Mercurial > hg > truffle
annotate src/cpu/sparc/vm/interp_masm_sparc.cpp @ 1367:9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
Summary: Use BIS instruction for allocation prefetch on Sparc
Reviewed-by: twisti
| author | kvn |
|---|---|
| date | Wed, 07 Apr 2010 12:39:27 -0700 |
| parents | 3cf667df43ef |
| children | c640000b7cc1 |
| rev | line source |
|---|---|
| 0 | 1 /* |
|
1251
576e77447e3c
6923002: assert(false,"this call site should not be polymorphic")
kvn
parents:
1206
diff
changeset
|
2 * Copyright 1997-2010 Sun Microsystems, Inc. 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 * | |
| 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
| 20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
| 21 * have any questions. | |
| 22 * | |
| 23 */ | |
| 24 | |
| 25 #include "incls/_precompiled.incl" | |
| 26 #include "incls/_interp_masm_sparc.cpp.incl" | |
| 27 | |
| 28 #ifndef CC_INTERP | |
| 29 #ifndef FAST_DISPATCH | |
| 30 #define FAST_DISPATCH 1 | |
| 31 #endif | |
| 32 #undef FAST_DISPATCH | |
| 33 | |
| 34 // Implementation of InterpreterMacroAssembler | |
| 35 | |
| 36 // This file specializes the assember with interpreter-specific macros | |
| 37 | |
| 727 | 38 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS); |
| 39 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS); | |
| 0 | 40 |
| 41 #else // CC_INTERP | |
| 42 #ifndef STATE | |
| 43 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name)) | |
| 44 #endif // STATE | |
| 45 | |
| 46 #endif // CC_INTERP | |
| 47 | |
| 48 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) { | |
| 49 // Note: this algorithm is also used by C1's OSR entry sequence. | |
| 50 // Any changes should also be applied to CodeEmitter::emit_osr_entry(). | |
| 51 assert_different_registers(args_size, locals_size); | |
| 52 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted. | |
| 53 if (TaggedStackInterpreter) sll(locals_size, 1, locals_size); | |
| 54 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words | |
| 55 // Use br/mov combination because it works on both V8 and V9 and is | |
| 56 // faster. | |
| 57 Label skip_move; | |
| 58 br(Assembler::negative, true, Assembler::pt, skip_move); | |
| 59 delayed()->mov(G0, delta); | |
| 60 bind(skip_move); | |
| 61 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned) | |
| 62 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes | |
| 63 } | |
| 64 | |
| 65 #ifndef CC_INTERP | |
| 66 | |
| 67 // Dispatch code executed in the prolog of a bytecode which does not do it's | |
| 68 // own dispatch. The dispatch address is computed and placed in IdispatchAddress | |
| 69 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { | |
| 70 assert_not_delayed(); | |
| 71 #ifdef FAST_DISPATCH | |
| 72 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since | |
| 73 // they both use I2. | |
| 74 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive"); | |
| 75 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
| 76 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); | |
| 77 // add offset to correct dispatch table | |
| 78 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
| 79 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr | |
| 80 #else | |
| 727 | 81 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode |
| 0 | 82 // dispatch table to use |
| 727 | 83 AddressLiteral tbl(Interpreter::dispatch_table(state)); |
| 84 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
| 85 set(tbl, G3_scratch); // compute addr of table | |
| 86 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr | |
| 0 | 87 #endif |
| 88 } | |
| 89 | |
| 90 | |
| 91 // Dispatch code executed in the epilog of a bytecode which does not do it's | |
| 92 // own dispatch. The dispatch address in IdispatchAddress is used for the | |
| 93 // dispatch. | |
| 94 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { | |
| 95 assert_not_delayed(); | |
| 96 verify_FPU(1, state); | |
| 97 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 98 jmp( IdispatchAddress, 0 ); | |
| 99 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); | |
| 100 else delayed()->nop(); | |
| 101 } | |
| 102 | |
| 103 | |
| 104 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { | |
| 105 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) | |
| 106 assert_not_delayed(); | |
| 107 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
| 108 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr); | |
| 109 } | |
| 110 | |
| 111 | |
| 112 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) { | |
| 113 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr) | |
| 114 assert_not_delayed(); | |
| 115 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode | |
| 116 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false); | |
| 117 } | |
| 118 | |
| 119 | |
| 120 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { | |
| 121 // load current bytecode | |
| 122 assert_not_delayed(); | |
| 123 ldub( Lbcp, 0, Lbyte_code); // load next bytecode | |
| 124 dispatch_base(state, table); | |
| 125 } | |
| 126 | |
| 127 | |
| 128 void InterpreterMacroAssembler::call_VM_leaf_base( | |
| 129 Register java_thread, | |
| 130 address entry_point, | |
| 131 int number_of_arguments | |
| 132 ) { | |
| 133 if (!java_thread->is_valid()) | |
| 134 java_thread = L7_thread_cache; | |
| 135 // super call | |
| 136 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments); | |
| 137 } | |
| 138 | |
| 139 | |
| 140 void InterpreterMacroAssembler::call_VM_base( | |
| 141 Register oop_result, | |
| 142 Register java_thread, | |
| 143 Register last_java_sp, | |
| 144 address entry_point, | |
| 145 int number_of_arguments, | |
| 146 bool check_exception | |
| 147 ) { | |
| 148 if (!java_thread->is_valid()) | |
| 149 java_thread = L7_thread_cache; | |
| 150 // See class ThreadInVMfromInterpreter, which assumes that the interpreter | |
| 151 // takes responsibility for setting its own thread-state on call-out. | |
| 152 // However, ThreadInVMfromInterpreter resets the state to "in_Java". | |
| 153 | |
| 154 //save_bcp(); // save bcp | |
| 155 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception); | |
| 156 //restore_bcp(); // restore bcp | |
| 157 //restore_locals(); // restore locals pointer | |
| 158 } | |
| 159 | |
| 160 | |
| 161 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { | |
| 162 if (JvmtiExport::can_pop_frame()) { | |
| 163 Label L; | |
| 164 | |
| 165 // Check the "pending popframe condition" flag in the current thread | |
| 727 | 166 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg); |
| 0 | 167 |
| 168 // Initiate popframe handling only if it is not already being processed. If the flag | |
| 169 // has the popframe_processing bit set, it means that this code is called *during* popframe | |
| 170 // handling - we don't want to reenter. | |
| 171 btst(JavaThread::popframe_pending_bit, scratch_reg); | |
| 172 br(zero, false, pt, L); | |
| 173 delayed()->nop(); | |
| 174 btst(JavaThread::popframe_processing_bit, scratch_reg); | |
| 175 br(notZero, false, pt, L); | |
| 176 delayed()->nop(); | |
| 177 | |
| 178 // Call Interpreter::remove_activation_preserving_args_entry() to get the | |
| 179 // address of the same-named entrypoint in the generated interpreter code. | |
| 180 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); | |
| 181 | |
| 182 // Jump to Interpreter::_remove_activation_preserving_args_entry | |
| 183 jmpl(O0, G0, G0); | |
| 184 delayed()->nop(); | |
| 185 bind(L); | |
| 186 } | |
| 187 } | |
| 188 | |
| 189 | |
| 190 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { | |
| 191 Register thr_state = G4_scratch; | |
| 727 | 192 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); |
| 193 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset()); | |
| 194 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset()); | |
| 195 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset()); | |
| 0 | 196 switch (state) { |
| 197 case ltos: ld_long(val_addr, Otos_l); break; | |
| 198 case atos: ld_ptr(oop_addr, Otos_l); | |
| 199 st_ptr(G0, oop_addr); break; | |
| 200 case btos: // fall through | |
| 201 case ctos: // fall through | |
| 202 case stos: // fall through | |
| 203 case itos: ld(val_addr, Otos_l1); break; | |
| 204 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break; | |
| 205 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break; | |
| 206 case vtos: /* nothing to do */ break; | |
| 207 default : ShouldNotReachHere(); | |
| 208 } | |
| 209 // Clean up tos value in the jvmti thread state | |
| 210 or3(G0, ilgl, G3_scratch); | |
| 211 stw(G3_scratch, tos_addr); | |
| 212 st_long(G0, val_addr); | |
| 213 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 214 } | |
| 215 | |
| 216 | |
| 217 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { | |
| 218 if (JvmtiExport::can_force_early_return()) { | |
| 219 Label L; | |
| 220 Register thr_state = G3_scratch; | |
| 727 | 221 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state); |
| 0 | 222 tst(thr_state); |
| 223 br(zero, false, pt, L); // if (thread->jvmti_thread_state() == NULL) exit; | |
| 224 delayed()->nop(); | |
| 225 | |
| 226 // Initiate earlyret handling only if it is not already being processed. | |
| 227 // If the flag has the earlyret_processing bit set, it means that this code | |
| 228 // is called *during* earlyret handling - we don't want to reenter. | |
| 727 | 229 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch); |
| 0 | 230 cmp(G4_scratch, JvmtiThreadState::earlyret_pending); |
| 231 br(Assembler::notEqual, false, pt, L); | |
| 232 delayed()->nop(); | |
| 233 | |
| 234 // Call Interpreter::remove_activation_early_entry() to get the address of the | |
| 235 // same-named entrypoint in the generated interpreter code | |
| 727 | 236 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1); |
| 0 | 237 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1); |
| 238 | |
| 239 // Jump to Interpreter::_remove_activation_early_entry | |
| 240 jmpl(O0, G0, G0); | |
| 241 delayed()->nop(); | |
| 242 bind(L); | |
| 243 } | |
| 244 } | |
| 245 | |
| 246 | |
| 1295 | 247 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { |
| 0 | 248 mov(arg_1, O0); |
| 1295 | 249 mov(arg_2, O1); |
| 250 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2); | |
| 0 | 251 } |
| 252 #endif /* CC_INTERP */ | |
| 253 | |
| 254 | |
| 255 #ifndef CC_INTERP | |
| 256 | |
| 257 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) { | |
| 258 assert_not_delayed(); | |
| 259 dispatch_Lbyte_code(state, table); | |
| 260 } | |
| 261 | |
| 262 | |
| 263 void InterpreterMacroAssembler::dispatch_normal(TosState state) { | |
| 264 dispatch_base(state, Interpreter::normal_table(state)); | |
| 265 } | |
| 266 | |
| 267 | |
| 268 void InterpreterMacroAssembler::dispatch_only(TosState state) { | |
| 269 dispatch_base(state, Interpreter::dispatch_table(state)); | |
| 270 } | |
| 271 | |
| 272 | |
| 273 // common code to dispatch and dispatch_only | |
| 274 // dispatch value in Lbyte_code and increment Lbcp | |
| 275 | |
| 276 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) { | |
| 277 verify_FPU(1, state); | |
| 278 // %%%%% maybe implement +VerifyActivationFrameSize here | |
| 279 //verify_thread(); //too slow; we will just verify on method entry & exit | |
| 280 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 281 #ifdef FAST_DISPATCH | |
| 282 if (table == Interpreter::dispatch_table(state)) { | |
| 283 // use IdispatchTables | |
| 284 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code); | |
| 285 // add offset to correct dispatch table | |
| 286 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize | |
| 287 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr | |
| 288 } else { | |
| 289 #endif | |
| 290 // dispatch table to use | |
| 727 | 291 AddressLiteral tbl(table); |
| 0 | 292 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize |
| 727 | 293 set(tbl, G3_scratch); // compute addr of table |
| 0 | 294 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr |
| 295 #ifdef FAST_DISPATCH | |
| 296 } | |
| 297 #endif | |
| 298 jmp( G3_scratch, 0 ); | |
| 299 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr); | |
| 300 else delayed()->nop(); | |
| 301 } | |
| 302 | |
| 303 | |
| 304 // Helpers for expression stack | |
| 305 | |
| 306 // Longs and doubles are Category 2 computational types in the | |
| 307 // JVM specification (section 3.11.1) and take 2 expression stack or | |
| 308 // local slots. | |
| 309 // Aligning them on 32 bit with tagged stacks is hard because the code generated | |
| 310 // for the dup* bytecodes depends on what types are already on the stack. | |
| 311 // If the types are split into the two stack/local slots, that is much easier | |
| 312 // (and we can use 0 for non-reference tags). | |
| 313 | |
| 314 // Known good alignment in _LP64 but unknown otherwise | |
| 315 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) { | |
| 316 assert_not_delayed(); | |
| 317 | |
| 318 #ifdef _LP64 | |
