1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/share/vm/opto/callnode.cpp Sat Dec 01 00:00:00 2007 +0000
1.3 @@ -0,0 +1,1311 @@
1.4 +/*
1.5 + * Copyright 1997-2006 Sun Microsystems, Inc. All Rights Reserved.
1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
1.7 + *
1.8 + * This code is free software; you can redistribute it and/or modify it
1.9 + * under the terms of the GNU General Public License version 2 only, as
1.10 + * published by the Free Software Foundation.
1.11 + *
1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT
1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
1.15 + * version 2 for more details (a copy is included in the LICENSE file that
1.16 + * accompanied this code).
1.17 + *
1.18 + * You should have received a copy of the GNU General Public License version
1.19 + * 2 along with this work; if not, write to the Free Software Foundation,
1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
1.21 + *
1.22 + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
1.23 + * CA 95054 USA or visit www.sun.com if you need additional information or
1.24 + * have any questions.
1.25 + *
1.26 + */
1.27 +
1.28 +// Portions of code courtesy of Clifford Click
1.29 +
1.30 +// Optimization - Graph Style
1.31 +
1.32 +#include "incls/_precompiled.incl"
1.33 +#include "incls/_callnode.cpp.incl"
1.34 +
1.35 +//=============================================================================
1.36 +uint StartNode::size_of() const { return sizeof(*this); }
1.37 +uint StartNode::cmp( const Node &n ) const
1.38 +{ return _domain == ((StartNode&)n)._domain; }
1.39 +const Type *StartNode::bottom_type() const { return _domain; }
1.40 +const Type *StartNode::Value(PhaseTransform *phase) const { return _domain; }
1.41 +#ifndef PRODUCT
1.42 +void StartNode::dump_spec(outputStream *st) const { st->print(" #"); _domain->dump_on(st);}
1.43 +#endif
1.44 +
1.45 +//------------------------------Ideal------------------------------------------
1.46 +Node *StartNode::Ideal(PhaseGVN *phase, bool can_reshape){
1.47 + return remove_dead_region(phase, can_reshape) ? this : NULL;
1.48 +}
1.49 +
1.50 +//------------------------------calling_convention-----------------------------
1.51 +void StartNode::calling_convention( BasicType* sig_bt, VMRegPair *parm_regs, uint argcnt ) const {
1.52 + Matcher::calling_convention( sig_bt, parm_regs, argcnt, false );
1.53 +}
1.54 +
1.55 +//------------------------------Registers--------------------------------------
1.56 +const RegMask &StartNode::in_RegMask(uint) const {
1.57 + return RegMask::Empty;
1.58 +}
1.59 +
1.60 +//------------------------------match------------------------------------------
1.61 +// Construct projections for incoming parameters, and their RegMask info
1.62 +Node *StartNode::match( const ProjNode *proj, const Matcher *match ) {
1.63 + switch (proj->_con) {
1.64 + case TypeFunc::Control:
1.65 + case TypeFunc::I_O:
1.66 + case TypeFunc::Memory:
1.67 + return new (match->C, 1) MachProjNode(this,proj->_con,RegMask::Empty,MachProjNode::unmatched_proj);
1.68 + case TypeFunc::FramePtr:
1.69 + return new (match->C, 1) MachProjNode(this,proj->_con,Matcher::c_frame_ptr_mask, Op_RegP);
1.70 + case TypeFunc::ReturnAdr:
1.71 + return new (match->C, 1) MachProjNode(this,proj->_con,match->_return_addr_mask,Op_RegP);
1.72 + case TypeFunc::Parms:
1.73 + default: {
1.74 + uint parm_num = proj->_con - TypeFunc::Parms;
1.75 + const Type *t = _domain->field_at(proj->_con);
1.76 + if (t->base() == Type::Half) // 2nd half of Longs and Doubles
1.77 + return new (match->C, 1) ConNode(Type::TOP);
1.78 + uint ideal_reg = Matcher::base2reg[t->base()];
1.79 + RegMask &rm = match->_calling_convention_mask[parm_num];
1.80 + return new (match->C, 1) MachProjNode(this,proj->_con,rm,ideal_reg);
1.81 + }
1.82 + }
1.83 + return NULL;
1.84 +}
1.85 +
1.86 +//------------------------------StartOSRNode----------------------------------
1.87 +// The method start node for an on stack replacement adapter
1.88 +
1.89 +//------------------------------osr_domain-----------------------------
1.90 +const TypeTuple *StartOSRNode::osr_domain() {
1.91 + const Type **fields = TypeTuple::fields(2);
1.92 + fields[TypeFunc::Parms+0] = TypeRawPtr::BOTTOM; // address of osr buffer
1.93 +
1.94 + return TypeTuple::make(TypeFunc::Parms+1, fields);
1.95 +}
1.96 +
1.97 +//=============================================================================
1.98 +const char * const ParmNode::names[TypeFunc::Parms+1] = {
1.99 + "Control", "I_O", "Memory", "FramePtr", "ReturnAdr", "Parms"
1.100 +};
1.101 +
1.102 +#ifndef PRODUCT
1.103 +void ParmNode::dump_spec(outputStream *st) const {
1.104 + if( _con < TypeFunc::Parms ) {
1.105 + st->print(names[_con]);
1.106 + } else {
1.107 + st->print("Parm%d: ",_con-TypeFunc::Parms);
1.108 + // Verbose and WizardMode dump bottom_type for all nodes
1.109 + if( !Verbose && !WizardMode ) bottom_type()->dump_on(st);
1.110 + }
1.111 +}
1.112 +#endif
1.113 +
1.114 +uint ParmNode::ideal_reg() const {
1.115 + switch( _con ) {
1.116 + case TypeFunc::Control : // fall through
1.117 + case TypeFunc::I_O : // fall through
1.118 + case TypeFunc::Memory : return 0;
1.119 + case TypeFunc::FramePtr : // fall through
1.120 + case TypeFunc::ReturnAdr: return Op_RegP;
1.121 + default : assert( _con > TypeFunc::Parms, "" );
1.122 + // fall through
1.123 + case TypeFunc::Parms : {
1.124 + // Type of argument being passed
1.125 + const Type *t = in(0)->as_Start()->_domain->field_at(_con);
1.126 + return Matcher::base2reg[t->base()];
1.127 + }
1.128 + }
1.129 + ShouldNotReachHere();
1.130 + return 0;
1.131 +}
1.132 +
1.133 +//=============================================================================
1.134 +ReturnNode::ReturnNode(uint edges, Node *cntrl, Node *i_o, Node *memory, Node *frameptr, Node *retadr ) : Node(edges) {
1.135 + init_req(TypeFunc::Control,cntrl);
1.136 + init_req(TypeFunc::I_O,i_o);
1.137 + init_req(TypeFunc::Memory,memory);
1.138 + init_req(TypeFunc::FramePtr,frameptr);
1.139 + init_req(TypeFunc::ReturnAdr,retadr);
1.140 +}
1.141 +
1.142 +Node *ReturnNode::Ideal(PhaseGVN *phase, bool can_reshape){
1.143 + return remove_dead_region(phase, can_reshape) ? this : NULL;
1.144 +}
1.145 +
