Mon, 23 Jun 2008 14:11:12 -0700
6708714: Optimize long LShift on 32-bits x86
Summary: For small (1-3 bits) left long shifts in 32-bits VM use sets of add+addc instructions instead of shld+shl on new AMD cpus.
Reviewed-by: never
Contributed-by: shrinivas.joshi@amd.com
duke@435 | 1 | /* |
duke@435 | 2 | * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
duke@435 | 19 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
duke@435 | 20 | * CA 95054 USA or visit www.sun.com if you need additional information or |
duke@435 | 21 | * have any questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
duke@435 | 25 | # include "incls/_precompiled.incl" |
duke@435 | 26 | # include "incls/_vframeArray.cpp.incl" |
duke@435 | 27 | |
duke@435 | 28 | |
duke@435 | 29 | int vframeArrayElement:: bci(void) const { return (_bci == SynchronizationEntryBCI ? 0 : _bci); } |
duke@435 | 30 | |
duke@435 | 31 | void vframeArrayElement::free_monitors(JavaThread* jt) { |
duke@435 | 32 | if (_monitors != NULL) { |
duke@435 | 33 | MonitorChunk* chunk = _monitors; |
duke@435 | 34 | _monitors = NULL; |
duke@435 | 35 | jt->remove_monitor_chunk(chunk); |
duke@435 | 36 | delete chunk; |
duke@435 | 37 | } |
duke@435 | 38 | } |
duke@435 | 39 | |
duke@435 | 40 | void vframeArrayElement::fill_in(compiledVFrame* vf) { |
duke@435 | 41 | |
duke@435 | 42 | // Copy the information from the compiled vframe to the |
duke@435 | 43 | // interpreter frame we will be creating to replace vf |
duke@435 | 44 | |
duke@435 | 45 | _method = vf->method(); |
duke@435 | 46 | _bci = vf->raw_bci(); |
duke@435 | 47 | |
duke@435 | 48 | int index; |
duke@435 | 49 | |
duke@435 | 50 | // Get the monitors off-stack |
duke@435 | 51 | |
duke@435 | 52 | GrowableArray<MonitorInfo*>* list = vf->monitors(); |
duke@435 | 53 | if (list->is_empty()) { |
duke@435 | 54 | _monitors = NULL; |
duke@435 | 55 | } else { |
duke@435 | 56 | |
duke@435 | 57 | // Allocate monitor chunk |
duke@435 | 58 | _monitors = new MonitorChunk(list->length()); |
duke@435 | 59 | vf->thread()->add_monitor_chunk(_monitors); |
duke@435 | 60 | |
duke@435 | 61 | // Migrate the BasicLocks from the stack to the monitor chunk |
duke@435 | 62 | for (index = 0; index < list->length(); index++) { |
duke@435 | 63 | MonitorInfo* monitor = list->at(index); |
duke@435 | 64 | assert(monitor->owner() == NULL || (!monitor->owner()->is_unlocked() && !monitor->owner()->has_bias_pattern()), "object must be null or locked, and unbiased"); |
duke@435 | 65 | BasicObjectLock* dest = _monitors->at(index); |
duke@435 | 66 | dest->set_obj(monitor->owner()); |
duke@435 | 67 | monitor->lock()->move_to(monitor->owner(), dest->lock()); |
duke@435 | 68 | } |
duke@435 | 69 | } |
duke@435 | 70 | |
duke@435 | 71 | // Convert the vframe locals and expressions to off stack |
duke@435 | 72 | // values. Because we will not gc all oops can be converted to |
duke@435 | 73 | // intptr_t (i.e. a stack slot) and we are fine. This is |
duke@435 | 74 | // good since we are inside a HandleMark and the oops in our |
duke@435 | 75 | // collection would go away between packing them here and |
duke@435 | 76 | // unpacking them in unpack_on_stack. |
duke@435 | 77 | |
duke@435 | 78 | // First the locals go off-stack |
duke@435 | 79 | |
duke@435 | 80 | // FIXME this seems silly it creates a StackValueCollection |
duke@435 | 81 | // in order to get the size to then copy them and |
duke@435 | 82 | // convert the types to intptr_t size slots. Seems like it |
duke@435 | 83 | // could do it in place... Still uses less memory than the |
duke@435 | 84 | // old way though |
duke@435 | 85 | |
duke@435 | 86 | StackValueCollection *locs = vf->locals(); |