| 319 ldf(FloatRegisterImpl::D, r1, offset, d); | |
| 320 #else | |
| 321 ldf(FloatRegisterImpl::S, r1, offset, d); | |
| 322 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize(), d->successor()); | |
| 323 #endif | |
| 324 } | |
| 325 | |
| 326 // Known good alignment in _LP64 but unknown otherwise | |
| 327 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) { | |
| 328 assert_not_delayed(); | |
| 329 | |
| 330 #ifdef _LP64 | |
| 331 stf(FloatRegisterImpl::D, d, r1, offset); | |
| 332 // store something more useful here | |
| 333 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize());) | |
| 334 #else | |
| 335 stf(FloatRegisterImpl::S, d, r1, offset); | |
| 336 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize()); | |
| 337 #endif | |
| 338 } | |
| 339 | |
| 340 | |
| 341 // Known good alignment in _LP64 but unknown otherwise | |
| 342 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) { | |
| 343 assert_not_delayed(); | |
| 344 #ifdef _LP64 | |
| 345 ldx(r1, offset, rd); | |
| 346 #else | |
| 347 ld(r1, offset, rd); | |
| 348 ld(r1, offset + Interpreter::stackElementSize(), rd->successor()); | |
| 349 #endif | |
| 350 } | |
| 351 | |
| 352 // Known good alignment in _LP64 but unknown otherwise | |
| 353 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) { | |
| 354 assert_not_delayed(); | |
| 355 | |
| 356 #ifdef _LP64 | |
| 357 stx(l, r1, offset); | |
| 358 // store something more useful here | |
| 359 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize());) | |
| 360 #else | |
| 361 st(l, r1, offset); | |
| 362 st(l->successor(), r1, offset + Interpreter::stackElementSize()); | |
| 363 #endif | |
| 364 } | |
| 365 | |
| 366 #ifdef ASSERT | |
| 367 void InterpreterMacroAssembler::verify_stack_tag(frame::Tag t, | |
| 368 Register r, | |
| 369 Register scratch) { | |
| 370 if (TaggedStackInterpreter) { | |
| 371 Label ok, long_ok; | |
| 372 ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(0), r); | |
| 373 if (t == frame::TagCategory2) { | |
| 374 cmp(r, G0); | |
| 375 brx(Assembler::equal, false, Assembler::pt, long_ok); | |
| 376 delayed()->ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(1), r); | |
| 377 stop("stack long/double tag value bad"); | |
| 378 bind(long_ok); | |
| 379 cmp(r, G0); | |
| 380 } else if (t == frame::TagValue) { | |
| 381 cmp(r, G0); | |
| 382 } else { | |
| 383 assert_different_registers(r, scratch); | |
| 384 mov(t, scratch); | |
| 385 cmp(r, scratch); | |
| 386 } | |
| 387 brx(Assembler::equal, false, Assembler::pt, ok); | |
| 388 delayed()->nop(); | |
| 389 // Also compare if the stack value is zero, then the tag might | |
| 390 // not have been set coming from deopt. | |
| 391 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
| 392 cmp(r, G0); | |
| 393 brx(Assembler::equal, false, Assembler::pt, ok); | |
| 394 delayed()->nop(); | |
| 395 stop("Stack tag value is bad"); | |
| 396 bind(ok); | |
| 397 } | |
| 398 } | |
| 399 #endif // ASSERT | |
| 400 | |
| 401 void InterpreterMacroAssembler::pop_i(Register r) { | |
| 402 assert_not_delayed(); | |
| 403 // Uses destination register r for scratch | |
| 404 debug_only(verify_stack_tag(frame::TagValue, r)); | |
| 405 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
| 406 inc(Lesp, Interpreter::stackElementSize()); | |
| 407 debug_only(verify_esp(Lesp)); | |
| 408 } | |
| 409 | |
| 410 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) { | |
| 411 assert_not_delayed(); | |
| 412 // Uses destination register r for scratch | |
| 413 debug_only(verify_stack_tag(frame::TagReference, r, scratch)); | |
| 414 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
| 415 inc(Lesp, Interpreter::stackElementSize()); | |
| 416 debug_only(verify_esp(Lesp)); | |
| 417 } | |
| 418 | |
| 419 void InterpreterMacroAssembler::pop_l(Register r) { | |
| 420 assert_not_delayed(); | |
| 421 // Uses destination register r for scratch | |
| 422 debug_only(verify_stack_tag(frame::TagCategory2, r)); | |
| 423 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r); | |
| 424 inc(Lesp, 2*Interpreter::stackElementSize()); | |
| 425 debug_only(verify_esp(Lesp)); | |
| 426 } | |
| 427 | |
| 428 | |
| 429 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) { | |
| 430 assert_not_delayed(); | |
| 431 debug_only(verify_stack_tag(frame::TagValue, scratch)); | |
| 432 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f); | |
| 433 inc(Lesp, Interpreter::stackElementSize()); | |
| 434 debug_only(verify_esp(Lesp)); | |
| 435 } | |
| 436 | |
| 437 | |
| 438 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) { | |
| 439 assert_not_delayed(); | |
| 440 debug_only(verify_stack_tag(frame::TagCategory2, scratch)); | |
| 441 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f); | |
| 442 inc(Lesp, 2*Interpreter::stackElementSize()); | |
| 443 debug_only(verify_esp(Lesp)); | |
| 444 } | |
| 445 | |
| 446 | |
| 447 // (Note use register first, then decrement so dec can be done during store stall) | |
| 448 void InterpreterMacroAssembler::tag_stack(Register r) { | |
| 449 if (TaggedStackInterpreter) { | |
| 450 st_ptr(r, Lesp, Interpreter::tag_offset_in_bytes()); | |
| 451 } | |
| 452 } | |
| 453 | |
| 454 void InterpreterMacroAssembler::tag_stack(frame::Tag t, Register r) { | |
| 455 if (TaggedStackInterpreter) { | |
| 456 assert (frame::TagValue == 0, "TagValue must be zero"); | |
| 457 if (t == frame::TagValue) { | |
| 458 st_ptr(G0, Lesp, Interpreter::tag_offset_in_bytes()); | |
| 459 } else if (t == frame::TagCategory2) { | |
| 460 st_ptr(G0, Lesp, Interpreter::tag_offset_in_bytes()); | |
| 461 // Tag next slot down too | |
| 462 st_ptr(G0, Lesp, -Interpreter::stackElementSize() + Interpreter::tag_offset_in_bytes()); | |
| 463 } else { | |
| 464 assert_different_registers(r, O3); | |
| 465 mov(t, O3); | |
| 466 st_ptr(O3, Lesp, Interpreter::tag_offset_in_bytes()); | |
| 467 } | |
| 468 } | |
| 469 } | |
| 470 | |
| 471 void InterpreterMacroAssembler::push_i(Register r) { | |
| 472 assert_not_delayed(); | |
| 473 debug_only(verify_esp(Lesp)); | |
| 474 tag_stack(frame::TagValue, r); | |
| 475 st( r, Lesp, Interpreter::value_offset_in_bytes()); | |
| 476 dec( Lesp, Interpreter::stackElementSize()); | |
| 477 } | |
| 478 | |
| 479 void InterpreterMacroAssembler::push_ptr(Register r) { | |
| 480 assert_not_delayed(); | |
| 481 tag_stack(frame::TagReference, r); | |
| 482 st_ptr( r, Lesp, Interpreter::value_offset_in_bytes()); | |
| 483 dec( Lesp, Interpreter::stackElementSize()); | |
| 484 } | |
| 485 | |
| 486 void InterpreterMacroAssembler::push_ptr(Register r, Register tag) { | |
| 487 assert_not_delayed(); | |
| 488 tag_stack(tag); | |
| 489 st_ptr(r, Lesp, Interpreter::value_offset_in_bytes()); | |
| 490 dec( Lesp, Interpreter::stackElementSize()); | |
| 491 } | |
| 492 | |
| 493 // remember: our convention for longs in SPARC is: | |
| 494 // O0 (Otos_l1) has high-order part in first word, | |
| 495 // O1 (Otos_l2) has low-order part in second word | |
| 496 | |
| 497 void InterpreterMacroAssembler::push_l(Register r) { | |
| 498 assert_not_delayed(); | |
| 499 debug_only(verify_esp(Lesp)); | |
| 500 tag_stack(frame::TagCategory2, r); | |
| 501 // Longs are in stored in memory-correct order, even if unaligned. | |
| 502 // and may be separated by stack tags. | |
| 503 int offset = -Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); | |
| 504 store_unaligned_long(r, Lesp, offset); | |
| 505 dec(Lesp, 2 * Interpreter::stackElementSize()); | |
| 506 } | |
| 507 | |
| 508 | |
| 509 void InterpreterMacroAssembler::push_f(FloatRegister f) { | |
| 510 assert_not_delayed(); | |
| 511 debug_only(verify_esp(Lesp)); | |
| 512 tag_stack(frame::TagValue, Otos_i); | |
| 513 stf(FloatRegisterImpl::S, f, Lesp, Interpreter::value_offset_in_bytes()); | |
| 514 dec(Lesp, Interpreter::stackElementSize()); | |
| 515 } | |
| 516 | |
| 517 | |
| 518 void InterpreterMacroAssembler::push_d(FloatRegister d) { | |
| 519 assert_not_delayed(); | |
| 520 debug_only(verify_esp(Lesp)); | |
| 521 tag_stack(frame::TagCategory2, Otos_i); | |
| 522 // Longs are in stored in memory-correct order, even if unaligned. | |
| 523 // and may be separated by stack tags. | |
| 524 int offset = -Interpreter::stackElementSize() + Interpreter::value_offset_in_bytes(); | |
| 525 store_unaligned_double(d, Lesp, offset); | |
| 526 dec(Lesp, 2 * Interpreter::stackElementSize()); | |
| 527 } | |
| 528 | |
| 529 | |
| 530 void InterpreterMacroAssembler::push(TosState state) { | |
| 531 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 532 switch (state) { | |
| 533 case atos: push_ptr(); break; | |
| 534 case btos: push_i(); break; | |
| 535 case ctos: | |
| 536 case stos: push_i(); break; | |
| 537 case itos: push_i(); break; | |
| 538 case ltos: push_l(); break; | |
| 539 case ftos: push_f(); break; | |
| 540 case dtos: push_d(); break; | |
| 541 case vtos: /* nothing to do */ break; | |
| 542 default : ShouldNotReachHere(); | |
| 543 } | |
| 544 } | |
| 545 | |
| 546 | |
| 547 void InterpreterMacroAssembler::pop(TosState state) { | |
| 548 switch (state) { | |
| 549 case atos: pop_ptr(); break; | |
| 550 case btos: pop_i(); break; | |
| 551 case ctos: | |
| 552 case stos: pop_i(); break; | |
| 553 case itos: pop_i(); break; | |
| 554 case ltos: pop_l(); break; | |
| 555 case ftos: pop_f(); break; | |
| 556 case dtos: pop_d(); break; | |
| 557 case vtos: /* nothing to do */ break; | |
| 558 default : ShouldNotReachHere(); | |
| 559 } | |
| 560 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 561 } | |
| 562 | |
| 563 | |
| 564 // Tagged stack helpers for swap and dup | |
| 565 void InterpreterMacroAssembler::load_ptr_and_tag(int n, Register val, | |
| 566 Register tag) { | |
| 567 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val); | |
| 568 if (TaggedStackInterpreter) { | |
| 569 ld_ptr(Lesp, Interpreter::expr_tag_offset_in_bytes(n), tag); | |
| 570 } | |
| 571 } | |
| 572 void InterpreterMacroAssembler::store_ptr_and_tag(int n, Register val, | |
| 573 Register tag) { | |
| 574 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n)); | |
| 575 if (TaggedStackInterpreter) { | |
| 576 st_ptr(tag, Lesp, Interpreter::expr_tag_offset_in_bytes(n)); | |
| 577 } | |
| 578 } | |
| 579 | |
| 580 | |
| 581 void InterpreterMacroAssembler::load_receiver(Register param_count, | |
| 582 Register recv) { | |
| 583 | |
| 584 sll(param_count, Interpreter::logStackElementSize(), param_count); | |
| 585 if (TaggedStackInterpreter) { | |
| 586 add(param_count, Interpreter::value_offset_in_bytes(), param_count); // get obj address | |
| 587 } | |
| 588 ld_ptr(Lesp, param_count, recv); // gets receiver Oop | |
| 589 } | |
| 590 | |
| 591 void InterpreterMacroAssembler::empty_expression_stack() { | |
| 592 // Reset Lesp. | |
| 593 sub( Lmonitors, wordSize, Lesp ); | |
| 594 | |
| 595 // Reset SP by subtracting more space from Lesp. | |
| 596 Label done; | |
| 597 verify_oop(Lmethod); | |
| 727 | 598 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!"); |
| 0 | 599 |
| 600 // A native does not need to do this, since its callee does not change SP. | |
| 727 | 601 ld(Lmethod, methodOopDesc::access_flags_offset(), Gframe_size); // Load access flags. |
| 0 | 602 btst(JVM_ACC_NATIVE, Gframe_size); |
| 603 br(Assembler::notZero, false, Assembler::pt, done); | |
| 604 delayed()->nop(); | |
| 605 | |
| 606 // Compute max expression stack+register save area | |
| 727 | 607 lduh(Lmethod, in_bytes(methodOopDesc::max_stack_offset()), Gframe_size); // Load max stack. |
| 0 | 608 if (TaggedStackInterpreter) sll ( Gframe_size, 1, Gframe_size); // max_stack * 2 for TAGS |
| 609 add( Gframe_size, frame::memory_parameter_word_sp_offset, Gframe_size ); | |
| 610 | |
| 611 // | |
| 612 // now set up a stack frame with the size computed above | |
| 613 // | |
| 614 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below | |
| 615 sll( Gframe_size, LogBytesPerWord, Gframe_size ); | |
| 616 sub( Lesp, Gframe_size, Gframe_size ); | |
| 617 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary | |
| 618 debug_only(verify_sp(Gframe_size, G4_scratch)); | |
| 619 #ifdef _LP64 | |
| 620 sub(Gframe_size, STACK_BIAS, Gframe_size ); | |
| 621 #endif | |
| 622 mov(Gframe_size, SP); | |
| 623 | |
| 624 bind(done); | |
| 625 } | |
| 626 | |
| 627 | |
| 628 #ifdef ASSERT | |
| 629 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) { | |
| 630 Label Bad, OK; | |
| 631 | |
| 632 // Saved SP must be aligned. | |
| 633 #ifdef _LP64 | |
| 634 btst(2*BytesPerWord-1, Rsp); | |
| 635 #else | |
| 636 btst(LongAlignmentMask, Rsp); | |
| 637 #endif | |
| 638 br(Assembler::notZero, false, Assembler::pn, Bad); | |
| 639 delayed()->nop(); | |
| 640 | |
| 641 // Saved SP, plus register window size, must not be above FP. | |
| 642 add(Rsp, frame::register_save_words * wordSize, Rtemp); | |
| 643 #ifdef _LP64 | |
| 644 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP | |
| 645 #endif | |
| 646 cmp(Rtemp, FP); | |
| 647 brx(Assembler::greaterUnsigned, false, Assembler::pn, Bad); | |
| 648 delayed()->nop(); | |
| 649 | |
| 650 // Saved SP must not be ridiculously below current SP. | |