1.146 +const Type *ReturnNode::Value( PhaseTransform *phase ) const {
1.147 + return ( phase->type(in(TypeFunc::Control)) == Type::TOP)
1.148 + ? Type::TOP
1.149 + : Type::BOTTOM;
1.150 +}
1.151 +
1.152 +// Do we Match on this edge index or not? No edges on return nodes
1.153 +uint ReturnNode::match_edge(uint idx) const {
1.154 + return 0;
1.155 +}
1.156 +
1.157 +
1.158 +#ifndef PRODUCT
1.159 +void ReturnNode::dump_req() const {
1.160 + // Dump the required inputs, enclosed in '(' and ')'
1.161 + uint i; // Exit value of loop
1.162 + for( i=0; i<req(); i++ ) { // For all required inputs
1.163 + if( i == TypeFunc::Parms ) tty->print("returns");
1.164 + if( in(i) ) tty->print("%c%d ", Compile::current()->node_arena()->contains(in(i)) ? ' ' : 'o', in(i)->_idx);
1.165 + else tty->print("_ ");
1.166 + }
1.167 +}
1.168 +#endif
1.169 +
1.170 +//=============================================================================
1.171 +RethrowNode::RethrowNode(
1.172 + Node* cntrl,
1.173 + Node* i_o,
1.174 + Node* memory,
1.175 + Node* frameptr,
1.176 + Node* ret_adr,
1.177 + Node* exception
1.178 +) : Node(TypeFunc::Parms + 1) {
1.179 + init_req(TypeFunc::Control , cntrl );
1.180 + init_req(TypeFunc::I_O , i_o );
1.181 + init_req(TypeFunc::Memory , memory );
1.182 + init_req(TypeFunc::FramePtr , frameptr );
1.183 + init_req(TypeFunc::ReturnAdr, ret_adr);
1.184 + init_req(TypeFunc::Parms , exception);
1.185 +}
1.186 +
1.187 +Node *RethrowNode::Ideal(PhaseGVN *phase, bool can_reshape){
1.188 + return remove_dead_region(phase, can_reshape) ? this : NULL;
1.189 +}
1.190 +
1.191 +const Type *RethrowNode::Value( PhaseTransform *phase ) const {
1.192 + return (phase->type(in(TypeFunc::Control)) == Type::TOP)
1.193 + ? Type::TOP
1.194 + : Type::BOTTOM;
1.195 +}
1.196 +
1.197 +uint RethrowNode::match_edge(uint idx) const {
1.198 + return 0;
1.199 +}
1.200 +
1.201 +#ifndef PRODUCT
1.202 +void RethrowNode::dump_req() const {
1.203 + // Dump the required inputs, enclosed in '(' and ')'
1.204 + uint i; // Exit value of loop
1.205 + for( i=0; i<req(); i++ ) { // For all required inputs
1.206 + if( i == TypeFunc::Parms ) tty->print("exception");
1.207 + if( in(i) ) tty->print("%c%d ", Compile::current()->node_arena()->contains(in(i)) ? ' ' : 'o', in(i)->_idx);
1.208 + else tty->print("_ ");
1.209 + }
1.210 +}
1.211 +#endif
1.212 +
1.213 +//=============================================================================
1.214 +// Do we Match on this edge index or not? Match only target address & method
1.215 +uint TailCallNode::match_edge(uint idx) const {
1.216 + return TypeFunc::Parms <= idx && idx <= TypeFunc::Parms+1;
1.217 +}
1.218 +
1.219 +//=============================================================================
1.220 +// Do we Match on this edge index or not? Match only target address & oop
1.221 +uint TailJumpNode::match_edge(uint idx) const {
1.222 + return TypeFunc::Parms <= idx && idx <= TypeFunc::Parms+1;
1.223 +}
1.224 +
1.225 +//=============================================================================
1.226 +JVMState::JVMState(ciMethod* method, JVMState* caller) {
1.227 + assert(method != NULL, "must be valid call site");
1.228 + _method = method;
1.229 + debug_only(_bci = -99); // random garbage value
1.230 + debug_only(_map = (SafePointNode*)-1);
1.231 + _caller = caller;
1.232 + _depth = 1 + (caller == NULL ? 0 : caller->depth());
1.233 + _locoff = TypeFunc::Parms;
1.234 + _stkoff = _locoff + _method->max_locals();
1.235 + _monoff = _stkoff + _method->max_stack();
1.236 + _endoff = _monoff;
1.237 + _sp = 0;
1.238 +}
1.239 +JVMState::JVMState(int stack_size) {
1.240 + _method = NULL;
1.241 + _bci = InvocationEntryBci;
1.242 + debug_only(_map = (SafePointNode*)-1);
1.243 + _caller = NULL;
1.244 + _depth = 1;
1.245 + _locoff = TypeFunc::Parms;
1.246 + _stkoff = _locoff;
1.247 + _monoff = _stkoff + stack_size;
1.248 + _endoff = _monoff;
1.249 + _sp = 0;
1.250 +}
1.251 +
1.252 +//--------------------------------of_depth-------------------------------------
1.253 +JVMState* JVMState::of_depth(int d) const {
1.254 + const JVMState* jvmp = this;
1.255 + assert(0 < d && (uint)d <= depth(), "oob");
1.256 + for (int skip = depth() - d; skip > 0; skip--) {
1.257 + jvmp = jvmp->caller();
1.258 + }
1.259 + assert(jvmp->depth() == (uint)d, "found the right one");
1.260 + return (JVMState*)jvmp;
1.261 +}
1.262 +
1.263 +//-----------------------------same_calls_as-----------------------------------
1.264 +bool JVMState::same_calls_as(const JVMState* that) const {
1.265 + if (this == that) return true;
1.266 + if (this->depth() != that->depth()) return false;
1.267 + const JVMState* p = this;
1.268 + const JVMState* q = that;
1.269 + for (;;) {
1.270 + if (p->_method != q->_method) return false;
1.271 + if (p->_method == NULL) return true; // bci is irrelevant
1.272 + if (p->_bci != q->_bci) return false;
1.273 + p = p->caller();
1.274 + q = q->caller();
1.275 + if (p == q) return true;
1.276 + assert(p != NULL && q != NULL, "depth check ensures we don't run off end");
1.277 + }
1.278 +}
1.279 +
1.280 +//------------------------------debug_start------------------------------------
1.281 +uint JVMState::debug_start() const {
1.282 + debug_only(JVMState* jvmroot = of_depth(1));
1.283 + assert(jvmroot->locoff() <= this->locoff(), "youngest JVMState must be last");
1.284 + return of_depth(1)->locoff();
1.285 +}
1.286 +
1.287 +//-------------------------------debug_end-------------------------------------
1.288 +uint JVMState::debug_end() const {
1.289 + debug_only(JVMState* jvmroot = of_depth(1));
1.290 + assert(jvmroot->endoff() <= this->endoff(), "youngest JVMState must be last");
1.291 + return endoff();
1.292 +}
1.293 +
1.294 +//------------------------------debug_depth------------------------------------
1.295 +uint JVMState::debug_depth() const {
1.296 + uint total = 0;
1.297 + for (const JVMState* jvmp = this; jvmp != NULL; jvmp = jvmp->caller()) {
1.298 + total += jvmp->debug_size();
1.299 + }
1.300 + return total;
1.301 +}
1.302 +
1.303 +//------------------------------format_helper----------------------------------
1.304 +// Given an allocation (a Chaitin object) and a Node decide if the Node carries
1.305 +// any defined value or not. If it does, print out the register or constant.
