Google Git
Sign in
android / platform / external / jetbrains / jdk8u_hotspot / a482fc01a461b60a024295f544eaa063285fee2d / . / src / share / vm / opto / callnode.cpp
blob: 0320e312d784e044a4be8863eab563e4231450d0 [file] [log] [blame]
/*
* Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*
*/
#include "precompiled.hpp"
#include "compiler/compileLog.hpp"
#include "ci/bcEscapeAnalyzer.hpp"
#include "compiler/oopMap.hpp"
#include "opto/callGenerator.hpp"
#include "opto/callnode.hpp"
#include "opto/escape.hpp"
#include "opto/locknode.hpp"
#include "opto/machnode.hpp"
#include "opto/matcher.hpp"
#include "opto/parse.hpp"
#include "opto/regalloc.hpp"
#include "opto/regmask.hpp"
#include "opto/rootnode.hpp"
#include "opto/runtime.hpp"
// Portions of code courtesy of Clifford Click
// Optimization - Graph Style
//=============================================================================
uint StartNode::size_of() const { return sizeof(*this); }
uint StartNode::cmp( const Node &n ) const
{ return _domain == ((StartNode&)n)._domain; }
const Type *StartNode::bottom_type() const { return _domain; }
const Type *StartNode::Value(PhaseTransform *phase) const { return _domain; }
#ifndef PRODUCT
void StartNode::dump_spec(outputStream *st) const { st->print(" #"); _domain->dump_on(st);}
#endif
//------------------------------Ideal------------------------------------------
Node *StartNode::Ideal(PhaseGVN *phase, bool can_reshape){
return remove_dead_region(phase, can_reshape) ? this : NULL;
}
//------------------------------calling_convention-----------------------------
void StartNode::calling_convention( BasicType* sig_bt, VMRegPair *parm_regs, uint argcnt ) const {
Matcher::calling_convention( sig_bt, parm_regs, argcnt, false );
}
//------------------------------Registers--------------------------------------
const RegMask &StartNode::in_RegMask(uint) const {
return RegMask::Empty;
}
//------------------------------match------------------------------------------
// Construct projections for incoming parameters, and their RegMask info
Node *StartNode::match( const ProjNode *proj, const Matcher *match ) {
switch (proj->_con) {
case TypeFunc::Control:
case TypeFunc::I_O:
case TypeFunc::Memory:
return new (match->C) MachProjNode(this,proj->_con,RegMask::Empty,MachProjNode::unmatched_proj);
case TypeFunc::FramePtr:
return new (match->C) MachProjNode(this,proj->_con,Matcher::c_frame_ptr_mask, Op_RegP);
case TypeFunc::ReturnAdr:
return new (match->C) MachProjNode(this,proj->_con,match->_return_addr_mask,Op_RegP);
case TypeFunc::Parms:
default: {
uint parm_num = proj->_con - TypeFunc::Parms;
const Type *t = _domain->field_at(proj->_con);
if (t->base() == Type::Half) // 2nd half of Longs and Doubles
return new (match->C) ConNode(Type::TOP);
uint ideal_reg = t->ideal_reg();
RegMask &rm = match->_calling_convention_mask[parm_num];
return new (match->C) MachProjNode(this,proj->_con,rm,ideal_reg);
}
}
return NULL;
}
//------------------------------StartOSRNode----------------------------------
// The method start node for an on stack replacement adapter
//------------------------------osr_domain-----------------------------
const TypeTuple *StartOSRNode::osr_domain() {
const Type **fields = TypeTuple::fields(2);
fields[TypeFunc::Parms+0] = TypeRawPtr::BOTTOM; // address of osr buffer
return TypeTuple::make(TypeFunc::Parms+1, fields);
}
//=============================================================================
const char * const ParmNode::names[TypeFunc::Parms+1] = {
"Control", "I_O", "Memory", "FramePtr", "ReturnAdr", "Parms"
};
#ifndef PRODUCT
void ParmNode::dump_spec(outputStream *st) const {
if( _con < TypeFunc::Parms ) {
st->print("%s", names[_con]);
} else {
st->print("Parm%d: ",_con-TypeFunc::Parms);
// Verbose and WizardMode dump bottom_type for all nodes
if( !Verbose && !WizardMode ) bottom_type()->dump_on(st);
}
}
#endif
uint ParmNode::ideal_reg() const {
switch( _con ) {
case TypeFunc::Control : // fall through
case TypeFunc::I_O : // fall through
case TypeFunc::Memory : return 0;
case TypeFunc::FramePtr : // fall through
case TypeFunc::ReturnAdr: return Op_RegP;
default : assert( _con > TypeFunc::Parms, "" );
// fall through
case TypeFunc::Parms : {
// Type of argument being passed
const Type *t = in(0)->as_Start()->_domain->field_at(_con);
return t->ideal_reg();
}
}
ShouldNotReachHere();
return 0;
}
//=============================================================================
ReturnNode::ReturnNode(uint edges, Node *cntrl, Node *i_o, Node *memory, Node *frameptr, Node *retadr ) : Node(edges) {
init_req(TypeFunc::Control,cntrl);
init_req(TypeFunc::I_O,i_o);
init_req(TypeFunc::Memory,memory);
init_req(TypeFunc::FramePtr,frameptr);
init_req(TypeFunc::ReturnAdr,retadr);
}
Node *ReturnNode::Ideal(PhaseGVN *phase, bool can_reshape){
return remove_dead_region(phase, can_reshape) ? this : NULL;
}
const Type *ReturnNode::Value( PhaseTransform *phase ) const {
return ( phase->type(in(TypeFunc::Control)) == Type::TOP)
? Type::TOP
: Type::BOTTOM;
}
// Do we Match on this edge index or not? No edges on return nodes
uint ReturnNode::match_edge(uint idx) const {
return 0;
}
#ifndef PRODUCT
void ReturnNode::dump_req(outputStream *st) const {
// Dump the required inputs, enclosed in '(' and ')'
uint i; // Exit value of loop
for (i = 0; i < req(); i++) { // For all required inputs
if (i == TypeFunc::Parms) st->print("returns");
if (in(i)) st->print("%c%d ", Compile::current()->node_arena()->contains(in(i)) ? ' ' : 'o', in(i)->_idx);
else st->print("_ ");
}
}
#endif
//=============================================================================
RethrowNode::RethrowNode(
Node* cntrl,
Node* i_o,
Node* memory,
Node* frameptr,
Node* ret_adr,
Node* exception
) : Node(TypeFunc::Parms + 1) {
init_req(TypeFunc::Control , cntrl );
init_req(TypeFunc::I_O , i_o );
init_req(TypeFunc::Memory , memory );
init_req(TypeFunc::FramePtr , frameptr );
init_req(TypeFunc::ReturnAdr, ret_adr);
init_req(TypeFunc::Parms , exception);
}
Node *RethrowNode::Ideal(PhaseGVN *phase, bool can_reshape){
return remove_dead_region(phase, can_reshape) ? this : NULL;
}
const Type *RethrowNode::Value( PhaseTransform *phase ) const {
return (phase->type(in(TypeFunc::Control)) == Type::TOP)
? Type::TOP
: Type::BOTTOM;
}
uint RethrowNode::match_edge(uint idx) const {
return 0;
}
#ifndef PRODUCT
void RethrowNode::dump_req(outputStream *st) const {
// Dump the required inputs, enclosed in '(' and ')'
uint i; // Exit value of loop
for (i = 0; i < req(); i++) { // For all required inputs
if (i == TypeFunc::Parms) st->print("exception");
if (in(i)) st->print("%c%d ", Compile::current()->node_arena()->contains(in(i)) ? ' ' : 'o', in(i)->_idx);
else st->print("_ ");
}
}
#endif
//=============================================================================