| 651 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K); | |
| 652 set(maxstack, Rtemp); | |
| 653 sub(SP, Rtemp, Rtemp); | |
| 654 #ifdef _LP64 | |
| 655 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp | |
| 656 #endif | |
| 657 cmp(Rsp, Rtemp); | |
| 658 brx(Assembler::lessUnsigned, false, Assembler::pn, Bad); | |
| 659 delayed()->nop(); | |
| 660 | |
| 661 br(Assembler::always, false, Assembler::pn, OK); | |
| 662 delayed()->nop(); | |
| 663 | |
| 664 bind(Bad); | |
| 665 stop("on return to interpreted call, restored SP is corrupted"); | |
| 666 | |
| 667 bind(OK); | |
| 668 } | |
| 669 | |
| 670 | |
| 671 void InterpreterMacroAssembler::verify_esp(Register Resp) { | |
| 672 // about to read or write Resp[0] | |
| 673 // make sure it is not in the monitors or the register save area | |
| 674 Label OK1, OK2; | |
| 675 | |
| 676 cmp(Resp, Lmonitors); | |
| 677 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1); | |
| 678 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp); | |
| 679 stop("too many pops: Lesp points into monitor area"); | |
| 680 bind(OK1); | |
| 681 #ifdef _LP64 | |
| 682 sub(Resp, STACK_BIAS, Resp); | |
| 683 #endif | |
| 684 cmp(Resp, SP); | |
| 685 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2); | |
| 686 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp); | |
| 687 stop("too many pushes: Lesp points into register window"); | |
| 688 bind(OK2); | |
| 689 } | |
| 690 #endif // ASSERT | |
| 691 | |
| 692 // Load compiled (i2c) or interpreter entry when calling from interpreted and | |
| 693 // do the call. Centralized so that all interpreter calls will do the same actions. | |
| 694 // If jvmti single stepping is on for a thread we must not call compiled code. | |
| 695 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) { | |
| 696 | |
| 697 // Assume we want to go compiled if available | |
| 698 | |
| 699 ld_ptr(G5_method, in_bytes(methodOopDesc::from_interpreted_offset()), target); | |
| 700 | |
| 701 if (JvmtiExport::can_post_interpreter_events()) { | |
| 702 // JVMTI events, such as single-stepping, are implemented partly by avoiding running | |
| 703 // compiled code in threads for which the event is enabled. Check here for | |
| 704 // interp_only_mode if these events CAN be enabled. | |
| 705 verify_thread(); | |
| 706 Label skip_compiled_code; | |
| 707 | |
| 727 | 708 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
| 0 | 709 ld(interp_only, scratch); |
| 710 tst(scratch); | |
| 711 br(Assembler::notZero, true, Assembler::pn, skip_compiled_code); | |
| 712 delayed()->ld_ptr(G5_method, in_bytes(methodOopDesc::interpreter_entry_offset()), target); | |
| 713 bind(skip_compiled_code); | |
| 714 } | |
| 715 | |
| 716 // the i2c_adapters need methodOop in G5_method (right? %%%) | |
| 717 // do the call | |
| 718 #ifdef ASSERT | |
| 719 { | |
| 720 Label ok; | |
| 721 br_notnull(target, false, Assembler::pt, ok); | |
| 722 delayed()->nop(); | |
| 723 stop("null entry point"); | |
| 724 bind(ok); | |
| 725 } | |
| 726 #endif // ASSERT | |
| 727 | |
| 728 // Adjust Rret first so Llast_SP can be same as Rret | |
| 729 add(Rret, -frame::pc_return_offset, O7); | |
| 730 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer | |
| 731 // Record SP so we can remove any stack space allocated by adapter transition | |
| 732 jmp(target, 0); | |
| 733 delayed()->mov(SP, Llast_SP); | |
| 734 } | |
| 735 | |
| 736 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) { | |
| 737 assert_not_delayed(); | |
| 738 | |
| 739 Label not_taken; | |
| 740 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken); | |
| 741 else br (cc, false, Assembler::pn, not_taken); | |
| 742 delayed()->nop(); | |
| 743 | |
| 744 TemplateTable::branch(false,false); | |
| 745 | |
| 746 bind(not_taken); | |
| 747 | |
| 748 profile_not_taken_branch(G3_scratch); | |
| 749 } | |
| 750 | |
| 751 | |
| 752 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp( | |
| 753 int bcp_offset, | |
| 754 Register Rtmp, | |
| 755 Register Rdst, | |
| 756 signedOrNot is_signed, | |
| 757 setCCOrNot should_set_CC ) { | |
| 758 assert(Rtmp != Rdst, "need separate temp register"); | |
| 759 assert_not_delayed(); | |
| 760 switch (is_signed) { | |
| 761 default: ShouldNotReachHere(); | |
| 762 | |
| 763 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte | |
| 764 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte | |
| 765 } | |
| 766 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte | |
| 767 sll( Rdst, BitsPerByte, Rdst); | |
| 768 switch (should_set_CC ) { | |
| 769 default: ShouldNotReachHere(); | |
| 770 | |
| 771 case set_CC: orcc( Rdst, Rtmp, Rdst ); break; | |
| 772 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break; | |
| 773 } | |
| 774 } | |
| 775 | |
| 776 | |
| 777 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp( | |
| 778 int bcp_offset, | |
| 779 Register Rtmp, | |
| 780 Register Rdst, | |
| 781 setCCOrNot should_set_CC ) { | |
| 782 assert(Rtmp != Rdst, "need separate temp register"); | |
| 783 assert_not_delayed(); | |
| 784 add( Lbcp, bcp_offset, Rtmp); | |
| 785 andcc( Rtmp, 3, G0); | |
| 786 Label aligned; | |
| 787 switch (should_set_CC ) { | |
| 788 default: ShouldNotReachHere(); | |
| 789 | |
| 790 case set_CC: break; | |
| 791 case dont_set_CC: break; | |
| 792 } | |
| 793 | |
| 794 br(Assembler::zero, true, Assembler::pn, aligned); | |
| 795 #ifdef _LP64 | |
| 796 delayed()->ldsw(Rtmp, 0, Rdst); | |
| 797 #else | |
| 798 delayed()->ld(Rtmp, 0, Rdst); | |
| 799 #endif | |
| 800 | |
| 801 ldub(Lbcp, bcp_offset + 3, Rdst); | |
| 802 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst); | |
| 803 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst); | |
| 804 #ifdef _LP64 | |
| 805 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); | |
| 806 #else | |
| 807 // Unsigned load is faster than signed on some implementations | |
| 808 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp); | |
| 809 #endif | |
| 810 or3(Rtmp, Rdst, Rdst ); | |
| 811 | |
| 812 bind(aligned); | |
| 813 if (should_set_CC == set_CC) tst(Rdst); | |
| 814 } | |
| 815 | |
| 816 | |
| 817 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp, int bcp_offset) { | |
| 818 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); | |
| 819 assert_different_registers(cache, tmp); | |
| 820 assert_not_delayed(); | |
| 821 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); | |
| 822 // convert from field index to ConstantPoolCacheEntry index | |
| 823 // and from word index to byte offset | |
| 824 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); | |
| 825 add(LcpoolCache, tmp, cache); | |
| 826 } | |
| 827 | |
| 828 | |
| 829 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp, int bcp_offset) { | |
| 830 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); | |
| 831 assert_different_registers(cache, tmp); | |
| 832 assert_not_delayed(); | |
| 833 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned); | |
| 834 // convert from field index to ConstantPoolCacheEntry index | |
| 835 // and from word index to byte offset | |
| 836 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp); | |
| 837 // skip past the header | |
| 838 add(tmp, in_bytes(constantPoolCacheOopDesc::base_offset()), tmp); | |
| 839 // construct pointer to cache entry | |
| 840 add(LcpoolCache, tmp, cache); | |
| 841 } | |
| 842 | |
| 843 | |
| 844 // 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
|
845 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2. |
| 0 | 846 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, |
| 847 Register Rsuper_klass, | |
| 848 Register Rtmp1, | |
| 849 Register Rtmp2, | |
| 850 Register Rtmp3, | |
| 851 Label &ok_is_subtype ) { | |
|
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
852 Label not_subtype; |
| 0 | 853 |
| 854 // Profile the not-null value's klass. | |
| 855 profile_typecheck(Rsub_klass, Rtmp1); | |
| 856 | |
|
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
857 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
|
858 Rtmp1, Rtmp2, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
859 &ok_is_subtype, ¬_subtype, NULL); |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
860 |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
861 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
|
862 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, |
|
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
614
diff
changeset
|
863 &ok_is_subtype, NULL); |
| 0 | 864 |
| 865 bind(not_subtype); | |
| 866 profile_typecheck_failed(Rtmp1); | |
| 867 } | |
| 868 | |
| 869 // Separate these two to allow for delay slot in middle | |
| 870 // These are used to do a test and full jump to exception-throwing code. | |
| 871 | |
| 872 // %%%%% Could possibly reoptimize this by testing to see if could use | |
| 873 // a single conditional branch (i.e. if span is small enough. | |
| 874 // If you go that route, than get rid of the split and give up | |
| 875 // on the delay-slot hack. | |
| 876 | |
| 877 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition, | |
| 878 Label& ok ) { | |
| 879 assert_not_delayed(); | |
| 880 br(ok_condition, true, pt, ok); | |
| 881 // DELAY SLOT | |
| 882 } | |
| 883 | |
| 884 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition, | |
| 885 Label& ok ) { | |
| 886 assert_not_delayed(); | |
| 887 bp( ok_condition, true, Assembler::xcc, pt, ok); | |
| 888 // DELAY SLOT | |
| 889 } | |
| 890 | |
| 891 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition, | |
| 892 Label& ok ) { | |
| 893 assert_not_delayed(); | |
| 894 brx(ok_condition, true, pt, ok); | |
| 895 // DELAY SLOT | |
| 896 } | |
| 897 | |
| 898 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point, | |
| 899 Register Rscratch, | |
| 900 Label& ok ) { | |
| 901 assert(throw_entry_point != NULL, "entry point must be generated by now"); | |
| 727 | 902 AddressLiteral dest(throw_entry_point); |
| 903 jump_to(dest, Rscratch); | |
| 0 | 904 delayed()->nop(); |
| 905 bind(ok); | |
| 906 } | |
| 907 | |
| 908 | |
| 909 // And if you cannot use the delay slot, here is a shorthand: | |
| 910 | |
| 911 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition, | |
| 912 address throw_entry_point, | |
| 913 Register Rscratch ) { | |
| 914 Label ok; | |
| 915 if (ok_condition != never) { | |
| 916 throw_if_not_1_icc( ok_condition, ok); | |
| 917 delayed()->nop(); | |
| 918 } | |
| 919 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
| 920 } | |
| 921 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition, | |
| 922 address throw_entry_point, | |
| 923 Register Rscratch ) { | |
| 924 Label ok; | |
| 925 if (ok_condition != never) { | |
| 926 throw_if_not_1_xcc( ok_condition, ok); | |
| 927 delayed()->nop(); | |
| 928 } | |
| 929 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
| 930 } | |
| 931 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition, | |
| 932 address throw_entry_point, | |
| 933 Register Rscratch ) { | |
| 934 Label ok; | |
| 935 if (ok_condition != never) { | |
| 936 throw_if_not_1_x( ok_condition, ok); | |
| 937 delayed()->nop(); | |
| 938 } | |
| 939 throw_if_not_2( throw_entry_point, Rscratch, ok); | |
| 940 } | |
| 941 | |
| 942 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res | |
| 943 // Note: res is still shy of address by array offset into object. | |
| 944 | |
| 945 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) { | |
| 946 assert_not_delayed(); | |
| 947 | |
| 948 verify_oop(array); | |
| 949 #ifdef _LP64 | |
| 950 // sign extend since tos (index) can be a 32bit value | |
| 951 sra(index, G0, index); | |
| 952 #endif // _LP64 | |
| 953 | |
| 954 // check array | |
| 955 Label ptr_ok; | |
| 956 tst(array); | |
| 957 throw_if_not_1_x( notZero, ptr_ok ); | |
| 958 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index | |
| 959 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok); | |
| 960 | |
| 961 Label index_ok; | |
| 962 cmp(index, tmp); | |
| 963 throw_if_not_1_icc( lessUnsigned, index_ok ); | |
| 964 if (index_shift > 0) delayed()->sll(index, index_shift, index); | |
| 965 else delayed()->add(array, index, res); // addr - const offset in index | |
| 966 // convention: move aberrant index into G3_scratch for exception message | |
| 967 mov(index, G3_scratch); | |
| 968 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok); | |
| 969 | |
| 970 // add offset if didn't do it in delay slot | |
| 971 if (index_shift > 0) add(array, index, res); // addr - const offset in index | |
| 972 } | |
| 973 | |
| 974 | |
| 975 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { | |
| 976 assert_not_delayed(); | |
| 977 | |
| 978 // pop array | |
| 979 pop_ptr(array); | |
| 980 | |
| 981 // check array | |
| 982 index_check_without_pop(array, index, index_shift, tmp, res); | |
| 983 } | |
| 984 | |
| 985 | |
| 986 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { | |
| 987 ld_ptr(Lmethod, in_bytes(methodOopDesc::constants_offset()), Rdst); | |
| 988 } | |
| 989 | |
| 990 | |