1.306 +#ifndef PRODUCT
1.307 +static void format_helper( PhaseRegAlloc *regalloc, outputStream* st, Node *n, const char *msg, uint i ) {
1.308 + if (n == NULL) { st->print(" NULL"); return; }
1.309 + if( OptoReg::is_valid(regalloc->get_reg_first(n))) { // Check for undefined
1.310 + char buf[50];
1.311 + regalloc->dump_register(n,buf);
1.312 + st->print(" %s%d]=%s",msg,i,buf);
1.313 + } else { // No register, but might be constant
1.314 + const Type *t = n->bottom_type();
1.315 + switch (t->base()) {
1.316 + case Type::Int:
1.317 + st->print(" %s%d]=#"INT32_FORMAT,msg,i,t->is_int()->get_con());
1.318 + break;
1.319 + case Type::AnyPtr:
1.320 + assert( t == TypePtr::NULL_PTR, "" );
1.321 + st->print(" %s%d]=#NULL",msg,i);
1.322 + break;
1.323 + case Type::AryPtr:
1.324 + case Type::KlassPtr:
1.325 + case Type::InstPtr:
1.326 + st->print(" %s%d]=#Ptr" INTPTR_FORMAT,msg,i,t->isa_oopptr()->const_oop());
1.327 + break;
1.328 + case Type::RawPtr:
1.329 + st->print(" %s%d]=#Raw" INTPTR_FORMAT,msg,i,t->is_rawptr());
1.330 + break;
1.331 + case Type::DoubleCon:
1.332 + st->print(" %s%d]=#%fD",msg,i,t->is_double_constant()->_d);
1.333 + break;
1.334 + case Type::FloatCon:
1.335 + st->print(" %s%d]=#%fF",msg,i,t->is_float_constant()->_f);
1.336 + break;
1.337 + case Type::Long:
1.338 + st->print(" %s%d]=#"INT64_FORMAT,msg,i,t->is_long()->get_con());
1.339 + break;
1.340 + case Type::Half:
1.341 + case Type::Top:
1.342 + st->print(" %s%d]=_",msg,i);
1.343 + break;
1.344 + default: ShouldNotReachHere();
1.345 + }
1.346 + }
1.347 +}
1.348 +#endif
1.349 +
1.350 +//------------------------------format-----------------------------------------
1.351 +#ifndef PRODUCT
1.352 +void JVMState::format(PhaseRegAlloc *regalloc, const Node *n, outputStream* st) const {
1.353 + st->print(" #");
1.354 + if( _method ) {
1.355 + _method->print_short_name(st);
1.356 + st->print(" @ bci:%d ",_bci);
1.357 + } else {
1.358 + st->print_cr(" runtime stub ");
1.359 + return;
1.360 + }
1.361 + if (n->is_MachSafePoint()) {
1.362 + MachSafePointNode *mcall = n->as_MachSafePoint();
1.363 + uint i;
1.364 + // Print locals
1.365 + for( i = 0; i < (uint)loc_size(); i++ )
1.366 + format_helper( regalloc, st, mcall->local(this, i), "L[", i );
1.367 + // Print stack
1.368 + for (i = 0; i < (uint)stk_size(); i++) {
1.369 + if ((uint)(_stkoff + i) >= mcall->len())
1.370 + st->print(" oob ");
1.371 + else
1.372 + format_helper( regalloc, st, mcall->stack(this, i), "STK[", i );
1.373 + }
1.374 + for (i = 0; (int)i < nof_monitors(); i++) {
1.375 + Node *box = mcall->monitor_box(this, i);
1.376 + Node *obj = mcall->monitor_obj(this, i);
1.377 + if ( OptoReg::is_valid(regalloc->get_reg_first(box)) ) {
1.378 + while( !box->is_BoxLock() ) box = box->in(1);
1.379 + format_helper( regalloc, st, box, "MON-BOX[", i );
1.380 + } else {
1.381 + OptoReg::Name box_reg = BoxLockNode::stack_slot(box);
1.382 + st->print(" MON-BOX%d=%s+%d",
1.383 + i,
1.384 + OptoReg::regname(OptoReg::c_frame_pointer),
1.385 + regalloc->reg2offset(box_reg));
1.386 + }
1.387 + format_helper( regalloc, st, obj, "MON-OBJ[", i );
1.388 + }
1.389 + }
1.390 + st->print_cr("");
1.391 + if (caller() != NULL) caller()->format(regalloc, n, st);
1.392 +}
1.393 +#endif
1.394 +
1.395 +#ifndef PRODUCT
1.396 +void JVMState::dump_spec(outputStream *st) const {
1.397 + if (_method != NULL) {
1.398 + bool printed = false;
1.399 + if (!Verbose) {
1.400 + // The JVMS dumps make really, really long lines.
1.401 + // Take out the most boring parts, which are the package prefixes.
1.402 + char buf[500];
1.403 + stringStream namest(buf, sizeof(buf));
1.404 + _method->print_short_name(&namest);
1.405 + if (namest.count() < sizeof(buf)) {
1.406 + const char* name = namest.base();
1.407 + if (name[0] == ' ') ++name;
1.408 + const char* endcn = strchr(name, ':'); // end of class name
1.409 + if (endcn == NULL) endcn = strchr(name, '(');
1.410 + if (endcn == NULL) endcn = name + strlen(name);
1.411 + while (endcn > name && endcn[-1] != '.' && endcn[-1] != '/')
1.412 + --endcn;
1.413 + st->print(" %s", endcn);
1.414 + printed = true;
1.415 + }
1.416 + }
1.417 + if (!printed)
1.418 + _method->print_short_name(st);
1.419 + st->print(" @ bci:%d",_bci);
1.420 + } else {
1.421 + st->print(" runtime stub");
1.422 + }
1.423 + if (caller() != NULL) caller()->dump_spec(st);
1.424 +}
1.425 +#endif
1.426 +
1.427 +#ifndef PRODUCT
1.428 +void JVMState::dump_on(outputStream* st) const {
1.429 + if (_map && !((uintptr_t)_map & 1)) {
1.430 + if (_map->len() > _map->req()) { // _map->has_exceptions()
1.431 + Node* ex = _map->in(_map->req()); // _map->next_exception()
1.432 + // skip the first one; it's already being printed
1.433 + while (ex != NULL && ex->len() > ex->req()) {
1.434 + ex = ex->in(ex->req()); // ex->next_exception()
1.435 + ex->dump(1);
1.436 + }
1.437 + }
1.438 + _map->dump(2);
1.439 + }
1.440 + st->print("JVMS depth=%d loc=%d stk=%d mon=%d end=%d mondepth=%d sp=%d bci=%d method=",
1.441 + depth(), locoff(), stkoff(), monoff(), endoff(), monitor_depth(), sp(), bci());
1.442 + if (_method == NULL) {
1.443 + st->print_cr("(none)");
1.444 + } else {
1.445 + _method->print_name(st);
1.446 + st->cr();
1.447 + if (bci() >= 0 && bci() < _method->code_size()) {
1.448 + st->print(" bc: ");
1.449 + _method->print_codes_on(bci(), bci()+1, st);
1.450 + }
1.451 + }
1.452 + if (caller() != NULL) {
1.453 + caller()->dump_on(st);
1.454 + }
1.455 +}
1.456 +
1.457 +// Extra way to dump a jvms from the debugger,
1.458 +// to avoid a bug with C++ member function calls.