duke@435 | 87 | _locals = new StackValueCollection(locs->size()); |
duke@435 | 88 | for(index = 0; index < locs->size(); index++) { |
duke@435 | 89 | StackValue* value = locs->at(index); |
duke@435 | 90 | switch(value->type()) { |
duke@435 | 91 | case T_OBJECT: |
duke@435 | 92 | // preserve object type |
duke@435 | 93 | _locals->add( new StackValue((intptr_t) (value->get_obj()()), T_OBJECT )); |
duke@435 | 94 | break; |
duke@435 | 95 | case T_CONFLICT: |
duke@435 | 96 | // A dead local. Will be initialized to null/zero. |
duke@435 | 97 | _locals->add( new StackValue()); |
duke@435 | 98 | break; |
duke@435 | 99 | case T_INT: |
duke@435 | 100 | _locals->add( new StackValue(value->get_int())); |
duke@435 | 101 | break; |
duke@435 | 102 | default: |
duke@435 | 103 | ShouldNotReachHere(); |
duke@435 | 104 | } |
duke@435 | 105 | } |
duke@435 | 106 | |
duke@435 | 107 | // Now the expressions off-stack |
duke@435 | 108 | // Same silliness as above |
duke@435 | 109 | |
duke@435 | 110 | StackValueCollection *exprs = vf->expressions(); |
duke@435 | 111 | _expressions = new StackValueCollection(exprs->size()); |
duke@435 | 112 | for(index = 0; index < exprs->size(); index++) { |
duke@435 | 113 | StackValue* value = exprs->at(index); |
duke@435 | 114 | switch(value->type()) { |
duke@435 | 115 | case T_OBJECT: |
duke@435 | 116 | // preserve object type |
duke@435 | 117 | _expressions->add( new StackValue((intptr_t) (value->get_obj()()), T_OBJECT )); |
duke@435 | 118 | break; |
duke@435 | 119 | case T_CONFLICT: |
duke@435 | 120 | // A dead stack element. Will be initialized to null/zero. |
duke@435 | 121 | // This can occur when the compiler emits a state in which stack |
duke@435 | 122 | // elements are known to be dead (because of an imminent exception). |
duke@435 | 123 | _expressions->add( new StackValue()); |
duke@435 | 124 | break; |
duke@435 | 125 | case T_INT: |
duke@435 | 126 | _expressions->add( new StackValue(value->get_int())); |
duke@435 | 127 | break; |
duke@435 | 128 | default: |
duke@435 | 129 | ShouldNotReachHere(); |
duke@435 | 130 | } |
duke@435 | 131 | } |
duke@435 | 132 | } |
duke@435 | 133 | |
duke@435 | 134 | int unpack_counter = 0; |
duke@435 | 135 | |
duke@435 | 136 | void vframeArrayElement::unpack_on_stack(int callee_parameters, |
duke@435 | 137 | int callee_locals, |
duke@435 | 138 | frame* caller, |
duke@435 | 139 | bool is_top_frame, |
duke@435 | 140 | int exec_mode) { |
duke@435 | 141 | JavaThread* thread = (JavaThread*) Thread::current(); |
duke@435 | 142 | |
duke@435 | 143 | // Look at bci and decide on bcp and continuation pc |
duke@435 | 144 | address bcp; |
duke@435 | 145 | // C++ interpreter doesn't need a pc since it will figure out what to do when it |
duke@435 | 146 | // begins execution |
duke@435 | 147 | address pc; |
duke@435 | 148 | bool use_next_mdp; // true if we should use the mdp associated with the next bci |
duke@435 | 149 | // rather than the one associated with bcp |
duke@435 | 150 | if (raw_bci() == SynchronizationEntryBCI) { |
duke@435 | 151 | // We are deoptimizing while hanging in prologue code for synchronized method |
duke@435 | 152 | bcp = method()->bcp_from(0); // first byte code |
duke@435 | 153 | pc = Interpreter::deopt_entry(vtos, 0); // step = 0 since we don't skip current bytecode |
duke@435 | 154 | use_next_mdp = false; |
duke@435 | 155 | } else { |
duke@435 | 156 | bcp = method()->bcp_from(bci()); |
duke@435 | 157 | pc = Interpreter::continuation_for(method(), bcp, callee_parameters, is_top_frame, use_next_mdp); |
duke@435 | 158 | } |
duke@435 | 159 | assert(Bytecodes::is_defined(*bcp), "must be a valid bytecode"); |