| 991 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { | |
| 992 get_constant_pool(Rdst); | |
| 993 ld_ptr(Rdst, constantPoolOopDesc::cache_offset_in_bytes(), Rdst); | |
| 994 } | |
| 995 | |
| 996 | |
| 997 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { | |
| 998 get_constant_pool(Rcpool); | |
| 999 ld_ptr(Rcpool, constantPoolOopDesc::tags_offset_in_bytes(), Rtags); | |
| 1000 } | |
| 1001 | |
| 1002 | |
| 1003 // unlock if synchronized method | |
| 1004 // | |
| 1005 // Unlock the receiver if this is a synchronized method. | |
| 1006 // Unlock any Java monitors from syncronized blocks. | |
| 1007 // | |
| 1008 // If there are locked Java monitors | |
| 1009 // If throw_monitor_exception | |
| 1010 // throws IllegalMonitorStateException | |
| 1011 // Else if install_monitor_exception | |
| 1012 // installs IllegalMonitorStateException | |
| 1013 // Else | |
| 1014 // no error processing | |
| 1015 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, | |
| 1016 bool throw_monitor_exception, | |
| 1017 bool install_monitor_exception) { | |
| 1018 Label unlocked, unlock, no_unlock; | |
| 1019 | |
| 1020 // get the value of _do_not_unlock_if_synchronized into G1_scratch | |
| 727 | 1021 const Address do_not_unlock_if_synchronized(G2_thread, |
| 1022 JavaThread::do_not_unlock_if_synchronized_offset()); | |
| 0 | 1023 ldbool(do_not_unlock_if_synchronized, G1_scratch); |
| 1024 stbool(G0, do_not_unlock_if_synchronized); // reset the flag | |
| 1025 | |
| 1026 // check if synchronized method | |
| 727 | 1027 const Address access_flags(Lmethod, methodOopDesc::access_flags_offset()); |
| 0 | 1028 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); |
| 1029 push(state); // save tos | |
| 727 | 1030 ld(access_flags, G3_scratch); // Load access flags. |
| 0 | 1031 btst(JVM_ACC_SYNCHRONIZED, G3_scratch); |
| 727 | 1032 br(zero, false, pt, unlocked); |
| 0 | 1033 delayed()->nop(); |
| 1034 | |
| 1035 // Don't unlock anything if the _do_not_unlock_if_synchronized flag | |
| 1036 // is set. | |
| 1037 tstbool(G1_scratch); | |
| 1038 br(Assembler::notZero, false, pn, no_unlock); | |
| 1039 delayed()->nop(); | |
| 1040 | |
| 1041 // BasicObjectLock will be first in list, since this is a synchronized method. However, need | |
| 1042 // to check that the object has not been unlocked by an explicit monitorexit bytecode. | |
| 1043 | |
| 1044 //Intel: if (throw_monitor_exception) ... else ... | |
| 1045 // Entry already unlocked, need to throw exception | |
| 1046 //... | |
| 1047 | |
| 1048 // pass top-most monitor elem | |
| 1049 add( top_most_monitor(), O1 ); | |
| 1050 | |
| 1051 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch); | |
| 1052 br_notnull(G3_scratch, false, pt, unlock); | |
| 1053 delayed()->nop(); | |
| 1054 | |
| 1055 if (throw_monitor_exception) { | |
| 1056 // Entry already unlocked need to throw an exception | |
| 1057 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
| 1058 should_not_reach_here(); | |
| 1059 } else { | |
| 1060 // Monitor already unlocked during a stack unroll. | |
| 1061 // If requested, install an illegal_monitor_state_exception. | |
| 1062 // Continue with stack unrolling. | |
| 1063 if (install_monitor_exception) { | |
| 1064 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); | |
| 1065 } | |
| 1066 ba(false, unlocked); | |
| 1067 delayed()->nop(); | |
| 1068 } | |
| 1069 | |
| 1070 bind(unlock); | |
| 1071 | |
| 1072 unlock_object(O1); | |
| 1073 | |
| 1074 bind(unlocked); | |
| 1075 | |
| 1076 // I0, I1: Might contain return value | |
| 1077 | |
| 1078 // Check that all monitors are unlocked | |
| 1079 { Label loop, exception, entry, restart; | |
| 1080 | |
| 1081 Register Rmptr = O0; | |
| 1082 Register Rtemp = O1; | |
| 1083 Register Rlimit = Lmonitors; | |
| 1084 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
| 1085 assert( (delta & LongAlignmentMask) == 0, | |
| 1086 "sizeof BasicObjectLock must be even number of doublewords"); | |
| 1087 | |
| 1088 #ifdef ASSERT | |
| 1089 add(top_most_monitor(), Rmptr, delta); | |
| 1090 { Label L; | |
| 1091 // ensure that Rmptr starts out above (or at) Rlimit | |
| 1092 cmp(Rmptr, Rlimit); | |
| 1093 brx(Assembler::greaterEqualUnsigned, false, pn, L); | |
| 1094 delayed()->nop(); | |
| 1095 stop("monitor stack has negative size"); | |
| 1096 bind(L); | |
| 1097 } | |
| 1098 #endif | |
| 1099 bind(restart); | |
| 1100 ba(false, entry); | |
| 1101 delayed()-> | |
| 1102 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry | |
| 1103 | |
| 1104 // Entry is still locked, need to throw exception | |
| 1105 bind(exception); | |
| 1106 if (throw_monitor_exception) { | |
| 1107 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); | |
| 1108 should_not_reach_here(); | |
| 1109 } else { | |
| 1110 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. | |
| 1111 // Unlock does not block, so don't have to worry about the frame | |
| 1112 unlock_object(Rmptr); | |
| 1113 if (install_monitor_exception) { | |
| 1114 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); | |
| 1115 } | |
| 1116 ba(false, restart); | |
| 1117 delayed()->nop(); | |
| 1118 } | |
| 1119 | |
| 1120 bind(loop); | |
| 1121 cmp(Rtemp, G0); // check if current entry is used | |
| 1122 brx(Assembler::notEqual, false, pn, exception); | |
| 1123 delayed()-> | |
| 1124 dec(Rmptr, delta); // otherwise advance to next entry | |
| 1125 #ifdef ASSERT | |
| 1126 { Label L; | |
| 1127 // ensure that Rmptr has not somehow stepped below Rlimit | |
| 1128 cmp(Rmptr, Rlimit); | |
| 1129 brx(Assembler::greaterEqualUnsigned, false, pn, L); | |
| 1130 delayed()->nop(); | |
| 1131 stop("ran off the end of the monitor stack"); | |
| 1132 bind(L); | |
| 1133 } | |
| 1134 #endif | |
| 1135 bind(entry); | |
| 1136 cmp(Rmptr, Rlimit); // check if bottom reached | |
| 1137 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry | |
| 1138 delayed()-> | |
| 1139 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp); | |
| 1140 } | |
| 1141 | |
| 1142 bind(no_unlock); | |
| 1143 pop(state); | |
| 1144 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 1145 } | |
| 1146 | |
| 1147 | |
| 1148 // remove activation | |
| 1149 // | |
| 1150 // Unlock the receiver if this is a synchronized method. | |
| 1151 // Unlock any Java monitors from syncronized blocks. | |
| 1152 // Remove the activation from the stack. | |
| 1153 // | |
| 1154 // If there are locked Java monitors | |
| 1155 // If throw_monitor_exception | |
| 1156 // throws IllegalMonitorStateException | |
| 1157 // Else if install_monitor_exception | |
| 1158 // installs IllegalMonitorStateException | |
| 1159 // Else | |
| 1160 // no error processing | |
| 1161 void InterpreterMacroAssembler::remove_activation(TosState state, | |
| 1162 bool throw_monitor_exception, | |
| 1163 bool install_monitor_exception) { | |
| 1164 | |
| 1165 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); | |
| 1166 | |
| 1167 // save result (push state before jvmti call and pop it afterwards) and notify jvmti | |
| 1168 notify_method_exit(false, state, NotifyJVMTI); | |
| 1169 | |
| 1170 interp_verify_oop(Otos_i, state, __FILE__, __LINE__); | |
| 1171 verify_oop(Lmethod); | |
| 1172 verify_thread(); | |
| 1173 | |
| 1174 // return tos | |
| 1175 assert(Otos_l1 == Otos_i, "adjust code below"); | |
| 1176 switch (state) { | |
| 1177 #ifdef _LP64 | |
| 1178 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0 | |
| 1179 #else | |
| 1180 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1 | |
| 1181 #endif | |
| 1182 case btos: // fall through | |
| 1183 case ctos: | |
| 1184 case stos: // fall through | |
| 1185 case atos: // fall through | |
| 1186 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0 | |
| 1187 case ftos: // fall through | |
| 1188 case dtos: // fall through | |
| 1189 case vtos: /* nothing to do */ break; | |
| 1190 default : ShouldNotReachHere(); | |
| 1191 } | |
| 1192 | |
| 1193 #if defined(COMPILER2) && !defined(_LP64) | |
| 1194 if (state == ltos) { | |
| 1195 // C2 expects long results in G1 we can't tell if we're returning to interpreted | |
| 1196 // or compiled so just be safe use G1 and O0/O1 | |
| 1197 | |
| 1198 // Shift bits into high (msb) of G1 | |
| 1199 sllx(Otos_l1->after_save(), 32, G1); | |
| 1200 // Zero extend low bits | |
| 1201 srl (Otos_l2->after_save(), 0, Otos_l2->after_save()); | |
| 1202 or3 (Otos_l2->after_save(), G1, G1); | |
| 1203 } | |
| 1204 #endif /* COMPILER2 */ | |
| 1205 | |
| 1206 } | |
| 1207 #endif /* CC_INTERP */ | |
| 1208 | |
| 1209 | |
| 1210 // Lock object | |
| 1211 // | |
| 1212 // Argument - lock_reg points to the BasicObjectLock to be used for locking, | |
| 1213 // it must be initialized with the object to lock | |
| 1214 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) { | |
| 1215 if (UseHeavyMonitors) { | |
| 1216 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); | |
| 1217 } | |
| 1218 else { | |
| 1219 Register obj_reg = Object; | |
| 1220 Register mark_reg = G4_scratch; | |
| 1221 Register temp_reg = G1_scratch; | |
| 727 | 1222 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes()); |
| 1223 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); | |
| 0 | 1224 Label done; |
| 1225 | |
| 1226 Label slow_case; | |
| 1227 | |
| 1228 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg); | |
| 1229 | |
| 1230 // load markOop from object into mark_reg | |
| 1231 ld_ptr(mark_addr, mark_reg); | |
| 1232 | |
| 1233 if (UseBiasedLocking) { | |
| 1234 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case); | |
| 1235 } | |
| 1236 | |
| 1237 // get the address of basicLock on stack that will be stored in the object | |
| 1238 // we need a temporary register here as we do not want to clobber lock_reg | |
| 1239 // (cas clobbers the destination register) | |
| 1240 mov(lock_reg, temp_reg); | |
| 1241 // set mark reg to be (markOop of object | UNLOCK_VALUE) | |
| 1242 or3(mark_reg, markOopDesc::unlocked_value, mark_reg); | |
| 1243 // initialize the box (Must happen before we update the object mark!) | |
| 1244 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes()); | |
| 1245 // compare and exchange object_addr, markOop | 1, stack address of basicLock | |
| 1246 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
| 1247 casx_under_lock(mark_addr.base(), mark_reg, temp_reg, | |
| 1248 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
| 1249 | |
| 1250 // if the compare and exchange succeeded we are done (we saw an unlocked object) | |
| 1251 cmp(mark_reg, temp_reg); | |
| 1252 brx(Assembler::equal, true, Assembler::pt, done); | |
| 1253 delayed()->nop(); | |
| 1254 | |
| 1255 // We did not see an unlocked object so try the fast recursive case | |
| 1256 | |
| 1257 // Check if owner is self by comparing the value in the markOop of object | |
| 1258 // with the stack pointer | |
| 1259 sub(temp_reg, SP, temp_reg); | |
| 1260 #ifdef _LP64 | |
| 1261 sub(temp_reg, STACK_BIAS, temp_reg); | |
| 1262 #endif | |
| 1263 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
| 1264 | |
| 1265 // Composite "andcc" test: | |
| 1266 // (a) %sp -vs- markword proximity check, and, | |
| 1267 // (b) verify mark word LSBs == 0 (Stack-locked). | |
| 1268 // | |
| 1269 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size()) | |
| 1270 // Note that the page size used for %sp proximity testing is arbitrary and is | |
| 1271 // unrelated to the actual MMU page size. We use a 'logical' page size of | |
| 1272 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate | |
| 1273 // field of the andcc instruction. | |
| 1274 andcc (temp_reg, 0xFFFFF003, G0) ; | |
| 1275 | |
| 1276 // if condition is true we are done and hence we can store 0 in the displaced | |
| 1277 // header indicating it is a recursive lock and be done | |
| 1278 brx(Assembler::zero, true, Assembler::pt, done); | |
| 1279 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes()); | |
| 1280 | |
| 1281 // none of the above fast optimizations worked so we have to get into the | |
| 1282 // slow case of monitor enter | |
| 1283 bind(slow_case); | |
| 1284 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg); | |
| 1285 | |
| 1286 bind(done); | |
| 1287 } | |
| 1288 } | |
| 1289 | |
| 1290 // Unlocks an object. Used in monitorexit bytecode and remove_activation. | |
| 1291 // | |
| 1292 // Argument - lock_reg points to the BasicObjectLock for lock | |
| 1293 // Throw IllegalMonitorException if object is not locked by current thread | |
| 1294 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { | |
| 1295 if (UseHeavyMonitors) { | |
| 1296 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); | |
| 1297 } else { | |
| 1298 Register obj_reg = G3_scratch; | |
| 1299 Register mark_reg = G4_scratch; | |
| 1300 Register displaced_header_reg = G1_scratch; | |
| 727 | 1301 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes()); |
| 1302 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes()); | |
| 0 | 1303 Label done; |
| 1304 | |
| 1305 if (UseBiasedLocking) { | |
| 1306 // load the object out of the BasicObjectLock | |