// Do we Match on this edge index or not? Match only target address & method
uint TailCallNode::match_edge(uint idx) const {
return TypeFunc::Parms <= idx && idx <= TypeFunc::Parms+1;
}
//=============================================================================
// Do we Match on this edge index or not? Match only target address & oop
uint TailJumpNode::match_edge(uint idx) const {
return TypeFunc::Parms <= idx && idx <= TypeFunc::Parms+1;
}
//=============================================================================
JVMState::JVMState(ciMethod* method, JVMState* caller) :
_method(method) {
assert(method != NULL, "must be valid call site");
_reexecute = Reexecute_Undefined;
debug_only(_bci = -99); // random garbage value
debug_only(_map = (SafePointNode*)-1);
_caller = caller;
_depth = 1 + (caller == NULL ? 0 : caller->depth());
_locoff = TypeFunc::Parms;
_stkoff = _locoff + _method->max_locals();
_monoff = _stkoff + _method->max_stack();
_scloff = _monoff;
_endoff = _monoff;
_sp = 0;
}
JVMState::JVMState(int stack_size) :
_method(NULL) {
_bci = InvocationEntryBci;
_reexecute = Reexecute_Undefined;
debug_only(_map = (SafePointNode*)-1);
_caller = NULL;
_depth = 1;
_locoff = TypeFunc::Parms;
_stkoff = _locoff;
_monoff = _stkoff + stack_size;
_scloff = _monoff;
_endoff = _monoff;
_sp = 0;
}
//--------------------------------of_depth-------------------------------------
JVMState* JVMState::of_depth(int d) const {
const JVMState* jvmp = this;
assert(0 < d && (uint)d <= depth(), "oob");
for (int skip = depth() - d; skip > 0; skip--) {
jvmp = jvmp->caller();
}
assert(jvmp->depth() == (uint)d, "found the right one");
return (JVMState*)jvmp;
}
//-----------------------------same_calls_as-----------------------------------
bool JVMState::same_calls_as(const JVMState* that) const {
if (this == that) return true;
if (this->depth() != that->depth()) return false;
const JVMState* p = this;
const JVMState* q = that;
for (;;) {
if (p->_method != q->_method) return false;
if (p->_method == NULL) return true; // bci is irrelevant
if (p->_bci != q->_bci) return false;
if (p->_reexecute != q->_reexecute) return false;
p = p->caller();
q = q->caller();
if (p == q) return true;
assert(p != NULL && q != NULL, "depth check ensures we don't run off end");
}
}
//------------------------------debug_start------------------------------------
uint JVMState::debug_start() const {
debug_only(JVMState* jvmroot = of_depth(1));
assert(jvmroot->locoff() <= this->locoff(), "youngest JVMState must be last");
return of_depth(1)->locoff();
}
//-------------------------------debug_end-------------------------------------
uint JVMState::debug_end() const {
debug_only(JVMState* jvmroot = of_depth(1));
assert(jvmroot->endoff() <= this->endoff(), "youngest JVMState must be last");
return endoff();
}
//------------------------------debug_depth------------------------------------
uint JVMState::debug_depth() const {
uint total = 0;
for (const JVMState* jvmp = this; jvmp != NULL; jvmp = jvmp->caller()) {
total += jvmp->debug_size();
}
return total;
}
#ifndef PRODUCT
//------------------------------format_helper----------------------------------
// Given an allocation (a Chaitin object) and a Node decide if the Node carries
// any defined value or not. If it does, print out the register or constant.
static void format_helper( PhaseRegAlloc *regalloc, outputStream* st, Node *n, const char *msg, uint i, GrowableArray<SafePointScalarObjectNode*> *scobjs ) {
if (n == NULL) { st->print(" NULL"); return; }
if (n->is_SafePointScalarObject()) {
// Scalar replacement.
SafePointScalarObjectNode* spobj = n->as_SafePointScalarObject();
scobjs->append_if_missing(spobj);
int sco_n = scobjs->find(spobj);
assert(sco_n >= 0, "");
st->print(" %s%d]=#ScObj" INT32_FORMAT, msg, i, sco_n);
return;
}
if (regalloc->node_regs_max_index() > 0 &&
OptoReg::is_valid(regalloc->get_reg_first(n))) { // Check for undefined
char buf[50];
regalloc->dump_register(n,buf);
st->print(" %s%d]=%s",msg,i,buf);
} else { // No register, but might be constant
const Type *t = n->bottom_type();
switch (t->base()) {
case Type::Int:
st->print(" %s%d]=#"INT32_FORMAT,msg,i,t->is_int()->get_con());
break;
case Type::AnyPtr:
assert( t == TypePtr::NULL_PTR || n->in_dump(), "" );
st->print(" %s%d]=#NULL",msg,i);
break;
case Type::AryPtr:
case Type::InstPtr:
st->print(" %s%d]=#Ptr" INTPTR_FORMAT,msg,i,p2i(t->isa_oopptr()->const_oop()));
break;
case Type::KlassPtr:
st->print(" %s%d]=#Ptr" INTPTR_FORMAT,msg,i,p2i(t->make_ptr()->isa_klassptr()->klass()));
break;
case Type::MetadataPtr:
st->print(" %s%d]=#Ptr" INTPTR_FORMAT,msg,i,p2i(t->make_ptr()->isa_metadataptr()->metadata()));
break;
case Type::NarrowOop:
st->print(" %s%d]=#Ptr" INTPTR_FORMAT,msg,i,p2i(t->make_ptr()->isa_oopptr()->const_oop()));
break;
case Type::RawPtr:
st->print(" %s%d]=#Raw" INTPTR_FORMAT,msg,i,p2i(t->is_rawptr()));
break;
case Type::DoubleCon:
st->print(" %s%d]=#%fD",msg,i,t->is_double_constant()->_d);
break;
case Type::FloatCon:
st->print(" %s%d]=#%fF",msg,i,t->is_float_constant()->_f);
break;
case Type::Long:
st->print(" %s%d]=#"INT64_FORMAT,msg,i,(int64_t)(t->is_long()->get_con()));
break;
case Type::Half:
case Type::Top:
st->print(" %s%d]=_",msg,i);
break;
default: ShouldNotReachHere();
}
}
}
//------------------------------format-----------------------------------------
void JVMState::format(PhaseRegAlloc *regalloc, const Node *n, outputStream* st) const {
st->print(" #");
if (_method) {
_method->print_short_name(st);
st->print(" @ bci:%d ",_bci);
} else {
st->print_cr(" runtime stub ");
return;
}
if (n->is_MachSafePoint()) {
GrowableArray<SafePointScalarObjectNode*> scobjs;
MachSafePointNode *mcall = n->as_MachSafePoint();
uint i;
// Print locals
for (i = 0; i < (uint)loc_size(); i++)
format_helper(regalloc, st, mcall->local(this, i), "L[", i, &scobjs);
// Print stack
for (i = 0; i < (uint)stk_size(); i++) {
if ((uint)(_stkoff + i) >= mcall->len())
st->print(" oob ");
else
format_helper(regalloc, st, mcall->stack(this, i), "STK[", i, &scobjs);
}
for (i = 0; (int)i < nof_monitors(); i++) {
Node *box = mcall->monitor_box(this, i);
Node *obj = mcall->monitor_obj(this, i);
if (regalloc->node_regs_max_index() > 0 &&
OptoReg::is_valid(regalloc->get_reg_first(box))) {
box = BoxLockNode::box_node(box);
format_helper(regalloc, st, box, "MON-BOX[", i, &scobjs);
} else {
OptoReg::Name box_reg = BoxLockNode::reg(box);
st->print(" MON-BOX%d=%s+%d",
i,
OptoReg::regname(OptoReg::c_frame_pointer),
regalloc->reg2offset(box_reg));
}
const char* obj_msg = "MON-OBJ[";
if (EliminateLocks) {
if (BoxLockNode::box_node(box)->is_eliminated())
obj_msg = "MON-OBJ(LOCK ELIMINATED)[";
}
format_helper(regalloc, st, obj, obj_msg, i, &scobjs);
}
for (i = 0; i < (uint)scobjs.length(); i++) {
// Scalar replaced objects.