duke@435 | 160 | |
duke@435 | 161 | // Monitorenter and pending exceptions: |
duke@435 | 162 | // |
duke@435 | 163 | // For Compiler2, there should be no pending exception when deoptimizing at monitorenter |
duke@435 | 164 | // because there is no safepoint at the null pointer check (it is either handled explicitly |
duke@435 | 165 | // or prior to the monitorenter) and asynchronous exceptions are not made "pending" by the |
duke@435 | 166 | // runtime interface for the slow case (see JRT_ENTRY_FOR_MONITORENTER). If an asynchronous |
duke@435 | 167 | // exception was processed, the bytecode pointer would have to be extended one bytecode beyond |
duke@435 | 168 | // the monitorenter to place it in the proper exception range. |
duke@435 | 169 | // |
duke@435 | 170 | // For Compiler1, deoptimization can occur while throwing a NullPointerException at monitorenter, |
duke@435 | 171 | // in which case bcp should point to the monitorenter since it is within the exception's range. |
duke@435 | 172 | |
duke@435 | 173 | assert(*bcp != Bytecodes::_monitorenter || is_top_frame, "a _monitorenter must be a top frame"); |
duke@435 | 174 | // TIERED Must know the compiler of the deoptee QQQ |
duke@435 | 175 | COMPILER2_PRESENT(guarantee(*bcp != Bytecodes::_monitorenter || exec_mode != Deoptimization::Unpack_exception, |
duke@435 | 176 | "shouldn't get exception during monitorenter");) |
duke@435 | 177 | |
duke@435 | 178 | int popframe_preserved_args_size_in_bytes = 0; |
duke@435 | 179 | int popframe_preserved_args_size_in_words = 0; |
duke@435 | 180 | if (is_top_frame) { |
duke@435 | 181 | JvmtiThreadState *state = thread->jvmti_thread_state(); |
duke@435 | 182 | if (JvmtiExport::can_pop_frame() && |
duke@435 | 183 | (thread->has_pending_popframe() || thread->popframe_forcing_deopt_reexecution())) { |
duke@435 | 184 | if (thread->has_pending_popframe()) { |
duke@435 | 185 | // Pop top frame after deoptimization |
duke@435 | 186 | #ifndef CC_INTERP |
duke@435 | 187 | pc = Interpreter::remove_activation_preserving_args_entry(); |
duke@435 | 188 | #else |
duke@435 | 189 | // Do an uncommon trap type entry. c++ interpreter will know |
duke@435 | 190 | // to pop frame and preserve the args |
duke@435 | 191 | pc = Interpreter::deopt_entry(vtos, 0); |
duke@435 | 192 | use_next_mdp = false; |
duke@435 | 193 | #endif |
duke@435 | 194 | } else { |
duke@435 | 195 | // Reexecute invoke in top frame |
duke@435 | 196 | pc = Interpreter::deopt_entry(vtos, 0); |
duke@435 | 197 | use_next_mdp = false; |
duke@435 | 198 | popframe_preserved_args_size_in_bytes = in_bytes(thread->popframe_preserved_args_size()); |
duke@435 | 199 | // Note: the PopFrame-related extension of the expression stack size is done in |
duke@435 | 200 | // Deoptimization::fetch_unroll_info_helper |
duke@435 | 201 | popframe_preserved_args_size_in_words = in_words(thread->popframe_preserved_args_size_in_words()); |
duke@435 | 202 | } |
duke@435 | 203 | } else if (JvmtiExport::can_force_early_return() && state != NULL && state->is_earlyret_pending()) { |
duke@435 | 204 | // Force early return from top frame after deoptimization |
duke@435 | 205 | #ifndef CC_INTERP |
duke@435 | 206 | pc = Interpreter::remove_activation_early_entry(state->earlyret_tos()); |
duke@435 | 207 | #else |
duke@435 | 208 | // TBD: Need to implement ForceEarlyReturn for CC_INTERP (ia64) |
duke@435 | 209 | #endif |
duke@435 | 210 | } else { |
duke@435 | 211 | // Possibly override the previous pc computation of the top (youngest) frame |
duke@435 | 212 | switch (exec_mode) { |
duke@435 | 213 | case Deoptimization::Unpack_deopt: |
duke@435 | 214 | // use what we've got |