1.459 +void dump_jvms(JVMState* jvms) {
1.460 + jvms->dump();
1.461 +}
1.462 +#endif
1.463 +
1.464 +//--------------------------clone_shallow--------------------------------------
1.465 +JVMState* JVMState::clone_shallow(Compile* C) const {
1.466 + JVMState* n = has_method() ? new (C) JVMState(_method, _caller) : new (C) JVMState(0);
1.467 + n->set_bci(_bci);
1.468 + n->set_locoff(_locoff);
1.469 + n->set_stkoff(_stkoff);
1.470 + n->set_monoff(_monoff);
1.471 + n->set_endoff(_endoff);
1.472 + n->set_sp(_sp);
1.473 + n->set_map(_map);
1.474 + return n;
1.475 +}
1.476 +
1.477 +//---------------------------clone_deep----------------------------------------
1.478 +JVMState* JVMState::clone_deep(Compile* C) const {
1.479 + JVMState* n = clone_shallow(C);
1.480 + for (JVMState* p = n; p->_caller != NULL; p = p->_caller) {
1.481 + p->_caller = p->_caller->clone_shallow(C);
1.482 + }
1.483 + assert(n->depth() == depth(), "sanity");
1.484 + assert(n->debug_depth() == debug_depth(), "sanity");
1.485 + return n;
1.486 +}
1.487 +
1.488 +//=============================================================================
1.489 +uint CallNode::cmp( const Node &n ) const
1.490 +{ return _tf == ((CallNode&)n)._tf && _jvms == ((CallNode&)n)._jvms; }
1.491 +#ifndef PRODUCT
1.492 +void CallNode::dump_req() const {
1.493 + // Dump the required inputs, enclosed in '(' and ')'
1.494 + uint i; // Exit value of loop
1.495 + for( i=0; i<req(); i++ ) { // For all required inputs
1.496 + if( i == TypeFunc::Parms ) tty->print("(");
1.497 + if( in(i) ) tty->print("%c%d ", Compile::current()->node_arena()->contains(in(i)) ? ' ' : 'o', in(i)->_idx);
1.498 + else tty->print("_ ");
1.499 + }
1.500 + tty->print(")");
1.501 +}
1.502 +
1.503 +void CallNode::dump_spec(outputStream *st) const {
1.504 + st->print(" ");
1.505 + tf()->dump_on(st);
1.506 + if (_cnt != COUNT_UNKNOWN) st->print(" C=%f",_cnt);
1.507 + if (jvms() != NULL) jvms()->dump_spec(st);
1.508 +}
1.509 +#endif
1.510 +
1.511 +const Type *CallNode::bottom_type() const { return tf()->range(); }
1.512 +const Type *CallNode::Value(PhaseTransform *phase) const {
1.513 + if (phase->type(in(0)) == Type::TOP) return Type::TOP;
1.514 + return tf()->range();
1.515 +}
1.516 +
1.517 +//------------------------------calling_convention-----------------------------
1.518 +void CallNode::calling_convention( BasicType* sig_bt, VMRegPair *parm_regs, uint argcnt ) const {
1.519 + // Use the standard compiler calling convention
1.520 + Matcher::calling_convention( sig_bt, parm_regs, argcnt, true );
1.521 +}
1.522 +
1.523 +
1.524 +//------------------------------match------------------------------------------
1.525 +// Construct projections for control, I/O, memory-fields, ..., and
1.526 +// return result(s) along with their RegMask info
1.527 +Node *CallNode::match( const ProjNode *proj, const Matcher *match ) {
1.528 + switch (proj->_con) {
1.529 + case TypeFunc::Control:
1.530 + case TypeFunc::I_O:
1.531 + case TypeFunc::Memory:
1.532 + return new (match->C, 1) MachProjNode(this,proj->_con,RegMask::Empty,MachProjNode::unmatched_proj);
1.533 +
1.534 + case TypeFunc::Parms+1: // For LONG & DOUBLE returns
1.535 + assert(tf()->_range->field_at(TypeFunc::Parms+1) == Type::HALF, "");
1.536 + // 2nd half of doubles and longs
1.537 + return new (match->C, 1) MachProjNode(this,proj->_con, RegMask::Empty, (uint)OptoReg::Bad);
1.538 +
1.539 + case TypeFunc::Parms: { // Normal returns
1.540 + uint ideal_reg = Matcher::base2reg[tf()->range()->field_at(TypeFunc::Parms)->base()];
1.541 + OptoRegPair regs = is_CallRuntime()
1.542 + ? match->c_return_value(ideal_reg,true) // Calls into C runtime
1.543 + : match-> return_value(ideal_reg,true); // Calls into compiled Java code
1.544 + RegMask rm = RegMask(regs.first());
1.545 + if( OptoReg::is_valid(regs.second()) )
1.546 + rm.Insert( regs.second() );
1.547 + return new (match->C, 1) MachProjNode(this,proj->_con,rm,ideal_reg);
1.548 + }
1.549 +
1.550 + case TypeFunc::ReturnAdr:
1.551 + case TypeFunc::FramePtr:
1.552 + default:
1.553 + ShouldNotReachHere();
1.554 + }
1.555 + return NULL;
1.556 +}
1.557 +
1.558 +// Do we Match on this edge index or not? Match no edges
1.559 +uint CallNode::match_edge(uint idx) const {
1.560 + return 0;
1.561 +}
1.562 +
1.563 +//=============================================================================
1.564 +uint CallJavaNode::size_of() const { return sizeof(*this); }
1.565 +uint CallJavaNode::cmp( const Node &n ) const {
1.566 + CallJavaNode &call = (CallJavaNode&)n;
1.567 + return CallNode::cmp(call) && _method == call._method;
1.568 +}
1.569 +#ifndef PRODUCT
1.570 +void CallJavaNode::dump_spec(outputStream *st) const {
1.571 + if( _method ) _method->print_short_name(st);
1.572 + CallNode::dump_spec(st);
1.573 +}
1.574 +#endif
1.575 +
1.576 +//=============================================================================
1.577 +uint CallStaticJavaNode::size_of() const { return sizeof(*this); }
1.578 +uint CallStaticJavaNode::cmp( const Node &n ) const {
1.579 + CallStaticJavaNode &call = (CallStaticJavaNode&)n;
1.580 + return CallJavaNode::cmp(call);
1.581 +}
1.582 +
1.583 +//----------------------------uncommon_trap_request----------------------------
1.584 +// If this is an uncommon trap, return the request code, else zero.
1.585 +int CallStaticJavaNode::uncommon_trap_request() const {
1.586 + if (_name != NULL && !strcmp(_name, "uncommon_trap")) {
1.587 + return extract_uncommon_trap_request(this);
1.588 + }
1.589 + return 0;
1.590 +}
1.591 +int CallStaticJavaNode::extract_uncommon_trap_request(const Node* call) {
1.592 +#ifndef PRODUCT
1.593 + if (!(call->req() > TypeFunc::Parms &&
1.594 + call->in(TypeFunc::Parms) != NULL &&
1.595 + call->in(TypeFunc::Parms)->is_Con())) {
1.596 + assert(_in_dump_cnt != 0, "OK if dumping");
1.597 + tty->print("[bad uncommon trap]");
1.598 + return 0;
1.599 + }
1.600 +#endif
1.601 + return call->in(TypeFunc::Parms)->bottom_type()->is_int()->get_con();
1.602 +}
1.603 +
1.604 +#ifndef PRODUCT
1.605 +void CallStaticJavaNode::dump_spec(outputStream *st) const {
1.606 + st->print("# Static ");
1.607 + if (_name != NULL) {
1.608 + st->print("%s", _name);
1.609 + int trap_req = uncommon_trap_request();
1.610 + if (trap_req != 0) {
1.611 + char buf[100];
1.612 + st->print("(%s)",
1.613 + Deoptimization::format_trap_request(buf, sizeof(buf),
1.614 + trap_req));
1.615 + }
1.616 + st->print(" ");
1.617 + }
1.618 + CallJavaNode::dump_spec(st);
1.619 +}
1.620 +#endif
1.621 +
1.622 +//=============================================================================
1.623 +uint CallDynamicJavaNode::size_of() const { return sizeof(*this); }
1.624 +uint CallDynamicJavaNode::cmp( const Node &n ) const {
1.625 + CallDynamicJavaNode &call = (CallDynamicJavaNode&)n;
1.626 + return CallJavaNode::cmp(call);
1.627 +}
1.628 +#ifndef PRODUCT
1.629 +void CallDynamicJavaNode::dump_spec(outputStream *st) const {
1.630 + st->print("# Dynamic ");
1.631 + CallJavaNode::dump_spec(st);
1.632 +}
1.633 +#endif
1.634 +
1.635 +//=============================================================================
1.636 +uint CallRuntimeNode::size_of() const { return sizeof(*this); }
1.637 +uint CallRuntimeNode::cmp( const Node &n ) const {
1.638 + CallRuntimeNode &call = (CallRuntimeNode&)n;
1.639 + return CallNode::cmp(call) && !strcmp(_name,call._name);
1.640 +}
1.641 +#ifndef PRODUCT
1.642 +void CallRuntimeNode::dump_spec(outputStream *st) const {
1.643 + st->print("# ");
1.644 + st->print(_name);
1.645 + CallNode::dump_spec(st);
1.646 +}
1.647 +#endif
1.648 +
1.649 +//------------------------------calling_convention-----------------------------
1.650 +void CallRuntimeNode::calling_convention( BasicType* sig_bt, VMRegPair *parm_regs, uint argcnt ) const {
1.651 + Matcher::c_calling_convention( sig_bt, parm_regs, argcnt );
1.652 +}
1.653 +
1.654 +//=============================================================================
1.655 +//------------------------------calling_convention-----------------------------
1.656 +
1.657 +
1.658 +//=============================================================================
1.659 +#ifndef PRODUCT
1.660 +void CallLeafNode::dump_spec(outputStream *st) const {
1.661 + st->print("# ");
1.662 + st->print(_name);
1.663 + CallNode::dump_spec(st);
1.664 +}
1.665 +#endif
1.666 +
1.667 +//=============================================================================
1.668 +
1.669 +void SafePointNode::set_local(JVMState* jvms, uint idx, Node *c) {
1.670 + assert(verify_jvms(jvms), "jvms must match");
1.671 + int loc = jvms->locoff() + idx;
1.672 + if (in(loc)->is_top() && idx > 0 && !c->is_top() ) {
1.673 + // If current local idx is top then local idx - 1 could
1.674 + // be a long/double that needs to be killed since top could
1.675 + // represent the 2nd half ofthe long/double.