st->cr();
st->print(" # ScObj" INT32_FORMAT " ", i);
SafePointScalarObjectNode* spobj = scobjs.at(i);
ciKlass* cik = spobj->bottom_type()->is_oopptr()->klass();
assert(cik->is_instance_klass() ||
cik->is_array_klass(), "Not supported allocation.");
ciInstanceKlass *iklass = NULL;
if (cik->is_instance_klass()) {
cik->print_name_on(st);
iklass = cik->as_instance_klass();
} else if (cik->is_type_array_klass()) {
cik->as_array_klass()->base_element_type()->print_name_on(st);
st->print("[%d]", spobj->n_fields());
} else if (cik->is_obj_array_klass()) {
ciKlass* cie = cik->as_obj_array_klass()->base_element_klass();
if (cie->is_instance_klass()) {
cie->print_name_on(st);
} else if (cie->is_type_array_klass()) {
cie->as_array_klass()->base_element_type()->print_name_on(st);
} else {
ShouldNotReachHere();
}
st->print("[%d]", spobj->n_fields());
int ndim = cik->as_array_klass()->dimension() - 1;
while (ndim-- > 0) {
st->print("[]");
}
}
st->print("={");
uint nf = spobj->n_fields();
if (nf > 0) {
uint first_ind = spobj->first_index(mcall->jvms());
Node* fld_node = mcall->in(first_ind);
ciField* cifield;
if (iklass != NULL) {
st->print(" [");
cifield = iklass->nonstatic_field_at(0);
cifield->print_name_on(st);
format_helper(regalloc, st, fld_node, ":", 0, &scobjs);
} else {
format_helper(regalloc, st, fld_node, "[", 0, &scobjs);
}
for (uint j = 1; j < nf; j++) {
fld_node = mcall->in(first_ind+j);
if (iklass != NULL) {
st->print(", [");
cifield = iklass->nonstatic_field_at(j);
cifield->print_name_on(st);
format_helper(regalloc, st, fld_node, ":", j, &scobjs);
} else {
format_helper(regalloc, st, fld_node, ", [", j, &scobjs);
}
}
}
st->print(" }");
}
}
st->cr();
if (caller() != NULL) caller()->format(regalloc, n, st);
}
void JVMState::dump_spec(outputStream *st) const {
if (_method != NULL) {
bool printed = false;
if (!Verbose) {
// The JVMS dumps make really, really long lines.
// Take out the most boring parts, which are the package prefixes.
char buf[500];
stringStream namest(buf, sizeof(buf));
_method->print_short_name(&namest);
if (namest.count() < sizeof(buf)) {
const char* name = namest.base();
if (name[0] == ' ') ++name;
const char* endcn = strchr(name, ':'); // end of class name
if (endcn == NULL) endcn = strchr(name, '(');
if (endcn == NULL) endcn = name + strlen(name);
while (endcn > name && endcn[-1] != '.' && endcn[-1] != '/')
--endcn;
st->print(" %s", endcn);
printed = true;
}
}
if (!printed)
_method->print_short_name(st);
st->print(" @ bci:%d",_bci);
if(_reexecute == Reexecute_True)
st->print(" reexecute");
} else {
st->print(" runtime stub");
}
if (caller() != NULL) caller()->dump_spec(st);
}
void JVMState::dump_on(outputStream* st) const {
bool print_map = _map && !((uintptr_t)_map & 1) &&
((caller() == NULL) || (caller()->map() != _map));
if (print_map) {
if (_map->len() > _map->req()) { // _map->has_exceptions()
Node* ex = _map->in(_map->req()); // _map->next_exception()
// skip the first one; it's already being printed
while (ex != NULL && ex->len() > ex->req()) {
ex = ex->in(ex->req()); // ex->next_exception()
ex->dump(1);
}
}
_map->dump(Verbose ? 2 : 1);
}
if (caller() != NULL) {
caller()->dump_on(st);
}
st->print("JVMS depth=%d loc=%d stk=%d arg=%d mon=%d scalar=%d end=%d mondepth=%d sp=%d bci=%d reexecute=%s method=",
depth(), locoff(), stkoff(), argoff(), monoff(), scloff(), endoff(), monitor_depth(), sp(), bci(), should_reexecute()?"true":"false");
if (_method == NULL) {
st->print_cr("(none)");
} else {
_method->print_name(st);
st->cr();
if (bci() >= 0 && bci() < _method->code_size()) {
st->print(" bc: ");
_method->print_codes_on(bci(), bci()+1, st);
}
}
}
// Extra way to dump a jvms from the debugger,
// to avoid a bug with C++ member function calls.
void dump_jvms(JVMState* jvms) {
jvms->dump();
}
#endif
//--------------------------clone_shallow--------------------------------------
JVMState* JVMState::clone_shallow(Compile* C) const {
JVMState* n = has_method() ? new (C) JVMState(_method, _caller) : new (C) JVMState(0);
n->set_bci(_bci);
n->_reexecute = _reexecute;
n->set_locoff(_locoff);
n->set_stkoff(_stkoff);
n->set_monoff(_monoff);
n->set_scloff(_scloff);
n->set_endoff(_endoff);
n->set_sp(_sp);
n->set_map(_map);
return n;
}
//---------------------------clone_deep----------------------------------------
JVMState* JVMState::clone_deep(Compile* C) const {
JVMState* n = clone_shallow(C);
for (JVMState* p = n; p->_caller != NULL; p = p->_caller) {
p->_caller = p->_caller->clone_shallow(C);
}
assert(n->depth() == depth(), "sanity");
assert(n->debug_depth() == debug_depth(), "sanity");
return n;
}
/**
* Reset map for all callers
*/
void JVMState::set_map_deep(SafePointNode* map) {
for (JVMState* p = this; p->_caller != NULL; p = p->_caller) {
p->set_map(map);
}
}
// Adapt offsets in in-array after adding or removing an edge.
// Prerequisite is that the JVMState is used by only one node.
void JVMState::adapt_position(int delta) {
for (JVMState* jvms = this; jvms != NULL; jvms = jvms->caller()) {
jvms->set_locoff(jvms->locoff() + delta);
jvms->set_stkoff(jvms->stkoff() + delta);
jvms->set_monoff(jvms->monoff() + delta);
jvms->set_scloff(jvms->scloff() + delta);
jvms->set_endoff(jvms->endoff() + delta);
}
}
// Mirror the stack size calculation in the deopt code
// How much stack space would we need at this point in the program in
// case of deoptimization?
int JVMState::interpreter_frame_size() const {
const JVMState* jvms = this;
int size = 0;
int callee_parameters = 0;
int callee_locals = 0;
int extra_args = method()->max_stack() - stk_size();
while (jvms != NULL) {
int locks = jvms->nof_monitors();
int temps = jvms->stk_size();
bool is_top_frame = (jvms == this);
ciMethod* method = jvms->method();
int frame_size = BytesPerWord * Interpreter::size_activation(method->max_stack(),
temps + callee_parameters,
extra_args,
locks,
callee_parameters,
callee_locals,
is_top_frame);
size += frame_size;
callee_parameters = method->size_of_parameters();
callee_locals = method->max_locals();
extra_args = 0;
jvms = jvms->caller();
}
return size + Deoptimization::last_frame_adjust(0, callee_locals) * BytesPerWord;
}
//=============================================================================
uint CallNode::cmp( const Node &n ) const
{ return _tf == ((CallNode&)n)._tf && _jvms == ((CallNode&)n)._jvms; }
#ifndef PRODUCT
void CallNode::dump_req(outputStream *st) const {
// Dump the required inputs, enclosed in '(' and ')'
uint i; // Exit value of loop
for (i = 0; i < req(); i++) { // For all required inputs
if (i == TypeFunc::Parms) st->print("(");
if (in(i)) st->print("%c%d ", Compile::current()->node_arena()->contains(in(i)) ? ' ' : 'o', in(i)->_idx);
else st->print("_ ");
}
st->print(")");
}
void CallNode::dump_spec(outputStream *st) const {
st->print(" ");
tf()->dump_on(st);
if (_cnt != COUNT_UNKNOWN) st->print(" C=%f",_cnt);
if (jvms() != NULL) jvms()->dump_spec(st);
}
#endif
const Type *CallNode::bottom_type() const { return tf()->range(); }
const Type *CallNode::Value(PhaseTransform *phase) const {
if (phase->type(in(0)) == Type::TOP) return Type::TOP;
return tf()->range();
}
//------------------------------calling_convention-----------------------------
void CallNode::calling_convention( BasicType* sig_bt, VMRegPair *parm_regs, uint argcnt ) const {
// Use the standard compiler calling convention
Matcher::calling_convention( sig_bt, parm_regs, argcnt, true );
}
//------------------------------match------------------------------------------
// Construct projections for control, I/O, memory-fields, ..., and
// return result(s) along with their RegMask info
Node *CallNode::match( const ProjNode *proj, const Matcher *match ) {
switch (proj->_con) {
case TypeFunc::Control:
case TypeFunc::I_O:
case TypeFunc::Memory:
return new (match->C) MachProjNode(this,proj->_con,RegMask::Empty,MachProjNode::unmatched_proj);
case TypeFunc::Parms+1: // For LONG & DOUBLE returns
assert(tf()->_range->field_at(TypeFunc::Parms+1) == Type::HALF, "");
// 2nd half of doubles and longs
return new (match->C) MachProjNode(this,proj->_con, RegMask::Empty, (uint)OptoReg::Bad);
case TypeFunc::Parms: { // Normal returns
uint ideal_reg = tf()->range()->field_at(TypeFunc::Parms)->ideal_reg();
OptoRegPair regs = is_CallRuntime()
? match->c_return_value(ideal_reg,true) // Calls into C runtime
: match-> return_value(ideal_reg,true); // Calls into compiled Java code
RegMask rm = RegMask(regs.first());
if( OptoReg::is_valid(regs.second()) )
rm.Insert( regs.second() );
return new (match->C) MachProjNode(this,proj->_con,rm,ideal_reg);
}
case TypeFunc::ReturnAdr:
case TypeFunc::FramePtr:
default:
ShouldNotReachHere();
}
return NULL;
}
// Do we Match on this edge index or not? Match no edges
uint CallNode::match_edge(uint idx) const {
return 0;
}
//
// Determine whether the call could modify the field of the specified
// instance at the specified offset.