duke@435 | 215 | break; |
duke@435 | 216 | case Deoptimization::Unpack_exception: |
duke@435 | 217 | // exception is pending |
duke@435 | 218 | pc = SharedRuntime::raw_exception_handler_for_return_address(pc); |
duke@435 | 219 | // [phh] We're going to end up in some handler or other, so it doesn't |
duke@435 | 220 | // matter what mdp we point to. See exception_handler_for_exception() |
duke@435 | 221 | // in interpreterRuntime.cpp. |
duke@435 | 222 | break; |
duke@435 | 223 | case Deoptimization::Unpack_uncommon_trap: |
duke@435 | 224 | case Deoptimization::Unpack_reexecute: |
duke@435 | 225 | // redo last byte code |
duke@435 | 226 | pc = Interpreter::deopt_entry(vtos, 0); |
duke@435 | 227 | use_next_mdp = false; |
duke@435 | 228 | break; |
duke@435 | 229 | default: |
duke@435 | 230 | ShouldNotReachHere(); |
duke@435 | 231 | } |
duke@435 | 232 | } |
duke@435 | 233 | } |
duke@435 | 234 | |
duke@435 | 235 | // Setup the interpreter frame |
duke@435 | 236 | |
duke@435 | 237 | assert(method() != NULL, "method must exist"); |
duke@435 | 238 | int temps = expressions()->size(); |
duke@435 | 239 | |
duke@435 | 240 | int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors(); |
duke@435 | 241 | |
duke@435 | 242 | Interpreter::layout_activation(method(), |
duke@435 | 243 | temps + callee_parameters, |
duke@435 | 244 | popframe_preserved_args_size_in_words, |
duke@435 | 245 | locks, |
duke@435 | 246 | callee_parameters, |
duke@435 | 247 | callee_locals, |
duke@435 | 248 | caller, |
duke@435 | 249 | iframe(), |
duke@435 | 250 | is_top_frame); |
duke@435 | 251 | |
duke@435 | 252 | // Update the pc in the frame object and overwrite the temporary pc |
duke@435 | 253 | // we placed in the skeletal frame now that we finally know the |
duke@435 | 254 | // exact interpreter address we should use. |
duke@435 | 255 | |
duke@435 | 256 | _frame.patch_pc(thread, pc); |
duke@435 | 257 | |
duke@435 | 258 | assert (!method()->is_synchronized() || locks > 0, "synchronized methods must have monitors"); |
duke@435 | 259 | |
duke@435 | 260 | BasicObjectLock* top = iframe()->interpreter_frame_monitor_begin(); |
duke@435 | 261 | for (int index = 0; index < locks; index++) { |
duke@435 | 262 | top = iframe()->previous_monitor_in_interpreter_frame(top); |
duke@435 | 263 | BasicObjectLock* src = _monitors->at(index); |
duke@435 | 264 | top->set_obj(src->obj()); |
duke@435 | 265 | src->lock()->move_to(src->obj(), top->lock()); |
duke@435 | 266 | } |
duke@435 | 267 | if (ProfileInterpreter) { |
duke@435 | 268 | iframe()->interpreter_frame_set_mdx(0); // clear out the mdp. |
duke@435 | 269 | } |
duke@435 | 270 | iframe()->interpreter_frame_set_bcx((intptr_t)bcp); // cannot use bcp because frame is not initialized yet |
duke@435 | 271 | if (ProfileInterpreter) { |
duke@435 | 272 | methodDataOop mdo = method()->method_data(); |
duke@435 | 273 | if (mdo != NULL) { |
duke@435 | 274 | int bci = iframe()->interpreter_frame_bci(); |
duke@435 | 275 | if (use_next_mdp) ++bci; |
duke@435 | 276 | address mdp = mdo->bci_to_dp(bci); |
duke@435 | 277 | iframe()->interpreter_frame_set_mdp(mdp); |
duke@435 | 278 | } |
duke@435 | 279 | } |
duke@435 | 280 | |
duke@435 | 281 | // Unpack expression stack |
duke@435 | 282 | // If this is an intermediate frame (i.e. not top frame) then this |
duke@435 | 283 | // only unpacks the part of the expression stack not used by callee |
duke@435 | 284 | // as parameters. The callee parameters are unpacked as part of the |
duke@435 | 285 | // callee locals. |
duke@435 | 286 | int i; |
duke@435 | 287 | for(i = 0; i < expressions()->size(); i++) { |