1.676 + uint ideal = in(loc -1)->ideal_reg();
1.677 + if (ideal == Op_RegD || ideal == Op_RegL) {
1.678 + // set other (low index) half to top
1.679 + set_req(loc - 1, in(loc));
1.680 + }
1.681 + }
1.682 + set_req(loc, c);
1.683 +}
1.684 +
1.685 +uint SafePointNode::size_of() const { return sizeof(*this); }
1.686 +uint SafePointNode::cmp( const Node &n ) const {
1.687 + return (&n == this); // Always fail except on self
1.688 +}
1.689 +
1.690 +//-------------------------set_next_exception----------------------------------
1.691 +void SafePointNode::set_next_exception(SafePointNode* n) {
1.692 + assert(n == NULL || n->Opcode() == Op_SafePoint, "correct value for next_exception");
1.693 + if (len() == req()) {
1.694 + if (n != NULL) add_prec(n);
1.695 + } else {
1.696 + set_prec(req(), n);
1.697 + }
1.698 +}
1.699 +
1.700 +
1.701 +//----------------------------next_exception-----------------------------------
1.702 +SafePointNode* SafePointNode::next_exception() const {
1.703 + if (len() == req()) {
1.704 + return NULL;
1.705 + } else {
1.706 + Node* n = in(req());
1.707 + assert(n == NULL || n->Opcode() == Op_SafePoint, "no other uses of prec edges");
1.708 + return (SafePointNode*) n;
1.709 + }
1.710 +}
1.711 +
1.712 +
1.713 +//------------------------------Ideal------------------------------------------
1.714 +// Skip over any collapsed Regions
1.715 +Node *SafePointNode::Ideal(PhaseGVN *phase, bool can_reshape) {
1.716 + if (remove_dead_region(phase, can_reshape)) return this;
1.717 +
1.718 + return NULL;
1.719 +}
1.720 +
1.721 +//------------------------------Identity---------------------------------------
1.722 +// Remove obviously duplicate safepoints
1.723 +Node *SafePointNode::Identity( PhaseTransform *phase ) {
1.724 +
1.725 + // If you have back to back safepoints, remove one
1.726 + if( in(TypeFunc::Control)->is_SafePoint() )
1.727 + return in(TypeFunc::Control);
1.728 +
1.729 + if( in(0)->is_Proj() ) {
1.730 + Node *n0 = in(0)->in(0);
1.731 + // Check if he is a call projection (except Leaf Call)
1.732 + if( n0->is_Catch() ) {
1.733 + n0 = n0->in(0)->in(0);
1.734 + assert( n0->is_Call(), "expect a call here" );
1.735 + }
1.736 + if( n0->is_Call() && n0->as_Call()->guaranteed_safepoint() ) {
1.737 + // Useless Safepoint, so remove it
1.738 + return in(TypeFunc::Control);
1.739 + }
1.740 + }
1.741 +
1.742 + return this;
1.743 +}
1.744 +
1.745 +//------------------------------Value------------------------------------------
1.746 +const Type *SafePointNode::Value( PhaseTransform *phase ) const {
1.747 + if( phase->type(in(0)) == Type::TOP ) return Type::TOP;
1.748 + if( phase->eqv( in(0), this ) ) return Type::TOP; // Dead infinite loop
1.749 + return Type::CONTROL;
1.750 +}
1.751 +
1.752 +#ifndef PRODUCT
1.753 +void SafePointNode::dump_spec(outputStream *st) const {
1.754 + st->print(" SafePoint ");
1.755 +}
1.756 +#endif
1.757 +
1.758 +const RegMask &SafePointNode::in_RegMask(uint idx) const {
1.759 + if( idx < TypeFunc::Parms ) return RegMask::Empty;
1.760 + // Values outside the domain represent debug info
1.761 + return *(Compile::current()->matcher()->idealreg2debugmask[in(idx)->ideal_reg()]);
1.762 +}
1.763 +const RegMask &SafePointNode::out_RegMask() const {
1.764 + return RegMask::Empty;
1.765 +}
1.766 +
1.767 +
1.768 +void SafePointNode::grow_stack(JVMState* jvms, uint grow_by) {
1.769 + assert((int)grow_by > 0, "sanity");
1.770 + int monoff = jvms->monoff();
1.771 + int endoff = jvms->endoff();
1.772 + assert(endoff == (int)req(), "no other states or debug info after me");
1.773 + Node* top = Compile::current()->top();
1.774 + for (uint i = 0; i < grow_by; i++) {
1.775 + ins_req(monoff, top);
1.776 + }
1.777 + jvms->set_monoff(monoff + grow_by);
1.778 + jvms->set_endoff(endoff + grow_by);
1.779 +}
1.780 +
1.781 +void SafePointNode::push_monitor(const FastLockNode *lock) {
1.782 + // Add a LockNode, which points to both the original BoxLockNode (the
1.783 + // stack space for the monitor) and the Object being locked.