//
bool CallNode::may_modify(const TypeOopPtr *t_oop, PhaseTransform *phase) {
assert((t_oop != NULL), "sanity");
if (t_oop->is_known_instance()) {
// The instance_id is set only for scalar-replaceable allocations which
// are not passed as arguments according to Escape Analysis.
return false;
}
if (t_oop->is_ptr_to_boxed_value()) {
ciKlass* boxing_klass = t_oop->klass();
if (is_CallStaticJava() && as_CallStaticJava()->is_boxing_method()) {
// Skip unrelated boxing methods.
Node* proj = proj_out(TypeFunc::Parms);
if ((proj == NULL) || (phase->type(proj)->is_instptr()->klass() != boxing_klass)) {
return false;
}
}
if (is_CallJava() && as_CallJava()->method() != NULL) {
ciMethod* meth = as_CallJava()->method();
if (meth->is_accessor()) {
return false;
}
// May modify (by reflection) if an boxing object is passed
// as argument or returned.
Node* proj = returns_pointer() ? proj_out(TypeFunc::Parms) : NULL;
if (proj != NULL) {
const TypeInstPtr* inst_t = phase->type(proj)->isa_instptr();
if ((inst_t != NULL) && (!inst_t->klass_is_exact() ||
(inst_t->klass() == boxing_klass))) {
return true;
}
}
const TypeTuple* d = tf()->domain();
for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
const TypeInstPtr* inst_t = d->field_at(i)->isa_instptr();
if ((inst_t != NULL) && (!inst_t->klass_is_exact() ||
(inst_t->klass() == boxing_klass))) {
return true;
}
}
return false;
}
}
return true;
}
// Does this call have a direct reference to n other than debug information?
bool CallNode::has_non_debug_use(Node *n) {
const TypeTuple * d = tf()->domain();
for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
Node *arg = in(i);
if (arg == n) {
return true;
}
}
return false;
}
// Returns the unique CheckCastPP of a call
// or 'this' if there are several CheckCastPP or unexpected uses
// or returns NULL if there is no one.
Node *CallNode::result_cast() {
Node *cast = NULL;
Node *p = proj_out(TypeFunc::Parms);
if (p == NULL)
return NULL;
for (DUIterator_Fast imax, i = p->fast_outs(imax); i < imax; i++) {
Node *use = p->fast_out(i);
if (use->is_CheckCastPP()) {
if (cast != NULL) {
return this; // more than 1 CheckCastPP
}
cast = use;
} else if (!use->is_Initialize() &&
!use->is_AddP()) {
// Expected uses are restricted to a CheckCastPP, an Initialize
// node, and AddP nodes. If we encounter any other use (a Phi
// node can be seen in rare cases) return this to prevent
// incorrect optimizations.
return this;
}
}
return cast;
}
void CallNode::extract_projections(CallProjections* projs, bool separate_io_proj) {
projs->fallthrough_proj = NULL;
projs->fallthrough_catchproj = NULL;
projs->fallthrough_ioproj = NULL;
projs->catchall_ioproj = NULL;
projs->catchall_catchproj = NULL;
projs->fallthrough_memproj = NULL;
projs->catchall_memproj = NULL;
projs->resproj = NULL;
projs->exobj = NULL;
for (DUIterator_Fast imax, i = fast_outs(imax); i < imax; i++) {
ProjNode *pn = fast_out(i)->as_Proj();
if (pn->outcnt() == 0) continue;
switch (pn->_con) {
case TypeFunc::Control:
{
// For Control (fallthrough) and I_O (catch_all_index) we have CatchProj -> Catch -> Proj
projs->fallthrough_proj = pn;
DUIterator_Fast jmax, j = pn->fast_outs(jmax);
const Node *cn = pn->fast_out(j);
if (cn->is_Catch()) {
ProjNode *cpn = NULL;
for (DUIterator_Fast kmax, k = cn->fast_outs(kmax); k < kmax; k++) {
cpn = cn->fast_out(k)->as_Proj();
assert(cpn->is_CatchProj(), "must be a CatchProjNode");
if (cpn->_con == CatchProjNode::fall_through_index)
projs->fallthrough_catchproj = cpn;
else {
assert(cpn->_con == CatchProjNode::catch_all_index, "must be correct index.");
projs->catchall_catchproj = cpn;
}
}
}
break;
}
case TypeFunc::I_O:
if (pn->_is_io_use)
projs->catchall_ioproj = pn;
else
projs->fallthrough_ioproj = pn;
for (DUIterator j = pn->outs(); pn->has_out(j); j++) {
Node* e = pn->out(j);
if (e->Opcode() == Op_CreateEx && e->in(0)->is_CatchProj() && e->outcnt() > 0) {
assert(projs->exobj == NULL, "only one");
projs->exobj = e;
}
}
break;
case TypeFunc::Memory:
if (pn->_is_io_use)
projs->catchall_memproj = pn;
else
projs->fallthrough_memproj = pn;
break;
case TypeFunc::Parms:
projs->resproj = pn;
break;
default:
assert(false, "unexpected projection from allocation node.");
}
}
// The resproj may not exist because the result couuld be ignored
// and the exception object may not exist if an exception handler
// swallows the exception but all the other must exist and be found.
assert(projs->fallthrough_proj != NULL, "must be found");
assert(Compile::current()->inlining_incrementally() || projs->fallthrough_catchproj != NULL, "must be found");
assert(Compile::current()->inlining_incrementally() || projs->fallthrough_memproj != NULL, "must be found");
assert(Compile::current()->inlining_incrementally() || projs->fallthrough_ioproj != NULL, "must be found");
assert(Compile::current()->inlining_incrementally() || projs->catchall_catchproj != NULL, "must be found");
if (separate_io_proj) {
assert(Compile::current()->inlining_incrementally() || projs->catchall_memproj != NULL, "must be found");
assert(Compile::current()->inlining_incrementally() || projs->catchall_ioproj != NULL, "must be found");
}
}
Node *CallNode::Ideal(PhaseGVN *phase, bool can_reshape) {
CallGenerator* cg = generator();
if (can_reshape && cg != NULL && cg->is_mh_late_inline() && !cg->already_attempted()) {
// Check whether this MH handle call becomes a candidate for inlining
ciMethod* callee = cg->method();
vmIntrinsics::ID iid = callee->intrinsic_id();
if (iid == vmIntrinsics::_invokeBasic) {
if (in(TypeFunc::Parms)->Opcode() == Op_ConP) {
phase->C->prepend_late_inline(cg);
set_generator(NULL);
}
} else {
assert(callee->has_member_arg(), "wrong type of call?");
if (in(TypeFunc::Parms + callee->arg_size() - 1)->Opcode() == Op_ConP) {
phase->C->prepend_late_inline(cg);
set_generator(NULL);
}
}
}
return SafePointNode::Ideal(phase, can_reshape);
}
//=============================================================================
uint CallJavaNode::size_of() const { return sizeof(*this); }
uint CallJavaNode::cmp( const Node &n ) const {
CallJavaNode &call = (CallJavaNode&)n;
return CallNode::cmp(call) && _method == call._method;
}
#ifndef PRODUCT
void CallJavaNode::dump_spec(outputStream *st) const {
if( _method ) _method->print_short_name(st);
CallNode::dump_spec(st);
}
#endif
//=============================================================================
uint CallStaticJavaNode::size_of() const { return sizeof(*this); }
uint CallStaticJavaNode::cmp( const Node &n ) const {
CallStaticJavaNode &call = (CallStaticJavaNode&)n;
return CallJavaNode::cmp(call);
}
//----------------------------uncommon_trap_request----------------------------
// If this is an uncommon trap, return the request code, else zero.
