6912521: System.arraycopy works slower than the simple loop for little lengths
Summary: convert small array copies to series of loads and stores
Reviewed-by: kvn, vlivanov
/*
* Copyright (c) 2001, 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 "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
#include "gc_implementation/g1/heapRegion.hpp"
#include "gc_interface/collectedHeap.hpp"
#include "memory/barrierSet.hpp"
#include "memory/cardTableModRefBS.hpp"
#include "opto/addnode.hpp"
#include "opto/castnode.hpp"
#include "opto/convertnode.hpp"
#include "opto/graphKit.hpp"
#include "opto/idealKit.hpp"
#include "opto/intrinsicnode.hpp"
#include "opto/locknode.hpp"
#include "opto/machnode.hpp"
#include "opto/opaquenode.hpp"
#include "opto/parse.hpp"
#include "opto/rootnode.hpp"
#include "opto/runtime.hpp"
#include "runtime/deoptimization.hpp"
#include "runtime/sharedRuntime.hpp"
//----------------------------GraphKit-----------------------------------------
// Main utility constructor.
GraphKit::GraphKit(JVMState* jvms)
: Phase(Phase::Parser),
_env(C->env()),
_gvn(*C->initial_gvn())
{
_exceptions = jvms->map()->next_exception();
if (_exceptions != NULL) jvms->map()->set_next_exception(NULL);
set_jvms(jvms);
}
// Private constructor for parser.
GraphKit::GraphKit()
: Phase(Phase::Parser),
_env(C->env()),
_gvn(*C->initial_gvn())
{
_exceptions = NULL;
set_map(NULL);
debug_only(_sp = -99);
debug_only(set_bci(-99));
}
//---------------------------clean_stack---------------------------------------
// Clear away rubbish from the stack area of the JVM state.
// This destroys any arguments that may be waiting on the stack.
void GraphKit::clean_stack(int from_sp) {
SafePointNode* map = this->map();
JVMState* jvms = this->jvms();
int stk_size = jvms->stk_size();
int stkoff = jvms->stkoff();
Node* top = this->top();
for (int i = from_sp; i < stk_size; i++) {
if (map->in(stkoff + i) != top) {
map->set_req(stkoff + i, top);
}
}
}
//--------------------------------sync_jvms-----------------------------------
// Make sure our current jvms agrees with our parse state.
JVMState* GraphKit::sync_jvms() const {
JVMState* jvms = this->jvms();
jvms->set_bci(bci()); // Record the new bci in the JVMState
jvms->set_sp(sp()); // Record the new sp in the JVMState
assert(jvms_in_sync(), "jvms is now in sync");
return jvms;
}
//--------------------------------sync_jvms_for_reexecute---------------------
// Make sure our current jvms agrees with our parse state. This version
// uses the reexecute_sp for reexecuting bytecodes.
JVMState* GraphKit::sync_jvms_for_reexecute() {
JVMState* jvms = this->jvms();
jvms->set_bci(bci()); // Record the new bci in the JVMState
jvms->set_sp(reexecute_sp()); // Record the new sp in the JVMState
return jvms;
}
#ifdef ASSERT
bool GraphKit::jvms_in_sync() const {
Parse* parse = is_Parse();
if (parse == NULL) {
if (bci() != jvms()->bci()) return false;
if (sp() != (int)jvms()->sp()) return false;
return true;
}
if (jvms()->method() != parse->method()) return false;
if (jvms()->bci() != parse->bci()) return false;
int jvms_sp = jvms()->sp();
if (jvms_sp != parse->sp()) return false;
int jvms_depth = jvms()->depth();
if (jvms_depth != parse->depth()) return false;
return true;
}
// Local helper checks for special internal merge points
// used to accumulate and merge exception states.
// They are marked by the region's in(0) edge being the map itself.
// Such merge points must never "escape" into the parser at large,
// until they have been handed to gvn.transform.
static bool is_hidden_merge(Node* reg) {
if (reg == NULL) return false;
if (reg->is_Phi()) {
reg = reg->in(0);
if (reg == NULL) return false;
}
return reg->is_Region() && reg->in(0) != NULL && reg->in(0)->is_Root();
}
void GraphKit::verify_map() const {
if (map() == NULL) return; // null map is OK
assert(map()->req() <= jvms()->endoff(), "no extra garbage on map");
assert(!map()->has_exceptions(), "call add_exception_states_from 1st");
assert(!is_hidden_merge(control()), "call use_exception_state, not set_map");
}
void GraphKit::verify_exception_state(SafePointNode* ex_map) {
assert(ex_map->next_exception() == NULL, "not already part of a chain");
assert(has_saved_ex_oop(ex_map), "every exception state has an ex_oop");
}
#endif
//---------------------------stop_and_kill_map---------------------------------
// Set _map to NULL, signalling a stop to further bytecode execution.
// First smash the current map's control to a constant, to mark it dead.
void GraphKit::stop_and_kill_map() {
SafePointNode* dead_map = stop();
if (dead_map != NULL) {
dead_map->disconnect_inputs(NULL, C); // Mark the map as killed.
assert(dead_map->is_killed(), "must be so marked");
}
}
//--------------------------------stopped--------------------------------------
// Tell if _map is NULL, or control is top.
bool GraphKit::stopped() {
if (map() == NULL) return true;
else if (control() == top()) return true;
else return false;
}
//-----------------------------has_ex_handler----------------------------------
// Tell if this method or any caller method has exception handlers.
bool GraphKit::has_ex_handler() {
for (JVMState* jvmsp = jvms(); jvmsp != NULL; jvmsp = jvmsp->caller()) {
if (jvmsp->has_method() && jvmsp->method()->has_exception_handlers()) {
return true;
}
}
return false;
}
//------------------------------save_ex_oop------------------------------------
// Save an exception without blowing stack contents or other JVM state.
void GraphKit::set_saved_ex_oop(SafePointNode* ex_map, Node* ex_oop) {
assert(!has_saved_ex_oop(ex_map), "clear ex-oop before setting again");
ex_map->add_req(ex_oop);
debug_only(verify_exception_state(ex_map));
}
inline static Node* common_saved_ex_oop(SafePointNode* ex_map, bool clear_it) {
assert(GraphKit::has_saved_ex_oop(ex_map), "ex_oop must be there");
Node* ex_oop = ex_map->in(ex_map->req()-1);
if (clear_it) ex_map->del_req(ex_map->req()-1);
return ex_oop;
}
//-----------------------------saved_ex_oop------------------------------------
// Recover a saved exception from its map.
Node* GraphKit::saved_ex_oop(SafePointNode* ex_map) {
return common_saved_ex_oop(ex_map, false);
}
//--------------------------clear_saved_ex_oop---------------------------------
// Erase a previously saved exception from its map.
Node* GraphKit::clear_saved_ex_oop(SafePointNode* ex_map) {
return common_saved_ex_oop(ex_map, true);
}
#ifdef ASSERT
//---------------------------has_saved_ex_oop----------------------------------
// Erase a previously saved exception from its map.
bool GraphKit::has_saved_ex_oop(SafePointNode* ex_map) {
return ex_map->req() == ex_map->jvms()->endoff()+1;
}
#endif
//-------------------------make_exception_state--------------------------------
// Turn the current JVM state into an exception state, appending the ex_oop.
SafePointNode* GraphKit::make_exception_state(Node* ex_oop) {
sync_jvms();
SafePointNode* ex_map = stop(); // do not manipulate this map any more
set_saved_ex_oop(ex_map, ex_oop);
return ex_map;
}
//--------------------------add_exception_state--------------------------------
// Add an exception to my list of exceptions.
void GraphKit::add_exception_state(SafePointNode* ex_map) {
if (ex_map == NULL || ex_map->control() == top()) {
return;
}
#ifdef ASSERT
verify_exception_state(ex_map);
if (has_exceptions()) {
assert(ex_map->jvms()->same_calls_as(_exceptions->jvms()), "all collected exceptions must come from the same place");
}
#endif
// If there is already an exception of exactly this type, merge with it.
// In particular, null-checks and other low-level exceptions common up here.
Node* ex_oop = saved_ex_oop(ex_map);
const Type* ex_type = _gvn.type(ex_oop);
if (ex_oop == top()) {
// No action needed.
return;
}
assert(ex_type->isa_instptr(), "exception must be an instance");
for (SafePointNode* e2 = _exceptions; e2 != NULL; e2 = e2->next_exception()) {
const Type* ex_type2 = _gvn.type(saved_ex_oop(e2));
// We check sp also because call bytecodes can generate exceptions
// both before and after arguments are popped!
if (ex_type2 == ex_type
&& e2->_jvms->sp() == ex_map->_jvms->sp()) {
combine_exception_states(ex_map, e2);
return;
}
}
// No pre-existing exception of the same type. Chain it on the list.
push_exception_state(ex_map);
}
//-----------------------add_exception_states_from-----------------------------
void GraphKit::add_exception_states_from(JVMState* jvms) {
SafePointNode* ex_map = jvms->map()->next_exception();
if (ex_map != NULL) {
jvms->map()->set_next_exception(NULL);
for (SafePointNode* next_map; ex_map != NULL; ex_map = next_map) {
next_map = ex_map->next_exception();
ex_map->set_next_exception(NULL);
add_exception_state(ex_map);
}
}
}
//-----------------------transfer_exceptions_into_jvms-------------------------
JVMState* GraphKit::transfer_exceptions_into_jvms() {
if (map() == NULL) {
// We need a JVMS to carry the exceptions, but the map has gone away.
// Create a scratch JVMS, cloned from any of the exception states...
if (has_exceptions()) {
_map = _exceptions;
_map = clone_map();
_map->set_next_exception(NULL);
clear_saved_ex_oop(_map);
debug_only(verify_map());
} else {
// ...or created from scratch
JVMState* jvms = new (C) JVMState(_method, NULL);
jvms->set_bci(_bci);
jvms->set_sp(_sp);
jvms->set_map(new SafePointNode(TypeFunc::Parms, jvms));
set_jvms(jvms);
for (uint i = 0; i < map()->req(); i++) map()->init_req(i, top());
set_all_memory(top());
while (map()->req() < jvms->endoff()) map()->add_req(top());
}
// (This is a kludge, in case you didn't notice.)
set_control(top());
}
JVMState* jvms = sync_jvms();
assert(!jvms->map()->has_exceptions(), "no exceptions on this map yet");
jvms->map()->set_next_exception(_exceptions);
_exceptions = NULL; // done with this set of exceptions
return jvms;
}
static inline void add_n_reqs(Node* dstphi, Node* srcphi) {
assert(is_hidden_merge(dstphi), "must be a special merge node");
assert(is_hidden_merge(srcphi), "must be a special merge node");
uint limit = srcphi->req();
for (uint i = PhiNode::Input; i < limit; i++) {
dstphi->add_req(srcphi->in(i));
}
}
static inline void add_one_req(Node* dstphi, Node* src) {
assert(is_hidden_merge(dstphi), "must be a special merge node");
assert(!is_hidden_merge(src), "must not be a special merge node");
dstphi->add_req(src);
}
//-----------------------combine_exception_states------------------------------
// This helper function combines exception states by building phis on a
// specially marked state-merging region. These regions and phis are
// untransformed, and can build up gradually. The region is marked by
// having a control input of its exception map, rather than NULL. Such
// regions do not appear except in this function, and in use_exception_state.
void GraphKit::combine_exception_states(SafePointNode* ex_map, SafePointNode* phi_map) {
if (failing()) return; // dying anyway...
JVMState* ex_jvms = ex_map->_jvms;
assert(ex_jvms->same_calls_as(phi_map->_jvms), "consistent call chains");
assert(ex_jvms->stkoff() == phi_map->_jvms->stkoff(), "matching locals");
assert(ex_jvms->sp() == phi_map->_jvms->sp(), "matching stack sizes");
assert(ex_jvms->monoff() == phi_map->_jvms->monoff(), "matching JVMS");
assert(ex_jvms->scloff() == phi_map->_jvms->scloff(), "matching scalar replaced objects");
assert(ex_map->req() == phi_map->req(), "matching maps");
uint tos = ex_jvms->stkoff() + ex_jvms->sp();
Node* hidden_merge_mark = root();
Node* region = phi_map->control();
MergeMemNode* phi_mem = phi_map->merged_memory();
MergeMemNode* ex_mem = ex_map->merged_memory();
if (region->in(0) != hidden_merge_mark) {
// The control input is not (yet) a specially-marked region in phi_map.
// Make it so, and build some phis.
region = new RegionNode(2);
_gvn.set_type(region, Type::CONTROL);
region->set_req(0, hidden_merge_mark); // marks an internal ex-state
region->init_req(1, phi_map->control());
phi_map->set_control(region);
Node* io_phi = PhiNode::make(region, phi_map->i_o(), Type::ABIO);
record_for_igvn(io_phi);
_gvn.set_type(io_phi, Type::ABIO);
phi_map->set_i_o(io_phi);
for (MergeMemStream mms(phi_mem); mms.next_non_empty(); ) {
Node* m = mms.memory();
Node* m_phi = PhiNode::make(region, m, Type::MEMORY, mms.adr_type(C));
record_for_igvn(m_phi);
_gvn.set_type(m_phi, Type::MEMORY);
mms.set_memory(m_phi);
}
}
// Either or both of phi_map and ex_map might already be converted into phis.
Node* ex_control = ex_map->control();
// if there is special marking on ex_map also, we add multiple edges from src
bool add_multiple = (ex_control->in(0) == hidden_merge_mark);
// how wide was the destination phi_map, originally?
uint orig_width = region->req();
if (add_multiple) {
add_n_reqs(region, ex_control);
add_n_reqs(phi_map->i_o(), ex_map->i_o());
} else {
// ex_map has no merges, so we just add single edges everywhere
add_one_req(region, ex_control);
add_one_req(phi_map->i_o(), ex_map->i_o());
}
for (MergeMemStream mms(phi_mem, ex_mem); mms.next_non_empty2(); ) {
if (mms.is_empty()) {
// get a copy of the base memory, and patch some inputs into it
const TypePtr* adr_type = mms.adr_type(C);
Node* phi = mms.force_memory()->as_Phi()->slice_memory(adr_type);
assert(phi->as_Phi()->region() == mms.base_memory()->in(0), "");
mms.set_memory(phi);
// Prepare to append interesting stuff onto the newly sliced phi:
while (phi->req() > orig_width) phi->del_req(phi->req()-1);
}
// Append stuff from ex_map:
if (add_multiple) {
add_n_reqs(mms.memory(), mms.memory2());
} else {
add_one_req(mms.memory(), mms.memory2());
}
}
uint limit = ex_map->req();
for (uint i = TypeFunc::Parms; i < limit; i++) {
// Skip everything in the JVMS after tos. (The ex_oop follows.)
if (i == tos) i = ex_jvms->monoff();
Node* src = ex_map->in(i);
Node* dst = phi_map->in(i);
if (src != dst) {
PhiNode* phi;
if (dst->in(0) != region) {
dst = phi = PhiNode::make(region, dst, _gvn.type(dst));
record_for_igvn(phi);
_gvn.set_type(phi, phi->type());
phi_map->set_req(i, dst);
// Prepare to append interesting stuff onto the new phi:
while (dst->req() > orig_width) dst->del_req(dst->req()-1);
} else {
assert(dst->is_Phi(), "nobody else uses a hidden region");
phi = dst->as_Phi();
}
if (add_multiple && src->in(0) == ex_control) {
// Both are phis.
add_n_reqs(dst, src);
} else {
while (dst->req() < region->req()) add_one_req(dst, src);
}
const Type* srctype = _gvn.type(src);
if (phi->type() != srctype) {
const Type* dsttype = phi->type()->meet_speculative(srctype);
if (phi->type() != dsttype) {
phi->set_type(dsttype);
_gvn.set_type(phi, dsttype);
}
}
}
}
phi_map->merge_replaced_nodes_with(ex_map);
}
//--------------------------use_exception_state--------------------------------
Node* GraphKit::use_exception_state(SafePointNode* phi_map) {
if (failing()) { stop(); return top(); }
Node* region = phi_map->control();
Node* hidden_merge_mark = root();
assert(phi_map->jvms()->map() == phi_map, "sanity: 1-1 relation");
Node* ex_oop = clear_saved_ex_oop(phi_map);
if (region->in(0) == hidden_merge_mark) {
// Special marking for internal ex-states. Process the phis now.
region->set_req(0, region); // now it's an ordinary region
set_jvms(phi_map->jvms()); // ...so now we can use it as a map
// Note: Setting the jvms also sets the bci and sp.
set_control(_gvn.transform(region));
uint tos = jvms()->stkoff() + sp();
for (uint i = 1; i < tos; i++) {
Node* x = phi_map->in(i);
if (x->in(0) == region) {
assert(x->is_Phi(), "expected a special phi");
phi_map->set_req(i, _gvn.transform(x));
}
}
for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
Node* x = mms.memory();
if (x->in(0) == region) {
assert(x->is_Phi(), "nobody else uses a hidden region");
mms.set_memory(_gvn.transform(x));
}
}
if (ex_oop->in(0) == region) {
assert(ex_oop->is_Phi(), "expected a special phi");
ex_oop = _gvn.transform(ex_oop);
}
} else {
set_jvms(phi_map->jvms());
}
assert(!is_hidden_merge(phi_map->control()), "hidden ex. states cleared");
assert(!is_hidden_merge(phi_map->i_o()), "hidden ex. states cleared");
return ex_oop;
}
//---------------------------------java_bc-------------------------------------
Bytecodes::Code GraphKit::java_bc() const {
ciMethod* method = this->method();
int bci = this->bci();
if (method != NULL && bci != InvocationEntryBci)
return method->java_code_at_bci(bci);
else
return Bytecodes::_illegal;
}
void GraphKit::uncommon_trap_if_should_post_on_exceptions(Deoptimization::DeoptReason reason,
bool must_throw) {
// if the exception capability is set, then we will generate code
// to check the JavaThread.should_post_on_exceptions flag to see
// if we actually need to report exception events (for this
// thread). If we don't need to report exception events, we will
// take the normal fast path provided by add_exception_events. If
// exception event reporting is enabled for this thread, we will
// take the uncommon_trap in the BuildCutout below.