1.784 + const int MonitorEdges = 2;
1.785 + assert(JVMState::logMonitorEdges == exact_log2(MonitorEdges), "correct MonitorEdges");
1.786 + assert(req() == jvms()->endoff(), "correct sizing");
1.787 + if (GenerateSynchronizationCode) {
1.788 + add_req(lock->box_node());
1.789 + add_req(lock->obj_node());
1.790 + } else {
1.791 + add_req(NULL);
1.792 + add_req(NULL);
1.793 + }
1.794 + jvms()->set_endoff(req());
1.795 +}
1.796 +
1.797 +void SafePointNode::pop_monitor() {
1.798 + // Delete last monitor from debug info
1.799 + debug_only(int num_before_pop = jvms()->nof_monitors());
1.800 + const int MonitorEdges = (1<<JVMState::logMonitorEdges);
1.801 + int endoff = jvms()->endoff();
1.802 + int new_endoff = endoff - MonitorEdges;
1.803 + jvms()->set_endoff(new_endoff);
1.804 + while (endoff > new_endoff) del_req(--endoff);
1.805 + assert(jvms()->nof_monitors() == num_before_pop-1, "");
1.806 +}
1.807 +
1.808 +Node *SafePointNode::peek_monitor_box() const {
1.809 + int mon = jvms()->nof_monitors() - 1;
1.810 + assert(mon >= 0, "most have a monitor");
1.811 + return monitor_box(jvms(), mon);
1.812 +}
1.813 +
1.814 +Node *SafePointNode::peek_monitor_obj() const {
1.815 + int mon = jvms()->nof_monitors() - 1;
1.816 + assert(mon >= 0, "most have a monitor");
1.817 + return monitor_obj(jvms(), mon);
1.818 +}
1.819 +
1.820 +// Do we Match on this edge index or not? Match no edges
1.821 +uint SafePointNode::match_edge(uint idx) const {
1.822 + if( !needs_polling_address_input() )
1.823 + return 0;
1.824 +
1.825 + return (TypeFunc::Parms == idx);
1.826 +}
1.827 +
1.828 +//=============================================================================
1.829 +uint AllocateNode::size_of() const { return sizeof(*this); }
1.830 +
1.831 +AllocateNode::AllocateNode(Compile* C, const TypeFunc *atype,
1.832 + Node *ctrl, Node *mem, Node *abio,
1.833 + Node *size, Node *klass_node, Node *initial_test)
1.834 + : CallNode(atype, NULL, TypeRawPtr::BOTTOM)
1.835 +{
1.836 + init_class_id(Class_Allocate);
1.837 + init_flags(Flag_is_macro);
1.838 + Node *topnode = C->top();
1.839 +
1.840 + init_req( TypeFunc::Control , ctrl );
1.841 + init_req( TypeFunc::I_O , abio );
1.842 + init_req( TypeFunc::Memory , mem );
1.843 + init_req( TypeFunc::ReturnAdr, topnode );
1.844 + init_req( TypeFunc::FramePtr , topnode );
1.845 + init_req( AllocSize , size);
1.846 + init_req( KlassNode , klass_node);
1.847 + init_req( InitialTest , initial_test);
1.848 + init_req( ALength , topnode);
1.849 + C->add_macro_node(this);
1.850 +}
1.851 +
1.852 +//=============================================================================
1.853 +uint AllocateArrayNode::size_of() const { return sizeof(*this); }
1.854 +
1.855 +//=============================================================================
1.856 +uint LockNode::size_of() const { return sizeof(*this); }
1.857 +
1.858 +// Redundant lock elimination
1.859 +//
1.860 +// There are various patterns of locking where we release and
1.861 +// immediately reacquire a lock in a piece of code where no operations
1.862 +// occur in between that would be observable. In those cases we can
1.863 +// skip releasing and reacquiring the lock without violating any
1.864 +// fairness requirements. Doing this around a loop could cause a lock
1.865 +// to be held for a very long time so we concentrate on non-looping
1.866 +// control flow. We also require that the operations are fully
1.867 +// redundant meaning that we don't introduce new lock operations on
1.868 +// some paths so to be able to eliminate it on others ala PRE. This
1.869 +// would probably require some more extensive graph manipulation to
1.870 +// guarantee that the memory edges were all handled correctly.
1.871 +//
1.872 +// Assuming p is a simple predicate which can't trap in any way and s
1.873 +// is a synchronized method consider this code:
1.874 +//
1.875 +// s();
1.876 +// if (p)
1.877 +// s();
1.878 +// else
1.879 +// s();
1.880 +// s();
1.881 +//
1.882 +// 1. The unlocks of the first call to s can be eliminated if the
1.883 +// locks inside the then and else branches are eliminated.
1.884 +//
1.885 +// 2. The unlocks of the then and else branches can be eliminated if
1.886 +// the lock of the final call to s is eliminated.
1.887 +//
1.888 +// Either of these cases subsumes the simple case of sequential control flow
1.889 +//
1.890 +// Addtionally we can eliminate versions without the else case:
1.891 +//
1.892 +// s();
1.893 +// if (p)
1.894 +// s();
1.895 +// s();
1.896 +//
1.897 +// 3. In this case we eliminate the unlock of the first s, the lock
1.898 +// and unlock in the then case and the lock in the final s.
1.899 +//
1.900 +// Note also that in all these cases the then/else pieces don't have
1.901 +// to be trivial as long as they begin and end with synchronization
1.902 +// operations.
1.903 +//
1.904 +// s();
1.905 +// if (p)
1.906 +// s();
1.907 +// f();
1.908 +// s();
1.909 +// s();
1.910 +//
1.911 +// The code will work properly for this case, leaving in the unlock
1.912 +// before the call to f and the relock after it.
1.913 +//
1.914 +// A potentially interesting case which isn't handled here is when the
1.915 +// locking is partially redundant.
1.916 +//
1.917 +// s();
1.918 +// if (p)
1.919 +// s();
1.920 +//
1.921 +// This could be eliminated putting unlocking on the else case and
1.922 +// eliminating the first unlock and the lock in the then side.
1.923 +// Alternatively the unlock could be moved out of the then side so it
1.924 +// was after the merge and the first unlock and second lock
1.925 +// eliminated. This might require less manipulation of the memory
1.926 +// state to get correct.
1.927 +//
1.928 +// Additionally we might allow work between a unlock and lock before
1.929 +// giving up eliminating the locks. The current code disallows any
1.930 +// conditional control flow between these operations. A formulation
1.931 +// similar to partial redundancy elimination computing the
1.932 +// availability of unlocking and the anticipatability of locking at a
1.933 +// program point would allow detection of fully redundant locking with
1.934 +// some amount of work in between. I'm not sure how often I really
1.935 +// think that would occur though. Most of the cases I've seen
1.936 +// indicate it's likely non-trivial work would occur in between.
1.937 +// There may be other more complicated constructs where we could
1.938 +// eliminate locking but I haven't seen any others appear as hot or
1.939 +// interesting.
1.940 +//
1.941 +// Locking and unlocking have a canonical form in ideal that looks
1.942 +// roughly like this:
1.943 +//
1.944 +// <obj>
1.945 +// | \\------+
1.946 +// | \ \
1.947 +// | BoxLock \
1.948 +// | | | \
1.949 +// | | \ \
1.950 +// | | FastLock
1.951 +// | | /
1.952 +// | | /
1.953 +// | | |
1.954 +//
1.955 +// Lock
1.956 +// |
1.957 +// Proj #0
1.958 +// |
1.959 +// MembarAcquire
1.960 +// |
1.961 +// Proj #0
1.962 +//
1.963 +// MembarRelease
1.964 +// |
1.965 +// Proj #0
1.966 +// |
1.967 +// Unlock
1.968 +// |
1.969 +// Proj #0
1.970 +//
1.971 +//
1.972 +// This code proceeds by processing Lock nodes during PhaseIterGVN
1.973 +// and searching back through its control for the proper code
1.974 +// patterns. Once it finds a set of lock and unlock operations to
1.975 +// eliminate they are marked as eliminatable which causes the
1.976 +// expansion of the Lock and Unlock macro nodes to make the operation a NOP
1.977 +//
1.978 +//=============================================================================
1.979 +
1.980 +//
1.981 +// Utility function to skip over uninteresting control nodes. Nodes skipped are:
1.982 +// - copy regions. (These may not have been optimized away yet.)