int CallStaticJavaNode::uncommon_trap_request() const {
if (_name != NULL && !strcmp(_name, "uncommon_trap")) {
return extract_uncommon_trap_request(this);
}
return 0;
}
int CallStaticJavaNode::extract_uncommon_trap_request(const Node* call) {
#ifndef PRODUCT
if (!(call->req() > TypeFunc::Parms &&
call->in(TypeFunc::Parms) != NULL &&
call->in(TypeFunc::Parms)->is_Con())) {
assert(in_dump() != 0, "OK if dumping");
tty->print("[bad uncommon trap]");
return 0;
}
#endif
return call->in(TypeFunc::Parms)->bottom_type()->is_int()->get_con();
}
#ifndef PRODUCT
void CallStaticJavaNode::dump_spec(outputStream *st) const {
st->print("# Static ");
if (_name != NULL) {
st->print("%s", _name);
int trap_req = uncommon_trap_request();
if (trap_req != 0) {
char buf[100];
st->print("(%s)",
Deoptimization::format_trap_request(buf, sizeof(buf),
trap_req));
}
st->print(" ");
}
CallJavaNode::dump_spec(st);
}
#endif
//=============================================================================
uint CallDynamicJavaNode::size_of() const { return sizeof(*this); }
uint CallDynamicJavaNode::cmp( const Node &n ) const {
CallDynamicJavaNode &call = (CallDynamicJavaNode&)n;
return CallJavaNode::cmp(call);
}
#ifndef PRODUCT
void CallDynamicJavaNode::dump_spec(outputStream *st) const {
st->print("# Dynamic ");
CallJavaNode::dump_spec(st);
}
#endif
//=============================================================================
uint CallRuntimeNode::size_of() const { return sizeof(*this); }
uint CallRuntimeNode::cmp( const Node &n ) const {
CallRuntimeNode &call = (CallRuntimeNode&)n;
return CallNode::cmp(call) && !strcmp(_name,call._name);
}
#ifndef PRODUCT
void CallRuntimeNode::dump_spec(outputStream *st) const {
st->print("# ");
st->print("%s", _name);
CallNode::dump_spec(st);
}
#endif
//------------------------------calling_convention-----------------------------
void CallRuntimeNode::calling_convention( BasicType* sig_bt, VMRegPair *parm_regs, uint argcnt ) const {
Matcher::c_calling_convention( sig_bt, parm_regs, argcnt );
}
//=============================================================================
//------------------------------calling_convention-----------------------------
//=============================================================================
#ifndef PRODUCT
void CallLeafNode::dump_spec(outputStream *st) const {
st->print("# ");
st->print("%s", _name);
CallNode::dump_spec(st);
}
#endif
//=============================================================================
void SafePointNode::set_local(JVMState* jvms, uint idx, Node *c) {
assert(verify_jvms(jvms), "jvms must match");
int loc = jvms->locoff() + idx;
if (in(loc)->is_top() && idx > 0 && !c->is_top() ) {
// If current local idx is top then local idx - 1 could
// be a long/double that needs to be killed since top could
// represent the 2nd half ofthe long/double.
uint ideal = in(loc -1)->ideal_reg();
if (ideal == Op_RegD || ideal == Op_RegL) {
// set other (low index) half to top
set_req(loc - 1, in(loc));
}
}
set_req(loc, c);
}
uint SafePointNode::size_of() const { return sizeof(*this); }
uint SafePointNode::cmp( const Node &n ) const {
return (&n == this); // Always fail except on self
}
//-------------------------set_next_exception----------------------------------
void SafePointNode::set_next_exception(SafePointNode* n) {
assert(n == NULL || n->Opcode() == Op_SafePoint, "correct value for next_exception");
if (len() == req()) {
if (n != NULL) add_prec(n);
} else {
set_prec(req(), n);
}
}
//----------------------------next_exception-----------------------------------
SafePointNode* SafePointNode::next_exception() const {
if (len() == req()) {
return NULL;
} else {
Node* n = in(req());
assert(n == NULL || n->Opcode() == Op_SafePoint, "no other uses of prec edges");
return (SafePointNode*) n;
}
}
//------------------------------Ideal------------------------------------------
// Skip over any collapsed Regions
Node *SafePointNode::Ideal(PhaseGVN *phase, bool can_reshape) {
return remove_dead_region(phase, can_reshape) ? this : NULL;
}
//------------------------------Identity---------------------------------------
// Remove obviously duplicate safepoints
Node *SafePointNode::Identity( PhaseTransform *phase ) {
// If you have back to back safepoints, remove one
if( in(TypeFunc::Control)->is_SafePoint() )
return in(TypeFunc::Control);
if( in(0)->is_Proj() ) {
Node *n0 = in(0)->in(0);
// Check if he is a call projection (except Leaf Call)
if( n0->is_Catch() ) {
n0 = n0->in(0)->in(0);
assert( n0->is_Call(), "expect a call here" );
}
if( n0->is_Call() && n0->as_Call()->guaranteed_safepoint() ) {
// Useless Safepoint, so remove it
return in(TypeFunc::Control);
}
}
return this;
}
//------------------------------Value------------------------------------------
const Type *SafePointNode::Value( PhaseTransform *phase ) const {
if( phase->type(in(0)) == Type::TOP ) return Type::TOP;
if( phase->eqv( in(0), this ) ) return Type::TOP; // Dead infinite loop
return Type::CONTROL;
}
#ifndef PRODUCT
void SafePointNode::dump_spec(outputStream *st) const {
st->print(" SafePoint ");
_replaced_nodes.dump(st);
}
#endif
const RegMask &SafePointNode::in_RegMask(uint idx) const {
if( idx < TypeFunc::Parms ) return RegMask::Empty;
// Values outside the domain represent debug info
return *(Compile::current()->matcher()->idealreg2debugmask[in(idx)->ideal_reg()]);
}
const RegMask &SafePointNode::out_RegMask() const {
return RegMask::Empty;
}
void SafePointNode::grow_stack(JVMState* jvms, uint grow_by) {
assert((int)grow_by > 0, "sanity");
int monoff = jvms->monoff();
int scloff = jvms->scloff();
int endoff = jvms->endoff();
assert(endoff == (int)req(), "no other states or debug info after me");
Node* top = Compile::current()->top();
for (uint i = 0; i < grow_by; i++) {
ins_req(monoff, top);
}
jvms->set_monoff(monoff + grow_by);
jvms->set_scloff(scloff + grow_by);
jvms->set_endoff(endoff + grow_by);
}
void SafePointNode::push_monitor(const FastLockNode *lock) {
// Add a LockNode, which points to both the original BoxLockNode (the
// stack space for the monitor) and the Object being locked.
