 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 2024 sub(Llocals, index, index); |
| 1506 | 2025 #ifdef ASSERT |
| 0 | 2026 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); |
| 1506 | 2027 #endif |
| 0 | 2028 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 |
| 2029 } | |
| 2030 | |
| 2031 | |
| 2032 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { | |
| 2033 assert_not_delayed(); | |
| 1506 | 2034 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 2035 sub(Llocals, index, index); |
| 1506 | 2036 #ifdef ASSERT |
| 2037 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch); | |
| 2038 #endif | |
| 2039 stf(FloatRegisterImpl::S, src, index, 0); | |
| 0 | 2040 } |
| 2041 | |
| 2042 | |
| 2043 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { | |
| 2044 assert_not_delayed(); | |
| 1506 | 2045 sll(index, Interpreter::logStackElementSize, index); |
| 0 | 2046 sub(Llocals, index, index); |
| 1506 | 2047 #ifdef ASSERT |
| 0 | 2048 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); |
| 1506 | 2049 #endif |
| 0 | 2050 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); |
| 2051 } | |
| 2052 | |
| 2053 | |
| 2054 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { | |
| 2055 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
| 2056 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong); | |
| 2057 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; | |
| 2058 } | |
| 2059 | |
| 2060 | |
| 2061 Address InterpreterMacroAssembler::top_most_monitor() { | |
| 727 | 2062 return Address(FP, top_most_monitor_byte_offset()); |
| 0 | 2063 } |
| 2064 | |
| 2065 | |
| 2066 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { | |
| 2067 add( Lesp, wordSize, Rdest ); | |
| 2068 } | |
| 2069 | |
| 2070 #endif /* CC_INTERP */ | |
| 2071 | |
| 2072 void InterpreterMacroAssembler::increment_invocation_counter( Register Rtmp, Register Rtmp2 ) { | |
| 2073 assert(UseCompiler, "incrementing must be useful"); | |
| 2074 #ifdef CC_INTERP | |
| 727 | 2075 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() + |
| 2076 InvocationCounter::counter_offset()); | |
| 2077 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() + | |
| 2078 InvocationCounter::counter_offset()); | |
| 0 | 2079 #else |
| 727 | 2080 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() + |
| 2081 InvocationCounter::counter_offset()); | |
| 2082 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() + | |
| 2083 InvocationCounter::counter_offset()); | |
| 0 | 2084 #endif /* CC_INTERP */ |
| 2085 int delta = InvocationCounter::count_increment; | |
| 2086 | |
| 2087 // Load each counter in a register | |
| 2088 ld( inv_counter, Rtmp ); | |
| 2089 ld( be_counter, Rtmp2 ); | |
| 2090 | |
| 2091 assert( is_simm13( delta ), " delta too large."); | |
| 2092 | |
| 2093 // Add the delta to the invocation counter and store the result | |
| 2094 add( Rtmp, delta, Rtmp ); | |
| 2095 | |
| 2096 // Mask the backedge counter | |
| 2097 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); | |
| 2098 | |
| 2099 // Store value | |
| 2100 st( Rtmp, inv_counter); | |
| 2101 | |
| 2102 // Add invocation counter + backedge counter | |
| 2103 add( Rtmp, Rtmp2, Rtmp); | |
| 2104 | |
| 2105 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! | |
| 2106 } | |
| 2107 | |
| 2108 void InterpreterMacroAssembler::increment_backedge_counter( Register Rtmp, Register Rtmp2 ) { | |
| 2109 assert(UseCompiler, "incrementing must be useful"); | |
| 2110 #ifdef CC_INTERP | |
| 727 | 2111 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() + |
| 2112 InvocationCounter::counter_offset()); | |
| 2113 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() + | |
| 2114 InvocationCounter::counter_offset()); | |
| 0 | 2115 #else |
| 727 | 2116 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() + |
| 2117 InvocationCounter::counter_offset()); | |
| 2118 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() + | |
| 2119 InvocationCounter::counter_offset()); | |
| 0 | 2120 #endif /* CC_INTERP */ |
| 2121 int delta = InvocationCounter::count_increment; | |