1.983 +// - eliminated locking nodes
1.984 +//
1.985 +static Node *next_control(Node *ctrl) {
1.986 + if (ctrl == NULL)
1.987 + return NULL;
1.988 + while (1) {
1.989 + if (ctrl->is_Region()) {
1.990 + RegionNode *r = ctrl->as_Region();
1.991 + Node *n = r->is_copy();
1.992 + if (n == NULL)
1.993 + break; // hit a region, return it
1.994 + else
1.995 + ctrl = n;
1.996 + } else if (ctrl->is_Proj()) {
1.997 + Node *in0 = ctrl->in(0);
1.998 + if (in0->is_AbstractLock() && in0->as_AbstractLock()->is_eliminated()) {
1.999 + ctrl = in0->in(0);
1.1000 + } else {
1.1001 + break;
1.1002 + }
1.1003 + } else {
1.1004 + break; // found an interesting control
1.1005 + }
1.1006 + }
1.1007 + return ctrl;
1.1008 +}
1.1009 +//
1.1010 +// Given a control, see if it's the control projection of an Unlock which
1.1011 +// operating on the same object as lock.
1.1012 +//
1.1013 +bool AbstractLockNode::find_matching_unlock(const Node* ctrl, LockNode* lock,
1.1014 + GrowableArray<AbstractLockNode*> &lock_ops) {
1.1015 + ProjNode *ctrl_proj = (ctrl->is_Proj()) ? ctrl->as_Proj() : NULL;
1.1016 + if (ctrl_proj != NULL && ctrl_proj->_con == TypeFunc::Control) {
1.1017 + Node *n = ctrl_proj->in(0);
1.1018 + if (n != NULL && n->is_Unlock()) {
1.1019 + UnlockNode *unlock = n->as_Unlock();
1.1020 + if ((lock->obj_node() == unlock->obj_node()) &&
1.1021 + (lock->box_node() == unlock->box_node()) && !unlock->is_eliminated()) {
1.1022 + lock_ops.append(unlock);
1.1023 + return true;
1.1024 + }
1.1025 + }
1.1026 + }
1.1027 + return false;
1.1028 +}
1.1029 +
1.1030 +//
1.1031 +// Find the lock matching an unlock. Returns null if a safepoint
1.1032 +// or complicated control is encountered first.
1.1033 +LockNode *AbstractLockNode::find_matching_lock(UnlockNode* unlock) {
1.1034 + LockNode *lock_result = NULL;
1.1035 + // find the matching lock, or an intervening safepoint
1.1036 + Node *ctrl = next_control(unlock->in(0));
1.1037 + while (1) {
1.1038 + assert(ctrl != NULL, "invalid control graph");
1.1039 + assert(!ctrl->is_Start(), "missing lock for unlock");
1.1040 + if (ctrl->is_top()) break; // dead control path
1.1041 + if (ctrl->is_Proj()) ctrl = ctrl->in(0);
1.1042 + if (ctrl->is_SafePoint()) {
1.1043 + break; // found a safepoint (may be the lock we are searching for)
1.1044 + } else if (ctrl->is_Region()) {
1.1045 + // Check for a simple diamond pattern. Punt on anything more complicated
1.1046 + if (ctrl->req() == 3 && ctrl->in(1) != NULL && ctrl->in(2) != NULL) {
1.1047 + Node *in1 = next_control(ctrl->in(1));
1.1048 + Node *in2 = next_control(ctrl->in(2));
1.1049 + if (((in1->is_IfTrue() && in2->is_IfFalse()) ||
1.1050 + (in2->is_IfTrue() && in1->is_IfFalse())) && (in1->in(0) == in2->in(0))) {
1.1051 + ctrl = next_control(in1->in(0)->in(0));
1.1052 + } else {
1.1053 + break;
1.1054 + }
1.1055 + } else {
1.1056 + break;
1.1057 + }
1.1058 + } else {
1.1059 + ctrl = next_control(ctrl->in(0)); // keep searching
1.1060 + }
1.1061 + }
1.1062 + if (ctrl->is_Lock()) {
1.1063 + LockNode *lock = ctrl->as_Lock();
1.1064 + if ((lock->obj_node() == unlock->obj_node()) &&
1.1065 + (lock->box_node() == unlock->box_node())) {
1.1066 + lock_result = lock;
1.1067 + }
1.1068 + }
1.1069 + return lock_result;
1.1070 +}
1.1071 +
1.1072 +// This code corresponds to case 3 above.
1.1073 +
1.1074 +bool AbstractLockNode::find_lock_and_unlock_through_if(Node* node, LockNode* lock,
1.1075 + GrowableArray<AbstractLockNode*> &lock_ops) {
1.1076 + Node* if_node = node->in(0);
1.1077 + bool if_true = node->is_IfTrue();
1.1078 +
1.1079 + if (if_node->is_If() && if_node->outcnt() == 2 && (if_true || node->is_IfFalse())) {
1.1080 + Node *lock_ctrl = next_control(if_node->in(0));
1.1081 + if (find_matching_unlock(lock_ctrl, lock, lock_ops)) {
1.1082 + Node* lock1_node = NULL;
1.1083 + ProjNode* proj = if_node->as_If()->proj_out(!if_true);
1.1084 + if (if_true) {
1.1085 + if (proj->is_IfFalse() && proj->outcnt() == 1) {
1.1086 + lock1_node = proj->unique_out();
1.1087 + }
1.1088 + } else {
1.1089 + if (proj->is_IfTrue() && proj->outcnt() == 1) {
1.1090 + lock1_node = proj->unique_out();
1.1091 + }
1.1092 + }
1.1093 + if (lock1_node != NULL && lock1_node->is_Lock()) {
1.1094 + LockNode *lock1 = lock1_node->as_Lock();
1.1095 + if ((lock->obj_node() == lock1->obj_node()) &&
1.1096 + (lock->box_node() == lock1->box_node()) && !lock1->is_eliminated()) {
1.1097 + lock_ops.append(lock1);
1.1098 + return true;
1.1099 + }
1.1100 + }
1.1101 + }
1.1102 + }
1.1103 +
1.1104 + lock_ops.trunc_to(0);
1.1105 + return false;
1.1106 +}
1.1107 +
1.1108 +bool AbstractLockNode::find_unlocks_for_region(const RegionNode* region, LockNode* lock,
1.1109 + GrowableArray<AbstractLockNode*> &lock_ops) {
1.1110 + // check each control merging at this point for a matching unlock.
1.1111 + // in(0) should be self edge so skip it.
1.1112 + for (int i = 1; i < (int)region->req(); i++) {
1.1113 + Node *in_node = next_control(region->in(i));
1.1114 + if (in_node != NULL) {
1.1115 + if (find_matching_unlock(in_node, lock, lock_ops)) {
1.1116 + // found a match so keep on checking.
1.1117 + continue;
1.1118 + } else if (find_lock_and_unlock_through_if(in_node, lock, lock_ops)) {
1.1119 + continue;
1.1120 + }
1.1121 +
1.1122 + // If we fall through to here then it was some kind of node we
1.1123 + // don't understand or there wasn't a matching unlock, so give
1.1124 + // up trying to merge locks.
1.1125 + lock_ops.trunc_to(0);
1.1126 + return false;
1.1127 + }
1.1128 + }
1.1129 + return true;
1.1130 +
1.1131 +}
1.1132 +
1.1133 +#ifndef PRODUCT
1.1134 +//
1.1135 +// Create a counter which counts the number of times this lock is acquired
1.1136 +//
1.1137 +void AbstractLockNode::create_lock_counter(JVMState* state) {
1.1138 + _counter = OptoRuntime::new_named_counter(state, NamedCounter::LockCounter);
1.1139 +}
1.1140 +#endif
1.1141 +
1.1142 +void AbstractLockNode::set_eliminated() {
1.1143 + _eliminate = true;
1.1144 +#ifndef PRODUCT
1.1145 + if (_counter) {
1.1146 + // Update the counter to indicate that this lock was eliminated.
1.1147 + // The counter update code will stay around even though the
1.1148 + // optimizer will eliminate the lock operation itself.