const int MonitorEdges = 2;
assert(JVMState::logMonitorEdges == exact_log2(MonitorEdges), "correct MonitorEdges");
assert(req() == jvms()->endoff(), "correct sizing");
int nextmon = jvms()->scloff();
if (GenerateSynchronizationCode) {
ins_req(nextmon, lock->box_node());
ins_req(nextmon+1, lock->obj_node());
} else {
Node* top = Compile::current()->top();
ins_req(nextmon, top);
ins_req(nextmon, top);
}
jvms()->set_scloff(nextmon + MonitorEdges);
jvms()->set_endoff(req());
}
void SafePointNode::pop_monitor() {
// Delete last monitor from debug info
debug_only(int num_before_pop = jvms()->nof_monitors());
const int MonitorEdges = 2;
assert(JVMState::logMonitorEdges == exact_log2(MonitorEdges), "correct MonitorEdges");
int scloff = jvms()->scloff();
int endoff = jvms()->endoff();
int new_scloff = scloff - MonitorEdges;
int new_endoff = endoff - MonitorEdges;
jvms()->set_scloff(new_scloff);
jvms()->set_endoff(new_endoff);
while (scloff > new_scloff) del_req_ordered(--scloff);
assert(jvms()->nof_monitors() == num_before_pop-1, "");
}
Node *SafePointNode::peek_monitor_box() const {
int mon = jvms()->nof_monitors() - 1;
assert(mon >= 0, "most have a monitor");
return monitor_box(jvms(), mon);
}
Node *SafePointNode::peek_monitor_obj() const {
int mon = jvms()->nof_monitors() - 1;
assert(mon >= 0, "most have a monitor");
return monitor_obj(jvms(), mon);
}
// Do we Match on this edge index or not? Match no edges
uint SafePointNode::match_edge(uint idx) const {
if( !needs_polling_address_input() )
return 0;
return (TypeFunc::Parms == idx);
}
//============== SafePointScalarObjectNode ==============
SafePointScalarObjectNode::SafePointScalarObjectNode(const TypeOopPtr* tp,
#ifdef ASSERT
AllocateNode* alloc,
#endif
uint first_index,
uint n_fields) :
TypeNode(tp, 1), // 1 control input -- seems required. Get from root.
#ifdef ASSERT
_alloc(alloc),
#endif
_first_index(first_index),
_n_fields(n_fields)
{
init_class_id(Class_SafePointScalarObject);
}
// Do not allow value-numbering for SafePointScalarObject node.
uint SafePointScalarObjectNode::hash() const { return NO_HASH; }
uint SafePointScalarObjectNode::cmp( const Node &n ) const {
return (&n == this); // Always fail except on self
}
uint SafePointScalarObjectNode::ideal_reg() const {
return 0; // No matching to machine instruction
}
const RegMask &SafePointScalarObjectNode::in_RegMask(uint idx) const {
return *(Compile::current()->matcher()->idealreg2debugmask[in(idx)->ideal_reg()]);
}
const RegMask &SafePointScalarObjectNode::out_RegMask() const {
return RegMask::Empty;
}
uint SafePointScalarObjectNode::match_edge(uint idx) const {
return 0;
}
SafePointScalarObjectNode*
SafePointScalarObjectNode::clone(Dict* sosn_map) const {
void* cached = (*sosn_map)[(void*)this];
if (cached != NULL) {
return (SafePointScalarObjectNode*)cached;
}
SafePointScalarObjectNode* res = (SafePointScalarObjectNode*)Node::clone();
sosn_map->Insert((void*)this, (void*)res);
return res;
}
#ifndef PRODUCT
void SafePointScalarObjectNode::dump_spec(outputStream *st) const {
st->print(" # fields@[%d..%d]", first_index(),
first_index() + n_fields() - 1);
}
#endif
//=============================================================================
uint AllocateNode::size_of() const { return sizeof(*this); }
AllocateNode::AllocateNode(Compile* C, const TypeFunc *atype,
Node *ctrl, Node *mem, Node *abio,
Node *size, Node *klass_node, Node *initial_test)
: CallNode(atype, NULL, TypeRawPtr::BOTTOM)
{
init_class_id(Class_Allocate);
init_flags(Flag_is_macro);
_is_scalar_replaceable = false;
_is_non_escaping = false;
Node *topnode = C->top();
init_req( TypeFunc::Control , ctrl );
init_req( TypeFunc::I_O , abio );
init_req( TypeFunc::Memory , mem );
init_req( TypeFunc::ReturnAdr, topnode );
init_req( TypeFunc::FramePtr , topnode );
init_req( AllocSize , size);
init_req( KlassNode , klass_node);
init_req( InitialTest , initial_test);
init_req( ALength , topnode);
C->add_macro_node(this);
}
//=============================================================================
Node* AllocateArrayNode::Ideal(PhaseGVN *phase, bool can_reshape) {
if (remove_dead_region(phase, can_reshape)) return this;
// Don't bother trying to transform a dead node
if (in(0) && in(0)->is_top()) return NULL;
const Type* type = phase->type(Ideal_length());
if (type->isa_int() && type->is_int()->_hi < 0) {
if (can_reshape) {
PhaseIterGVN *igvn = phase->is_IterGVN();
// Unreachable fall through path (negative array length),
// the allocation can only throw so disconnect it.
Node* proj = proj_out(TypeFunc::Control);
Node* catchproj = NULL;
if (proj != NULL) {
for (DUIterator_Fast imax, i = proj->fast_outs(imax); i < imax; i++) {
Node *cn = proj->fast_out(i);
if (cn->is_Catch()) {
catchproj = cn->as_Multi()->proj_out(CatchProjNode::fall_through_index);
break;
}
}
}
if (catchproj != NULL && catchproj->outcnt() > 0 &&
(catchproj->outcnt() > 1 ||
catchproj->unique_out()->Opcode() != Op_Halt)) {
assert(catchproj->is_CatchProj(), "must be a CatchProjNode");
Node* nproj = catchproj->clone();
igvn->register_new_node_with_optimizer(nproj);
Node *frame = new (phase->C) ParmNode( phase->C->start(), TypeFunc::FramePtr );
frame = phase->transform(frame);
// Halt & Catch Fire
Node *halt = new (phase->C) HaltNode( nproj, frame );
phase->C->root()->add_req(halt);
phase->transform(halt);
igvn->replace_node(catchproj, phase->C->top());
return this;
}
} else {
// Can't correct it during regular GVN so register for IGVN
phase->C->record_for_igvn(this);
}
}
return NULL;
}
// Retrieve the length from the AllocateArrayNode. Narrow the type with a
// CastII, if appropriate. If we are not allowed to create new nodes, and
// a CastII is appropriate, return NULL.
Node *AllocateArrayNode::make_ideal_length(const TypeOopPtr* oop_type, PhaseTransform *phase, bool allow_new_nodes) {
Node *length = in(AllocateNode::ALength);
assert(length != NULL, "length is not null");
const TypeInt* length_type = phase->find_int_type(length);
const TypeAryPtr* ary_type = oop_type->isa_aryptr();
if (ary_type != NULL && length_type != NULL) {
const TypeInt* narrow_length_type = ary_type->narrow_size_type(length_type);
if (narrow_length_type != length_type) {
// Assert one of:
// - the narrow_length is 0
// - the narrow_length is not wider than length
assert(narrow_length_type == TypeInt::ZERO ||
length_type->is_con() && narrow_length_type->is_con() &&
(narrow_length_type->_hi <= length_type->_lo) ||
(narrow_length_type->_hi <= length_type->_hi &&
narrow_length_type->_lo >= length_type->_lo),
"narrow type must be narrower than length type");
// Return NULL if new nodes are not allowed
if (!allow_new_nodes) return NULL;
// Create a cast which is control dependent on the initialization to
// propagate the fact that the array length must be positive.
length = new (phase->C) CastIINode(length, narrow_length_type);
length->set_req(0, initialization()->proj_out(0));
}
}
return length;
}
//=============================================================================
uint LockNode::size_of() const { return sizeof(*this); }
// Redundant lock elimination
//
// There are various patterns of locking where we release and
// immediately reacquire a lock in a piece of code where no operations
// occur in between that would be observable. In those cases we can
// skip releasing and reacquiring the lock without violating any
// fairness requirements. Doing this around a loop could cause a lock
// to be held for a very long time so we concentrate on non-looping
// control flow. We also require that the operations are fully
// redundant meaning that we don't introduce new lock operations on
// some paths so to be able to eliminate it on others ala PRE. This
// would probably require some more extensive graph manipulation to
// guarantee that the memory edges were all handled correctly.