duke@435 | 288 | StackValue *value = expressions()->at(i); |
duke@435 | 289 | intptr_t* addr = iframe()->interpreter_frame_expression_stack_at(i); |
duke@435 | 290 | switch(value->type()) { |
duke@435 | 291 | case T_INT: |
duke@435 | 292 | *addr = value->get_int(); |
duke@435 | 293 | break; |
duke@435 | 294 | case T_OBJECT: |
duke@435 | 295 | *addr = value->get_int(T_OBJECT); |
duke@435 | 296 | break; |
duke@435 | 297 | case T_CONFLICT: |
duke@435 | 298 | // A dead stack slot. Initialize to null in case it is an oop. |
duke@435 | 299 | *addr = NULL_WORD; |
duke@435 | 300 | break; |
duke@435 | 301 | default: |
duke@435 | 302 | ShouldNotReachHere(); |
duke@435 | 303 | } |
duke@435 | 304 | if (TaggedStackInterpreter) { |
duke@435 | 305 | // Write tag to the stack |
duke@435 | 306 | iframe()->interpreter_frame_set_expression_stack_tag(i, |
duke@435 | 307 | frame::tag_for_basic_type(value->type())); |
duke@435 | 308 | } |
duke@435 | 309 | } |
duke@435 | 310 | |
duke@435 | 311 | |
duke@435 | 312 | // Unpack the locals |
duke@435 | 313 | for(i = 0; i < locals()->size(); i++) { |
duke@435 | 314 | StackValue *value = locals()->at(i); |
duke@435 | 315 | intptr_t* addr = iframe()->interpreter_frame_local_at(i); |
duke@435 | 316 | switch(value->type()) { |
duke@435 | 317 | case T_INT: |
duke@435 | 318 | *addr = value->get_int(); |
duke@435 | 319 | break; |
duke@435 | 320 | case T_OBJECT: |
duke@435 | 321 | *addr = value->get_int(T_OBJECT); |
duke@435 | 322 | break; |
duke@435 | 323 | case T_CONFLICT: |
duke@435 | 324 | // A dead location. If it is an oop then we need a NULL to prevent GC from following it |
duke@435 | 325 | *addr = NULL_WORD; |
duke@435 | 326 | break; |
duke@435 | 327 | default: |
duke@435 | 328 | ShouldNotReachHere(); |
duke@435 | 329 | } |
duke@435 | 330 | if (TaggedStackInterpreter) { |
duke@435 | 331 | // Write tag to stack |
duke@435 | 332 | iframe()->interpreter_frame_set_local_tag(i, |
duke@435 | 333 | frame::tag_for_basic_type(value->type())); |
duke@435 | 334 | } |
duke@435 | 335 | } |
duke@435 | 336 | |
duke@435 | 337 | if (is_top_frame && JvmtiExport::can_pop_frame() && thread->popframe_forcing_deopt_reexecution()) { |
duke@435 | 338 | // An interpreted frame was popped but it returns to a deoptimized |
duke@435 | 339 | // frame. The incoming arguments to the interpreted activation |
duke@435 | 340 | // were preserved in thread-local storage by the |
duke@435 | 341 | // remove_activation_preserving_args_entry in the interpreter; now |
duke@435 | 342 | // we put them back into the just-unpacked interpreter frame. |
duke@435 | 343 | // Note that this assumes that the locals arena grows toward lower |
duke@435 | 344 | // addresses. |
duke@435 | 345 | if (popframe_preserved_args_size_in_words != 0) { |
duke@435 | 346 | void* saved_args = thread->popframe_preserved_args(); |
duke@435 | 347 | assert(saved_args != NULL, "must have been saved by interpreter"); |
duke@435 | 348 | #ifdef ASSERT |
duke@435 | 349 | int stack_words = Interpreter::stackElementWords(); |
duke@435 | 350 | assert(popframe_preserved_args_size_in_words <= |
duke@435 | 351 | iframe()->interpreter_frame_expression_stack_size()*stack_words, |
duke@435 | 352 | "expression stack size should have been extended"); |
duke@435 | 353 | #endif // ASSERT |
duke@435 | 354 | int top_element = iframe()->interpreter_frame_expression_stack_size()-1; |
duke@435 | 355 | intptr_t* base; |
duke@435 | 356 | if (frame::interpreter_frame_expression_stack_direction() < 0) { |
duke@435 | 357 | base = iframe()->interpreter_frame_expression_stack_at(top_element); |
duke@435 | 358 | } else { |
duke@435 | 359 | base = iframe()->interpreter_frame_expression_stack(); |