1.1149 + _counter->set_tag(NamedCounter::EliminatedLockCounter);
1.1150 + }
1.1151 +#endif
1.1152 +}
1.1153 +
1.1154 +//=============================================================================
1.1155 +Node *LockNode::Ideal(PhaseGVN *phase, bool can_reshape) {
1.1156 +
1.1157 + // perform any generic optimizations first
1.1158 + Node *result = SafePointNode::Ideal(phase, can_reshape);
1.1159 +
1.1160 + // Now see if we can optimize away this lock. We don't actually
1.1161 + // remove the locking here, we simply set the _eliminate flag which
1.1162 + // prevents macro expansion from expanding the lock. Since we don't
1.1163 + // modify the graph, the value returned from this function is the
1.1164 + // one computed above.
1.1165 + if (EliminateLocks && !is_eliminated()) {
1.1166 + //
1.1167 + // Try lock coarsening
1.1168 + //
1.1169 + PhaseIterGVN* iter = phase->is_IterGVN();
1.1170 + if (iter != NULL) {
1.1171 +
1.1172 + GrowableArray<AbstractLockNode*> lock_ops;
1.1173 +
1.1174 + Node *ctrl = next_control(in(0));
1.1175 +
1.1176 + // now search back for a matching Unlock
1.1177 + if (find_matching_unlock(ctrl, this, lock_ops)) {
1.1178 + // found an unlock directly preceding this lock. This is the
1.1179 + // case of single unlock directly control dependent on a
1.1180 + // single lock which is the trivial version of case 1 or 2.
1.1181 + } else if (ctrl->is_Region() ) {
1.1182 + if (find_unlocks_for_region(ctrl->as_Region(), this, lock_ops)) {
1.1183 + // found lock preceded by multiple unlocks along all paths
1.1184 + // joining at this point which is case 3 in description above.
1.1185 + }
1.1186 + } else {
1.1187 + // see if this lock comes from either half of an if and the
1.1188 + // predecessors merges unlocks and the other half of the if
1.1189 + // performs a lock.
1.1190 + if (find_lock_and_unlock_through_if(ctrl, this, lock_ops)) {
1.1191 + // found unlock splitting to an if with locks on both branches.
1.1192 + }
1.1193 + }
1.1194 +
1.1195 + if (lock_ops.length() > 0) {
1.1196 + // add ourselves to the list of locks to be eliminated.
1.1197 + lock_ops.append(this);
1.1198 +
1.1199 + #ifndef PRODUCT
1.1200 + if (PrintEliminateLocks) {
1.1201 + int locks = 0;
1.1202 + int unlocks = 0;
1.1203 + for (int i = 0; i < lock_ops.length(); i++) {
1.1204 + AbstractLockNode* lock = lock_ops.at(i);
1.1205 + if (lock->Opcode() == Op_Lock) locks++;
1.1206 + else unlocks++;
1.1207 + if (Verbose) {
1.1208 + lock->dump(1);
1.1209 + }
1.1210 + }
1.1211 + tty->print_cr("***Eliminated %d unlocks and %d locks", unlocks, locks);
1.1212 + }
1.1213 + #endif
1.1214 +
1.1215 + // for each of the identified locks, mark them
1.1216 + // as eliminatable
1.1217 + for (int i = 0; i < lock_ops.length(); i++) {
1.1218 + AbstractLockNode* lock = lock_ops.at(i);
1.1219 +
1.1220 + // Mark it eliminated to update any counters
1.1221 + lock->set_eliminated();
1.1222 + }
1.1223 + } else if (result != NULL && ctrl->is_Region() &&
1.1224 + iter->_worklist.member(ctrl)) {
1.1225 + // We weren't able to find any opportunities but the region this
1.1226 + // lock is control dependent on hasn't been processed yet so put
1.1227 + // this lock back on the worklist so we can check again once any
1.1228 + // region simplification has occurred.
1.1229 + iter->_worklist.push(this);
1.1230 + }
1.1231 + }
1.1232 + }
1.1233 +
1.1234 + return result;
1.1235 +}
1.1236 +
1.1237 +//=============================================================================
1.1238 +uint UnlockNode::size_of() const { return sizeof(*this); }
1.1239 +
1.1240 +//=============================================================================
1.1241 +Node *UnlockNode::Ideal(PhaseGVN *phase, bool can_reshape) {
1.1242 +
1.1243 + // perform any generic optimizations first
1.1244 + Node * result = SafePointNode::Ideal(phase, can_reshape);
1.1245 +
1.1246 + // Now see if we can optimize away this unlock. We don't actually
1.1247 + // remove the unlocking here, we simply set the _eliminate flag which
1.1248 + // prevents macro expansion from expanding the unlock. Since we don't
1.1249 + // modify the graph, the value returned from this function is the
1.1250 + // one computed above.
1.1251 + if (EliminateLocks && !is_eliminated()) {
1.1252 + //
1.1253 + // If we are unlocking an unescaped object, the lock/unlock is unnecessary
1.1254 + // We can eliminate them if there are no safepoints in the locked region.
1.1255 + //
1.1256 + ConnectionGraph *cgr = Compile::current()->congraph();
1.1257 + if (cgr != NULL && cgr->escape_state(obj_node(), phase) == PointsToNode::NoEscape) {
1.1258 + GrowableArray<AbstractLockNode*> lock_ops;
1.1259 + LockNode *lock = find_matching_lock(this);
1.1260 + if (lock != NULL) {
1.1261 + lock_ops.append(this);
1.1262 + lock_ops.append(lock);
1.1263 + // find other unlocks which pair with the lock we found and add them
1.1264 + // to the list
1.1265 + Node * box = box_node();
1.1266 +
1.1267 + for (DUIterator_Fast imax, i = box->fast_outs(imax); i < imax; i++) {
1.1268 + Node *use = box->fast_out(i);
1.1269 + if (use->is_Unlock() && use != this) {
1.1270 + UnlockNode *unlock1 = use->as_Unlock();
1.1271 + if (!unlock1->is_eliminated()) {
1.1272 + LockNode *lock1 = find_matching_lock(unlock1);
1.1273 + if (lock == lock1)
1.1274 + lock_ops.append(unlock1);
1.1275 + else if (lock1 == NULL) {
1.1276 + // we can't find a matching lock, we must assume the worst
1.1277 + lock_ops.trunc_to(0);
1.1278 + break;
1.1279 + }
1.1280 + }
1.1281 + }
1.1282 + }
1.1283 + if (lock_ops.length() > 0) {
1.1284 +
1.1285 + #ifndef PRODUCT
1.1286 + if (PrintEliminateLocks) {
1.1287 + int locks = 0;
1.1288 + int unlocks = 0;
1.1289 + for (int i = 0; i < lock_ops.length(); i++) {
1.1290 + AbstractLockNode* lock = lock_ops.at(i);
1.1291 + if (lock->Opcode() == Op_Lock) locks++;
1.1292 + else unlocks++;
1.1293 + if (Verbose) {
1.1294 + lock->dump(1);
1.1295 + }
1.1296 + }
1.1297 + tty->print_cr("***Eliminated %d unescaped unlocks and %d unescaped locks", unlocks, locks);
1.1298 + }
1.1299 + #endif
1.1300 +
1.1301 + // for each of the identified locks, mark them
1.1302 + // as eliminatable
1.1303 + for (int i = 0; i < lock_ops.length(); i++) {
1.1304 + AbstractLockNode* lock = lock_ops.at(i);
1.1305 +
1.1306 + // Mark it eliminated to update any counters
1.1307 + lock->set_eliminated();
1.1308 + }
1.1309 + }
1.1310 + }
1.1311 + }
1.1312 + }
1.1313 + return result;
1.1314 +}