//
// Assuming p is a simple predicate which can't trap in any way and s
// is a synchronized method consider this code:
//
// s();
// if (p)
// s();
// else
// s();
// s();
//
// 1. The unlocks of the first call to s can be eliminated if the
// locks inside the then and else branches are eliminated.
//
// 2. The unlocks of the then and else branches can be eliminated if
// the lock of the final call to s is eliminated.
//
// Either of these cases subsumes the simple case of sequential control flow
//
// Addtionally we can eliminate versions without the else case:
//
// s();
// if (p)
// s();
// s();
//
// 3. In this case we eliminate the unlock of the first s, the lock
// and unlock in the then case and the lock in the final s.
//
// Note also that in all these cases the then/else pieces don't have
// to be trivial as long as they begin and end with synchronization
// operations.
//
// s();
// if (p)
// s();
// f();
// s();
// s();
//
// The code will work properly for this case, leaving in the unlock
// before the call to f and the relock after it.
//
// A potentially interesting case which isn't handled here is when the
// locking is partially redundant.
//
// s();
// if (p)
// s();
//
// This could be eliminated putting unlocking on the else case and
// eliminating the first unlock and the lock in the then side.
// Alternatively the unlock could be moved out of the then side so it
// was after the merge and the first unlock and second lock
// eliminated. This might require less manipulation of the memory
// state to get correct.
//
// Additionally we might allow work between a unlock and lock before
// giving up eliminating the locks. The current code disallows any
// conditional control flow between these operations. A formulation
// similar to partial redundancy elimination computing the
// availability of unlocking and the anticipatability of locking at a
// program point would allow detection of fully redundant locking with
// some amount of work in between. I'm not sure how often I really
// think that would occur though. Most of the cases I've seen
// indicate it's likely non-trivial work would occur in between.
// There may be other more complicated constructs where we could
// eliminate locking but I haven't seen any others appear as hot or
// interesting.
//
// Locking and unlocking have a canonical form in ideal that looks
// roughly like this:
//
// <obj>
// | \\------+
// | \ \
// | BoxLock \
// | | | \
// | | \ \
// | | FastLock
// | | /
// | | /
// | | |
//
// Lock
// |
// Proj #0
// |
// MembarAcquire
// |
// Proj #0
//
// MembarRelease
// |
// Proj #0
// |
// Unlock
// |
// Proj #0
//
//
// This code proceeds by processing Lock nodes during PhaseIterGVN
// and searching back through its control for the proper code
// patterns. Once it finds a set of lock and unlock operations to
// eliminate they are marked as eliminatable which causes the
// expansion of the Lock and Unlock macro nodes to make the operation a NOP
//
//=============================================================================
//
// Utility function to skip over uninteresting control nodes. Nodes skipped are:
// - copy regions. (These may not have been optimized away yet.)
// - eliminated locking nodes
//
static Node *next_control(Node *ctrl) {
if (ctrl == NULL)
return NULL;
while (1) {
if (ctrl->is_Region()) {
RegionNode *r = ctrl->as_Region();
Node *n = r->is_copy();
if (n == NULL)
break; // hit a region, return it
else
ctrl = n;
} else if (ctrl->is_Proj()) {
Node *in0 = ctrl->in(0);
if (in0->is_AbstractLock() && in0->as_AbstractLock()->is_eliminated()) {
ctrl = in0->in(0);
} else {
break;
}
} else {
break; // found an interesting control
}
}
return ctrl;
}
//
// Given a control, see if it's the control projection of an Unlock which
// operating on the same object as lock.
//
bool AbstractLockNode::find_matching_unlock(const Node* ctrl, LockNode* lock,
GrowableArray<AbstractLockNode*> &lock_ops) {
ProjNode *ctrl_proj = (ctrl->is_Proj()) ? ctrl->as_Proj() : NULL;
if (ctrl_proj != NULL && ctrl_proj->_con == TypeFunc::Control) {
Node *n = ctrl_proj->in(0);
if (n != NULL && n->is_Unlock()) {
UnlockNode *unlock = n->as_Unlock();
if (lock->obj_node()->eqv_uncast(unlock->obj_node()) &&
BoxLockNode::same_slot(lock->box_node(), unlock->box_node()) &&
!unlock->is_eliminated()) {
lock_ops.append(unlock);
return true;
}
}
}
return false;
}
//
// Find the lock matching an unlock. Returns null if a safepoint
// or complicated control is encountered first.
LockNode *AbstractLockNode::find_matching_lock(UnlockNode* unlock) {
LockNode *lock_result = NULL;
// find the matching lock, or an intervening safepoint
Node *ctrl = next_control(unlock->in(0));
while (1) {
assert(ctrl != NULL, "invalid control graph");
assert(!ctrl->is_Start(), "missing lock for unlock");
if (ctrl->is_top()) break; // dead control path
if (ctrl->is_Proj()) ctrl = ctrl->in(0);
if (ctrl->is_SafePoint()) {
break; // found a safepoint (may be the lock we are searching for)
} else if (ctrl->is_Region()) {
// Check for a simple diamond pattern. Punt on anything more complicated
if (ctrl->req() == 3 && ctrl->in(1) != NULL && ctrl->in(2) != NULL) {
Node *in1 = next_control(ctrl->in(1));
Node *in2 = next_control(ctrl->in(2));
if (((in1->is_IfTrue() && in2->is_IfFalse()) ||
(in2->is_IfTrue() && in1->is_IfFalse())) && (in1->in(0) == in2->in(0))) {
ctrl = next_control(in1->in(0)->in(0));
} else {
break;
}
} else {
break;
}
} else {
ctrl = next_control(ctrl->in(0)); // keep searching
}
}
if (ctrl->is_Lock()) {
LockNode *lock = ctrl->as_Lock();
if (lock->obj_node()->eqv_uncast(unlock->obj_node()) &&
BoxLockNode::same_slot(lock->box_node(), unlock->box_node())) {
lock_result = lock;
}
}
return lock_result;
}
// This code corresponds to case 3 above.
bool AbstractLockNode::find_lock_and_unlock_through_if(Node* node, LockNode* lock,
GrowableArray<AbstractLockNode*> &lock_ops) {
Node* if_node = node->in(0);
bool if_true = node->is_IfTrue();
if (if_node->is_If() && if_node->outcnt() == 2 && (if_true || node->is_IfFalse())) {
Node *lock_ctrl = next_control(if_node->in(0));
if (find_matching_unlock(lock_ctrl, lock, lock_ops)) {
Node* lock1_node = NULL;
ProjNode* proj = if_node->as_If()->proj_out(!if_true);
if (if_true) {
if (proj->is_IfFalse() && proj->outcnt() == 1) {
lock1_node = proj->unique_out();
}
} else {
if (proj->is_IfTrue() && proj->outcnt() == 1) {
lock1_node = proj->unique_out();
}
}
if (lock1_node != NULL && lock1_node->is_Lock()) {
LockNode *lock1 = lock1_node->as_Lock();
if (lock->obj_node()->eqv_uncast(lock1->obj_node()) &&
BoxLockNode::same_slot(lock->box_node(), lock1->box_node()) &&
!lock1->is_eliminated()) {
lock_ops.append(lock1);
return true;
}
}
}
}
lock_ops.trunc_to(0);
return false;
}
bool AbstractLockNode::find_unlocks_for_region(const RegionNode* region, LockNode* lock,
GrowableArray<AbstractLockNode*> &lock_ops) {
// check each control merging at this point for a matching unlock.
// in(0) should be self edge so skip it.
for (int i = 1; i < (int)region->req(); i++) {
Node *in_node = next_control(region->in(i));
if (in_node != NULL) {
if (find_matching_unlock(in_node, lock, lock_ops)) {
// found a match so keep on checking.
continue;
} else if (find_lock_and_unlock_through_if(in_node, lock, lock_ops)) {
continue;
}
// If we fall through to here then it was some kind of node we
// don't understand or there wasn't a matching unlock, so give
// up trying to merge locks.
lock_ops.trunc_to(0);
return false;
}
}
return true;
}
#ifndef PRODUCT
//
// Create a counter which counts the number of times this lock is acquired
//
void AbstractLockNode::create_lock_counter(JVMState* state) {
_counter = OptoRuntime::new_named_counter(state, NamedCounter::LockCounter);
}
void AbstractLockNode::set_eliminated_lock_counter() {
if (_counter) {
// Update the counter to indicate that this lock was eliminated.