| 1307 ld_ptr(lockobj_addr, obj_reg); | |
| 1308 biased_locking_exit(mark_addr, mark_reg, done, true); | |
| 1309 st_ptr(G0, lockobj_addr); // free entry | |
| 1310 } | |
| 1311 | |
| 1312 // Test first if we are in the fast recursive case | |
| 727 | 1313 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes()); |
| 1314 ld_ptr(lock_addr, displaced_header_reg); | |
| 0 | 1315 br_null(displaced_header_reg, true, Assembler::pn, done); |
| 1316 delayed()->st_ptr(G0, lockobj_addr); // free entry | |
| 1317 | |
| 1318 // See if it is still a light weight lock, if so we just unlock | |
| 1319 // the object and we are done | |
| 1320 | |
| 1321 if (!UseBiasedLocking) { | |
| 1322 // load the object out of the BasicObjectLock | |
| 1323 ld_ptr(lockobj_addr, obj_reg); | |
| 1324 } | |
| 1325 | |
| 1326 // we have the displaced header in displaced_header_reg | |
| 1327 // we expect to see the stack address of the basicLock in case the | |
| 1328 // lock is still a light weight lock (lock_reg) | |
| 1329 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
| 1330 casx_under_lock(mark_addr.base(), lock_reg, displaced_header_reg, | |
| 1331 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
| 1332 cmp(lock_reg, displaced_header_reg); | |
| 1333 brx(Assembler::equal, true, Assembler::pn, done); | |
| 1334 delayed()->st_ptr(G0, lockobj_addr); // free entry | |
| 1335 | |
| 1336 // The lock has been converted into a heavy lock and hence | |
| 1337 // we need to get into the slow case | |
| 1338 | |
| 1339 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); | |
| 1340 | |
| 1341 bind(done); | |
| 1342 } | |
| 1343 } | |
| 1344 | |
| 1345 #ifndef CC_INTERP | |
| 1346 | |
| 1347 // Get the method data pointer from the methodOop and set the | |
| 1348 // specified register to its value. | |
| 1349 | |
| 1350 void InterpreterMacroAssembler::set_method_data_pointer_offset(Register Roff) { | |
| 1351 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1352 Label get_continue; | |
| 1353 | |
| 1354 ld_ptr(Lmethod, in_bytes(methodOopDesc::method_data_offset()), ImethodDataPtr); | |
| 1355 test_method_data_pointer(get_continue); | |
| 1356 add(ImethodDataPtr, in_bytes(methodDataOopDesc::data_offset()), ImethodDataPtr); | |
| 1357 if (Roff != noreg) | |
| 1358 // Roff contains a method data index ("mdi"). It defaults to zero. | |
| 1359 add(ImethodDataPtr, Roff, ImethodDataPtr); | |
| 1360 bind(get_continue); | |
| 1361 } | |
| 1362 | |
| 1363 // Set the method data pointer for the current bcp. | |
| 1364 | |
| 1365 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { | |
| 1366 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1367 Label zero_continue; | |
| 1368 | |
| 1369 // Test MDO to avoid the call if it is NULL. | |
| 727 | 1370 ld_ptr(Lmethod, methodOopDesc::method_data_offset(), ImethodDataPtr); |
| 0 | 1371 test_method_data_pointer(zero_continue); |
| 1372 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp); | |
| 1373 set_method_data_pointer_offset(O0); | |
| 1374 bind(zero_continue); | |
| 1375 } | |
| 1376 | |
| 1377 // Test ImethodDataPtr. If it is null, continue at the specified label | |
| 1378 | |
| 1379 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { | |
| 1380 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1381 #ifdef _LP64 | |
| 1382 bpr(Assembler::rc_z, false, Assembler::pn, ImethodDataPtr, zero_continue); | |
| 1383 #else | |
| 1384 tst(ImethodDataPtr); | |
| 1385 br(Assembler::zero, false, Assembler::pn, zero_continue); | |
| 1386 #endif | |
| 1387 delayed()->nop(); | |
| 1388 } | |
| 1389 | |
| 1390 void InterpreterMacroAssembler::verify_method_data_pointer() { | |
| 1391 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1392 #ifdef ASSERT | |
| 1393 Label verify_continue; | |
| 1394 test_method_data_pointer(verify_continue); | |
| 1395 | |
| 1396 // If the mdp is valid, it will point to a DataLayout header which is | |
| 1397 // consistent with the bcp. The converse is highly probable also. | |
| 1398 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch); | |
| 727 | 1399 ld_ptr(Lmethod, methodOopDesc::const_offset(), O5); |
| 0 | 1400 add(G3_scratch, in_bytes(constMethodOopDesc::codes_offset()), G3_scratch); |
| 1401 add(G3_scratch, O5, G3_scratch); | |
| 1402 cmp(Lbcp, G3_scratch); | |
| 1403 brx(Assembler::equal, false, Assembler::pt, verify_continue); | |
| 1404 | |
| 1405 Register temp_reg = O5; | |
| 1406 delayed()->mov(ImethodDataPtr, temp_reg); | |
| 1407 // %%% should use call_VM_leaf here? | |
| 1408 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr); | |
| 1409 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1); | |
| 727 | 1410 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS); |
| 0 | 1411 stf(FloatRegisterImpl::D, Ftos_d, d_save); |
| 1412 mov(temp_reg->after_save(), O2); | |
| 1413 save_thread(L7_thread_cache); | |
| 1414 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none); | |
| 1415 delayed()->nop(); | |
| 1416 restore_thread(L7_thread_cache); | |
| 1417 ldf(FloatRegisterImpl::D, d_save, Ftos_d); | |
| 1418 restore(); | |
| 1419 bind(verify_continue); | |
| 1420 #endif // ASSERT | |
| 1421 } | |
| 1422 | |
| 1423 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, | |
| 1424 Register cur_bcp, | |
| 1425 Register Rtmp, | |
| 1426 Label &profile_continue) { | |
| 1427 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1428 // Control will flow to "profile_continue" if the counter is less than the | |
| 1429 // limit or if we call profile_method() | |
| 1430 | |
| 1431 Label done; | |
| 1432 | |
| 1433 // if no method data exists, and the counter is high enough, make one | |
| 1434 #ifdef _LP64 | |
| 1435 bpr(Assembler::rc_nz, false, Assembler::pn, ImethodDataPtr, done); | |
| 1436 #else | |
| 1437 tst(ImethodDataPtr); | |
| 1438 br(Assembler::notZero, false, Assembler::pn, done); | |
| 1439 #endif | |
| 1440 | |
| 1441 // Test to see if we should create a method data oop | |
| 727 | 1442 AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit); |
| 0 | 1443 #ifdef _LP64 |
| 1444 delayed()->nop(); | |
| 727 | 1445 sethi(profile_limit, Rtmp); |
| 0 | 1446 #else |
| 727 | 1447 delayed()->sethi(profile_limit, Rtmp); |
| 0 | 1448 #endif |
| 727 | 1449 ld(Rtmp, profile_limit.low10(), Rtmp); |
| 0 | 1450 cmp(invocation_count, Rtmp); |
| 1451 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue); | |
| 1452 delayed()->nop(); | |
| 1453 | |
| 1454 // Build it now. | |
| 1455 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), cur_bcp); | |
| 1456 set_method_data_pointer_offset(O0); | |
| 1457 ba(false, profile_continue); | |
| 1458 delayed()->nop(); | |
| 1459 bind(done); | |
| 1460 } | |
| 1461 | |
| 1462 // Store a value at some constant offset from the method data pointer. | |
| 1463 | |
| 1464 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { | |
| 1465 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1466 st_ptr(value, ImethodDataPtr, constant); | |
| 1467 } | |
| 1468 | |
| 1469 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter, | |
| 1470 Register bumped_count, | |
| 1471 bool decrement) { | |
| 1472 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1473 | |
| 1474 // Load the counter. | |
| 1475 ld_ptr(counter, bumped_count); | |
| 1476 | |
| 1477 if (decrement) { | |
| 1478 // Decrement the register. Set condition codes. | |
| 1479 subcc(bumped_count, DataLayout::counter_increment, bumped_count); | |
| 1480 | |
| 1481 // If the decrement causes the counter to overflow, stay negative | |
| 1482 Label L; | |
| 1483 brx(Assembler::negative, true, Assembler::pn, L); | |
| 1484 | |
| 1485 // Store the decremented counter, if it is still negative. | |
| 1486 delayed()->st_ptr(bumped_count, counter); | |
| 1487 bind(L); | |
| 1488 } else { | |
| 1489 // Increment the register. Set carry flag. | |
| 1490 addcc(bumped_count, DataLayout::counter_increment, bumped_count); | |
| 1491 | |
| 1492 // If the increment causes the counter to overflow, pull back by 1. | |
| 1493 assert(DataLayout::counter_increment == 1, "subc works"); | |
| 1494 subc(bumped_count, G0, bumped_count); | |
| 1495 | |
| 1496 // Store the incremented counter. | |
| 1497 st_ptr(bumped_count, counter); | |
| 1498 } | |
| 1499 } | |
| 1500 | |
| 1501 // Increment the value at some constant offset from the method data pointer. | |
| 1502 | |
| 1503 void InterpreterMacroAssembler::increment_mdp_data_at(int constant, | |
| 1504 Register bumped_count, | |
| 1505 bool decrement) { | |
| 1506 // Locate the counter at a fixed offset from the mdp: | |
| 727 | 1507 Address counter(ImethodDataPtr, constant); |
| 0 | 1508 increment_mdp_data_at(counter, bumped_count, decrement); |
| 1509 } | |
| 1510 | |
| 1511 // Increment the value at some non-fixed (reg + constant) offset from | |
| 1512 // the method data pointer. | |
| 1513 | |
| 1514 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, | |
| 1515 int constant, | |
| 1516 Register bumped_count, | |
| 1517 Register scratch2, | |
| 1518 bool decrement) { | |
| 1519 // Add the constant to reg to get the offset. | |
| 1520 add(ImethodDataPtr, reg, scratch2); | |
| 727 | 1521 Address counter(scratch2, constant); |
| 0 | 1522 increment_mdp_data_at(counter, bumped_count, decrement); |
| 1523 } | |
| 1524 | |
| 1525 // Set a flag value at the current method data pointer position. | |
| 1526 // Updates a single byte of the header, to avoid races with other header bits. | |
| 1527 | |
| 1528 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, | |
| 1529 Register scratch) { | |
| 1530 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1531 // Load the data header | |
| 1532 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch); | |
| 1533 | |
| 1534 // Set the flag | |
| 1535 or3(scratch, flag_constant, scratch); | |
| 1536 | |
| 1537 // Store the modified header. | |
| 1538 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset())); | |
| 1539 } | |
| 1540 | |
| 1541 // Test the location at some offset from the method data pointer. | |
| 1542 // If it is not equal to value, branch to the not_equal_continue Label. | |
| 1543 // Set condition codes to match the nullness of the loaded value. | |
| 1544 | |
| 1545 void InterpreterMacroAssembler::test_mdp_data_at(int offset, | |
| 1546 Register value, | |
| 1547 Label& not_equal_continue, | |
| 1548 Register scratch) { | |
| 1549 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1550 ld_ptr(ImethodDataPtr, offset, scratch); | |
| 1551 cmp(value, scratch); | |
| 1552 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue); | |
| 1553 delayed()->tst(scratch); | |
| 1554 } | |
| 1555 | |
| 1556 // Update the method data pointer by the displacement located at some fixed | |
| 1557 // offset from the method data pointer. | |
| 1558 | |
| 1559 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, | |
| 1560 Register scratch) { | |
| 1561 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1562 ld_ptr(ImethodDataPtr, offset_of_disp, scratch); | |
| 1563 add(ImethodDataPtr, scratch, ImethodDataPtr); | |
| 1564 } | |
| 1565 | |
| 1566 // Update the method data pointer by the displacement located at the | |
| 1567 // offset (reg + offset_of_disp). | |
| 1568 | |
| 1569 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, | |
| 1570 int offset_of_disp, | |
| 1571 Register scratch) { | |
| 1572 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1573 add(reg, offset_of_disp, scratch); | |
| 1574 ld_ptr(ImethodDataPtr, scratch, scratch); | |
| 1575 add(ImethodDataPtr, scratch, ImethodDataPtr); | |
| 1576 } | |
| 1577 | |
| 1578 // Update the method data pointer by a simple constant displacement. | |
| 1579 | |
| 1580 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { | |
| 1581 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1582 add(ImethodDataPtr, constant, ImethodDataPtr); | |
| 1583 } | |
| 1584 | |
| 1585 // Update the method data pointer for a _ret bytecode whose target | |
| 1586 // was not among our cached targets. | |
| 1587 | |
| 1588 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, | |
| 1589 Register return_bci) { | |
| 1590 assert(ProfileInterpreter, "must be profiling interpreter"); | |
| 1591 push(state); | |
| 1592 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile | |
| 1593 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); | |
| 1594 ld_ptr(l_tmp, return_bci); | |
| 1595 pop(state); | |
| 1596 } | |
| 1597 | |
| 1598 // Count a taken branch in the bytecodes. | |
| 1599 | |
| 1600 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { | |
| 1601 if (ProfileInterpreter) { | |
| 1602 Label profile_continue; | |
| 1603 | |
| 1604 // If no method data exists, go to profile_continue. | |
| 1605 test_method_data_pointer(profile_continue); | |
| 1606 | |
| 1607 // We are taking a branch. Increment the taken count. | |
| 1608 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count); | |
| 1609 | |
| 1610 // The method data pointer needs to be updated to reflect the new target. | |
| 1611 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); | |
| 1612 bind (profile_continue); | |
| 1613 } | |