| 2122 // Load each counter in a register | |
| 2123 ld( be_counter, Rtmp ); | |
| 2124 ld( inv_counter, Rtmp2 ); | |
| 2125 | |
| 2126 // Add the delta to the backedge counter | |
| 2127 add( Rtmp, delta, Rtmp ); | |
| 2128 | |
| 2129 // Mask the invocation counter, add to backedge counter | |
| 2130 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); | |
| 2131 | |
| 2132 // and store the result to memory | |
| 2133 st( Rtmp, be_counter ); | |
| 2134 | |
| 2135 // Add backedge + invocation counter | |
| 2136 add( Rtmp, Rtmp2, Rtmp ); | |
| 2137 | |
| 2138 // Note that this macro must leave backedge_count + invocation_count in Rtmp! | |
| 2139 } | |
| 2140 | |
| 2141 #ifndef CC_INTERP | |
| 2142 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, | |
| 2143 Register branch_bcp, | |
| 2144 Register Rtmp ) { | |
| 2145 Label did_not_overflow; | |
| 2146 Label overflow_with_error; | |
| 2147 assert_different_registers(backedge_count, Rtmp, branch_bcp); | |
| 2148 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); | |
| 2149 | |
| 727 | 2150 AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit); |
| 0 | 2151 load_contents(limit, Rtmp); |
| 3839 | 2152 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow); |
| 0 | 2153 |
| 2154 // When ProfileInterpreter is on, the backedge_count comes from the | |
| 2155 // methodDataOop, which value does not get reset on the call to | |
| 2156 // frequency_counter_overflow(). To avoid excessive calls to the overflow | |
| 2157 // routine while the method is being compiled, add a second test to make sure | |
| 2158 // the overflow function is called only once every overflow_frequency. | |
| 2159 if (ProfileInterpreter) { | |
| 2160 const int overflow_frequency = 1024; | |
| 2161 andcc(backedge_count, overflow_frequency-1, Rtmp); | |
| 2162 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); | |
| 2163 delayed()->nop(); | |
| 2164 } | |
| 2165 | |
| 2166 // overflow in loop, pass branch bytecode | |
| 2167 set(6,Rtmp); | |
| 2168 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); | |
| 2169 | |
| 2170 // Was an OSR adapter generated? | |
| 2171 // O0 = osr nmethod | |
| 3839 | 2172 br_null_short(O0, Assembler::pn, overflow_with_error); |
| 0 | 2173 |
| 2174 // Has the nmethod been invalidated already? | |
| 2175 ld(O0, nmethod::entry_bci_offset(), O2); | |
| 3839 | 2176 cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, overflow_with_error); |
| 0 | 2177 |
| 2178 // migrate the interpreter frame off of the stack | |
| 2179 | |
| 2180 mov(G2_thread, L7); | |
| 2181 // save nmethod | |
| 2182 mov(O0, L6); | |
| 2183 set_last_Java_frame(SP, noreg); | |
| 2184 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); | |
| 2185 reset_last_Java_frame(); | |
| 2186 mov(L7, G2_thread); | |
| 2187 | |
| 2188 // move OSR nmethod to I1 | |
| 2189 mov(L6, I1); | |
| 2190 | |
| 2191 // OSR buffer to I0 | |
| 2192 mov(O0, I0); | |
| 2193 | |
| 2194 // remove the interpreter frame | |
| 2195 restore(I5_savedSP, 0, SP); | |
| 2196 | |
| 2197 // Jump to the osr code. | |
| 2198 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); | |
| 2199 jmp(O2, G0); | |
| 2200 delayed()->nop(); | |
| 2201 | |
| 2202 bind(overflow_with_error); | |
| 2203 | |
| 2204 bind(did_not_overflow); | |
| 2205 } | |
| 2206 | |
| 2207 | |
| 2208 | |
| 2209 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { | |
| 2210 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } | |
| 2211 } | |
| 2212 | |
| 2213 | |
| 2214 // local helper function for the verify_oop_or_return_address macro | |
| 2215 static bool verify_return_address(methodOopDesc* m, int bci) { | |
| 2216 #ifndef PRODUCT | |
| 2217 address pc = (address)(m->constMethod()) | |
| 2218 + in_bytes(constMethodOopDesc::codes_offset()) + bci; | |
| 2219 // assume it is a valid return address if it is inside m and is preceded by a jsr | |
| 2220 if (!m->contains(pc)) return false; | |
| 2221 address jsr_pc; | |
| 2222 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); | |
| 2223 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; | |