// first must access the should_post_on_exceptions_flag in this thread's JavaThread
Node* jthread = _gvn.transform(new ThreadLocalNode());
Node* adr = basic_plus_adr(top(), jthread, in_bytes(JavaThread::should_post_on_exceptions_flag_offset()));
Node* should_post_flag = make_load(control(), adr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, MemNode::unordered);
// Test the should_post_on_exceptions_flag vs. 0
Node* chk = _gvn.transform( new CmpINode(should_post_flag, intcon(0)) );
Node* tst = _gvn.transform( new BoolNode(chk, BoolTest::eq) );
// Branch to slow_path if should_post_on_exceptions_flag was true
{ BuildCutout unless(this, tst, PROB_MAX);
// Do not try anything fancy if we're notifying the VM on every throw.
// Cf. case Bytecodes::_athrow in parse2.cpp.
uncommon_trap(reason, Deoptimization::Action_none,
(ciKlass*)NULL, (char*)NULL, must_throw);
}
}
//------------------------------builtin_throw----------------------------------
void GraphKit::builtin_throw(Deoptimization::DeoptReason reason, Node* arg) {
bool must_throw = true;
if (env()->jvmti_can_post_on_exceptions()) {
// check if we must post exception events, take uncommon trap if so
uncommon_trap_if_should_post_on_exceptions(reason, must_throw);
// here if should_post_on_exceptions is false
// continue on with the normal codegen
}
// If this particular condition has not yet happened at this
// bytecode, then use the uncommon trap mechanism, and allow for
// a future recompilation if several traps occur here.
// If the throw is hot, try to use a more complicated inline mechanism
// which keeps execution inside the compiled code.
bool treat_throw_as_hot = false;
ciMethodData* md = method()->method_data();
if (ProfileTraps) {
if (too_many_traps(reason)) {
treat_throw_as_hot = true;
}
// (If there is no MDO at all, assume it is early in
// execution, and that any deopts are part of the
// startup transient, and don't need to be remembered.)
// Also, if there is a local exception handler, treat all throws
// as hot if there has been at least one in this method.
if (C->trap_count(reason) != 0
&& method()->method_data()->trap_count(reason) != 0
&& has_ex_handler()) {
treat_throw_as_hot = true;
}
}
// If this throw happens frequently, an uncommon trap might cause
// a performance pothole. If there is a local exception handler,
// and if this particular bytecode appears to be deoptimizing often,
// let us handle the throw inline, with a preconstructed instance.
// Note: If the deopt count has blown up, the uncommon trap
// runtime is going to flush this nmethod, not matter what.
if (treat_throw_as_hot
&& (!StackTraceInThrowable || OmitStackTraceInFastThrow)) {
// If the throw is local, we use a pre-existing instance and
// punt on the backtrace. This would lead to a missing backtrace
// (a repeat of 4292742) if the backtrace object is ever asked
// for its backtrace.
// Fixing this remaining case of 4292742 requires some flavor of
// escape analysis. Leave that for the future.
ciInstance* ex_obj = NULL;
switch (reason) {
case Deoptimization::Reason_null_check:
ex_obj = env()->NullPointerException_instance();
break;
case Deoptimization::Reason_div0_check:
ex_obj = env()->ArithmeticException_instance();
break;
case Deoptimization::Reason_range_check:
ex_obj = env()->ArrayIndexOutOfBoundsException_instance();
break;
case Deoptimization::Reason_class_check:
if (java_bc() == Bytecodes::_aastore) {
ex_obj = env()->ArrayStoreException_instance();
} else {
ex_obj = env()->ClassCastException_instance();
}
break;
}
if (failing()) { stop(); return; } // exception allocation might fail
if (ex_obj != NULL) {
// Cheat with a preallocated exception object.
if (C->log() != NULL)
C->log()->elem("hot_throw preallocated='1' reason='%s'",
Deoptimization::trap_reason_name(reason));
const TypeInstPtr* ex_con = TypeInstPtr::make(ex_obj);
Node* ex_node = _gvn.transform(ConNode::make(ex_con));
// Clear the detail message of the preallocated exception object.
// Weblogic sometimes mutates the detail message of exceptions
// using reflection.
int offset = java_lang_Throwable::get_detailMessage_offset();
const TypePtr* adr_typ = ex_con->add_offset(offset);
Node *adr = basic_plus_adr(ex_node, ex_node, offset);
const TypeOopPtr* val_type = TypeOopPtr::make_from_klass(env()->String_klass());
// Conservatively release stores of object references.
Node *store = store_oop_to_object(control(), ex_node, adr, adr_typ, null(), val_type, T_OBJECT, MemNode::release);
add_exception_state(make_exception_state(ex_node));
return;
}
}
// %%% Maybe add entry to OptoRuntime which directly throws the exc.?
// It won't be much cheaper than bailing to the interp., since we'll
// have to pass up all the debug-info, and the runtime will have to
// create the stack trace.
// Usual case: Bail to interpreter.
// Reserve the right to recompile if we haven't seen anything yet.
ciMethod* m = Deoptimization::reason_is_speculate(reason) ? C->method() : NULL;
Deoptimization::DeoptAction action = Deoptimization::Action_maybe_recompile;
if (treat_throw_as_hot
&& (method()->method_data()->trap_recompiled_at(bci(), m)
|| C->too_many_traps(reason))) {
// We cannot afford to take more traps here. Suffer in the interpreter.
if (C->log() != NULL)
C->log()->elem("hot_throw preallocated='0' reason='%s' mcount='%d'",
Deoptimization::trap_reason_name(reason),
C->trap_count(reason));
action = Deoptimization::Action_none;
}
// "must_throw" prunes the JVM state to include only the stack, if there
// are no local exception handlers. This should cut down on register
// allocation time and code size, by drastically reducing the number
// of in-edges on the call to the uncommon trap.
uncommon_trap(reason, action, (ciKlass*)NULL, (char*)NULL, must_throw);
}
//----------------------------PreserveJVMState---------------------------------
PreserveJVMState::PreserveJVMState(GraphKit* kit, bool clone_map) {
debug_only(kit->verify_map());
_kit = kit;
_map = kit->map(); // preserve the map
_sp = kit->sp();
kit->set_map(clone_map ? kit->clone_map() : NULL);
#ifdef ASSERT
_bci = kit->bci();
Parse* parser = kit->is_Parse();
int block = (parser == NULL || parser->block() == NULL) ? -1 : parser->block()->rpo();
_block = block;
#endif
}
PreserveJVMState::~PreserveJVMState() {
GraphKit* kit = _kit;
#ifdef ASSERT
assert(kit->bci() == _bci, "bci must not shift");
Parse* parser = kit->is_Parse();
int block = (parser == NULL || parser->block() == NULL) ? -1 : parser->block()->rpo();
assert(block == _block, "block must not shift");
#endif
kit->set_map(_map);
kit->set_sp(_sp);
}
//-----------------------------BuildCutout-------------------------------------
BuildCutout::BuildCutout(GraphKit* kit, Node* p, float prob, float cnt)
: PreserveJVMState(kit)
{
assert(p->is_Con() || p->is_Bool(), "test must be a bool");
SafePointNode* outer_map = _map; // preserved map is caller's
SafePointNode* inner_map = kit->map();
IfNode* iff = kit->create_and_map_if(outer_map->control(), p, prob, cnt);
outer_map->set_control(kit->gvn().transform( new IfTrueNode(iff) ));
inner_map->set_control(kit->gvn().transform( new IfFalseNode(iff) ));
}
BuildCutout::~BuildCutout() {
GraphKit* kit = _kit;
assert(kit->stopped(), "cutout code must stop, throw, return, etc.");
}
//---------------------------PreserveReexecuteState----------------------------
PreserveReexecuteState::PreserveReexecuteState(GraphKit* kit) {
assert(!kit->stopped(), "must call stopped() before");
_kit = kit;
_sp = kit->sp();
_reexecute = kit->jvms()->_reexecute;
}
PreserveReexecuteState::~PreserveReexecuteState() {
if (_kit->stopped()) return;
_kit->jvms()->_reexecute = _reexecute;
_kit->set_sp(_sp);
}
//------------------------------clone_map--------------------------------------
// Implementation of PreserveJVMState
//
// Only clone_map(...) here. If this function is only used in the
// PreserveJVMState class we may want to get rid of this extra
// function eventually and do it all there.
SafePointNode* GraphKit::clone_map() {
if (map() == NULL) return NULL;
// Clone the memory edge first
Node* mem = MergeMemNode::make(map()->memory());
gvn().set_type_bottom(mem);
SafePointNode *clonemap = (SafePointNode*)map()->clone();
JVMState* jvms = this->jvms();
JVMState* clonejvms = jvms->clone_shallow(C);
clonemap->set_memory(mem);
clonemap->set_jvms(clonejvms);
clonejvms->set_map(clonemap);
record_for_igvn(clonemap);
gvn().set_type_bottom(clonemap);
return clonemap;
}
//-----------------------------set_map_clone-----------------------------------
void GraphKit::set_map_clone(SafePointNode* m) {
_map = m;
_map = clone_map();
_map->set_next_exception(NULL);
debug_only(verify_map());
}
//----------------------------kill_dead_locals---------------------------------
// Detect any locals which are known to be dead, and force them to top.
void GraphKit::kill_dead_locals() {
// Consult the liveness information for the locals. If any
// of them are unused, then they can be replaced by top(). This
// should help register allocation time and cut down on the size
// of the deoptimization information.
// This call is made from many of the bytecode handling
// subroutines called from the Big Switch in do_one_bytecode.
// Every bytecode which might include a slow path is responsible
// for killing its dead locals. The more consistent we
// are about killing deads, the fewer useless phis will be
// constructed for them at various merge points.
// bci can be -1 (InvocationEntryBci). We return the entry
// liveness for the method.
if (method() == NULL || method()->code_size() == 0) {
// We are building a graph for a call to a native method.
// All locals are live.
return;
}
ResourceMark rm;
// Consult the liveness information for the locals. If any
// of them are unused, then they can be replaced by top(). This
// should help register allocation time and cut down on the size
// of the deoptimization information.
MethodLivenessResult live_locals = method()->liveness_at_bci(bci());
int len = (int)live_locals.size();
assert(len <= jvms()->loc_size(), "too many live locals");
for (int local = 0; local < len; local++) {
if (!live_locals.at(local)) {
set_local(local, top());
}
}
}
#ifdef ASSERT
//-------------------------dead_locals_are_killed------------------------------
// Return true if all dead locals are set to top in the map.
// Used to assert "clean" debug info at various points.
bool GraphKit::dead_locals_are_killed() {
if (method() == NULL || method()->code_size() == 0) {
// No locals need to be dead, so all is as it should be.
return true;
}
// Make sure somebody called kill_dead_locals upstream.
ResourceMark rm;
for (JVMState* jvms = this->jvms(); jvms != NULL; jvms = jvms->caller()) {
if (jvms->loc_size() == 0) continue; // no locals to consult
SafePointNode* map = jvms->map();
ciMethod* method = jvms->method();
int bci = jvms->bci();
if (jvms == this->jvms()) {
bci = this->bci(); // it might not yet be synched
}
MethodLivenessResult live_locals = method->liveness_at_bci(bci);
int len = (int)live_locals.size();
if (!live_locals.is_valid() || len == 0)
// This method is trivial, or is poisoned by a breakpoint.
return true;
assert(len == jvms->loc_size(), "live map consistent with locals map");
for (int local = 0; local < len; local++) {
if (!live_locals.at(local) && map->local(jvms, local) != top()) {
if (PrintMiscellaneous && (Verbose || WizardMode)) {
tty->print_cr("Zombie local %d: ", local);
jvms->dump();
}
return false;
}
}
}
return true;
}
#endif //ASSERT
// Helper function for enforcing certain bytecodes to reexecute if
// deoptimization happens
static bool should_reexecute_implied_by_bytecode(JVMState *jvms, bool is_anewarray) {
ciMethod* cur_method = jvms->method();
int cur_bci = jvms->bci();
if (cur_method != NULL && cur_bci != InvocationEntryBci) {
Bytecodes::Code code = cur_method->java_code_at_bci(cur_bci);
return Interpreter::bytecode_should_reexecute(code) ||
is_anewarray && code == Bytecodes::_multianewarray;
// Reexecute _multianewarray bytecode which was replaced with
// sequence of [a]newarray. See Parse::do_multianewarray().
//
// Note: interpreter should not have it set since this optimization
// is limited by dimensions and guarded by flag so in some cases
// multianewarray() runtime calls will be generated and
// the bytecode should not be reexecutes (stack will not be reset).
} else
return false;
}
// Helper function for adding JVMState and debug information to node
void GraphKit::add_safepoint_edges(SafePointNode* call, bool must_throw) {
// Add the safepoint edges to the call (or other safepoint).
// Make sure dead locals are set to top. This
// should help register allocation time and cut down on the size
// of the deoptimization information.
assert(dead_locals_are_killed(), "garbage in debug info before safepoint");
// Walk the inline list to fill in the correct set of JVMState's
// Also fill in the associated edges for each JVMState.