| 1614 } | |
| 1615 | |
| 1616 | |
| 1617 // Count a not-taken branch in the bytecodes. | |
| 1618 | |
| 1619 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) { | |
| 1620 if (ProfileInterpreter) { | |
| 1621 Label profile_continue; | |
| 1622 | |
| 1623 // If no method data exists, go to profile_continue. | |
| 1624 test_method_data_pointer(profile_continue); | |
| 1625 | |
| 1626 // We are taking a branch. Increment the not taken count. | |
| 1627 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch); | |
| 1628 | |
| 1629 // The method data pointer needs to be updated to correspond to the | |
| 1630 // next bytecode. | |
| 1631 update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); | |
| 1632 bind (profile_continue); | |
| 1633 } | |
| 1634 } | |
| 1635 | |
| 1636 | |
| 1637 // Count a non-virtual call in the bytecodes. | |
| 1638 | |
| 1639 void InterpreterMacroAssembler::profile_call(Register scratch) { | |
| 1640 if (ProfileInterpreter) { | |
| 1641 Label profile_continue; | |
| 1642 | |
| 1643 // If no method data exists, go to profile_continue. | |
| 1644 test_method_data_pointer(profile_continue); | |
| 1645 | |
| 1646 // We are making a call. Increment the count. | |
| 1647 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
| 1648 | |
| 1649 // The method data pointer needs to be updated to reflect the new target. | |
| 1650 update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); | |
| 1651 bind (profile_continue); | |
| 1652 } | |
| 1653 } | |
| 1654 | |
| 1655 | |
| 1656 // Count a final call in the bytecodes. | |
| 1657 | |
| 1658 void InterpreterMacroAssembler::profile_final_call(Register scratch) { | |
| 1659 if (ProfileInterpreter) { | |
| 1660 Label profile_continue; | |
| 1661 | |
| 1662 // If no method data exists, go to profile_continue. | |
| 1663 test_method_data_pointer(profile_continue); | |
| 1664 | |
| 1665 // We are making a call. Increment the count. | |
| 1666 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
| 1667 | |
| 1668 // The method data pointer needs to be updated to reflect the new target. | |
| 1669 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); | |
| 1670 bind (profile_continue); | |
| 1671 } | |
| 1672 } | |
| 1673 | |
| 1674 | |
| 1675 // Count a virtual call in the bytecodes. | |
| 1676 | |
| 1677 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, | |
| 1678 Register scratch) { | |
| 1679 if (ProfileInterpreter) { | |
| 1680 Label profile_continue; | |
| 1681 | |
| 1682 // If no method data exists, go to profile_continue. | |
| 1683 test_method_data_pointer(profile_continue); | |
| 1684 | |
| 1685 // Record the receiver type. | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1686 record_klass_in_profile(receiver, scratch, true); |
| 0 | 1687 |
| 1688 // The method data pointer needs to be updated to reflect the new target. | |
| 1689 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); | |
| 1690 bind (profile_continue); | |
| 1691 } | |
| 1692 } | |
| 1693 | |
| 1694 void InterpreterMacroAssembler::record_klass_in_profile_helper( | |
| 1695 Register receiver, Register scratch, | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1696 int start_row, Label& done, bool is_virtual_call) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1697 if (TypeProfileWidth == 0) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1698 if (is_virtual_call) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1699 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1700 } |
|
967
6918603297f7
6858208: jvm crash when specifying TypeProfileWidth=0 on jdk 6.0
poonam
parents:
727
diff
changeset
|
1701 return; |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1702 } |
|
967
6918603297f7
6858208: jvm crash when specifying TypeProfileWidth=0 on jdk 6.0
poonam
parents:
727
diff
changeset
|
1703 |
| 0 | 1704 int last_row = VirtualCallData::row_limit() - 1; |
| 1705 assert(start_row <= last_row, "must be work left to do"); | |
| 1706 // Test this row for both the receiver and for null. | |
| 1707 // Take any of three different outcomes: | |
| 1708 // 1. found receiver => increment count and goto done | |
| 1709 // 2. found null => keep looking for case 1, maybe allocate this cell | |
| 1710 // 3. found something else => keep looking for cases 1 and 2 | |
| 1711 // Case 3 is handled by a recursive call. | |
| 1712 for (int row = start_row; row <= last_row; row++) { | |
| 1713 Label next_test; | |
| 1714 bool test_for_null_also = (row == start_row); | |
| 1715 | |
| 1716 // See if the receiver is receiver[n]. | |
| 1717 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); | |
| 1718 test_mdp_data_at(recvr_offset, receiver, next_test, scratch); | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1719 // delayed()->tst(scratch); |
| 0 | 1720 |
| 1721 // The receiver is receiver[n]. Increment count[n]. | |
| 1722 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); | |
| 1723 increment_mdp_data_at(count_offset, scratch); | |
| 1724 ba(false, done); | |
| 1725 delayed()->nop(); | |
| 1726 bind(next_test); | |
| 1727 | |
| 1728 if (test_for_null_also) { | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1729 Label found_null; |
| 0 | 1730 // Failed the equality check on receiver[n]... Test for null. |
| 1731 if (start_row == last_row) { | |
| 1732 // 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
|
1733 if (is_virtual_call) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1734 brx(Assembler::zero, false, Assembler::pn, found_null); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1735 delayed()->nop(); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1736 // 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
|
1737 // Increment total counter to indicate polymorphic case. |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1738 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1739 ba(false, done); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1740 delayed()->nop(); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1741 bind(found_null); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1742 } else { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1743 brx(Assembler::notZero, false, Assembler::pt, done); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1744 delayed()->nop(); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1745 } |
| 0 | 1746 break; |
| 1747 } | |
| 1748 // Since null is rare, make it be the branch-taken case. | |
| 1749 brx(Assembler::zero, false, Assembler::pn, found_null); | |
| 1750 delayed()->nop(); | |
| 1751 | |
| 1752 // Put all the "Case 3" tests here. | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1753 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call); |
| 0 | 1754 |
| 1755 // Found a null. Keep searching for a matching receiver, | |
| 1756 // but remember that this is an empty (unused) slot. | |
| 1757 bind(found_null); | |
| 1758 } | |
| 1759 } | |
| 1760 | |
| 1761 // In the fall-through case, we found no matching receiver, but we | |
| 1762 // observed the receiver[start_row] is NULL. | |
| 1763 | |
| 1764 // Fill in the receiver field and increment the count. | |
| 1765 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); | |
| 1766 set_mdp_data_at(recvr_offset, receiver); | |
| 1767 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); | |
| 1768 mov(DataLayout::counter_increment, scratch); | |
| 1769 set_mdp_data_at(count_offset, scratch); | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1770 if (start_row > 0) { |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1771 ba(false, done); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1772 delayed()->nop(); |
|
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1773 } |
| 0 | 1774 } |
| 1775 | |
| 1776 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1777 Register scratch, bool is_virtual_call) { |
| 0 | 1778 assert(ProfileInterpreter, "must be profiling"); |
| 1779 Label done; | |
| 1780 | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1781 record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call); |
| 0 | 1782 |
| 1783 bind (done); | |
| 1784 } | |
| 1785 | |
| 1786 | |
| 1787 // Count a ret in the bytecodes. | |
| 1788 | |
| 1789 void InterpreterMacroAssembler::profile_ret(TosState state, | |
| 1790 Register return_bci, | |
| 1791 Register scratch) { | |
| 1792 if (ProfileInterpreter) { | |
| 1793 Label profile_continue; | |
| 1794 uint row; | |
| 1795 | |
| 1796 // If no method data exists, go to profile_continue. | |
| 1797 test_method_data_pointer(profile_continue); | |
| 1798 | |
| 1799 // Update the total ret count. | |
| 1800 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch); | |
| 1801 | |
| 1802 for (row = 0; row < RetData::row_limit(); row++) { | |
| 1803 Label next_test; | |
| 1804 | |
| 1805 // See if return_bci is equal to bci[n]: | |
| 1806 test_mdp_data_at(in_bytes(RetData::bci_offset(row)), | |
| 1807 return_bci, next_test, scratch); | |
| 1808 | |
| 1809 // return_bci is equal to bci[n]. Increment the count. | |
| 1810 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch); | |
| 1811 | |
| 1812 // The method data pointer needs to be updated to reflect the new target. | |
| 1813 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch); | |
| 1814 ba(false, profile_continue); | |
| 1815 delayed()->nop(); | |
| 1816 bind(next_test); | |
| 1817 } | |
| 1818 | |
| 1819 update_mdp_for_ret(state, return_bci); | |
| 1820 | |
| 1821 bind (profile_continue); | |
| 1822 } | |
| 1823 } | |
| 1824 | |
| 1825 // Profile an unexpected null in the bytecodes. | |
| 1826 void InterpreterMacroAssembler::profile_null_seen(Register scratch) { | |
| 1827 if (ProfileInterpreter) { | |
| 1828 Label profile_continue; | |
| 1829 | |
| 1830 // If no method data exists, go to profile_continue. | |
| 1831 test_method_data_pointer(profile_continue); | |
| 1832 | |
| 1833 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch); | |
| 1834 | |
| 1835 // The method data pointer needs to be updated. | |
| 1836 int mdp_delta = in_bytes(BitData::bit_data_size()); | |
| 1837 if (TypeProfileCasts) { | |
| 1838 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); | |
| 1839 } | |
| 1840 update_mdp_by_constant(mdp_delta); | |
| 1841 | |
| 1842 bind (profile_continue); | |
| 1843 } | |
| 1844 } | |
| 1845 | |
| 1846 void InterpreterMacroAssembler::profile_typecheck(Register klass, | |
| 1847 Register scratch) { | |
| 1848 if (ProfileInterpreter) { | |
| 1849 Label profile_continue; | |
| 1850 | |
| 1851 // If no method data exists, go to profile_continue. | |
| 1852 test_method_data_pointer(profile_continue); | |
| 1853 | |
| 1854 int mdp_delta = in_bytes(BitData::bit_data_size()); | |
| 1855 if (TypeProfileCasts) { | |
| 1856 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); | |
| 1857 | |
| 1858 // Record the object type. | |
|
1206
87684f1a88b5
6614597: Performance variability in jvm2008 xml.validation
kvn
parents:
967
diff
changeset
|
1859 record_klass_in_profile(klass, scratch, false); |
| 0 | 1860 } |
| 1861 | |
| 1862 // The method data pointer needs to be updated. | |
| 1863 update_mdp_by_constant(mdp_delta); | |
| 1864 | |
| 1865 bind (profile_continue); | |
| 1866 } | |
| 1867 } | |
| 1868 | |
| 1869 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) { | |
| 1870 if (ProfileInterpreter && TypeProfileCasts) { | |
| 1871 Label profile_continue; | |
| 1872 | |
| 1873 // If no method data exists, go to profile_continue. | |
| 1874 test_method_data_pointer(profile_continue); | |
| 1875 | |
| 1876 int count_offset = in_bytes(CounterData::count_offset()); | |
| 1877 // Back up the address, since we have already bumped the mdp. | |
| 1878 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); | |
| 1879 | |
| 1880 // *Decrement* the counter. We expect to see zero or small negatives. | |
| 1881 increment_mdp_data_at(count_offset, scratch, true); | |
| 1882 | |
| 1883 bind (profile_continue); | |
| 1884 } | |
| 1885 } | |
| 1886 | |
| 1887 // Count the default case of a switch construct. | |
| 1888 | |
| 1889 void InterpreterMacroAssembler::profile_switch_default(Register scratch) { | |
| 1890 if (ProfileInterpreter) { | |
| 1891 Label profile_continue; | |
| 1892 | |
| 1893 // If no method data exists, go to profile_continue. | |
| 1894 test_method_data_pointer(profile_continue); | |
| 1895 | |
| 1896 // Update the default case count | |
| 1897 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), | |
| 1898 scratch); | |
| 1899 | |
| 1900 // The method data pointer needs to be updated. | |
| 1901 update_mdp_by_offset( | |
| 1902 in_bytes(MultiBranchData::default_displacement_offset()), | |
| 1903 scratch); | |
| 1904 | |
| 1905 bind (profile_continue); | |
| 1906 } | |
| 1907 } | |
| 1908 | |
| 1909 // Count the index'th case of a switch construct. | |
| 1910 | |
| 1911 void InterpreterMacroAssembler::profile_switch_case(Register index, | |
| 1912 Register scratch, | |
| 1913 Register scratch2, | |
| 1914 Register scratch3) { | |
| 1915 if (ProfileInterpreter) { | |
| 1916 Label profile_continue; | |
| 1917 | |