| 2224 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); | |
| 2225 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; | |
| 2226 #endif // PRODUCT | |
| 2227 return false; | |
| 2228 } | |
| 2229 | |
| 2230 | |
| 2231 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { | |
| 2232 if (!VerifyOops) return; | |
| 2233 // the VM documentation for the astore[_wide] bytecode allows | |
| 2234 // the TOS to be not only an oop but also a return address | |
| 2235 Label test; | |
| 2236 Label skip; | |
| 2237 // See if it is an address (in the current method): | |
| 2238 | |
| 2239 mov(reg, Rtmp); | |
| 2240 const int log2_bytecode_size_limit = 16; | |
| 2241 srl(Rtmp, log2_bytecode_size_limit, Rtmp); | |
| 3839 | 2242 br_notnull_short( Rtmp, pt, test ); |
| 0 | 2243 |
| 2244 // %%% should use call_VM_leaf here? | |
| 2245 save_frame_and_mov(0, Lmethod, O0, reg, O1); | |
| 2246 save_thread(L7_thread_cache); | |
| 2247 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); | |
| 2248 delayed()->nop(); | |
| 2249 restore_thread(L7_thread_cache); | |
| 2250 br_notnull( O0, false, pt, skip ); | |
| 2251 delayed()->restore(); | |
| 2252 | |
| 2253 // Perform a more elaborate out-of-line call | |
| 2254 // Not an address; verify it: | |
| 2255 bind(test); | |
| 2256 verify_oop(reg); | |
| 2257 bind(skip); | |
| 2258 } | |
| 2259 | |
| 2260 | |
| 2261 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { | |
| 2262 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); | |
| 2263 } | |
| 2264 #endif /* CC_INTERP */ | |
| 2265 | |
| 2266 // Inline assembly for: | |
| 2267 // | |
| 2268 // if (thread is in interp_only_mode) { | |
| 2269 // InterpreterRuntime::post_method_entry(); | |
| 2270 // } | |
| 2271 // if (DTraceMethodProbes) { | |
| 605 | 2272 // SharedRuntime::dtrace_method_entry(method, receiver); |
| 0 | 2273 // } |
|
610
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2274 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2275 // SharedRuntime::rc_trace_method_entry(method, receiver); |
| 0 | 2276 // } |
| 2277 | |
| 2278 void InterpreterMacroAssembler::notify_method_entry() { | |
| 2279 | |
| 2280 // C++ interpreter only uses this for native methods. | |
| 2281 | |
| 2282 // Whenever JVMTI puts a thread in interp_only_mode, method | |
| 2283 // entry/exit events are sent for that thread to track stack | |
| 2284 // depth. If it is possible to enter interp_only_mode we add | |
| 2285 // the code to check if the event should be sent. | |
| 2286 if (JvmtiExport::can_post_interpreter_events()) { | |
| 2287 Label L; | |
| 2288 Register temp_reg = O5; | |
| 727 | 2289 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
| 0 | 2290 ld(interp_only, temp_reg); |
| 3839 | 2291 cmp_and_br_short(temp_reg, 0, equal, pt, L); |
| 0 | 2292 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); |
| 2293 bind(L); | |
| 2294 } | |
| 2295 | |
| 2296 { | |
| 2297 Register temp_reg = O5; | |
| 2298 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); | |
| 2299 call_VM_leaf(noreg, | |
| 2300 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), | |
| 2301 G2_thread, Lmethod); | |
| 2302 } | |
|
610
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2303 |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2304 // RedefineClasses() tracing support for obsolete method entry |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2305 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) { |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2306 call_VM_leaf(noreg, |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2307 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2308 G2_thread, Lmethod); |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2309 } |
| 0 | 2310 } |
| 2311 | |
| 2312 | |
| 2313 // Inline assembly for: | |
| 2314 // | |
| 2315 // if (thread is in interp_only_mode) { | |
| 2316 // // save result | |
| 2317 // InterpreterRuntime::post_method_exit(); | |
| 2318 // // restore result | |
| 2319 // } | |
| 2320 // if (DTraceMethodProbes) { | |
| 2321 // SharedRuntime::dtrace_method_exit(thread, method); | |