// If the bytecode needs to be reexecuted we need to put
// the arguments back on the stack.
const bool should_reexecute = jvms()->should_reexecute();
JVMState* youngest_jvms = should_reexecute ? sync_jvms_for_reexecute() : sync_jvms();
// NOTE: set_bci (called from sync_jvms) might reset the reexecute bit to
// undefined if the bci is different. This is normal for Parse but it
// should not happen for LibraryCallKit because only one bci is processed.
assert(!is_LibraryCallKit() || (jvms()->should_reexecute() == should_reexecute),
"in LibraryCallKit the reexecute bit should not change");
// If we are guaranteed to throw, we can prune everything but the
// input to the current bytecode.
bool can_prune_locals = false;
uint stack_slots_not_pruned = 0;
int inputs = 0, depth = 0;
if (must_throw) {
assert(method() == youngest_jvms->method(), "sanity");
if (compute_stack_effects(inputs, depth)) {
can_prune_locals = true;
stack_slots_not_pruned = inputs;
}
}
if (env()->should_retain_local_variables()) {
// At any safepoint, this method can get breakpointed, which would
// then require an immediate deoptimization.
can_prune_locals = false; // do not prune locals
stack_slots_not_pruned = 0;
}
// do not scribble on the input jvms
JVMState* out_jvms = youngest_jvms->clone_deep(C);
call->set_jvms(out_jvms); // Start jvms list for call node
// For a known set of bytecodes, the interpreter should reexecute them if
// deoptimization happens. We set the reexecute state for them here
if (out_jvms->is_reexecute_undefined() && //don't change if already specified
should_reexecute_implied_by_bytecode(out_jvms, call->is_AllocateArray())) {
out_jvms->set_should_reexecute(true); //NOTE: youngest_jvms not changed
}
// Presize the call:
DEBUG_ONLY(uint non_debug_edges = call->req());
call->add_req_batch(top(), youngest_jvms->debug_depth());
assert(call->req() == non_debug_edges + youngest_jvms->debug_depth(), "");
// Set up edges so that the call looks like this:
// Call [state:] ctl io mem fptr retadr
// [parms:] parm0 ... parmN
// [root:] loc0 ... locN stk0 ... stkSP mon0 obj0 ... monN objN
// [...mid:] loc0 ... locN stk0 ... stkSP mon0 obj0 ... monN objN [...]
// [young:] loc0 ... locN stk0 ... stkSP mon0 obj0 ... monN objN
// Note that caller debug info precedes callee debug info.
// Fill pointer walks backwards from "young:" to "root:" in the diagram above:
uint debug_ptr = call->req();
// Loop over the map input edges associated with jvms, add them
// to the call node, & reset all offsets to match call node array.
for (JVMState* in_jvms = youngest_jvms; in_jvms != NULL; ) {
uint debug_end = debug_ptr;
uint debug_start = debug_ptr - in_jvms->debug_size();
debug_ptr = debug_start; // back up the ptr
uint p = debug_start; // walks forward in [debug_start, debug_end)
uint j, k, l;
SafePointNode* in_map = in_jvms->map();
out_jvms->set_map(call);
if (can_prune_locals) {
assert(in_jvms->method() == out_jvms->method(), "sanity");
// If the current throw can reach an exception handler in this JVMS,
// then we must keep everything live that can reach that handler.
// As a quick and dirty approximation, we look for any handlers at all.
if (in_jvms->method()->has_exception_handlers()) {
can_prune_locals = false;
}
}
// Add the Locals
k = in_jvms->locoff();
l = in_jvms->loc_size();
out_jvms->set_locoff(p);
if (!can_prune_locals) {
for (j = 0; j < l; j++)
call->set_req(p++, in_map->in(k+j));
} else {
p += l; // already set to top above by add_req_batch
}
// Add the Expression Stack
k = in_jvms->stkoff();
l = in_jvms->sp();
out_jvms->set_stkoff(p);
if (!can_prune_locals) {
for (j = 0; j < l; j++)
call->set_req(p++, in_map->in(k+j));
} else if (can_prune_locals && stack_slots_not_pruned != 0) {
// Divide stack into {S0,...,S1}, where S0 is set to top.
uint s1 = stack_slots_not_pruned;
stack_slots_not_pruned = 0; // for next iteration
if (s1 > l) s1 = l;
uint s0 = l - s1;
p += s0; // skip the tops preinstalled by add_req_batch
for (j = s0; j < l; j++)
call->set_req(p++, in_map->in(k+j));
} else {
p += l; // already set to top above by add_req_batch
}
// Add the Monitors
k = in_jvms->monoff();
l = in_jvms->mon_size();
out_jvms->set_monoff(p);
for (j = 0; j < l; j++)
call->set_req(p++, in_map->in(k+j));
// Copy any scalar object fields.
k = in_jvms->scloff();
l = in_jvms->scl_size();
out_jvms->set_scloff(p);
for (j = 0; j < l; j++)
call->set_req(p++, in_map->in(k+j));
// Finish the new jvms.
out_jvms->set_endoff(p);
assert(out_jvms->endoff() == debug_end, "fill ptr must match");
assert(out_jvms->depth() == in_jvms->depth(), "depth must match");
assert(out_jvms->loc_size() == in_jvms->loc_size(), "size must match");
assert(out_jvms->mon_size() == in_jvms->mon_size(), "size must match");
assert(out_jvms->scl_size() == in_jvms->scl_size(), "size must match");
assert(out_jvms->debug_size() == in_jvms->debug_size(), "size must match");
// Update the two tail pointers in parallel.
out_jvms = out_jvms->caller();
in_jvms = in_jvms->caller();
}
assert(debug_ptr == non_debug_edges, "debug info must fit exactly");
// Test the correctness of JVMState::debug_xxx accessors:
assert(call->jvms()->debug_start() == non_debug_edges, "");
assert(call->jvms()->debug_end() == call->req(), "");
assert(call->jvms()->debug_depth() == call->req() - non_debug_edges, "");
}
bool GraphKit::compute_stack_effects(int& inputs, int& depth) {
Bytecodes::Code code = java_bc();
if (code == Bytecodes::_wide) {
code = method()->java_code_at_bci(bci() + 1);
}
BasicType rtype = T_ILLEGAL;
int rsize = 0;
if (code != Bytecodes::_illegal) {
depth = Bytecodes::depth(code); // checkcast=0, athrow=-1
rtype = Bytecodes::result_type(code); // checkcast=P, athrow=V
if (rtype < T_CONFLICT)
rsize = type2size[rtype];
}
switch (code) {
case Bytecodes::_illegal:
return false;
case Bytecodes::_ldc:
case Bytecodes::_ldc_w:
case Bytecodes::_ldc2_w:
inputs = 0;
break;
case Bytecodes::_dup: inputs = 1; break;
case Bytecodes::_dup_x1: inputs = 2; break;
case Bytecodes::_dup_x2: inputs = 3; break;
case Bytecodes::_dup2: inputs = 2; break;
case Bytecodes::_dup2_x1: inputs = 3; break;
case Bytecodes::_dup2_x2: inputs = 4; break;
case Bytecodes::_swap: inputs = 2; break;
case Bytecodes::_arraylength: inputs = 1; break;
case Bytecodes::_getstatic:
case Bytecodes::_putstatic:
case Bytecodes::_getfield:
case Bytecodes::_putfield:
{
bool ignored_will_link;
ciField* field = method()->get_field_at_bci(bci(), ignored_will_link);
int size = field->type()->size();
bool is_get = (depth >= 0), is_static = (depth & 1);
inputs = (is_static ? 0 : 1);
if (is_get) {
depth = size - inputs;
} else {
inputs += size; // putxxx pops the value from the stack
depth = - inputs;
}
}
break;
case Bytecodes::_invokevirtual:
case Bytecodes::_invokespecial:
case Bytecodes::_invokestatic:
case Bytecodes::_invokedynamic:
case Bytecodes::_invokeinterface:
{
bool ignored_will_link;
ciSignature* declared_signature = NULL;
ciMethod* ignored_callee = method()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
assert(declared_signature != NULL, "cannot be null");
inputs = declared_signature->arg_size_for_bc(code);
int size = declared_signature->return_type()->size();
depth = size - inputs;
}
break;
case Bytecodes::_multianewarray:
{
ciBytecodeStream iter(method());
iter.reset_to_bci(bci());
iter.next();
inputs = iter.get_dimensions();
assert(rsize == 1, "");
depth = rsize - inputs;
}
break;
case Bytecodes::_ireturn:
case Bytecodes::_lreturn:
case Bytecodes::_freturn:
case Bytecodes::_dreturn:
case Bytecodes::_areturn:
assert(rsize = -depth, "");
inputs = rsize;
break;
case Bytecodes::_jsr:
case Bytecodes::_jsr_w:
inputs = 0;
depth = 1; // S.B. depth=1, not zero
break;
default:
// bytecode produces a typed result
inputs = rsize - depth;
assert(inputs >= 0, "");
break;
}
#ifdef ASSERT
// spot check
int outputs = depth + inputs;
assert(outputs >= 0, "sanity");
switch (code) {
case Bytecodes::_checkcast: assert(inputs == 1 && outputs == 1, ""); break;
case Bytecodes::_athrow: assert(inputs == 1 && outputs == 0, ""); break;
case Bytecodes::_aload_0: assert(inputs == 0 && outputs == 1, ""); break;
case Bytecodes::_return: assert(inputs == 0 && outputs == 0, ""); break;
case Bytecodes::_drem: assert(inputs == 4 && outputs == 2, ""); break;
}
#endif //ASSERT
return true;
}
//------------------------------basic_plus_adr---------------------------------
Node* GraphKit::basic_plus_adr(Node* base, Node* ptr, Node* offset) {
// short-circuit a common case
if (offset == intcon(0)) return ptr;
return _gvn.transform( new AddPNode(base, ptr, offset) );
}
Node* GraphKit::ConvI2L(Node* offset) {
// short-circuit a common case
jint offset_con = find_int_con(offset, Type::OffsetBot);
if (offset_con != Type::OffsetBot) {
return longcon((jlong) offset_con);
}
return _gvn.transform( new ConvI2LNode(offset));
}
Node* GraphKit::ConvI2UL(Node* offset) {
juint offset_con = (juint) find_int_con(offset, Type::OffsetBot);
if (offset_con != (juint) Type::OffsetBot) {
return longcon((julong) offset_con);
}
Node* conv = _gvn.transform( new ConvI2LNode(offset));
Node* mask = _gvn.transform(ConLNode::make((julong) max_juint));
return _gvn.transform( new AndLNode(conv, mask) );
}
Node* GraphKit::ConvL2I(Node* offset) {
// short-circuit a common case
jlong offset_con = find_long_con(offset, (jlong)Type::OffsetBot);
if (offset_con != (jlong)Type::OffsetBot) {
return intcon((int) offset_con);
}
return _gvn.transform( new ConvL2INode(offset));
}
//-------------------------load_object_klass-----------------------------------
Node* GraphKit::load_object_klass(Node* obj) {
// Special-case a fresh allocation to avoid building nodes:
Node* akls = AllocateNode::Ideal_klass(obj, &_gvn);
if (akls != NULL) return akls;
Node* k_adr = basic_plus_adr(obj, oopDesc::klass_offset_in_bytes());
return _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), k_adr, TypeInstPtr::KLASS));
}
//-------------------------load_array_length-----------------------------------
Node* GraphKit::load_array_length(Node* array) {
// Special-case a fresh allocation to avoid building nodes:
AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(array, &_gvn);
Node *alen;
if (alloc == NULL) {
Node *r_adr = basic_plus_adr(array, arrayOopDesc::length_offset_in_bytes());
alen = _gvn.transform( new LoadRangeNode(0, immutable_memory(), r_adr, TypeInt::POS));
} else {
alen = alloc->Ideal_length();
Node* ccast = alloc->make_ideal_length(_gvn.type(array)->is_oopptr(), &_gvn);
if (ccast != alen) {
alen = _gvn.transform(ccast);
}
}
return alen;
}
//------------------------------do_null_check----------------------------------
// Helper function to do a NULL pointer check. Returned value is
// the incoming address with NULL casted away. You are allowed to use the
// not-null value only if you are control dependent on the test.
extern int explicit_null_checks_inserted,
explicit_null_checks_elided;
Node* GraphKit::null_check_common(Node* value, BasicType type,
// optional arguments for variations:
bool assert_null,
Node* *null_control,
bool speculative) {
assert(!assert_null || null_control == NULL, "not both at once");
if (stopped()) return top();
if (!GenerateCompilerNullChecks && !assert_null && null_control == NULL) {
// For some performance testing, we may wish to suppress null checking.
value = cast_not_null(value); // Make it appear to be non-null (4962416).
return value;
}
explicit_null_checks_inserted++;
// Construct NULL check
Node *chk = NULL;
switch(type) {
case T_LONG : chk = new CmpLNode(value, _gvn.zerocon(T_LONG)); break;
case T_INT : chk = new CmpINode(value, _gvn.intcon(0)); break;
case T_ARRAY : // fall through
type = T_OBJECT; // simplify further tests
case T_OBJECT : {
const Type *t = _gvn.type( value );
const TypeOopPtr* tp = t->isa_oopptr();
if (tp != NULL && tp->klass() != NULL && !tp->klass()->is_loaded()
// Only for do_null_check, not any of its siblings:
&& !assert_null && null_control == NULL) {
// Usually, any field access or invocation on an unloaded oop type
// will simply fail to link, since the statically linked class is
// likely also to be unloaded. However, in -Xcomp mode, sometimes
// the static class is loaded but the sharper oop type is not.
// Rather than checking for this obscure case in lots of places,
// we simply observe that a null check on an unloaded class
// will always be followed by a nonsense operation, so we
// can just issue the uncommon trap here.
// Our access to the unloaded class will only be correct
// after it has been loaded and initialized, which requires
// a trip through the interpreter.
#ifndef PRODUCT
if (WizardMode) { tty->print("Null check of unloaded "); tp->klass()->print(); tty->cr(); }
#endif
uncommon_trap(Deoptimization::Reason_unloaded,
Deoptimization::Action_reinterpret,
tp->klass(), "!loaded");
return top();
}
if (assert_null) {
// See if the type is contained in NULL_PTR.
// If so, then the value is already null.
if (t->higher_equal(TypePtr::NULL_PTR)) {
explicit_null_checks_elided++;
return value; // Elided null assert quickly!
}
} else {
// See if mixing in the NULL pointer changes type.
// If so, then the NULL pointer was not allowed in the original
// type. In other words, "value" was not-null.
if (t->meet(TypePtr::NULL_PTR) != t->remove_speculative()) {
// same as: if (!TypePtr::NULL_PTR->higher_equal(t)) ...
explicit_null_checks_elided++;
return value; // Elided null check quickly!
}
}
chk = new CmpPNode( value, null() );
break;
}
default:
fatal(err_msg_res("unexpected type: %s", type2name(type)));
}
assert(chk != NULL, "sanity check");
chk = _gvn.transform(chk);
BoolTest::mask btest = assert_null ? BoolTest::eq : BoolTest::ne;
BoolNode *btst = new BoolNode( chk, btest);
Node *tst = _gvn.transform( btst );
//-----------
// if peephole optimizations occurred, a prior test existed.
// If a prior test existed, maybe it dominates as we can avoid this test.
if (tst != btst && type == T_OBJECT) {
// At this point we want to scan up the CFG to see if we can
// find an identical test (and so avoid this test altogether).
Node *cfg = control();
int depth = 0;
while( depth < 16 ) { // Limit search depth for speed
if( cfg->Opcode() == Op_IfTrue &&
cfg->in(0)->in(1) == tst ) {
// Found prior test. Use "cast_not_null" to construct an identical
// CastPP (and hence hash to) as already exists for the prior test.
// Return that casted value.
if (assert_null) {
replace_in_map(value, null());
return null(); // do not issue the redundant test
}
Node *oldcontrol = control();
set_control(cfg);
Node *res = cast_not_null(value);
set_control(oldcontrol);
explicit_null_checks_elided++;
return res;
}
cfg = IfNode::up_one_dom(cfg, /*linear_only=*/ true);
if (cfg == NULL) break; // Quit at region nodes
depth++;
}
}
//-----------
// Branch to failure if null
float ok_prob = PROB_MAX; // a priori estimate: nulls never happen
Deoptimization::DeoptReason reason;
if (assert_null) {
reason = Deoptimization::Reason_null_assert;
} else if (type == T_OBJECT) {
reason = Deoptimization::reason_null_check(speculative);
} else {
reason = Deoptimization::Reason_div0_check;
}
// %%% Since Reason_unhandled is not recorded on a per-bytecode basis,
// ciMethodData::has_trap_at will return a conservative -1 if any
// must-be-null assertion has failed. This could cause performance
// problems for a method after its first do_null_assert failure.
// Consider using 'Reason_class_check' instead?