duke@435 | 360 | } |
duke@435 | 361 | Copy::conjoint_bytes(saved_args, |
duke@435 | 362 | base, |
duke@435 | 363 | popframe_preserved_args_size_in_bytes); |
duke@435 | 364 | thread->popframe_free_preserved_args(); |
duke@435 | 365 | } |
duke@435 | 366 | } |
duke@435 | 367 | |
duke@435 | 368 | #ifndef PRODUCT |
duke@435 | 369 | if (TraceDeoptimization && Verbose) { |
duke@435 | 370 | ttyLocker ttyl; |
duke@435 | 371 | tty->print_cr("[%d Interpreted Frame]", ++unpack_counter); |
duke@435 | 372 | iframe()->print_on(tty); |
duke@435 | 373 | RegisterMap map(thread); |
duke@435 | 374 | vframe* f = vframe::new_vframe(iframe(), &map, thread); |
duke@435 | 375 | f->print(); |
duke@435 | 376 | iframe()->interpreter_frame_print_on(tty); |
duke@435 | 377 | |
duke@435 | 378 | tty->print_cr("locals size %d", locals()->size()); |
duke@435 | 379 | tty->print_cr("expression size %d", expressions()->size()); |
duke@435 | 380 | |
duke@435 | 381 | method()->print_value(); |
duke@435 | 382 | tty->cr(); |
duke@435 | 383 | // method()->print_codes(); |
duke@435 | 384 | } else if (TraceDeoptimization) { |
duke@435 | 385 | tty->print(" "); |
duke@435 | 386 | method()->print_value(); |
duke@435 | 387 | Bytecodes::Code code = Bytecodes::java_code_at(bcp); |
duke@435 | 388 | int bci = method()->bci_from(bcp); |
duke@435 | 389 | tty->print(" - %s", Bytecodes::name(code)); |
duke@435 | 390 | tty->print(" @ bci %d ", bci); |
duke@435 | 391 | tty->print_cr("sp = " PTR_FORMAT, iframe()->sp()); |
duke@435 | 392 | } |
duke@435 | 393 | #endif // PRODUCT |
duke@435 | 394 | |
duke@435 | 395 | // The expression stack and locals are in the resource area don't leave |
duke@435 | 396 | // a dangling pointer in the vframeArray we leave around for debug |
duke@435 | 397 | // purposes |
duke@435 | 398 | |
duke@435 | 399 | _locals = _expressions = NULL; |
duke@435 | 400 | |
duke@435 | 401 | } |
duke@435 | 402 | |
duke@435 | 403 | int vframeArrayElement::on_stack_size(int callee_parameters, |
duke@435 | 404 | int callee_locals, |
duke@435 | 405 | bool is_top_frame, |
duke@435 | 406 | int popframe_extra_stack_expression_els) const { |
duke@435 | 407 | assert(method()->max_locals() == locals()->size(), "just checking"); |
duke@435 | 408 | int locks = monitors() == NULL ? 0 : monitors()->number_of_monitors(); |
duke@435 | 409 | int temps = expressions()->size(); |
duke@435 | 410 | return Interpreter::size_activation(method(), |
duke@435 | 411 | temps + callee_parameters, |
duke@435 | 412 | popframe_extra_stack_expression_els, |
duke@435 | 413 | locks, |
duke@435 | 414 | callee_parameters, |
duke@435 | 415 | callee_locals, |
duke@435 | 416 | is_top_frame); |
duke@435 | 417 | } |
duke@435 | 418 | |
duke@435 | 419 | |
duke@435 | 420 | |
duke@435 | 421 | vframeArray* vframeArray::allocate(JavaThread* thread, int frame_size, GrowableArray<compiledVFrame*>* chunk, |
duke@435 | 422 | RegisterMap *reg_map, frame sender, frame caller, frame self) { |
duke@435 | 423 | |
duke@435 | 424 | // Allocate the vframeArray |
duke@435 | 425 | vframeArray * result = (vframeArray*) AllocateHeap(sizeof(vframeArray) + // fixed part |
duke@435 | 426 | sizeof(vframeArrayElement) * (chunk->length() - 1), // variable part |
duke@435 | 427 | "vframeArray::allocate"); |
duke@435 | 428 | result->_frames = chunk->length(); |
duke@435 | 429 | result->_owner_thread = thread; |
duke@435 | 430 | result->_sender = sender; |
duke@435 | 431 | result->_caller = caller; |
duke@435 | 432 | result->_original = self; |
duke@435 | 433 | result->set_unroll_block(NULL); // initialize it |
duke@435 | 434 | result->fill_in(thread, frame_size, chunk, reg_map); |