| 1918 // If no method data exists, go to profile_continue. | |
| 1919 test_method_data_pointer(profile_continue); | |
| 1920 | |
| 1921 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes() | |
| 1922 set(in_bytes(MultiBranchData::per_case_size()), scratch); | |
| 1923 smul(index, scratch, scratch); | |
| 1924 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch); | |
| 1925 | |
| 1926 // Update the case count | |
| 1927 increment_mdp_data_at(scratch, | |
| 1928 in_bytes(MultiBranchData::relative_count_offset()), | |
| 1929 scratch2, | |
| 1930 scratch3); | |
| 1931 | |
| 1932 // The method data pointer needs to be updated. | |
| 1933 update_mdp_by_offset(scratch, | |
| 1934 in_bytes(MultiBranchData::relative_displacement_offset()), | |
| 1935 scratch2); | |
| 1936 | |
| 1937 bind (profile_continue); | |
| 1938 } | |
| 1939 } | |
| 1940 | |
| 1941 // add a InterpMonitorElem to stack (see frame_sparc.hpp) | |
| 1942 | |
| 1943 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty, | |
| 1944 Register Rtemp, | |
| 1945 Register Rtemp2 ) { | |
| 1946 | |
| 1947 Register Rlimit = Lmonitors; | |
| 1948 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
| 1949 assert( (delta & LongAlignmentMask) == 0, | |
| 1950 "sizeof BasicObjectLock must be even number of doublewords"); | |
| 1951 | |
| 1952 sub( SP, delta, SP); | |
| 1953 sub( Lesp, delta, Lesp); | |
| 1954 sub( Lmonitors, delta, Lmonitors); | |
| 1955 | |
| 1956 if (!stack_is_empty) { | |
| 1957 | |
| 1958 // must copy stack contents down | |
| 1959 | |
| 1960 Label start_copying, next; | |
| 1961 | |
| 1962 // untested("monitor stack expansion"); | |
| 1963 compute_stack_base(Rtemp); | |
| 1964 ba( false, start_copying ); | |
| 1965 delayed()->cmp( Rtemp, Rlimit); // done? duplicated below | |
| 1966 | |
| 1967 // note: must copy from low memory upwards | |
| 1968 // On entry to loop, | |
| 1969 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS) | |
| 1970 // Loop mutates Rtemp | |
| 1971 | |
| 1972 bind( next); | |
| 1973 | |
| 1974 st_ptr(Rtemp2, Rtemp, 0); | |
| 1975 inc(Rtemp, wordSize); | |
| 1976 cmp(Rtemp, Rlimit); // are we done? (duplicated above) | |
| 1977 | |
| 1978 bind( start_copying ); | |
| 1979 | |
| 1980 brx( notEqual, true, pn, next ); | |
| 1981 delayed()->ld_ptr( Rtemp, delta, Rtemp2 ); | |
| 1982 | |
| 1983 // done copying stack | |
| 1984 } | |
| 1985 } | |
| 1986 | |
| 1987 // Locals | |
| 1988 #ifdef ASSERT | |
| 1989 void InterpreterMacroAssembler::verify_local_tag(frame::Tag t, | |
| 1990 Register base, | |
| 1991 Register scratch, | |
| 1992 int n) { | |
| 1993 if (TaggedStackInterpreter) { | |
| 1994 Label ok, long_ok; | |
| 1995 // Use dst for scratch | |
| 1996 assert_different_registers(base, scratch); | |
| 1997 ld_ptr(base, Interpreter::local_tag_offset_in_bytes(n), scratch); | |
| 1998 if (t == frame::TagCategory2) { | |
| 1999 cmp(scratch, G0); | |
| 2000 brx(Assembler::equal, false, Assembler::pt, long_ok); | |
| 2001 delayed()->ld_ptr(base, Interpreter::local_tag_offset_in_bytes(n+1), scratch); | |
| 2002 stop("local long/double tag value bad"); | |
| 2003 bind(long_ok); | |
| 2004 // compare second half tag | |
| 2005 cmp(scratch, G0); | |
| 2006 } else if (t == frame::TagValue) { | |
| 2007 cmp(scratch, G0); | |
| 2008 } else { | |
| 2009 assert_different_registers(O3, base, scratch); | |
| 2010 mov(t, O3); | |
| 2011 cmp(scratch, O3); | |
| 2012 } | |
| 2013 brx(Assembler::equal, false, Assembler::pt, ok); | |
| 2014 delayed()->nop(); | |
| 2015 // Also compare if the local value is zero, then the tag might | |
| 2016 // not have been set coming from deopt. | |
| 2017 ld_ptr(base, Interpreter::local_offset_in_bytes(n), scratch); | |
| 2018 cmp(scratch, G0); | |
| 2019 brx(Assembler::equal, false, Assembler::pt, ok); | |
| 2020 delayed()->nop(); | |
| 2021 stop("Local tag value is bad"); | |
| 2022 bind(ok); | |
| 2023 } | |
| 2024 } | |
| 2025 #endif // ASSERT | |
| 2026 | |
| 2027 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) { | |
| 2028 assert_not_delayed(); | |
| 2029 sll(index, Interpreter::logStackElementSize(), index); | |
| 2030 sub(Llocals, index, index); | |
| 2031 debug_only(verify_local_tag(frame::TagReference, index, dst)); | |
| 2032 ld_ptr(index, Interpreter::value_offset_in_bytes(), dst); | |
| 2033 // Note: index must hold the effective address--the iinc template uses it | |
| 2034 } | |
| 2035 | |
| 2036 // Just like access_local_ptr but the tag is a returnAddress | |
| 2037 void InterpreterMacroAssembler::access_local_returnAddress(Register index, | |
| 2038 Register dst ) { | |
| 2039 assert_not_delayed(); | |
| 2040 sll(index, Interpreter::logStackElementSize(), index); | |
| 2041 sub(Llocals, index, index); | |
| 2042 debug_only(verify_local_tag(frame::TagValue, index, dst)); | |
| 2043 ld_ptr(index, Interpreter::value_offset_in_bytes(), dst); | |
| 2044 } | |
| 2045 | |
| 2046 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) { | |
| 2047 assert_not_delayed(); | |
| 2048 sll(index, Interpreter::logStackElementSize(), index); | |
| 2049 sub(Llocals, index, index); | |
| 2050 debug_only(verify_local_tag(frame::TagValue, index, dst)); | |
| 2051 ld(index, Interpreter::value_offset_in_bytes(), dst); | |
| 2052 // Note: index must hold the effective address--the iinc template uses it | |
| 2053 } | |
| 2054 | |
| 2055 | |
| 2056 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) { | |
| 2057 assert_not_delayed(); | |
| 2058 sll(index, Interpreter::logStackElementSize(), index); | |
| 2059 sub(Llocals, index, index); | |
| 2060 debug_only(verify_local_tag(frame::TagCategory2, index, dst)); | |
| 2061 // First half stored at index n+1 (which grows down from Llocals[n]) | |
| 2062 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst); | |
| 2063 } | |
| 2064 | |
| 2065 | |
| 2066 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) { | |
| 2067 assert_not_delayed(); | |
| 2068 sll(index, Interpreter::logStackElementSize(), index); | |
| 2069 sub(Llocals, index, index); | |
| 2070 debug_only(verify_local_tag(frame::TagValue, index, G1_scratch)); | |
| 2071 ldf(FloatRegisterImpl::S, index, Interpreter::value_offset_in_bytes(), dst); | |
| 2072 } | |
| 2073 | |
| 2074 | |
| 2075 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) { | |
| 2076 assert_not_delayed(); | |
| 2077 sll(index, Interpreter::logStackElementSize(), index); | |
| 2078 sub(Llocals, index, index); | |
| 2079 debug_only(verify_local_tag(frame::TagCategory2, index, G1_scratch)); | |
| 2080 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst); | |
| 2081 } | |
| 2082 | |
| 2083 | |
| 2084 #ifdef ASSERT | |
| 2085 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) { | |
| 2086 Label L; | |
| 2087 | |
| 2088 assert(Rindex != Rscratch, "Registers cannot be same"); | |
| 2089 assert(Rindex != Rscratch1, "Registers cannot be same"); | |
| 2090 assert(Rlimit != Rscratch, "Registers cannot be same"); | |
| 2091 assert(Rlimit != Rscratch1, "Registers cannot be same"); | |
| 2092 assert(Rscratch1 != Rscratch, "Registers cannot be same"); | |
| 2093 | |
| 2094 // untested("reg area corruption"); | |
| 2095 add(Rindex, offset, Rscratch); | |
| 2096 add(Rlimit, 64 + STACK_BIAS, Rscratch1); | |
| 2097 cmp(Rscratch, Rscratch1); | |
| 2098 brx(Assembler::greaterEqualUnsigned, false, pn, L); | |
| 2099 delayed()->nop(); | |
| 2100 stop("regsave area is being clobbered"); | |
| 2101 bind(L); | |
| 2102 } | |
| 2103 #endif // ASSERT | |
| 2104 | |
| 2105 void InterpreterMacroAssembler::tag_local(frame::Tag t, | |
| 2106 Register base, | |
| 2107 Register src, | |
| 2108 int n) { | |
| 2109 if (TaggedStackInterpreter) { | |
| 2110 // have to store zero because local slots can be reused (rats!) | |
| 2111 if (t == frame::TagValue) { | |
| 2112 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n)); | |
| 2113 } else if (t == frame::TagCategory2) { | |
| 2114 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n)); | |
| 2115 st_ptr(G0, base, Interpreter::local_tag_offset_in_bytes(n+1)); | |
| 2116 } else { | |
| 2117 // assert that we don't stomp the value in 'src' | |
| 2118 // O3 is arbitrary because it's not used. | |
| 2119 assert_different_registers(src, base, O3); | |
| 2120 mov( t, O3); | |
| 2121 st_ptr(O3, base, Interpreter::local_tag_offset_in_bytes(n)); | |
| 2122 } | |
| 2123 } | |
| 2124 } | |
| 2125 | |
| 2126 | |
| 2127 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) { | |
| 2128 assert_not_delayed(); | |
| 2129 sll(index, Interpreter::logStackElementSize(), index); | |
| 2130 sub(Llocals, index, index); | |
| 2131 debug_only(check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch);) | |
| 2132 tag_local(frame::TagValue, index, src); | |
| 2133 st(src, index, Interpreter::value_offset_in_bytes()); | |
| 2134 } | |
| 2135 | |
| 2136 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src, | |
| 2137 Register tag ) { | |
| 2138 assert_not_delayed(); | |
| 2139 sll(index, Interpreter::logStackElementSize(), index); | |
| 2140 sub(Llocals, index, index); | |
| 2141 #ifdef ASSERT | |
| 2142 check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch); | |
| 2143 #endif | |
| 2144 st_ptr(src, index, Interpreter::value_offset_in_bytes()); | |
| 2145 // Store tag register directly | |
| 2146 if (TaggedStackInterpreter) { | |
| 2147 st_ptr(tag, index, Interpreter::tag_offset_in_bytes()); | |
| 2148 } | |
| 2149 } | |
| 2150 | |
| 2151 | |
| 2152 | |
| 2153 void InterpreterMacroAssembler::store_local_ptr( int n, Register src, | |
| 2154 Register tag ) { | |
| 2155 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n)); | |
| 2156 if (TaggedStackInterpreter) { | |
| 2157 st_ptr(tag, Llocals, Interpreter::local_tag_offset_in_bytes(n)); | |
| 2158 } | |
| 2159 } | |
| 2160 | |
| 2161 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) { | |
| 2162 assert_not_delayed(); | |
| 2163 sll(index, Interpreter::logStackElementSize(), index); | |
| 2164 sub(Llocals, index, index); | |
| 2165 #ifdef ASSERT | |
| 2166 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); | |
| 2167 #endif | |
| 2168 tag_local(frame::TagCategory2, index, src); | |
| 2169 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1 | |
| 2170 } | |
| 2171 | |
| 2172 | |
| 2173 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) { | |
| 2174 assert_not_delayed(); | |
| 2175 sll(index, Interpreter::logStackElementSize(), index); | |
| 2176 sub(Llocals, index, index); | |
| 2177 #ifdef ASSERT | |
| 2178 check_for_regarea_stomp(index, Interpreter::value_offset_in_bytes(), FP, G1_scratch, G4_scratch); | |
| 2179 #endif | |
| 2180 tag_local(frame::TagValue, index, G1_scratch); | |
| 2181 stf(FloatRegisterImpl::S, src, index, Interpreter::value_offset_in_bytes()); | |
| 2182 } | |
| 2183 | |
| 2184 | |
| 2185 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) { | |
| 2186 assert_not_delayed(); | |
| 2187 sll(index, Interpreter::logStackElementSize(), index); | |
| 2188 sub(Llocals, index, index); | |
| 2189 #ifdef ASSERT | |
| 2190 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch); | |
| 2191 #endif | |
| 2192 tag_local(frame::TagCategory2, index, G1_scratch); | |
| 2193 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1)); | |
| 2194 } | |
| 2195 | |
| 2196 | |
| 2197 int InterpreterMacroAssembler::top_most_monitor_byte_offset() { | |
| 2198 const jint delta = frame::interpreter_frame_monitor_size() * wordSize; | |
| 2199 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong); | |
| 2200 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS; | |
| 2201 } | |
| 2202 | |
| 2203 | |
| 2204 Address InterpreterMacroAssembler::top_most_monitor() { | |
| 727 | 2205 return Address(FP, top_most_monitor_byte_offset()); |
| 0 | 2206 } |
| 2207 | |
| 2208 | |
| 2209 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) { | |
| 2210 add( Lesp, wordSize, Rdest ); | |
| 2211 } | |
| 2212 | |
| 2213 #endif /* CC_INTERP */ | |
| 2214 | |
| 2215 void InterpreterMacroAssembler::increment_invocation_counter( Register Rtmp, Register Rtmp2 ) { | |
| 2216 assert(UseCompiler, "incrementing must be useful"); | |
| 2217 #ifdef CC_INTERP | |
| 727 | 2218 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() + |
| 2219 InvocationCounter::counter_offset()); | |
| 2220 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() + | |
| 2221 InvocationCounter::counter_offset()); | |
| 0 | 2222 #else |
| 727 | 2223 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() + |
| 2224 InvocationCounter::counter_offset()); | |
| 2225 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() + | |
| 2226 InvocationCounter::counter_offset()); | |
| 0 | 2227 #endif /* CC_INTERP */ |
| 2228 int delta = InvocationCounter::count_increment; | |
| 2229 | |
| 2230 // Load each counter in a register | |
| 2231 ld( inv_counter, Rtmp ); | |
| 2232 ld( be_counter, Rtmp2 ); | |
| 2233 | |
| 2234 assert( is_simm13( delta ), " delta too large."); | |
| 2235 | |
| 2236 // Add the delta to the invocation counter and store the result | |
| 2237 add( Rtmp, delta, Rtmp ); | |
| 2238 | |
| 2239 // Mask the backedge counter | |