// To cause an implicit null check, we set the not-null probability
// to the maximum (PROB_MAX). For an explicit check the probability
// is set to a smaller value.
if (null_control != NULL || too_many_traps(reason)) {
// probability is less likely
ok_prob = PROB_LIKELY_MAG(3);
} else if (!assert_null &&
(ImplicitNullCheckThreshold > 0) &&
method() != NULL &&
(method()->method_data()->trap_count(reason)
>= (uint)ImplicitNullCheckThreshold)) {
ok_prob = PROB_LIKELY_MAG(3);
}
if (null_control != NULL) {
IfNode* iff = create_and_map_if(control(), tst, ok_prob, COUNT_UNKNOWN);
Node* null_true = _gvn.transform( new IfFalseNode(iff));
set_control( _gvn.transform( new IfTrueNode(iff)));
if (null_true == top())
explicit_null_checks_elided++;
(*null_control) = null_true;
} else {
BuildCutout unless(this, tst, ok_prob);
// Check for optimizer eliding test at parse time
if (stopped()) {
// Failure not possible; do not bother making uncommon trap.
explicit_null_checks_elided++;
} else if (assert_null) {
uncommon_trap(reason,
Deoptimization::Action_make_not_entrant,
NULL, "assert_null");
} else {
replace_in_map(value, zerocon(type));
builtin_throw(reason);
}
}
// Must throw exception, fall-thru not possible?
if (stopped()) {
return top(); // No result
}
if (assert_null) {
// Cast obj to null on this path.
replace_in_map(value, zerocon(type));
return zerocon(type);
}
// Cast obj to not-null on this path, if there is no null_control.
// (If there is a null_control, a non-null value may come back to haunt us.)
if (type == T_OBJECT) {
Node* cast = cast_not_null(value, false);
if (null_control == NULL || (*null_control) == top())
replace_in_map(value, cast);
value = cast;
}
return value;
}
//------------------------------cast_not_null----------------------------------
// Cast obj to not-null on this path
Node* GraphKit::cast_not_null(Node* obj, bool do_replace_in_map) {
const Type *t = _gvn.type(obj);
const Type *t_not_null = t->join_speculative(TypePtr::NOTNULL);
// Object is already not-null?
if( t == t_not_null ) return obj;
Node *cast = new CastPPNode(obj,t_not_null);
cast->init_req(0, control());
cast = _gvn.transform( cast );
// Scan for instances of 'obj' in the current JVM mapping.
// These instances are known to be not-null after the test.
if (do_replace_in_map)
replace_in_map(obj, cast);
return cast; // Return casted value
}
//--------------------------replace_in_map-------------------------------------
void GraphKit::replace_in_map(Node* old, Node* neww) {
if (old == neww) {
return;
}
map()->replace_edge(old, neww);
// Note: This operation potentially replaces any edge
// on the map. This includes locals, stack, and monitors
// of the current (innermost) JVM state.
// don't let inconsistent types from profiling escape this
// method
const Type* told = _gvn.type(old);
const Type* tnew = _gvn.type(neww);
if (!tnew->higher_equal(told)) {
return;
}
map()->record_replaced_node(old, neww);
}
//=============================================================================
//--------------------------------memory---------------------------------------
Node* GraphKit::memory(uint alias_idx) {
MergeMemNode* mem = merged_memory();
Node* p = mem->memory_at(alias_idx);
_gvn.set_type(p, Type::MEMORY); // must be mapped
return p;
}
//-----------------------------reset_memory------------------------------------
Node* GraphKit::reset_memory() {
Node* mem = map()->memory();
// do not use this node for any more parsing!
debug_only( map()->set_memory((Node*)NULL) );
return _gvn.transform( mem );
}
//------------------------------set_all_memory---------------------------------
void GraphKit::set_all_memory(Node* newmem) {
Node* mergemem = MergeMemNode::make(newmem);
gvn().set_type_bottom(mergemem);
map()->set_memory(mergemem);
}
//------------------------------set_all_memory_call----------------------------
void GraphKit::set_all_memory_call(Node* call, bool separate_io_proj) {
Node* newmem = _gvn.transform( new ProjNode(call, TypeFunc::Memory, separate_io_proj) );
set_all_memory(newmem);
}
//=============================================================================
//
// parser factory methods for MemNodes
//
// These are layered on top of the factory methods in LoadNode and StoreNode,
// and integrate with the parser's memory state and _gvn engine.
//
// factory methods in "int adr_idx"
Node* GraphKit::make_load(Node* ctl, Node* adr, const Type* t, BasicType bt,
int adr_idx,
MemNode::MemOrd mo, bool require_atomic_access) {
assert(adr_idx != Compile::AliasIdxTop, "use other make_load factory" );
const TypePtr* adr_type = NULL; // debug-mode-only argument
debug_only(adr_type = C->get_adr_type(adr_idx));
Node* mem = memory(adr_idx);
Node* ld;
if (require_atomic_access && bt == T_LONG) {
ld = LoadLNode::make_atomic(ctl, mem, adr, adr_type, t, mo);
} else if (require_atomic_access && bt == T_DOUBLE) {
ld = LoadDNode::make_atomic(ctl, mem, adr, adr_type, t, mo);
} else {
ld = LoadNode::make(_gvn, ctl, mem, adr, adr_type, t, bt, mo);
}
ld = _gvn.transform(ld);
if ((bt == T_OBJECT) && C->do_escape_analysis() || C->eliminate_boxing()) {
// Improve graph before escape analysis and boxing elimination.
record_for_igvn(ld);
}
return ld;
}
Node* GraphKit::store_to_memory(Node* ctl, Node* adr, Node *val, BasicType bt,
int adr_idx,
MemNode::MemOrd mo,
bool require_atomic_access) {
assert(adr_idx != Compile::AliasIdxTop, "use other store_to_memory factory" );
const TypePtr* adr_type = NULL;
debug_only(adr_type = C->get_adr_type(adr_idx));
Node *mem = memory(adr_idx);
Node* st;
if (require_atomic_access && bt == T_LONG) {
st = StoreLNode::make_atomic(ctl, mem, adr, adr_type, val, mo);
} else if (require_atomic_access && bt == T_DOUBLE) {
st = StoreDNode::make_atomic(ctl, mem, adr, adr_type, val, mo);
} else {
st = StoreNode::make(_gvn, ctl, mem, adr, adr_type, val, bt, mo);
}
st = _gvn.transform(st);
set_memory(st, adr_idx);
// Back-to-back stores can only remove intermediate store with DU info
// so push on worklist for optimizer.
if (mem->req() > MemNode::Address && adr == mem->in(MemNode::Address))
record_for_igvn(st);
return st;
}
void GraphKit::pre_barrier(bool do_load,
Node* ctl,
Node* obj,
Node* adr,
uint adr_idx,
Node* val,
const TypeOopPtr* val_type,
Node* pre_val,
BasicType bt) {
BarrierSet* bs = Universe::heap()->barrier_set();
set_control(ctl);
switch (bs->kind()) {
case BarrierSet::G1SATBCT:
case BarrierSet::G1SATBCTLogging:
g1_write_barrier_pre(do_load, obj, adr, adr_idx, val, val_type, pre_val, bt);
break;
case BarrierSet::CardTableModRef:
case BarrierSet::CardTableExtension:
case BarrierSet::ModRef:
break;
default :
ShouldNotReachHere();
}
}
bool GraphKit::can_move_pre_barrier() const {
BarrierSet* bs = Universe::heap()->barrier_set();
switch (bs->kind()) {
case BarrierSet::G1SATBCT:
case BarrierSet::G1SATBCTLogging:
return true; // Can move it if no safepoint
case BarrierSet::CardTableModRef:
case BarrierSet::CardTableExtension:
case BarrierSet::ModRef:
return true; // There is no pre-barrier
default :
ShouldNotReachHere();
}
return false;
}
void GraphKit::post_barrier(Node* ctl,
Node* store,
Node* obj,
Node* adr,
uint adr_idx,
Node* val,
BasicType bt,
bool use_precise) {
BarrierSet* bs = Universe::heap()->barrier_set();
set_control(ctl);
switch (bs->kind()) {
case BarrierSet::G1SATBCT:
case BarrierSet::G1SATBCTLogging:
g1_write_barrier_post(store, obj, adr, adr_idx, val, bt, use_precise);
break;
case BarrierSet::CardTableModRef:
case BarrierSet::CardTableExtension:
write_barrier_post(store, obj, adr, adr_idx, val, use_precise);
break;
case BarrierSet::ModRef:
break;
default :
ShouldNotReachHere();
}
}
Node* GraphKit::store_oop(Node* ctl,
Node* obj,
Node* adr,
const TypePtr* adr_type,
Node* val,
const TypeOopPtr* val_type,
BasicType bt,
bool use_precise,
MemNode::MemOrd mo) {
// Transformation of a value which could be NULL pointer (CastPP #NULL)
// could be delayed during Parse (for example, in adjust_map_after_if()).
// Execute transformation here to avoid barrier generation in such case.
if (_gvn.type(val) == TypePtr::NULL_PTR)
val = _gvn.makecon(TypePtr::NULL_PTR);
set_control(ctl);
if (stopped()) return top(); // Dead path ?
assert(bt == T_OBJECT, "sanity");
assert(val != NULL, "not dead path");
uint adr_idx = C->get_alias_index(adr_type);
assert(adr_idx != Compile::AliasIdxTop, "use other store_to_memory factory" );
pre_barrier(true /* do_load */,
control(), obj, adr, adr_idx, val, val_type,
NULL /* pre_val */,
bt);
Node* store = store_to_memory(control(), adr, val, bt, adr_idx, mo);
post_barrier(control(), store, obj, adr, adr_idx, val, bt, use_precise);
return store;
}
// Could be an array or object we don't know at compile time (unsafe ref.)
Node* GraphKit::store_oop_to_unknown(Node* ctl,
Node* obj, // containing obj
Node* adr, // actual adress to store val at
const TypePtr* adr_type,
Node* val,
BasicType bt,
MemNode::MemOrd mo) {
Compile::AliasType* at = C->alias_type(adr_type);
const TypeOopPtr* val_type = NULL;
if (adr_type->isa_instptr()) {
if (at->field() != NULL) {
// known field. This code is a copy of the do_put_xxx logic.
ciField* field = at->field();
if (!field->type()->is_loaded()) {
val_type = TypeInstPtr::BOTTOM;
} else {
val_type = TypeOopPtr::make_from_klass(field->type()->as_klass());
}
}
} else if (adr_type->isa_aryptr()) {
val_type = adr_type->is_aryptr()->elem()->make_oopptr();
}
if (val_type == NULL) {
val_type = TypeInstPtr::BOTTOM;
}
return store_oop(ctl, obj, adr, adr_type, val, val_type, bt, true, mo);
}
//-------------------------array_element_address-------------------------
Node* GraphKit::array_element_address(Node* ary, Node* idx, BasicType elembt,
const TypeInt* sizetype) {
uint shift = exact_log2(type2aelembytes(elembt));
uint header = arrayOopDesc::base_offset_in_bytes(elembt);
// short-circuit a common case (saves lots of confusing waste motion)
jint idx_con = find_int_con(idx, -1);
if (idx_con >= 0) {
intptr_t offset = header + ((intptr_t)idx_con << shift);
return basic_plus_adr(ary, offset);
}
// must be correct type for alignment purposes
Node* base = basic_plus_adr(ary, header);
idx = Compile::conv_I2X_index(&_gvn, idx, sizetype);
Node* scale = _gvn.transform( new LShiftXNode(idx, intcon(shift)) );
return basic_plus_adr(ary, base, scale);
}
//-------------------------load_array_element-------------------------
Node* GraphKit::load_array_element(Node* ctl, Node* ary, Node* idx, const TypeAryPtr* arytype) {
const Type* elemtype = arytype->elem();
BasicType elembt = elemtype->array_element_basic_type();
Node* adr = array_element_address(ary, idx, elembt, arytype->size());
Node* ld = make_load(ctl, adr, elemtype, elembt, arytype, MemNode::unordered);
return ld;
}
//-------------------------set_arguments_for_java_call-------------------------
// Arguments (pre-popped from the stack) are taken from the JVMS.
void GraphKit::set_arguments_for_java_call(CallJavaNode* call) {
// Add the call arguments:
uint nargs = call->method()->arg_size();
for (uint i = 0; i < nargs; i++) {
Node* arg = argument(i);
call->init_req(i + TypeFunc::Parms, arg);
}
}
//---------------------------set_edges_for_java_call---------------------------
// Connect a newly created call into the current JVMS.
// A return value node (if any) is returned from set_edges_for_java_call.
void GraphKit::set_edges_for_java_call(CallJavaNode* call, bool must_throw, bool separate_io_proj) {
// Add the predefined inputs:
call->init_req( TypeFunc::Control, control() );
call->init_req( TypeFunc::I_O , i_o() );
call->init_req( TypeFunc::Memory , reset_memory() );
call->init_req( TypeFunc::FramePtr, frameptr() );
call->init_req( TypeFunc::ReturnAdr, top() );
add_safepoint_edges(call, must_throw);
Node* xcall = _gvn.transform(call);
if (xcall == top()) {
set_control(top());
return;
}
assert(xcall == call, "call identity is stable");
// Re-use the current map to produce the result.
set_control(_gvn.transform(new ProjNode(call, TypeFunc::Control)));
set_i_o( _gvn.transform(new ProjNode(call, TypeFunc::I_O , separate_io_proj)));
set_all_memory_call(xcall, separate_io_proj);
//return xcall; // no need, caller already has it
}
Node* GraphKit::set_results_for_java_call(CallJavaNode* call, bool separate_io_proj) {
if (stopped()) return top(); // maybe the call folded up?
// Capture the return value, if any.
Node* ret;
if (call->method() == NULL ||
call->method()->return_type()->basic_type() == T_VOID)
ret = top();
else ret = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
// Note: Since any out-of-line call can produce an exception,
// we always insert an I_O projection from the call into the result.
make_slow_call_ex(call, env()->Throwable_klass(), separate_io_proj);
if (separate_io_proj) {
// The caller requested separate projections be used by the fall
// through and exceptional paths, so replace the projections for
// the fall through path.
set_i_o(_gvn.transform( new ProjNode(call, TypeFunc::I_O) ));
set_all_memory(_gvn.transform( new ProjNode(call, TypeFunc::Memory) ));
}
return ret;
}
//--------------------set_predefined_input_for_runtime_call--------------------
// Reading and setting the memory state is way conservative here.
// The real problem is that I am not doing real Type analysis on memory,
// so I cannot distinguish card mark stores from other stores. Across a GC
// point the Store Barrier and the card mark memory has to agree. I cannot
// have a card mark store and its barrier split across the GC point from
// either above or below. Here I get that to happen by reading ALL of memory.
// A better answer would be to separate out card marks from other memory.
// For now, return the input memory state, so that it can be reused
// after the call, if this call has restricted memory effects.
Node* GraphKit::set_predefined_input_for_runtime_call(SafePointNode* call) {
// Set fixed predefined input arguments
Node* memory = reset_memory();
call->init_req( TypeFunc::Control, control() );
call->init_req( TypeFunc::I_O, top() ); // does no i/o
call->init_req( TypeFunc::Memory, memory ); // may gc ptrs
call->init_req( TypeFunc::FramePtr, frameptr() );
call->init_req( TypeFunc::ReturnAdr, top() );
return memory;
}
//-------------------set_predefined_output_for_runtime_call--------------------
// Set control and memory (not i_o) from the call.
// If keep_mem is not NULL, use it for the output state,
// except for the RawPtr output of the call, if hook_mem is TypeRawPtr::BOTTOM.
// If hook_mem is NULL, this call produces no memory effects at all.
// If hook_mem is a Java-visible memory slice (such as arraycopy operands),
// then only that memory slice is taken from the call.
// In the last case, we must put an appropriate memory barrier before
// the call, so as to create the correct anti-dependencies on loads
// preceding the call.
void GraphKit::set_predefined_output_for_runtime_call(Node* call,
Node* keep_mem,
const TypePtr* hook_mem) {
// no i/o
set_control(_gvn.transform( new ProjNode(call,TypeFunc::Control) ));
if (keep_mem) {
// First clone the existing memory state
set_all_memory(keep_mem);
if (hook_mem != NULL) {
// Make memory for the call
Node* mem = _gvn.transform( new ProjNode(call, TypeFunc::Memory) );
// Set the RawPtr memory state only. This covers all the heap top/GC stuff
// We also use hook_mem to extract specific effects from arraycopy stubs.
set_memory(mem, hook_mem);
}
// ...else the call has NO memory effects.