duke@435 | 435 | return result; |
duke@435 | 436 | } |
duke@435 | 437 | |
duke@435 | 438 | void vframeArray::fill_in(JavaThread* thread, |
duke@435 | 439 | int frame_size, |
duke@435 | 440 | GrowableArray<compiledVFrame*>* chunk, |
duke@435 | 441 | const RegisterMap *reg_map) { |
duke@435 | 442 | // Set owner first, it is used when adding monitor chunks |
duke@435 | 443 | |
duke@435 | 444 | _frame_size = frame_size; |
duke@435 | 445 | for(int i = 0; i < chunk->length(); i++) { |
duke@435 | 446 | element(i)->fill_in(chunk->at(i)); |
duke@435 | 447 | } |
duke@435 | 448 | |
duke@435 | 449 | // Copy registers for callee-saved registers |
duke@435 | 450 | if (reg_map != NULL) { |
duke@435 | 451 | for(int i = 0; i < RegisterMap::reg_count; i++) { |
duke@435 | 452 | #ifdef AMD64 |
duke@435 | 453 | // The register map has one entry for every int (32-bit value), so |
duke@435 | 454 | // 64-bit physical registers have two entries in the map, one for |
duke@435 | 455 | // each half. Ignore the high halves of 64-bit registers, just like |
duke@435 | 456 | // frame::oopmapreg_to_location does. |
duke@435 | 457 | // |
duke@435 | 458 | // [phh] FIXME: this is a temporary hack! This code *should* work |
duke@435 | 459 | // correctly w/o this hack, possibly by changing RegisterMap::pd_location |
duke@435 | 460 | // in frame_amd64.cpp and the values of the phantom high half registers |
duke@435 | 461 | // in amd64.ad. |
duke@435 | 462 | // if (VMReg::Name(i) < SharedInfo::stack0 && is_even(i)) { |
duke@435 | 463 | intptr_t* src = (intptr_t*) reg_map->location(VMRegImpl::as_VMReg(i)); |
duke@435 | 464 | _callee_registers[i] = src != NULL ? *src : NULL_WORD; |
duke@435 | 465 | // } else { |
duke@435 | 466 | // jint* src = (jint*) reg_map->location(VMReg::Name(i)); |
duke@435 | 467 | // _callee_registers[i] = src != NULL ? *src : NULL_WORD; |
duke@435 | 468 | // } |
duke@435 | 469 | #else |
duke@435 | 470 | jint* src = (jint*) reg_map->location(VMRegImpl::as_VMReg(i)); |
duke@435 | 471 | _callee_registers[i] = src != NULL ? *src : NULL_WORD; |
duke@435 | 472 | #endif |
duke@435 | 473 | if (src == NULL) { |
duke@435 | 474 | set_location_valid(i, false); |
duke@435 | 475 | } else { |
duke@435 | 476 | set_location_valid(i, true); |
duke@435 | 477 | jint* dst = (jint*) register_location(i); |
duke@435 | 478 | *dst = *src; |
duke@435 | 479 | } |
duke@435 | 480 | } |
duke@435 | 481 | } |
duke@435 | 482 | } |
duke@435 | 483 | |
duke@435 | 484 | void vframeArray::unpack_to_stack(frame &unpack_frame, int exec_mode) { |
duke@435 | 485 | // stack picture |
duke@435 | 486 | // unpack_frame |
duke@435 | 487 | // [new interpreter frames ] (frames are skeletal but walkable) |
duke@435 | 488 | // caller_frame |
duke@435 | 489 | // |
duke@435 | 490 | // This routine fills in the missing data for the skeletal interpreter frames |
duke@435 | 491 | // in the above picture. |
duke@435 | 492 | |
duke@435 | 493 | // Find the skeletal interpreter frames to unpack into |
duke@435 | 494 | RegisterMap map(JavaThread::current(), false); |
duke@435 | 495 | // Get the youngest frame we will unpack (last to be unpacked) |
duke@435 | 496 | frame me = unpack_frame.sender(&map); |
duke@435 | 497 | int index; |
duke@435 | 498 | for (index = 0; index < frames(); index++ ) { |
duke@435 | 499 | *element(index)->iframe() = me; |
duke@435 | 500 | // Get the caller frame (possibly skeletal) |
duke@435 | 501 | me = me.sender(&map); |
duke@435 | 502 | } |
duke@435 | 503 | |
duke@435 | 504 | frame caller_frame = me; |
duke@435 | 505 | |
duke@435 | 506 | // Do the unpacking of interpreter frames; the frame at index 0 represents the top activation, so it has no callee |