// The counter update code will stay around even though the
// optimizer will eliminate the lock operation itself.
_counter->set_tag(NamedCounter::EliminatedLockCounter);
}
}
#endif
//=============================================================================
Node *LockNode::Ideal(PhaseGVN *phase, bool can_reshape) {
// perform any generic optimizations first (returns 'this' or NULL)
Node *result = SafePointNode::Ideal(phase, can_reshape);
if (result != NULL) return result;
// Don't bother trying to transform a dead node
if (in(0) && in(0)->is_top()) return NULL;
// Now see if we can optimize away this lock. We don't actually
// remove the locking here, we simply set the _eliminate flag which
// prevents macro expansion from expanding the lock. Since we don't
// modify the graph, the value returned from this function is the
// one computed above.
if (can_reshape && EliminateLocks && !is_non_esc_obj()) {
//
// If we are locking an unescaped object, the lock/unlock is unnecessary
//
ConnectionGraph *cgr = phase->C->congraph();
if (cgr != NULL && cgr->not_global_escape(obj_node())) {
assert(!is_eliminated() || is_coarsened(), "sanity");
// The lock could be marked eliminated by lock coarsening
// code during first IGVN before EA. Replace coarsened flag
// to eliminate all associated locks/unlocks.
#ifdef ASSERT
this->log_lock_optimization(phase->C,"eliminate_lock_set_non_esc1");
#endif
this->set_non_esc_obj();
return result;
}
//
// Try lock coarsening
//
PhaseIterGVN* iter = phase->is_IterGVN();
if (iter != NULL && !is_eliminated()) {
GrowableArray<AbstractLockNode*> lock_ops;
Node *ctrl = next_control(in(0));
// now search back for a matching Unlock
if (find_matching_unlock(ctrl, this, lock_ops)) {
// found an unlock directly preceding this lock. This is the
// case of single unlock directly control dependent on a
// single lock which is the trivial version of case 1 or 2.
} else if (ctrl->is_Region() ) {
if (find_unlocks_for_region(ctrl->as_Region(), this, lock_ops)) {
// found lock preceded by multiple unlocks along all paths
// joining at this point which is case 3 in description above.
}
} else {
// see if this lock comes from either half of an if and the
// predecessors merges unlocks and the other half of the if
// performs a lock.
if (find_lock_and_unlock_through_if(ctrl, this, lock_ops)) {
// found unlock splitting to an if with locks on both branches.
}
}
if (lock_ops.length() > 0) {
// add ourselves to the list of locks to be eliminated.
lock_ops.append(this);
#ifndef PRODUCT
if (PrintEliminateLocks) {
int locks = 0;
int unlocks = 0;
for (int i = 0; i < lock_ops.length(); i++) {
AbstractLockNode* lock = lock_ops.at(i);
if (lock->Opcode() == Op_Lock)
locks++;
else
unlocks++;
if (Verbose) {
lock->dump(1);
}
}
tty->print_cr("***Eliminated %d unlocks and %d locks", unlocks, locks);
}
#endif
// for each of the identified locks, mark them
// as eliminatable
for (int i = 0; i < lock_ops.length(); i++) {
AbstractLockNode* lock = lock_ops.at(i);
// Mark it eliminated by coarsening and update any counters
#ifdef ASSERT
lock->log_lock_optimization(phase->C, "eliminate_lock_set_coarsened");
#endif
lock->set_coarsened();
}
} else if (ctrl->is_Region() &&
iter->_worklist.member(ctrl)) {
// We weren't able to find any opportunities but the region this
// lock is control dependent on hasn't been processed yet so put
// this lock back on the worklist so we can check again once any
// region simplification has occurred.
iter->_worklist.push(this);
}
}
}
return result;
}
//=============================================================================
bool LockNode::is_nested_lock_region() {
return is_nested_lock_region(NULL);
}
// p is used for access to compilation log; no logging if NULL
bool LockNode::is_nested_lock_region(Compile * c) {
BoxLockNode* box = box_node()->as_BoxLock();
int stk_slot = box->stack_slot();
if (stk_slot <= 0) {
#ifdef ASSERT
this->log_lock_optimization(c, "eliminate_lock_INLR_1");
#endif
return false; // External lock or it is not Box (Phi node).
}
// Ignore complex cases: merged locks or multiple locks.
Node* obj = obj_node();
LockNode* unique_lock = NULL;
if (!box->is_simple_lock_region(&unique_lock, obj)) {
#ifdef ASSERT
this->log_lock_optimization(c, "eliminate_lock_INLR_2a");
#endif
return false;
}
if (unique_lock != this) {
#ifdef ASSERT
this->log_lock_optimization(c, "eliminate_lock_INLR_2b");
#endif
return false;
}
// Look for external lock for the same object.
SafePointNode* sfn = this->as_SafePoint();
JVMState* youngest_jvms = sfn->jvms();
int max_depth = youngest_jvms->depth();
for (int depth = 1; depth <= max_depth; depth++) {
JVMState* jvms = youngest_jvms->of_depth(depth);
int num_mon = jvms->nof_monitors();
// Loop over monitors
for (int idx = 0; idx < num_mon; idx++) {
Node* obj_node = sfn->monitor_obj(jvms, idx);
BoxLockNode* box_node = sfn->monitor_box(jvms, idx)->as_BoxLock();
if ((box_node->stack_slot() < stk_slot) && obj_node->eqv_uncast(obj)) {
return true;
}
}
}
#ifdef ASSERT
this->log_lock_optimization(c, "eliminate_lock_INLR_3");
#endif
return false;
}
//=============================================================================
uint UnlockNode::size_of() const { return sizeof(*this); }
//=============================================================================
Node *UnlockNode::Ideal(PhaseGVN *phase, bool can_reshape) {
// perform any generic optimizations first (returns 'this' or NULL)
Node *result = SafePointNode::Ideal(phase, can_reshape);
if (result != NULL) return result;
// Don't bother trying to transform a dead node
if (in(0) && in(0)->is_top()) return NULL;
// Now see if we can optimize away this unlock. We don't actually
// remove the unlocking here, we simply set the _eliminate flag which
// prevents macro expansion from expanding the unlock. Since we don't
// modify the graph, the value returned from this function is the
// one computed above.
// Escape state is defined after Parse phase.
if (can_reshape && EliminateLocks && !is_non_esc_obj()) {
//
// If we are unlocking an unescaped object, the lock/unlock is unnecessary.
//
ConnectionGraph *cgr = phase->C->congraph();
if (cgr != NULL && cgr->not_global_escape(obj_node())) {
assert(!is_eliminated() || is_coarsened(), "sanity");
// The lock could be marked eliminated by lock coarsening
// code during first IGVN before EA. Replace coarsened flag
// to eliminate all associated locks/unlocks.
#ifdef ASSERT
this->log_lock_optimization(phase->C, "eliminate_lock_set_non_esc2");
#endif
this->set_non_esc_obj();
}
}
return result;
}
const char * AbstractLockNode::kind_as_string() const {
return is_coarsened() ? "coarsened" :
is_nested() ? "nested" :
is_non_esc_obj() ? "non_escaping" :
"?";
}
void AbstractLockNode::log_lock_optimization(Compile *C, const char * tag) const {
if (C == NULL) {
return;
}
CompileLog* log = C->log();
if (log != NULL) {
log->begin_head("%s lock='%d' compile_id='%d' class_id='%s' kind='%s'",
tag, is_Lock(), C->compile_id(),
is_Unlock() ? "unlock" : is_Lock() ? "lock" : "?",
kind_as_string());
log->stamp();
log->end_head();
JVMState* p = is_Unlock() ? (as_Unlock()->dbg_jvms()) : jvms();
while (p != NULL) {
log->elem("jvms bci='%d' method='%d'", p->bci(), log->identify(p->method()));
p = p->caller();
}
log->tail(tag);
}
}