| 2322 // } | |
| 2323 // | |
| 2324 // Native methods have their result stored in d_tmp and l_tmp | |
| 2325 // Java methods have their result stored in the expression stack | |
| 2326 | |
| 2327 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, | |
| 2328 TosState state, | |
| 2329 NotifyMethodExitMode mode) { | |
| 2330 // C++ interpreter only uses this for native methods. | |
| 2331 | |
| 2332 // Whenever JVMTI puts a thread in interp_only_mode, method | |
| 2333 // entry/exit events are sent for that thread to track stack | |
| 2334 // depth. If it is possible to enter interp_only_mode we add | |
| 2335 // the code to check if the event should be sent. | |
| 2336 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { | |
| 2337 Label L; | |
| 2338 Register temp_reg = O5; | |
| 727 | 2339 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
| 0 | 2340 ld(interp_only, temp_reg); |
| 3839 | 2341 cmp_and_br_short(temp_reg, 0, equal, pt, L); |
| 0 | 2342 |
| 2343 // Note: frame::interpreter_frame_result has a dependency on how the | |
| 2344 // method result is saved across the call to post_method_exit. For | |
| 2345 // native methods it assumes the result registers are saved to | |
| 2346 // l_scratch and d_scratch. If this changes then the interpreter_frame_result | |
| 2347 // implementation will need to be updated too. | |
| 2348 | |
| 2349 save_return_value(state, is_native_method); | |
| 2350 call_VM(noreg, | |
| 2351 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); | |
| 2352 restore_return_value(state, is_native_method); | |
| 2353 bind(L); | |
| 2354 } | |
| 2355 | |
| 2356 { | |
| 2357 Register temp_reg = O5; | |
| 2358 // Dtrace notification | |
| 2359 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); | |
| 2360 save_return_value(state, is_native_method); | |
| 2361 call_VM_leaf( | |
| 2362 noreg, | |
| 2363 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), | |
| 2364 G2_thread, Lmethod); | |
| 2365 restore_return_value(state, is_native_method); | |
| 2366 } | |
| 2367 } | |
| 2368 | |
| 2369 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { | |
| 2370 #ifdef CC_INTERP | |
| 2371 // result potentially in O0/O1: save it across calls | |
| 2372 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult)); | |
| 2373 #ifdef _LP64 | |
| 2374 stx(O0, STATE(_native_lresult)); | |
| 2375 #else | |
| 2376 std(O0, STATE(_native_lresult)); | |
| 2377 #endif | |
| 2378 #else // CC_INTERP | |
| 2379 if (is_native_call) { | |
| 2380 stf(FloatRegisterImpl::D, F0, d_tmp); | |
| 2381 #ifdef _LP64 | |
| 2382 stx(O0, l_tmp); | |
| 2383 #else | |
| 2384 std(O0, l_tmp); | |
| 2385 #endif | |
| 2386 } else { | |
| 2387 push(state); | |
| 2388 } | |
| 2389 #endif // CC_INTERP | |
| 2390 } | |
| 2391 | |
| 2392 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { | |
| 2393 #ifdef CC_INTERP | |
| 2394 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0); | |
| 2395 #ifdef _LP64 | |
| 2396 ldx(STATE(_native_lresult), O0); | |
| 2397 #else | |
| 2398 ldd(STATE(_native_lresult), O0); | |
| 2399 #endif | |
| 2400 #else // CC_INTERP | |
| 2401 if (is_native_call) { | |
| 2402 ldf(FloatRegisterImpl::D, d_tmp, F0); | |
| 2403 #ifdef _LP64 | |
| 2404 ldx(l_tmp, O0); | |
| 2405 #else | |
| 2406 ldd(l_tmp, O0); | |
| 2407 #endif | |
| 2408 } else { | |
| 2409 pop(state); | |
| 2410 } | |
| 2411 #endif // CC_INTERP | |
| 2412 } | |
| 1783 | 2413 |
| 2414 // Jump if ((*counter_addr += increment) & mask) satisfies the condition. | |
| 2415 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, | |
| 2416 int increment, int mask, | |
| 2417 Register scratch1, Register scratch2, | |
| 2418 Condition cond, Label *where) { | |
| 2419 ld(counter_addr, scratch1); | |
| 2420 add(scratch1, increment, scratch1); | |
| 2421 if (is_simm13(mask)) { | |
| 2422 andcc(scratch1, mask, G0); | |
| 2423 } else { | |
| 2424 set(mask, scratch2); | |
| 2425 andcc(scratch1, scratch2, G0); | |
| 2426 } | |
| 2427 br(cond, false, Assembler::pn, *where); | |
| 2428 delayed()->st(scratch1, counter_addr); | |
| 2429 } |