duke@435 | 507 | |
duke@435 | 508 | // Unpack the frames from the oldest (frames() -1) to the youngest (0) |
duke@435 | 509 | |
duke@435 | 510 | for (index = frames() - 1; index >= 0 ; index--) { |
duke@435 | 511 | int callee_parameters = index == 0 ? 0 : element(index-1)->method()->size_of_parameters(); |
duke@435 | 512 | int callee_locals = index == 0 ? 0 : element(index-1)->method()->max_locals(); |
duke@435 | 513 | element(index)->unpack_on_stack(callee_parameters, |
duke@435 | 514 | callee_locals, |
duke@435 | 515 | &caller_frame, |
duke@435 | 516 | index == 0, |
duke@435 | 517 | exec_mode); |
duke@435 | 518 | if (index == frames() - 1) { |
duke@435 | 519 | Deoptimization::unwind_callee_save_values(element(index)->iframe(), this); |
duke@435 | 520 | } |
duke@435 | 521 | caller_frame = *element(index)->iframe(); |
duke@435 | 522 | } |
duke@435 | 523 | |
duke@435 | 524 | |
duke@435 | 525 | deallocate_monitor_chunks(); |
duke@435 | 526 | } |
duke@435 | 527 | |
duke@435 | 528 | void vframeArray::deallocate_monitor_chunks() { |
duke@435 | 529 | JavaThread* jt = JavaThread::current(); |
duke@435 | 530 | for (int index = 0; index < frames(); index++ ) { |
duke@435 | 531 | element(index)->free_monitors(jt); |
duke@435 | 532 | } |
duke@435 | 533 | } |
duke@435 | 534 | |
duke@435 | 535 | #ifndef PRODUCT |
duke@435 | 536 | |
duke@435 | 537 | bool vframeArray::structural_compare(JavaThread* thread, GrowableArray<compiledVFrame*>* chunk) { |
duke@435 | 538 | if (owner_thread() != thread) return false; |
duke@435 | 539 | int index = 0; |
duke@435 | 540 | #if 0 // FIXME can't do this comparison |
duke@435 | 541 | |
duke@435 | 542 | // Compare only within vframe array. |
duke@435 | 543 | for (deoptimizedVFrame* vf = deoptimizedVFrame::cast(vframe_at(first_index())); vf; vf = vf->deoptimized_sender_or_null()) { |
duke@435 | 544 | if (index >= chunk->length() || !vf->structural_compare(chunk->at(index))) return false; |
duke@435 | 545 | index++; |
duke@435 | 546 | } |
duke@435 | 547 | if (index != chunk->length()) return false; |
duke@435 | 548 | #endif |
duke@435 | 549 | |
duke@435 | 550 | return true; |
duke@435 | 551 | } |
duke@435 | 552 | |
duke@435 | 553 | #endif |
duke@435 | 554 | |
duke@435 | 555 | address vframeArray::register_location(int i) const { |
duke@435 | 556 | assert(0 <= i && i < RegisterMap::reg_count, "index out of bounds"); |
duke@435 | 557 | return (address) & _callee_registers[i]; |
duke@435 | 558 | } |
duke@435 | 559 | |
duke@435 | 560 | |
duke@435 | 561 | #ifndef PRODUCT |
duke@435 | 562 | |
duke@435 | 563 | // Printing |
duke@435 | 564 | |
duke@435 | 565 | // Note: we cannot have print_on as const, as we allocate inside the method |
duke@435 | 566 | void vframeArray::print_on_2(outputStream* st) { |
duke@435 | 567 | st->print_cr(" - sp: " INTPTR_FORMAT, sp()); |
duke@435 | 568 | st->print(" - thread: "); |
duke@435 | 569 | Thread::current()->print(); |
duke@435 | 570 | st->print_cr(" - frame size: %d", frame_size()); |
duke@435 | 571 | for (int index = 0; index < frames() ; index++ ) { |
duke@435 | 572 | element(index)->print(st); |
duke@435 | 573 | } |
duke@435 | 574 | } |
duke@435 | 575 | |
duke@435 | 576 | void vframeArrayElement::print(outputStream* st) { |
duke@435 | 577 | st->print_cr(" - interpreter_frame -> sp: ", INTPTR_FORMAT, iframe()->sp()); |
duke@435 | 578 | } |
duke@435 | 579 | |
duke@435 | 580 | void vframeArray::print_value_on(outputStream* st) const { |
duke@435 | 581 | st->print_cr("vframeArray [%d] ", frames()); |
duke@435 | 582 | } |
duke@435 | 583 | |
duke@435 | 584 | |
duke@435 | 585 | #endif |