// Make sure the call advertises its memory effects precisely.
// This lets us build accurate anti-dependences in gcm.cpp.
assert(C->alias_type(call->adr_type()) == C->alias_type(hook_mem),
"call node must be constructed correctly");
} else {
assert(hook_mem == NULL, "");
// This is not a "slow path" call; all memory comes from the call.
set_all_memory_call(call);
}
}
// Replace the call with the current state of the kit.
void GraphKit::replace_call(CallNode* call, Node* result, bool do_replaced_nodes) {
JVMState* ejvms = NULL;
if (has_exceptions()) {
ejvms = transfer_exceptions_into_jvms();
}
ReplacedNodes replaced_nodes = map()->replaced_nodes();
ReplacedNodes replaced_nodes_exception;
Node* ex_ctl = top();
SafePointNode* final_state = stop();
// Find all the needed outputs of this call
CallProjections callprojs;
call->extract_projections(&callprojs, true);
Node* init_mem = call->in(TypeFunc::Memory);
Node* final_mem = final_state->in(TypeFunc::Memory);
Node* final_ctl = final_state->in(TypeFunc::Control);
Node* final_io = final_state->in(TypeFunc::I_O);
// Replace all the old call edges with the edges from the inlining result
if (callprojs.fallthrough_catchproj != NULL) {
C->gvn_replace_by(callprojs.fallthrough_catchproj, final_ctl);
}
if (callprojs.fallthrough_memproj != NULL) {
if (final_mem->is_MergeMem()) {
// Parser's exits MergeMem was not transformed but may be optimized
final_mem = _gvn.transform(final_mem);
}
C->gvn_replace_by(callprojs.fallthrough_memproj, final_mem);
}
if (callprojs.fallthrough_ioproj != NULL) {
C->gvn_replace_by(callprojs.fallthrough_ioproj, final_io);
}
// Replace the result with the new result if it exists and is used
if (callprojs.resproj != NULL && result != NULL) {
C->gvn_replace_by(callprojs.resproj, result);
}
if (ejvms == NULL) {
// No exception edges to simply kill off those paths
if (callprojs.catchall_catchproj != NULL) {
C->gvn_replace_by(callprojs.catchall_catchproj, C->top());
}
if (callprojs.catchall_memproj != NULL) {
C->gvn_replace_by(callprojs.catchall_memproj, C->top());
}
if (callprojs.catchall_ioproj != NULL) {
C->gvn_replace_by(callprojs.catchall_ioproj, C->top());
}
// Replace the old exception object with top
if (callprojs.exobj != NULL) {
C->gvn_replace_by(callprojs.exobj, C->top());
}
} else {
GraphKit ekit(ejvms);
// Load my combined exception state into the kit, with all phis transformed:
SafePointNode* ex_map = ekit.combine_and_pop_all_exception_states();
replaced_nodes_exception = ex_map->replaced_nodes();
Node* ex_oop = ekit.use_exception_state(ex_map);
if (callprojs.catchall_catchproj != NULL) {
C->gvn_replace_by(callprojs.catchall_catchproj, ekit.control());
ex_ctl = ekit.control();
}
if (callprojs.catchall_memproj != NULL) {
C->gvn_replace_by(callprojs.catchall_memproj, ekit.reset_memory());
}
if (callprojs.catchall_ioproj != NULL) {
C->gvn_replace_by(callprojs.catchall_ioproj, ekit.i_o());
}
// Replace the old exception object with the newly created one
if (callprojs.exobj != NULL) {
C->gvn_replace_by(callprojs.exobj, ex_oop);
}
}
// Disconnect the call from the graph
call->disconnect_inputs(NULL, C);
C->gvn_replace_by(call, C->top());
// Clean up any MergeMems that feed other MergeMems since the
// optimizer doesn't like that.
if (final_mem->is_MergeMem()) {
Node_List wl;
for (SimpleDUIterator i(final_mem); i.has_next(); i.next()) {
Node* m = i.get();
if (m->is_MergeMem() && !wl.contains(m)) {
wl.push(m);
}
}
while (wl.size() > 0) {
_gvn.transform(wl.pop());
}
}
if (callprojs.fallthrough_catchproj != NULL && !final_ctl->is_top() && do_replaced_nodes) {
replaced_nodes.apply(C, final_ctl);
}
if (!ex_ctl->is_top() && do_replaced_nodes) {
replaced_nodes_exception.apply(C, ex_ctl);
}
}
//------------------------------increment_counter------------------------------
// for statistics: increment a VM counter by 1
void GraphKit::increment_counter(address counter_addr) {
Node* adr1 = makecon(TypeRawPtr::make(counter_addr));
increment_counter(adr1);
}
void GraphKit::increment_counter(Node* counter_addr) {
int adr_type = Compile::AliasIdxRaw;
Node* ctrl = control();
Node* cnt = make_load(ctrl, counter_addr, TypeInt::INT, T_INT, adr_type, MemNode::unordered);
Node* incr = _gvn.transform(new AddINode(cnt, _gvn.intcon(1)));
store_to_memory(ctrl, counter_addr, incr, T_INT, adr_type, MemNode::unordered);
}
//------------------------------uncommon_trap----------------------------------
// Bail out to the interpreter in mid-method. Implemented by calling the
// uncommon_trap blob. This helper function inserts a runtime call with the
// right debug info.
void GraphKit::uncommon_trap(int trap_request,
ciKlass* klass, const char* comment,
bool must_throw,
bool keep_exact_action) {
if (failing()) stop();
if (stopped()) return; // trap reachable?
// Note: If ProfileTraps is true, and if a deopt. actually
// occurs here, the runtime will make sure an MDO exists. There is
// no need to call method()->ensure_method_data() at this point.
// Set the stack pointer to the right value for reexecution:
set_sp(reexecute_sp());
#ifdef ASSERT
if (!must_throw) {
// Make sure the stack has at least enough depth to execute
// the current bytecode.
int inputs, ignored_depth;
if (compute_stack_effects(inputs, ignored_depth)) {
assert(sp() >= inputs, err_msg_res("must have enough JVMS stack to execute %s: sp=%d, inputs=%d",
Bytecodes::name(java_bc()), sp(), inputs));
}
}
#endif
Deoptimization::DeoptReason reason = Deoptimization::trap_request_reason(trap_request);
Deoptimization::DeoptAction action = Deoptimization::trap_request_action(trap_request);
switch (action) {
case Deoptimization::Action_maybe_recompile:
case Deoptimization::Action_reinterpret:
// Temporary fix for 6529811 to allow virtual calls to be sure they
// get the chance to go from mono->bi->mega
if (!keep_exact_action &&
Deoptimization::trap_request_index(trap_request) < 0 &&
too_many_recompiles(reason)) {
// This BCI is causing too many recompilations.
if (C->log() != NULL) {
C->log()->elem("observe that='trap_action_change' reason='%s' from='%s' to='none'",
Deoptimization::trap_reason_name(reason),
Deoptimization::trap_action_name(action));
}
action = Deoptimization::Action_none;
trap_request = Deoptimization::make_trap_request(reason, action);
} else {
C->set_trap_can_recompile(true);
}
break;
case Deoptimization::Action_make_not_entrant:
C->set_trap_can_recompile(true);
break;
#ifdef ASSERT
case Deoptimization::Action_none:
case Deoptimization::Action_make_not_compilable:
break;
default:
fatal(err_msg_res("unknown action %d: %s", action, Deoptimization::trap_action_name(action)));
break;
#endif
}
if (TraceOptoParse) {
char buf[100];
tty->print_cr("Uncommon trap %s at bci:%d",
Deoptimization::format_trap_request(buf, sizeof(buf),
trap_request), bci());
}
CompileLog* log = C->log();
if (log != NULL) {
int kid = (klass == NULL)? -1: log->identify(klass);
log->begin_elem("uncommon_trap bci='%d'", bci());
char buf[100];
log->print(" %s", Deoptimization::format_trap_request(buf, sizeof(buf),
trap_request));
if (kid >= 0) log->print(" klass='%d'", kid);
if (comment != NULL) log->print(" comment='%s'", comment);
log->end_elem();
}
// Make sure any guarding test views this path as very unlikely
Node *i0 = control()->in(0);
if (i0 != NULL && i0->is_If()) { // Found a guarding if test?
IfNode *iff = i0->as_If();
float f = iff->_prob; // Get prob
if (control()->Opcode() == Op_IfTrue) {
if (f > PROB_UNLIKELY_MAG(4))
iff->_prob = PROB_MIN;
} else {
if (f < PROB_LIKELY_MAG(4))
iff->_prob = PROB_MAX;
}
}
// Clear out dead values from the debug info.
kill_dead_locals();
// Now insert the uncommon trap subroutine call
address call_addr = SharedRuntime::uncommon_trap_blob()->entry_point();
const TypePtr* no_memory_effects = NULL;
// Pass the index of the class to be loaded
Node* call = make_runtime_call(RC_NO_LEAF | RC_UNCOMMON |
(must_throw ? RC_MUST_THROW : 0),
OptoRuntime::uncommon_trap_Type(),
call_addr, "uncommon_trap", no_memory_effects,
intcon(trap_request));
assert(call->as_CallStaticJava()->uncommon_trap_request() == trap_request,
"must extract request correctly from the graph");
assert(trap_request != 0, "zero value reserved by uncommon_trap_request");
call->set_req(TypeFunc::ReturnAdr, returnadr());
// The debug info is the only real input to this call.
// Halt-and-catch fire here. The above call should never return!
HaltNode* halt = new HaltNode(control(), frameptr());
_gvn.set_type_bottom(halt);
root()->add_req(halt);
stop_and_kill_map();
}
//--------------------------just_allocated_object------------------------------
// Report the object that was just allocated.
// It must be the case that there are no intervening safepoints.
// We use this to determine if an object is so "fresh" that
// it does not require card marks.
Node* GraphKit::just_allocated_object(Node* current_control) {
if (C->recent_alloc_ctl() == current_control)
return C->recent_alloc_obj();
return NULL;
}
void GraphKit::round_double_arguments(ciMethod* dest_method) {
// (Note: TypeFunc::make has a cache that makes this fast.)
const TypeFunc* tf = TypeFunc::make(dest_method);
int nargs = tf->domain()->cnt() - TypeFunc::Parms;
for (int j = 0; j < nargs; j++) {
const Type *targ = tf->domain()->field_at(j + TypeFunc::Parms);
if( targ->basic_type() == T_DOUBLE ) {
// If any parameters are doubles, they must be rounded before
// the call, dstore_rounding does gvn.transform
Node *arg = argument(j);
arg = dstore_rounding(arg);
set_argument(j, arg);
}
}
}
/**
* Record profiling data exact_kls for Node n with the type system so
* that it can propagate it (speculation)
*
* @param n node that the type applies to
* @param exact_kls type from profiling
* @param maybe_null did profiling see null?
*
* @return node with improved type
*/
Node* GraphKit::record_profile_for_speculation(Node* n, ciKlass* exact_kls, bool maybe_null) {
const Type* current_type = _gvn.type(n);
assert(UseTypeSpeculation, "type speculation must be on");
const TypePtr* speculative = current_type->speculative();
// Should the klass from the profile be recorded in the speculative type?
if (current_type->would_improve_type(exact_kls, jvms()->depth())) {
const TypeKlassPtr* tklass = TypeKlassPtr::make(exact_kls);
const TypeOopPtr* xtype = tklass->as_instance_type();
assert(xtype->klass_is_exact(), "Should be exact");
// Any reason to believe n is not null (from this profiling or a previous one)?
const TypePtr* ptr = (maybe_null && current_type->speculative_maybe_null()) ? TypePtr::BOTTOM : TypePtr::NOTNULL;
// record the new speculative type's depth
speculative = xtype->cast_to_ptr_type(ptr->ptr())->is_ptr();
speculative = speculative->with_inline_depth(jvms()->depth());
} else if (current_type->would_improve_ptr(maybe_null)) {
// Profiling report that null was never seen so we can change the
// speculative type to non null ptr.
assert(!maybe_null, "nothing to improve");
if (speculative == NULL) {
speculative = TypePtr::NOTNULL;
} else {
const TypePtr* ptr = TypePtr::NOTNULL;
speculative = speculative->cast_to_ptr_type(ptr->ptr())->is_ptr();
}
}
if (speculative != current_type->speculative()) {
// Build a type with a speculative type (what we think we know
// about the type but will need a guard when we use it)
const TypeOopPtr* spec_type = TypeOopPtr::make(TypePtr::BotPTR, Type::OffsetBot, TypeOopPtr::InstanceBot, speculative);
// We're changing the type, we need a new CheckCast node to carry
// the new type. The new type depends on the control: what
// profiling tells us is only valid from here as far as we can
// tell.
Node* cast = new CheckCastPPNode(control(), n, current_type->remove_speculative()->join_speculative(spec_type));
cast = _gvn.transform(cast);
replace_in_map(n, cast);
n = cast;
}
return n;
}
/**
* Record profiling data from receiver profiling at an invoke with the
* type system so that it can propagate it (speculation)
*
* @param n receiver node
*
* @return node with improved type
*/
Node* GraphKit::record_profiled_receiver_for_speculation(Node* n) {
if (!UseTypeSpeculation) {
return n;
}
ciKlass* exact_kls = profile_has_unique_klass();
bool maybe_null = true;
if (java_bc() == Bytecodes::_checkcast ||
java_bc() == Bytecodes::_instanceof ||
java_bc() == Bytecodes::_aastore) {
ciProfileData* data = method()->method_data()->bci_to_data(bci());
bool maybe_null = data == NULL ? true : data->as_BitData()->null_seen();
}
return record_profile_for_speculation(n, exact_kls, maybe_null);
return n;
}
/**
* Record profiling data from argument profiling at an invoke with the
* type system so that it can propagate it (speculation)
*
* @param dest_method target method for the call
* @param bc what invoke bytecode is this?
*/
void GraphKit::record_profiled_arguments_for_speculation(ciMethod* dest_method, Bytecodes::Code bc) {
if (!UseTypeSpeculation) {
return;
}
const TypeFunc* tf = TypeFunc::make(dest_method);
int nargs = tf->domain()->cnt() - TypeFunc::Parms;
int skip = Bytecodes::has_receiver(bc) ? 1 : 0;
for (int j = skip, i = 0; j < nargs && i < TypeProfileArgsLimit; j++) {
const Type *targ = tf->domain()->field_at(j + TypeFunc::Parms);
if (targ->basic_type() == T_OBJECT || targ->basic_type() == T_ARRAY) {
bool maybe_null = true;
ciKlass* better_type = NULL;
if (method()->argument_profiled_type(bci(), i, better_type, maybe_null)) {
record_profile_for_speculation(argument(j), better_type, maybe_null);
}
i++;
}
}
}
/**
* Record profiling data from parameter profiling at an invoke with
* the type system so that it can propagate it (speculation)
*/
void GraphKit::record_profiled_parameters_for_speculation() {
if (!UseTypeSpeculation) {
return;
}
for (int i = 0, j = 0; i < method()->arg_size() ; i++) {
if (_gvn.type(local(i))->isa_oopptr()) {
bool maybe_null = true;
ciKlass* better_type = NULL;
if (method()->parameter_profiled_type(j, better_type, maybe_null)) {
record_profile_for_speculation(local(i), better_type, maybe_null);
}
j++;
}
}
}
/**
* Record profiling data from return value profiling at an invoke with
* the type system so that it can propagate it (speculation)
*/
void GraphKit::record_profiled_return_for_speculation() {
if (!UseTypeSpeculation) {
return;
}
bool maybe_null = true;
ciKlass* better_type = NULL;
if (method()->return_profiled_type(bci(), better_type, maybe_null)) {
// If profiling reports a single type for the return value,
// feed it to the type system so it can propagate it as a
// speculative type
record_profile_for_speculation(stack(sp()-1), better_type, maybe_null);
}
}
void GraphKit::round_double_result(ciMethod* dest_method) {
// A non-strict method may return a double value which has an extended
// exponent, but this must not be visible in a caller which is 'strict'
// If a strict caller invokes a non-strict callee, round a double result
BasicType result_type = dest_method->return_type()->basic_type();
assert( method() != NULL, "must have caller context");
if( result_type == T_DOUBLE && method()->is_strict() && !dest_method->is_strict() ) {
// Destination method's return value is on top of stack
// dstore_rounding() does gvn.transform
Node *result = pop_pair();
result = dstore_rounding(result);
push_pair(result);
}
}
// rounding for strict float precision conformance
Node* GraphKit::precision_rounding(Node* n) {
return UseStrictFP && _method->flags().is_strict()
&& UseSSE == 0 && Matcher::strict_fp_requires_explicit_rounding
? _gvn.transform( new RoundFloatNode(0, n) )
: n;
}
// rounding for strict double precision conformance
Node* GraphKit::dprecision_rounding(Node *n) {
return UseStrictFP && _method->flags().is_strict()
&& UseSSE <= 1 && Matcher::strict_fp_requires_explicit_rounding
? _gvn.transform( new RoundDoubleNode(0, n) )
: n;
}
// rounding for non-strict double stores
Node* GraphKit::dstore_rounding(Node* n) {
return Matcher::strict_fp_requires_explicit_rounding
&& UseSSE <= 1
? _gvn.transform( new RoundDoubleNode(0, n) )
: n;
}
//=============================================================================
// Generate a fast path/slow path idiom. Graph looks like:
// [foo] indicates that 'foo' is a parameter
//
// [in] NULL
// \ /
// CmpP
// Bool ne
// If
// / \
// True False-<2>
// / |
// / cast_not_null
// Load | | ^
// [fast_test] | |
// gvn to opt_test | |
// / \ | <1>
// True False |
// | \\ |
// [slow_call] \[fast_result]
// Ctl Val \ \
// | \ \
// Catch <1> \ \
// / \ ^ \ \
// Ex No_Ex | \ \
// | \ \ | \ <2> \
// ... \ [slow_res] | | \ [null_result]
// \ \--+--+--- | |
// \ | / \ | /
// --------Region Phi
//
//=============================================================================
// Code is structured as a series of driver functions all called 'do_XXX' that
// call a set of helper functions. Helper functions first, then drivers.