| 2240 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); | |
| 2241 | |
| 2242 // Store value | |
| 2243 st( Rtmp, inv_counter); | |
| 2244 | |
| 2245 // Add invocation counter + backedge counter | |
| 2246 add( Rtmp, Rtmp2, Rtmp); | |
| 2247 | |
| 2248 // Note that this macro must leave the backedge_count + invocation_count in Rtmp! | |
| 2249 } | |
| 2250 | |
| 2251 void InterpreterMacroAssembler::increment_backedge_counter( Register Rtmp, Register Rtmp2 ) { | |
| 2252 assert(UseCompiler, "incrementing must be useful"); | |
| 2253 #ifdef CC_INTERP | |
| 727 | 2254 Address be_counter (G5_method, methodOopDesc::backedge_counter_offset() + |
| 2255 InvocationCounter::counter_offset()); | |
| 2256 Address inv_counter(G5_method, methodOopDesc::invocation_counter_offset() + | |
| 2257 InvocationCounter::counter_offset()); | |
| 0 | 2258 #else |
| 727 | 2259 Address be_counter (Lmethod, methodOopDesc::backedge_counter_offset() + |
| 2260 InvocationCounter::counter_offset()); | |
| 2261 Address inv_counter(Lmethod, methodOopDesc::invocation_counter_offset() + | |
| 2262 InvocationCounter::counter_offset()); | |
| 0 | 2263 #endif /* CC_INTERP */ |
| 2264 int delta = InvocationCounter::count_increment; | |
| 2265 // Load each counter in a register | |
| 2266 ld( be_counter, Rtmp ); | |
| 2267 ld( inv_counter, Rtmp2 ); | |
| 2268 | |
| 2269 // Add the delta to the backedge counter | |
| 2270 add( Rtmp, delta, Rtmp ); | |
| 2271 | |
| 2272 // Mask the invocation counter, add to backedge counter | |
| 2273 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 ); | |
| 2274 | |
| 2275 // and store the result to memory | |
| 2276 st( Rtmp, be_counter ); | |
| 2277 | |
| 2278 // Add backedge + invocation counter | |
| 2279 add( Rtmp, Rtmp2, Rtmp ); | |
| 2280 | |
| 2281 // Note that this macro must leave backedge_count + invocation_count in Rtmp! | |
| 2282 } | |
| 2283 | |
| 2284 #ifndef CC_INTERP | |
| 2285 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count, | |
| 2286 Register branch_bcp, | |
| 2287 Register Rtmp ) { | |
| 2288 Label did_not_overflow; | |
| 2289 Label overflow_with_error; | |
| 2290 assert_different_registers(backedge_count, Rtmp, branch_bcp); | |
| 2291 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); | |
| 2292 | |
| 727 | 2293 AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit); |
| 0 | 2294 load_contents(limit, Rtmp); |
| 2295 cmp(backedge_count, Rtmp); | |
| 2296 br(Assembler::lessUnsigned, false, Assembler::pt, did_not_overflow); | |
| 2297 delayed()->nop(); | |
| 2298 | |
| 2299 // When ProfileInterpreter is on, the backedge_count comes from the | |
| 2300 // methodDataOop, which value does not get reset on the call to | |
| 2301 // frequency_counter_overflow(). To avoid excessive calls to the overflow | |
| 2302 // routine while the method is being compiled, add a second test to make sure | |
| 2303 // the overflow function is called only once every overflow_frequency. | |
| 2304 if (ProfileInterpreter) { | |
| 2305 const int overflow_frequency = 1024; | |
| 2306 andcc(backedge_count, overflow_frequency-1, Rtmp); | |
| 2307 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow); | |
| 2308 delayed()->nop(); | |
| 2309 } | |
| 2310 | |
| 2311 // overflow in loop, pass branch bytecode | |
| 2312 set(6,Rtmp); | |
| 2313 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp); | |
| 2314 | |
| 2315 // Was an OSR adapter generated? | |
| 2316 // O0 = osr nmethod | |
| 2317 tst(O0); | |
| 2318 brx(Assembler::zero, false, Assembler::pn, overflow_with_error); | |
| 2319 delayed()->nop(); | |
| 2320 | |
| 2321 // Has the nmethod been invalidated already? | |
| 2322 ld(O0, nmethod::entry_bci_offset(), O2); | |
| 2323 cmp(O2, InvalidOSREntryBci); | |
| 2324 br(Assembler::equal, false, Assembler::pn, overflow_with_error); | |
| 2325 delayed()->nop(); | |
| 2326 | |
| 2327 // migrate the interpreter frame off of the stack | |
| 2328 | |
| 2329 mov(G2_thread, L7); | |
| 2330 // save nmethod | |
| 2331 mov(O0, L6); | |
| 2332 set_last_Java_frame(SP, noreg); | |
| 2333 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7); | |
| 2334 reset_last_Java_frame(); | |
| 2335 mov(L7, G2_thread); | |
| 2336 | |
| 2337 // move OSR nmethod to I1 | |
| 2338 mov(L6, I1); | |
| 2339 | |
| 2340 // OSR buffer to I0 | |
| 2341 mov(O0, I0); | |
| 2342 | |
| 2343 // remove the interpreter frame | |
| 2344 restore(I5_savedSP, 0, SP); | |
| 2345 | |
| 2346 // Jump to the osr code. | |
| 2347 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2); | |
| 2348 jmp(O2, G0); | |
| 2349 delayed()->nop(); | |
| 2350 | |
| 2351 bind(overflow_with_error); | |
| 2352 | |
| 2353 bind(did_not_overflow); | |
| 2354 } | |
| 2355 | |
| 2356 | |
| 2357 | |
| 2358 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) { | |
| 2359 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); } | |
| 2360 } | |
| 2361 | |
| 2362 | |
| 2363 // local helper function for the verify_oop_or_return_address macro | |
| 2364 static bool verify_return_address(methodOopDesc* m, int bci) { | |
| 2365 #ifndef PRODUCT | |
| 2366 address pc = (address)(m->constMethod()) | |
| 2367 + in_bytes(constMethodOopDesc::codes_offset()) + bci; | |
| 2368 // assume it is a valid return address if it is inside m and is preceded by a jsr | |
| 2369 if (!m->contains(pc)) return false; | |
| 2370 address jsr_pc; | |
| 2371 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); | |
| 2372 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; | |
| 2373 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); | |
| 2374 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; | |
| 2375 #endif // PRODUCT | |
| 2376 return false; | |
| 2377 } | |
| 2378 | |
| 2379 | |
| 2380 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { | |
| 2381 if (!VerifyOops) return; | |
| 2382 // the VM documentation for the astore[_wide] bytecode allows | |
| 2383 // the TOS to be not only an oop but also a return address | |
| 2384 Label test; | |
| 2385 Label skip; | |
| 2386 // See if it is an address (in the current method): | |
| 2387 | |
| 2388 mov(reg, Rtmp); | |
| 2389 const int log2_bytecode_size_limit = 16; | |
| 2390 srl(Rtmp, log2_bytecode_size_limit, Rtmp); | |
| 2391 br_notnull( Rtmp, false, pt, test ); | |
| 2392 delayed()->nop(); | |
| 2393 | |
| 2394 // %%% should use call_VM_leaf here? | |
| 2395 save_frame_and_mov(0, Lmethod, O0, reg, O1); | |
| 2396 save_thread(L7_thread_cache); | |
| 2397 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none); | |
| 2398 delayed()->nop(); | |
| 2399 restore_thread(L7_thread_cache); | |
| 2400 br_notnull( O0, false, pt, skip ); | |
| 2401 delayed()->restore(); | |
| 2402 | |
| 2403 // Perform a more elaborate out-of-line call | |
| 2404 // Not an address; verify it: | |
| 2405 bind(test); | |
| 2406 verify_oop(reg); | |
| 2407 bind(skip); | |
| 2408 } | |
| 2409 | |
| 2410 | |
| 2411 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { | |
| 2412 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth); | |
| 2413 } | |
| 2414 #endif /* CC_INTERP */ | |
| 2415 | |
| 2416 // Inline assembly for: | |
| 2417 // | |
| 2418 // if (thread is in interp_only_mode) { | |
| 2419 // InterpreterRuntime::post_method_entry(); | |
| 2420 // } | |
| 2421 // if (DTraceMethodProbes) { | |
| 605 | 2422 // SharedRuntime::dtrace_method_entry(method, receiver); |
| 0 | 2423 // } |
|
610
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2424 // 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
|
2425 // SharedRuntime::rc_trace_method_entry(method, receiver); |
| 0 | 2426 // } |
| 2427 | |
| 2428 void InterpreterMacroAssembler::notify_method_entry() { | |
| 2429 | |
| 2430 // C++ interpreter only uses this for native methods. | |
| 2431 | |
| 2432 // Whenever JVMTI puts a thread in interp_only_mode, method | |
| 2433 // entry/exit events are sent for that thread to track stack | |
| 2434 // depth. If it is possible to enter interp_only_mode we add | |
| 2435 // the code to check if the event should be sent. | |
| 2436 if (JvmtiExport::can_post_interpreter_events()) { | |
| 2437 Label L; | |
| 2438 Register temp_reg = O5; | |
| 727 | 2439 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
| 0 | 2440 ld(interp_only, temp_reg); |
| 2441 tst(temp_reg); | |
| 2442 br(zero, false, pt, L); | |
| 2443 delayed()->nop(); | |
| 2444 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry)); | |
| 2445 bind(L); | |
| 2446 } | |
| 2447 | |
| 2448 { | |
| 2449 Register temp_reg = O5; | |
| 2450 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); | |
| 2451 call_VM_leaf(noreg, | |
| 2452 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), | |
| 2453 G2_thread, Lmethod); | |
| 2454 } | |
|
610
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2455 |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2456 // 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
|
2457 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
|
2458 call_VM_leaf(noreg, |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2459 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
|
2460 G2_thread, Lmethod); |
|
70998f2e05ef
6805864: 4/3 Problem with jvmti->redefineClasses: some methods don't get redefined
dcubed
parents:
422
diff
changeset
|
2461 } |
| 0 | 2462 } |
| 2463 | |
| 2464 | |
| 2465 // Inline assembly for: | |
| 2466 // | |
| 2467 // if (thread is in interp_only_mode) { | |
| 2468 // // save result | |
| 2469 // InterpreterRuntime::post_method_exit(); | |
| 2470 // // restore result | |
| 2471 // } | |
| 2472 // if (DTraceMethodProbes) { | |
| 2473 // SharedRuntime::dtrace_method_exit(thread, method); | |
| 2474 // } | |
| 2475 // | |
| 2476 // Native methods have their result stored in d_tmp and l_tmp | |
| 2477 // Java methods have their result stored in the expression stack | |
| 2478 | |
| 2479 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, | |
| 2480 TosState state, | |
| 2481 NotifyMethodExitMode mode) { | |
| 2482 // C++ interpreter only uses this for native methods. | |
| 2483 | |
| 2484 // Whenever JVMTI puts a thread in interp_only_mode, method | |
| 2485 // entry/exit events are sent for that thread to track stack | |
| 2486 // depth. If it is possible to enter interp_only_mode we add | |
| 2487 // the code to check if the event should be sent. | |
| 2488 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { | |
| 2489 Label L; | |
| 2490 Register temp_reg = O5; | |
| 727 | 2491 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset()); |
| 0 | 2492 ld(interp_only, temp_reg); |
| 2493 tst(temp_reg); | |
| 2494 br(zero, false, pt, L); | |
| 2495 delayed()->nop(); | |
| 2496 | |
| 2497 // Note: frame::interpreter_frame_result has a dependency on how the | |
| 2498 // method result is saved across the call to post_method_exit. For | |
| 2499 // native methods it assumes the result registers are saved to | |
| 2500 // l_scratch and d_scratch. If this changes then the interpreter_frame_result | |
| 2501 // implementation will need to be updated too. | |
| 2502 | |
| 2503 save_return_value(state, is_native_method); | |
| 2504 call_VM(noreg, | |
| 2505 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); | |
| 2506 restore_return_value(state, is_native_method); | |
| 2507 bind(L); | |
| 2508 } | |
| 2509 | |
| 2510 { | |
| 2511 Register temp_reg = O5; | |
| 2512 // Dtrace notification | |
| 2513 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero); | |
| 2514 save_return_value(state, is_native_method); | |
| 2515 call_VM_leaf( | |
| 2516 noreg, | |
| 2517 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), | |
| 2518 G2_thread, Lmethod); | |
| 2519 restore_return_value(state, is_native_method); | |
| 2520 } | |
| 2521 } | |
| 2522 | |
| 2523 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) { | |
| 2524 #ifdef CC_INTERP | |
| 2525 // result potentially in O0/O1: save it across calls | |
| 2526 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult)); | |
| 2527 #ifdef _LP64 | |
| 2528 stx(O0, STATE(_native_lresult)); | |
| 2529 #else | |
| 2530 std(O0, STATE(_native_lresult)); | |
| 2531 #endif | |
| 2532 #else // CC_INTERP | |
| 2533 if (is_native_call) { | |
| 2534 stf(FloatRegisterImpl::D, F0, d_tmp); | |
| 2535 #ifdef _LP64 | |
| 2536 stx(O0, l_tmp); | |
| 2537 #else | |
| 2538 std(O0, l_tmp); | |
| 2539 #endif | |
| 2540 } else { | |
| 2541 push(state); | |
| 2542 } | |
| 2543 #endif // CC_INTERP | |
| 2544 } | |
| 2545 | |
| 2546 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) { | |
| 2547 #ifdef CC_INTERP | |
| 2548 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0); | |
| 2549 #ifdef _LP64 | |
| 2550 ldx(STATE(_native_lresult), O0); | |
| 2551 #else | |
| 2552 ldd(STATE(_native_lresult), O0); | |
| 2553 #endif | |
| 2554 #else // CC_INTERP | |
| 2555 if (is_native_call) { | |
| 2556 ldf(FloatRegisterImpl::D, d_tmp, F0); | |
| 2557 #ifdef _LP64 | |
| 2558 ldx(l_tmp, O0); | |
| 2559 #else | |
| 2560 ldd(l_tmp, O0); | |
| 2561 #endif | |
| 2562 } else { | |
| 2563 pop(state); | |
| 2564 } | |
| 2565 #endif // CC_INTERP | |
| 2566 } |