//------------------------------null_check_oop---------------------------------
// Null check oop. Set null-path control into Region in slot 3.
// Make a cast-not-nullness use the other not-null control. Return cast.
Node* GraphKit::null_check_oop(Node* value, Node* *null_control,
bool never_see_null,
bool safe_for_replace,
bool speculative) {
// Initial NULL check taken path
(*null_control) = top();
Node* cast = null_check_common(value, T_OBJECT, false, null_control, speculative);
// Generate uncommon_trap:
if (never_see_null && (*null_control) != top()) {
// If we see an unexpected null at a check-cast we record it and force a
// recompile; the offending check-cast will be compiled to handle NULLs.
// If we see more than one offending BCI, then all checkcasts in the
// method will be compiled to handle NULLs.
PreserveJVMState pjvms(this);
set_control(*null_control);
replace_in_map(value, null());
Deoptimization::DeoptReason reason = Deoptimization::reason_null_check(speculative);
uncommon_trap(reason,
Deoptimization::Action_make_not_entrant);
(*null_control) = top(); // NULL path is dead
}
if ((*null_control) == top() && safe_for_replace) {
replace_in_map(value, cast);
}
// Cast away null-ness on the result
return cast;
}
//------------------------------opt_iff----------------------------------------
// Optimize the fast-check IfNode. Set the fast-path region slot 2.
// Return slow-path control.
Node* GraphKit::opt_iff(Node* region, Node* iff) {
IfNode *opt_iff = _gvn.transform(iff)->as_If();
// Fast path taken; set region slot 2
Node *fast_taken = _gvn.transform( new IfFalseNode(opt_iff) );
region->init_req(2,fast_taken); // Capture fast-control
// Fast path not-taken, i.e. slow path
Node *slow_taken = _gvn.transform( new IfTrueNode(opt_iff) );
return slow_taken;
}
//-----------------------------make_runtime_call-------------------------------
Node* GraphKit::make_runtime_call(int flags,
const TypeFunc* call_type, address call_addr,
const char* call_name,
const TypePtr* adr_type,
// The following parms are all optional.
// The first NULL ends the list.
Node* parm0, Node* parm1,
Node* parm2, Node* parm3,
Node* parm4, Node* parm5,
Node* parm6, Node* parm7) {
// Slow-path call
bool is_leaf = !(flags & RC_NO_LEAF);
bool has_io = (!is_leaf && !(flags & RC_NO_IO));
if (call_name == NULL) {
assert(!is_leaf, "must supply name for leaf");
call_name = OptoRuntime::stub_name(call_addr);
}
CallNode* call;
if (!is_leaf) {
call = new CallStaticJavaNode(call_type, call_addr, call_name,
bci(), adr_type);
} else if (flags & RC_NO_FP) {
call = new CallLeafNoFPNode(call_type, call_addr, call_name, adr_type);
} else {
call = new CallLeafNode(call_type, call_addr, call_name, adr_type);
}
// The following is similar to set_edges_for_java_call,
// except that the memory effects of the call are restricted to AliasIdxRaw.
// Slow path call has no side-effects, uses few values
bool wide_in = !(flags & RC_NARROW_MEM);
bool wide_out = (C->get_alias_index(adr_type) == Compile::AliasIdxBot);
Node* prev_mem = NULL;
if (wide_in) {
prev_mem = set_predefined_input_for_runtime_call(call);
} else {
assert(!wide_out, "narrow in => narrow out");
Node* narrow_mem = memory(adr_type);
prev_mem = reset_memory();
map()->set_memory(narrow_mem);
set_predefined_input_for_runtime_call(call);
}
// Hook each parm in order. Stop looking at the first NULL.
if (parm0 != NULL) { call->init_req(TypeFunc::Parms+0, parm0);
if (parm1 != NULL) { call->init_req(TypeFunc::Parms+1, parm1);
if (parm2 != NULL) { call->init_req(TypeFunc::Parms+2, parm2);
if (parm3 != NULL) { call->init_req(TypeFunc::Parms+3, parm3);
if (parm4 != NULL) { call->init_req(TypeFunc::Parms+4, parm4);
if (parm5 != NULL) { call->init_req(TypeFunc::Parms+5, parm5);
if (parm6 != NULL) { call->init_req(TypeFunc::Parms+6, parm6);
if (parm7 != NULL) { call->init_req(TypeFunc::Parms+7, parm7);
/* close each nested if ===> */ } } } } } } } }
assert(call->in(call->req()-1) != NULL, "must initialize all parms");
if (!is_leaf) {
// Non-leaves can block and take safepoints:
add_safepoint_edges(call, ((flags & RC_MUST_THROW) != 0));
}
// Non-leaves can throw exceptions:
if (has_io) {
call->set_req(TypeFunc::I_O, i_o());
}
if (flags & RC_UNCOMMON) {
// Set the count to a tiny probability. Cf. Estimate_Block_Frequency.
// (An "if" probability corresponds roughly to an unconditional count.
// Sort of.)
call->set_cnt(PROB_UNLIKELY_MAG(4));
}
Node* c = _gvn.transform(call);
assert(c == call, "cannot disappear");
if (wide_out) {
// Slow path call has full side-effects.
set_predefined_output_for_runtime_call(call);
} else {
// Slow path call has few side-effects, and/or sets few values.
set_predefined_output_for_runtime_call(call, prev_mem, adr_type);
}
if (has_io) {
set_i_o(_gvn.transform(new ProjNode(call, TypeFunc::I_O)));
}
return call;
}
//------------------------------merge_memory-----------------------------------
// Merge memory from one path into the current memory state.
void GraphKit::merge_memory(Node* new_mem, Node* region, int new_path) {
for (MergeMemStream mms(merged_memory(), new_mem->as_MergeMem()); mms.next_non_empty2(); ) {
Node* old_slice = mms.force_memory();
Node* new_slice = mms.memory2();
if (old_slice != new_slice) {
PhiNode* phi;
if (old_slice->is_Phi() && old_slice->as_Phi()->region() == region) {
if (mms.is_empty()) {
// clone base memory Phi's inputs for this memory slice
assert(old_slice == mms.base_memory(), "sanity");
phi = PhiNode::make(region, NULL, Type::MEMORY, mms.adr_type(C));
_gvn.set_type(phi, Type::MEMORY);
for (uint i = 1; i < phi->req(); i++) {
phi->init_req(i, old_slice->in(i));
}
} else {
phi = old_slice->as_Phi(); // Phi was generated already
}
} else {
phi = PhiNode::make(region, old_slice, Type::MEMORY, mms.adr_type(C));
_gvn.set_type(phi, Type::MEMORY);
}
phi->set_req(new_path, new_slice);
mms.set_memory(phi);
}
}
}
//------------------------------make_slow_call_ex------------------------------
// Make the exception handler hookups for the slow call
void GraphKit::make_slow_call_ex(Node* call, ciInstanceKlass* ex_klass, bool separate_io_proj, bool deoptimize) {
if (stopped()) return;
// Make a catch node with just two handlers: fall-through and catch-all
Node* i_o = _gvn.transform( new ProjNode(call, TypeFunc::I_O, separate_io_proj) );
Node* catc = _gvn.transform( new CatchNode(control(), i_o, 2) );
Node* norm = _gvn.transform( new CatchProjNode(catc, CatchProjNode::fall_through_index, CatchProjNode::no_handler_bci) );
Node* excp = _gvn.transform( new CatchProjNode(catc, CatchProjNode::catch_all_index, CatchProjNode::no_handler_bci) );
{ PreserveJVMState pjvms(this);
set_control(excp);
set_i_o(i_o);
if (excp != top()) {
if (deoptimize) {
// Deoptimize if an exception is caught. Don't construct exception state in this case.
uncommon_trap(Deoptimization::Reason_unhandled,
Deoptimization::Action_none);
} else {
// Create an exception state also.
// Use an exact type if the caller has specified a specific exception.
const Type* ex_type = TypeOopPtr::make_from_klass_unique(ex_klass)->cast_to_ptr_type(TypePtr::NotNull);
Node* ex_oop = new CreateExNode(ex_type, control(), i_o);
add_exception_state(make_exception_state(_gvn.transform(ex_oop)));
}
}
}
// Get the no-exception control from the CatchNode.
set_control(norm);
}
static IfNode* gen_subtype_check_compare(Node* ctrl, Node* in1, Node* in2, BoolTest::mask test, float p, PhaseGVN* gvn, BasicType bt) {
Node* cmp = NULL;
switch(bt) {
case T_INT: cmp = new CmpINode(in1, in2); break;
case T_ADDRESS: cmp = new CmpPNode(in1, in2); break;
default: fatal(err_msg("unexpected comparison type %s", type2name(bt)));
}
gvn->transform(cmp);
Node* bol = gvn->transform(new BoolNode(cmp, test));
IfNode* iff = new IfNode(ctrl, bol, p, COUNT_UNKNOWN);
gvn->transform(iff);
if (!bol->is_Con()) gvn->record_for_igvn(iff);
return iff;
}
//-------------------------------gen_subtype_check-----------------------------
// Generate a subtyping check. Takes as input the subtype and supertype.
// Returns 2 values: sets the default control() to the true path and returns
// the false path. Only reads invariant memory; sets no (visible) memory.
// The PartialSubtypeCheckNode sets the hidden 1-word cache in the encoding
// but that's not exposed to the optimizer. This call also doesn't take in an
// Object; if you wish to check an Object you need to load the Object's class
// prior to coming here.
Node* Phase::gen_subtype_check(Node* subklass, Node* superklass, Node** ctrl, MergeMemNode* mem, PhaseGVN* gvn) {
Compile* C = gvn->C;
// Fast check for identical types, perhaps identical constants.
// The types can even be identical non-constants, in cases
// involving Array.newInstance, Object.clone, etc.
if (subklass == superklass)
return C->top(); // false path is dead; no test needed.
if (gvn->type(superklass)->singleton()) {
ciKlass* superk = gvn->type(superklass)->is_klassptr()->klass();
ciKlass* subk = gvn->type(subklass)->is_klassptr()->klass();
// In the common case of an exact superklass, try to fold up the
// test before generating code. You may ask, why not just generate
// the code and then let it fold up? The answer is that the generated
// code will necessarily include null checks, which do not always
// completely fold away. If they are also needless, then they turn
// into a performance loss. Example:
// Foo[] fa = blah(); Foo x = fa[0]; fa[1] = x;
// Here, the type of 'fa' is often exact, so the store check
// of fa[1]=x will fold up, without testing the nullness of x.
switch (C->static_subtype_check(superk, subk)) {
case Compile::SSC_always_false:
{
Node* always_fail = *ctrl;
*ctrl = gvn->C->top();
return always_fail;
}
case Compile::SSC_always_true:
return C->top();
case Compile::SSC_easy_test:
{
// Just do a direct pointer compare and be done.
IfNode* iff = gen_subtype_check_compare(*ctrl, subklass, superklass, BoolTest::eq, PROB_STATIC_FREQUENT, gvn, T_ADDRESS);
*ctrl = gvn->transform(new IfTrueNode(iff));
return gvn->transform(new IfFalseNode(iff));
}
case Compile::SSC_full_test:
break;
default:
ShouldNotReachHere();
}
}
// %%% Possible further optimization: Even if the superklass is not exact,
// if the subklass is the unique subtype of the superklass, the check
// will always succeed. We could leave a dependency behind to ensure this.
// First load the super-klass's check-offset
Node *p1 = gvn->transform(new AddPNode(superklass, superklass, gvn->MakeConX(in_bytes(Klass::super_check_offset_offset()))));
Node* m = mem->memory_at(C->get_alias_index(gvn->type(p1)->is_ptr()));
Node *chk_off = gvn->transform(new LoadINode(NULL, m, p1, gvn->type(p1)->is_ptr(), TypeInt::INT, MemNode::unordered));
int cacheoff_con = in_bytes(Klass::secondary_super_cache_offset());
bool might_be_cache = (gvn->find_int_con(chk_off, cacheoff_con) == cacheoff_con);
// Load from the sub-klass's super-class display list, or a 1-word cache of
// the secondary superclass list, or a failing value with a sentinel offset
// if the super-klass is an interface or exceptionally deep in the Java
// hierarchy and we have to scan the secondary superclass list the hard way.
// Worst-case type is a little odd: NULL is allowed as a result (usually
// klass loads can never produce a NULL).
Node *chk_off_X = chk_off;
#ifdef _LP64
chk_off_X = gvn->transform(new ConvI2LNode(chk_off_X));
#endif
Node *p2 = gvn->transform(new AddPNode(subklass,subklass,chk_off_X));
// For some types like interfaces the following loadKlass is from a 1-word
// cache which is mutable so can't use immutable memory. Other
// types load from the super-class display table which is immutable.
m = mem->memory_at(C->get_alias_index(gvn->type(p2)->is_ptr()));
Node *kmem = might_be_cache ? m : C->immutable_memory();
Node *nkls = gvn->transform(LoadKlassNode::make(*gvn, NULL, kmem, p2, gvn->type(p2)->is_ptr(), TypeKlassPtr::OBJECT_OR_NULL));
// Compile speed common case: ARE a subtype and we canNOT fail
if( superklass == nkls )
return C->top(); // false path is dead; no test needed.
// See if we get an immediate positive hit. Happens roughly 83% of the
// time. Test to see if the value loaded just previously from the subklass
// is exactly the superklass.
IfNode *iff1 = gen_subtype_check_compare(*ctrl, superklass, nkls, BoolTest::eq, PROB_LIKELY(0.83f), gvn, T_ADDRESS);
Node *iftrue1 = gvn->transform( new IfTrueNode (iff1));
*ctrl = gvn->transform(new IfFalseNode(iff1));
// Compile speed common case: Check for being deterministic right now. If
// chk_off is a constant and not equal to cacheoff then we are NOT a
// subklass. In this case we need exactly the 1 test above and we can
// return those results immediately.
if (!might_be_cache) {
Node* not_subtype_ctrl = *ctrl;
*ctrl = iftrue1; // We need exactly the 1 test above
return not_subtype_ctrl;
}
// Gather the various success & failures here
RegionNode *r_ok_subtype = new RegionNode(4);
gvn->record_for_igvn(r_ok_subtype);
RegionNode *r_not_subtype = new RegionNode(3);
gvn->record_for_igvn(r_not_subtype);
r_ok_subtype->init_req(1, iftrue1);
// Check for immediate negative hit. Happens roughly 11% of the time (which
// is roughly 63% of the remaining cases). Test to see if the loaded
// check-offset points into the subklass display list or the 1-element
// cache. If it points to the display (and NOT the cache) and the display
// missed then it's not a subtype.
Node *cacheoff = gvn->intcon(cacheoff_con);
IfNode *iff2 = gen_subtype_check_compare(*ctrl, chk_off, cacheoff, BoolTest::ne, PROB_LIKELY(0.63f), gvn, T_INT);
r_not_subtype->init_req(1, gvn->transform(new IfTrueNode (iff2)));
*ctrl = gvn->transform(new IfFalseNode(iff2));
// Check for self. Very rare to get here, but it is taken 1/3 the time.
// No performance impact (too rare) but allows sharing of secondary arrays
// which has some footprint reduction.
IfNode *iff3 = gen_subtype_check_compare(*ctrl, subklass, superklass, BoolTest::eq, PROB_LIKELY(0.36f), gvn, T_ADDRESS);
r_ok_subtype->init_req(2, gvn->transform(new IfTrueNode(iff3)));
*ctrl = gvn->transform(new IfFalseNode(iff3));
// -- Roads not taken here: --
// We could also have chosen to perform the self-check at the beginning
// of this code sequence, as the assembler does. This would not pay off
// the same way, since the optimizer, unlike the assembler, can perform
// static type analysis to fold away many successful self-checks.
// Non-foldable self checks work better here in second position, because
// the initial primary superclass check subsumes a self-check for most
// types. An exception would be a secondary type like array-of-interface,
// which does not appear in its own primary supertype display.
// Finally, we could have chosen to move the self-check into the
// PartialSubtypeCheckNode, and from there out-of-line in a platform
// dependent manner. But it is worthwhile to have the check here,
// where it can be perhaps be optimized. The cost in code space is
// small (register compare, branch).
// Now do a linear scan of the secondary super-klass array. Again, no real
// performance impact (too rare) but it's gotta be done.
// Since the code is rarely used, there is no penalty for moving it
// out of line, and it can only improve I-cache density.
// The decision to inline or out-of-line this final check is platform
// dependent, and is found in the AD file definition of PartialSubtypeCheck.
Node* psc = gvn->transform(
new PartialSubtypeCheckNode(*ctrl, subklass, superklass));
IfNode *iff4 = gen_subtype_check_compare(*ctrl, psc, gvn->zerocon(T_OBJECT), BoolTest::ne, PROB_FAIR, gvn, T_ADDRESS);
r_not_subtype->init_req(2, gvn->transform(new IfTrueNode (iff4)));
r_ok_subtype ->init_req(3, gvn->transform(new IfFalseNode(iff4)));
// Return false path; set default control to true path.
*ctrl = gvn->transform(r_ok_subtype);
return gvn->transform(r_not_subtype);
}
// Profile-driven exact type check:
Node* GraphKit::type_check_receiver(Node* receiver, ciKlass* klass,
float prob,
Node* *casted_receiver) {
const TypeKlassPtr* tklass = TypeKlassPtr::make(klass);
Node* recv_klass = load_object_klass(receiver);
Node* want_klass = makecon(tklass);
Node* cmp = _gvn.transform( new CmpPNode(recv_klass, want_klass) );
Node* bol = _gvn.transform( new BoolNode(cmp, BoolTest::eq) );
IfNode* iff = create_and_xform_if(control(), bol, prob, COUNT_UNKNOWN);
set_control( _gvn.transform( new IfTrueNode (iff) ));
Node* fail = _gvn.transform( new IfFalseNode(iff) );
const TypeOopPtr* recv_xtype = tklass->as_instance_type();
assert(recv_xtype->klass_is_exact(), "");
// Subsume downstream occurrences of receiver with a cast to
// recv_xtype, since now we know what the type will be.
Node* cast = new CheckCastPPNode(control(), receiver, recv_xtype);
(*casted_receiver) = _gvn.transform(cast);
// (User must make the replace_in_map call.)
return fail;
}
//------------------------------seems_never_null-------------------------------
// Use null_seen information if it is available from the profile.
// If we see an unexpected null at a type check we record it and force a
// recompile; the offending check will be recompiled to handle NULLs.
// If we see several offending BCIs, then all checks in the
// method will be recompiled.
bool GraphKit::seems_never_null(Node* obj, ciProfileData* data, bool& speculating) {
speculating = !_gvn.type(obj)->speculative_maybe_null();
Deoptimization::DeoptReason reason = Deoptimization::reason_null_check(speculating);
if (UncommonNullCast // Cutout for this technique
&& obj != null() // And not the -Xcomp stupid case?
&& !too_many_traps(reason)
) {
if (speculating) {
return true;
}
if (data == NULL)
// Edge case: no mature data. Be optimistic here.
return true;
// If the profile has not seen a null, assume it won't happen.
assert(java_bc() == Bytecodes::_checkcast ||
java_bc() == Bytecodes::_instanceof ||
java_bc() == Bytecodes::_aastore, "MDO must collect null_seen bit here");
return !data->as_BitData()->null_seen();
}
speculating = false;
return false;
}
//------------------------maybe_cast_profiled_receiver-------------------------
// If the profile has seen exactly one type, narrow to exactly that type.
// Subsequent type checks will always fold up.
Node* GraphKit::maybe_cast_profiled_receiver(Node* not_null_obj,
ciKlass* require_klass,
ciKlass* spec_klass,
bool safe_for_replace) {
if (!UseTypeProfile || !TypeProfileCasts) return NULL;
Deoptimization::DeoptReason reason = Deoptimization::reason_class_check(spec_klass != NULL);
// Make sure we haven't already deoptimized from this tactic.
if (too_many_traps(reason) || too_many_recompiles(reason))
return NULL;
// (No, this isn't a call, but it's enough like a virtual call
// to use the same ciMethod accessor to get the profile info...)
// If we have a speculative type use it instead of profiling (which
// may not help us)
ciKlass* exact_kls = spec_klass == NULL ? profile_has_unique_klass() : spec_klass;
if (exact_kls != NULL) {// no cast failures here
if (require_klass == NULL ||
C->static_subtype_check(require_klass, exact_kls) == Compile::SSC_always_true) {
// If we narrow the type to match what the type profile sees or
// the speculative type, we can then remove the rest of the
// cast.
// This is a win, even if the exact_kls is very specific,
// because downstream operations, such as method calls,
// will often benefit from the sharper type.
Node* exact_obj = not_null_obj; // will get updated in place...
Node* slow_ctl = type_check_receiver(exact_obj, exact_kls, 1.0,
&exact_obj);
{ PreserveJVMState pjvms(this);
set_control(slow_ctl);
uncommon_trap_exact(reason, Deoptimization::Action_maybe_recompile);
}
if (safe_for_replace) {
replace_in_map(not_null_obj, exact_obj);
}
return exact_obj;
}
// assert(ssc == Compile::SSC_always_true)... except maybe the profile lied to us.
}
return NULL;
}
/**
* Cast obj to type and emit guard unless we had too many traps here
* already
*
* @param obj node being casted
* @param type type to cast the node to
* @param not_null true if we know node cannot be null
*/
Node* GraphKit::maybe_cast_profiled_obj(Node* obj,
ciKlass* type,
bool not_null,
SafePointNode* sfpt) {
// type == NULL if profiling tells us this object is always null
if (type != NULL) {
Deoptimization::DeoptReason class_reason = Deoptimization::Reason_speculate_class_check;
Deoptimization::DeoptReason null_reason = Deoptimization::Reason_speculate_null_check;
ciMethod* trap_method = (sfpt == NULL) ? method() : sfpt->jvms()->method();
int trap_bci = (sfpt == NULL) ? bci() : sfpt->jvms()->bci();
if (!too_many_traps(null_reason) && !too_many_recompiles(null_reason) &&
!C->too_many_traps(trap_method, trap_bci, class_reason) &&
!C->too_many_recompiles(trap_method, trap_bci, class_reason)) {
Node* not_null_obj = NULL;
// not_null is true if we know the object is not null and
// there's no need for a null check
if (!not_null) {
Node* null_ctl = top();
not_null_obj = null_check_oop(obj, &null_ctl, true, true, true);
assert(null_ctl->is_top(), "no null control here");
} else {
not_null_obj = obj;
}
Node* exact_obj = not_null_obj;
ciKlass* exact_kls = type;
Node* slow_ctl = type_check_receiver(exact_obj, exact_kls, 1.0,
&exact_obj);
if (sfpt != NULL) {
GraphKit kit(sfpt->jvms());
PreserveJVMState pjvms(&kit);
kit.set_control(slow_ctl);
kit.uncommon_trap_exact(class_reason, Deoptimization::Action_maybe_recompile);
} else {
PreserveJVMState pjvms(this);
set_control(slow_ctl);
uncommon_trap_exact(class_reason, Deoptimization::Action_maybe_recompile);
}
replace_in_map(not_null_obj, exact_obj);
obj = exact_obj;
}
} else {
if (!too_many_traps(Deoptimization::Reason_null_assert) &&
!too_many_recompiles(Deoptimization::Reason_null_assert)) {
Node* exact_obj = null_assert(obj);
replace_in_map(obj, exact_obj);
obj = exact_obj;
}
}
return obj;
}
//-------------------------------gen_instanceof--------------------------------
// Generate an instance-of idiom. Used by both the instance-of bytecode
// and the reflective instance-of call.
Node* GraphKit::gen_instanceof(Node* obj, Node* superklass, bool safe_for_replace) {
kill_dead_locals(); // Benefit all the uncommon traps
assert( !stopped(), "dead parse path should be checked in callers" );
assert(!TypePtr::NULL_PTR->higher_equal(_gvn.type(superklass)->is_klassptr()),
"must check for not-null not-dead klass in callers");
// Make the merge point
enum { _obj_path = 1, _fail_path, _null_path, PATH_LIMIT };
RegionNode* region = new RegionNode(PATH_LIMIT);
Node* phi = new PhiNode(region, TypeInt::BOOL);
C->set_has_split_ifs(true); // Has chance for split-if optimization
ciProfileData* data = NULL;
if (java_bc() == Bytecodes::_instanceof) { // Only for the bytecode
data = method()->method_data()->bci_to_data(bci());
}
bool speculative_not_null = false;
bool never_see_null = (ProfileDynamicTypes // aggressive use of profile
&& seems_never_null(obj, data, speculative_not_null));
// Null check; get casted pointer; set region slot 3
Node* null_ctl = top();
Node* not_null_obj = null_check_oop(obj, &null_ctl, never_see_null, safe_for_replace, speculative_not_null);
// If not_null_obj is dead, only null-path is taken
if (stopped()) { // Doing instance-of on a NULL?
set_control(null_ctl);
return intcon(0);
}
region->init_req(_null_path, null_ctl);
phi ->init_req(_null_path, intcon(0)); // Set null path value
if (null_ctl == top()) {
// Do this eagerly, so that pattern matches like is_diamond_phi
// will work even during parsing.
assert(_null_path == PATH_LIMIT-1, "delete last");
region->del_req(_null_path);
phi ->del_req(_null_path);
}
// Do we know the type check always succeed?
bool known_statically = false;
if (_gvn.type(superklass)->singleton()) {
ciKlass* superk = _gvn.type(superklass)->is_klassptr()->klass();
ciKlass* subk = _gvn.type(obj)->is_oopptr()->klass();
if (subk != NULL && subk->is_loaded()) {
int static_res = C->static_subtype_check(superk, subk);
known_statically = (static_res == Compile::SSC_always_true || static_res == Compile::SSC_always_false);
}
}
if (known_statically && UseTypeSpeculation) {
// If we know the type check always succeeds then we don't use the
// profiling data at this bytecode. Don't lose it, feed it to the
// type system as a speculative type.
not_null_obj = record_profiled_receiver_for_speculation(not_null_obj);
} else {
const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
// We may not have profiling here or it may not help us. If we
// have a speculative type use it to perform an exact cast.
ciKlass* spec_obj_type = obj_type->speculative_type();
if (spec_obj_type != NULL || (ProfileDynamicTypes && data != NULL)) {
Node* cast_obj = maybe_cast_profiled_receiver(not_null_obj, NULL, spec_obj_type, safe_for_replace);
if (stopped()) { // Profile disagrees with this path.
set_control(null_ctl); // Null is the only remaining possibility.
return intcon(0);
}
if (cast_obj != NULL) {
not_null_obj = cast_obj;
}
}
}
// Load the object's klass
Node* obj_klass = load_object_klass(not_null_obj);
// Generate the subtype check
Node* not_subtype_ctrl = gen_subtype_check(obj_klass, superklass);
// Plug in the success path to the general merge in slot 1.
region->init_req(_obj_path, control());
phi ->init_req(_obj_path, intcon(1));
// Plug in the failing path to the general merge in slot 2.
region->init_req(_fail_path, not_subtype_ctrl);
phi ->init_req(_fail_path, intcon(0));
// Return final merged results
set_control( _gvn.transform(region) );
record_for_igvn(region);
return _gvn.transform(phi);
}
//-------------------------------gen_checkcast---------------------------------
// Generate a checkcast idiom. Used by both the checkcast bytecode and the
// array store bytecode. Stack must be as-if BEFORE doing the bytecode so the
// uncommon-trap paths work. Adjust stack after this call.
// If failure_control is supplied and not null, it is filled in with
// the control edge for the cast failure. Otherwise, an appropriate
// uncommon trap or exception is thrown.
Node* GraphKit::gen_checkcast(Node *obj, Node* superklass,
Node* *failure_control) {
kill_dead_locals(); // Benefit all the uncommon traps
const TypeKlassPtr *tk = _gvn.type(superklass)->is_klassptr();
const Type *toop = TypeOopPtr::make_from_klass(tk->klass());
// Fast cutout: Check the case that the cast is vacuously true.
// This detects the common cases where the test will short-circuit
// away completely. We do this before we perform the null check,
// because if the test is going to turn into zero code, we don't
// want a residual null check left around. (Causes a slowdown,
// for example, in some objArray manipulations, such as a[i]=a[j].)
if (tk->singleton()) {
const TypeOopPtr* objtp = _gvn.type(obj)->isa_oopptr();
if (objtp != NULL && objtp->klass() != NULL) {
switch (C->static_subtype_check(tk->klass(), objtp->klass())) {
case Compile::SSC_always_true:
// If we know the type check always succeed then we don't use
// the profiling data at this bytecode. Don't lose it, feed it
// to the type system as a speculative type.
return record_profiled_receiver_for_speculation(obj);
case Compile::SSC_always_false:
// It needs a null check because a null will *pass* the cast check.
// A non-null value will always produce an exception.
return null_assert(obj);
}
}
}
ciProfileData* data = NULL;
bool safe_for_replace = false;
if (failure_control == NULL) { // use MDO in regular case only
assert(java_bc() == Bytecodes::_aastore ||
java_bc() == Bytecodes::_checkcast,
"interpreter profiles type checks only for these BCs");
data = method()->method_data()->bci_to_data(bci());
safe_for_replace = true;
}
// Make the merge point
enum { _obj_path = 1, _null_path, PATH_LIMIT };
RegionNode* region = new RegionNode(PATH_LIMIT);
Node* phi = new PhiNode(region, toop);
C->set_has_split_ifs(true); // Has chance for split-if optimization
// Use null-cast information if it is available
bool speculative_not_null = false;
bool never_see_null = ((failure_control == NULL) // regular case only
&& seems_never_null(obj, data, speculative_not_null));
// Null check; get casted pointer; set region slot 3
Node* null_ctl = top();
Node* not_null_obj = null_check_oop(obj, &null_ctl, never_see_null, safe_for_replace, speculative_not_null);
// If not_null_obj is dead, only null-path is taken
if (stopped()) { // Doing instance-of on a NULL?
set_control(null_ctl);
return null();
}
region->init_req(_null_path, null_ctl);
phi ->init_req(_null_path, null()); // Set null path value
if (null_ctl == top()) {
// Do this eagerly, so that pattern matches like is_diamond_phi
// will work even during parsing.
assert(_null_path == PATH_LIMIT-1, "delete last");
region->del_req(_null_path);
phi ->del_req(_null_path);
}
Node* cast_obj = NULL;
if (tk->klass_is_exact()) {
// The following optimization tries to statically cast the speculative type of the object
// (for example obtained during profiling) to the type of the superklass and then do a
// dynamic check that the type of the object is what we expect. To work correctly
// for checkcast and aastore the type of superklass should be exact.
const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
// We may not have profiling here or it may not help us. If we have
// a speculative type use it to perform an exact cast.
ciKlass* spec_obj_type = obj_type->speculative_type();
if (spec_obj_type != NULL ||
(data != NULL &&
// Counter has never been decremented (due to cast failure).
// ...This is a reasonable thing to expect. It is true of
// all casts inserted by javac to implement generic types.
data->as_CounterData()->count() >= 0)) {
cast_obj = maybe_cast_profiled_receiver(not_null_obj, tk->klass(), spec_obj_type, safe_for_replace);
if (cast_obj != NULL) {
if (failure_control != NULL) // failure is now impossible
(*failure_control) = top();
// adjust the type of the phi to the exact klass:
phi->raise_bottom_type(_gvn.type(cast_obj)->meet_speculative(TypePtr::NULL_PTR));
}
}
}
if (cast_obj == NULL) {
// Load the object's klass
Node* obj_klass = load_object_klass(not_null_obj);
// Generate the subtype check
Node* not_subtype_ctrl = gen_subtype_check( obj_klass, superklass );
// Plug in success path into the merge
cast_obj = _gvn.transform(new CheckCastPPNode(control(), not_null_obj, toop));
// Failure path ends in uncommon trap (or may be dead - failure impossible)
if (failure_control == NULL) {
if (not_subtype_ctrl != top()) { // If failure is possible
PreserveJVMState pjvms(this);
set_control(not_subtype_ctrl);
builtin_throw(Deoptimization::Reason_class_check, obj_klass);
}
} else {
(*failure_control) = not_subtype_ctrl;
}
}
region->init_req(_obj_path, control());
phi ->init_req(_obj_path, cast_obj);
// A merge of NULL or Casted-NotNull obj
Node* res = _gvn.transform(phi);
// Note I do NOT always 'replace_in_map(obj,result)' here.
// if( tk->klass()->can_be_primary_super() )
// This means that if I successfully store an Object into an array-of-String
// I 'forget' that the Object is really now known to be a String. I have to
// do this because we don't have true union types for interfaces - if I store
// a Baz into an array-of-Interface and then tell the optimizer it's an
// Interface, I forget that it's also a Baz and cannot do Baz-like field
// references to it. FIX THIS WHEN UNION TYPES APPEAR!
// replace_in_map( obj, res );
// Return final merged results
set_control( _gvn.transform(region) );
record_for_igvn(region);
return res;
}
//------------------------------next_monitor-----------------------------------
// What number should be given to the next monitor?
int GraphKit::next_monitor() {
int current = jvms()->monitor_depth()* C->sync_stack_slots();
int next = current + C->sync_stack_slots();
// Keep the toplevel high water mark current:
if (C->fixed_slots() < next) C->set_fixed_slots(next);
return current;
}
//------------------------------insert_mem_bar---------------------------------
// Memory barrier to avoid floating things around
// The membar serves as a pinch point between both control and all memory slices.
Node* GraphKit::insert_mem_bar(int opcode, Node* precedent) {
MemBarNode* mb = MemBarNode::make(C, opcode, Compile::AliasIdxBot, precedent);
mb->init_req(TypeFunc::Control, control());
mb->init_req(TypeFunc::Memory, reset_memory());
Node* membar = _gvn.transform(mb);
set_control(_gvn.transform(new ProjNode(membar, TypeFunc::Control)));
set_all_memory_call(membar);
return membar;
}
//-------------------------insert_mem_bar_volatile----------------------------
// Memory barrier to avoid floating things around
// The membar serves as a pinch point between both control and memory(alias_idx).
// If you want to make a pinch point on all memory slices, do not use this
// function (even with AliasIdxBot); use insert_mem_bar() instead.
Node* GraphKit::insert_mem_bar_volatile(int opcode, int alias_idx, Node* precedent) {
// When Parse::do_put_xxx updates a volatile field, it appends a series
// of MemBarVolatile nodes, one for *each* volatile field alias category.
// The first membar is on the same memory slice as the field store opcode.
// This forces the membar to follow the store. (Bug 6500685 broke this.)
// All the other membars (for other volatile slices, including AliasIdxBot,
// which stands for all unknown volatile slices) are control-dependent
// on the first membar. This prevents later volatile loads or stores
// from sliding up past the just-emitted store.
MemBarNode* mb = MemBarNode::make(C, opcode, alias_idx, precedent);
mb->set_req(TypeFunc::Control,control());
if (alias_idx == Compile::AliasIdxBot) {
mb->set_req(TypeFunc::Memory, merged_memory()->base_memory());
} else {
assert(!(opcode == Op_Initialize && alias_idx != Compile::AliasIdxRaw), "fix caller");
mb->set_req(TypeFunc::Memory, memory(alias_idx));
}
Node* membar = _gvn.transform(mb);
set_control(_gvn.transform(new ProjNode(membar, TypeFunc::Control)));
if (alias_idx == Compile::AliasIdxBot) {
merged_memory()->set_base_memory(_gvn.transform(new ProjNode(membar, TypeFunc::Memory)));
} else {
set_memory(_gvn.transform(new ProjNode(membar, TypeFunc::Memory)),alias_idx);
}
return membar;
}
//------------------------------shared_lock------------------------------------
// Emit locking code.
FastLockNode* GraphKit::shared_lock(Node* obj) {
// bci is either a monitorenter bc or InvocationEntryBci
// %%% SynchronizationEntryBCI is redundant; use InvocationEntryBci in interfaces
assert(SynchronizationEntryBCI == InvocationEntryBci, "");
if( !GenerateSynchronizationCode )
return NULL; // Not locking things?
if (stopped()) // Dead monitor?
return NULL;
assert(dead_locals_are_killed(), "should kill locals before sync. point");
// Box the stack location
Node* box = _gvn.transform(new BoxLockNode(next_monitor()));
Node* mem = reset_memory();
FastLockNode * flock = _gvn.transform(new FastLockNode(0, obj, box) )->as_FastLock();
if (UseBiasedLocking && PrintPreciseBiasedLockingStatistics) {
// Create the counters for this fast lock.
flock->create_lock_counter(sync_jvms()); // sync_jvms used to get current bci
}
// Create the rtm counters for this fast lock if needed.
flock->create_rtm_lock_counter(sync_jvms()); // sync_jvms used to get current bci
// Add monitor to debug info for the slow path. If we block inside the
// slow path and de-opt, we need the monitor hanging around
map()->push_monitor( flock );
const TypeFunc *tf = LockNode::lock_type();
LockNode *lock = new LockNode(C, tf);
lock->init_req( TypeFunc::Control, control() );
lock->init_req( TypeFunc::Memory , mem );
lock->init_req( TypeFunc::I_O , top() ) ; // does no i/o
lock->init_req( TypeFunc::FramePtr, frameptr() );
lock->init_req( TypeFunc::ReturnAdr, top() );
lock->init_req(TypeFunc::Parms + 0, obj);
lock->init_req(TypeFunc::Parms + 1, box);
lock->init_req(TypeFunc::Parms + 2, flock);
add_safepoint_edges(lock);
lock = _gvn.transform( lock )->as_Lock();
// lock has no side-effects, sets few values
set_predefined_output_for_runtime_call(lock, mem, TypeRawPtr::BOTTOM);
insert_mem_bar(Op_MemBarAcquireLock);
// Add this to the worklist so that the lock can be eliminated
record_for_igvn(lock);
#ifndef PRODUCT
if (PrintLockStatistics) {
// Update the counter for this lock. Don't bother using an atomic
// operation since we don't require absolute accuracy.
lock->create_lock_counter(map()->jvms());
increment_counter(lock->counter()->addr());
}
#endif
return flock;
}
//------------------------------shared_unlock----------------------------------
// Emit unlocking code.
void GraphKit::shared_unlock(Node* box, Node* obj) {
// bci is either a monitorenter bc or InvocationEntryBci
// %%% SynchronizationEntryBCI is redundant; use InvocationEntryBci in interfaces
assert(SynchronizationEntryBCI == InvocationEntryBci, "");
if( !GenerateSynchronizationCode )
return;
if (stopped()) { // Dead monitor?
map()->pop_monitor(); // Kill monitor from debug info
return;
}
// Memory barrier to avoid floating things down past the locked region
insert_mem_bar(Op_MemBarReleaseLock);
const TypeFunc *tf = OptoRuntime::complete_monitor_exit_Type();
UnlockNode *unlock = new UnlockNode(C, tf);
#ifdef ASSERT
unlock->set_dbg_jvms(sync_jvms());
#endif
uint raw_idx = Compile::AliasIdxRaw;
unlock->init_req( TypeFunc::Control, control() );
unlock->init_req( TypeFunc::Memory , memory(raw_idx) );
unlock->init_req( TypeFunc::I_O , top() ) ; // does no i/o
unlock->init_req( TypeFunc::FramePtr, frameptr() );
unlock->init_req( TypeFunc::ReturnAdr, top() );
unlock->init_req(TypeFunc::Parms + 0, obj);
unlock->init_req(TypeFunc::Parms + 1, box);
unlock = _gvn.transform(unlock)->as_Unlock();
Node* mem = reset_memory();
// unlock has no side-effects, sets few values
set_predefined_output_for_runtime_call(unlock, mem, TypeRawPtr::BOTTOM);
// Kill monitor from debug info
map()->pop_monitor( );
}
//-------------------------------get_layout_helper-----------------------------
// If the given klass is a constant or known to be an array,
// fetch the constant layout helper value into constant_value
// and return (Node*)NULL. Otherwise, load the non-constant
// layout helper value, and return the node which represents it.
// This two-faced routine is useful because allocation sites
// almost always feature constant types.
Node* GraphKit::get_layout_helper(Node* klass_node, jint& constant_value) {
const TypeKlassPtr* inst_klass = _gvn.type(klass_node)->isa_klassptr();
if (!StressReflectiveCode && inst_klass != NULL) {
ciKlass* klass = inst_klass->klass();
bool xklass = inst_klass->klass_is_exact();
if (xklass || klass->is_array_klass()) {
jint lhelper = klass->layout_helper();
if (lhelper != Klass::_lh_neutral_value) {
constant_value = lhelper;
return (Node*) NULL;
}
}
}
constant_value = Klass::_lh_neutral_value; // put in a known value
Node* lhp = basic_plus_adr(klass_node, klass_node, in_bytes(Klass::layout_helper_offset()));
return make_load(NULL, lhp, TypeInt::INT, T_INT, MemNode::unordered);
}
// We just put in an allocate/initialize with a big raw-memory effect.
// Hook selected additional alias categories on the initialization.
static void hook_memory_on_init(GraphKit& kit, int alias_idx,
MergeMemNode* init_in_merge,
Node* init_out_raw) {
DEBUG_ONLY(Node* init_in_raw = init_in_merge->base_memory());
assert(init_in_merge->memory_at(alias_idx) == init_in_raw, "");
Node* prevmem = kit.memory(alias_idx);
init_in_merge->set_memory_at(alias_idx, prevmem);
kit.set_memory(init_out_raw, alias_idx);
}
//---------------------------set_output_for_allocation-------------------------
Node* GraphKit::set_output_for_allocation(AllocateNode* alloc,
const TypeOopPtr* oop_type,
bool deoptimize_on_exception) {
int rawidx = Compile::AliasIdxRaw;
alloc->set_req( TypeFunc::FramePtr, frameptr() );
add_safepoint_edges(alloc);
Node* allocx = _gvn.transform(alloc);
set_control( _gvn.transform(new ProjNode(allocx, TypeFunc::Control) ) );
// create memory projection for i_o
set_memory ( _gvn.transform( new ProjNode(allocx, TypeFunc::Memory, true) ), rawidx );
make_slow_call_ex(allocx, env()->Throwable_klass(), true, deoptimize_on_exception);
// create a memory projection as for the normal control path
Node* malloc = _gvn.transform(new ProjNode(allocx, TypeFunc::Memory));
set_memory(malloc, rawidx);
// a normal slow-call doesn't change i_o, but an allocation does
// we create a separate i_o projection for the normal control path
set_i_o(_gvn.transform( new ProjNode(allocx, TypeFunc::I_O, false) ) );
Node* rawoop = _gvn.transform( new ProjNode(allocx, TypeFunc::Parms) );
// put in an initialization barrier
InitializeNode* init = insert_mem_bar_volatile(Op_Initialize, rawidx,
rawoop)->as_Initialize();
assert(alloc->initialization() == init, "2-way macro link must work");
assert(init ->allocation() == alloc, "2-way macro link must work");
{
// Extract memory strands which may participate in the new object's
// initialization, and source them from the new InitializeNode.
// This will allow us to observe initializations when they occur,
// and link them properly (as a group) to the InitializeNode.
assert(init->in(InitializeNode::Memory) == malloc, "");
MergeMemNode* minit_in = MergeMemNode::make(malloc);
init->set_req(InitializeNode::Memory, minit_in);
record_for_igvn(minit_in); // fold it up later, if possible
Node* minit_out = memory(rawidx);
assert(minit_out->is_Proj() && minit_out->in(0) == init, "");
if (oop_type->isa_aryptr()) {
const TypePtr* telemref = oop_type->add_offset(Type::OffsetBot);
int elemidx = C->get_alias_index(telemref);
hook_memory_on_init(*this, elemidx, minit_in, minit_out);
} else if (oop_type->isa_instptr()) {
ciInstanceKlass* ik = oop_type->klass()->as_instance_klass();
for (int i = 0, len = ik->nof_nonstatic_fields(); i < len; i++) {
ciField* field = ik->nonstatic_field_at(i);
if (field->offset() >= TrackedInitializationLimit * HeapWordSize)
continue; // do not bother to track really large numbers of fields
// Find (or create) the alias category for this field:
int fieldidx = C->alias_type(field)->index();
hook_memory_on_init(*this, fieldidx, minit_in, minit_out);
}
}
}
// Cast raw oop to the real thing...
Node* javaoop = new CheckCastPPNode(control(), rawoop, oop_type);
javaoop = _gvn.transform(javaoop);
C->set_recent_alloc(control(), javaoop);
assert(just_allocated_object(control()) == javaoop, "just allocated");
#ifdef ASSERT
{ // Verify that the AllocateNode::Ideal_allocation recognizers work:
assert(AllocateNode::Ideal_allocation(rawoop, &_gvn) == alloc,
"Ideal_allocation works");
assert(AllocateNode::Ideal_allocation(javaoop, &_gvn) == alloc,
"Ideal_allocation works");
if (alloc->is_AllocateArray()) {
assert(AllocateArrayNode::Ideal_array_allocation(rawoop, &_gvn) == alloc->as_AllocateArray(),
"Ideal_allocation works");
assert(AllocateArrayNode::Ideal_array_allocation(javaoop, &_gvn) == alloc->as_AllocateArray(),
"Ideal_allocation works");
} else {
assert(alloc->in(AllocateNode::ALength)->is_top(), "no length, please");
}
}
#endif //ASSERT
return javaoop;
}
//---------------------------new_instance--------------------------------------
// This routine takes a klass_node which may be constant (for a static type)
// or may be non-constant (for reflective code). It will work equally well
// for either, and the graph will fold nicely if the optimizer later reduces
// the type to a constant.
// The optional arguments are for specialized use by intrinsics:
// - If 'extra_slow_test' if not null is an extra condition for the slow-path.
// - If 'return_size_val', report the the total object size to the caller.
// - deoptimize_on_exception controls how Java exceptions are handled (rethrow vs deoptimize)
Node* GraphKit::new_instance(Node* klass_node,
Node* extra_slow_test,
Node* *return_size_val,
bool deoptimize_on_exception) {
// Compute size in doublewords
// The size is always an integral number of doublewords, represented
// as a positive bytewise size stored in the klass's layout_helper.
// The layout_helper also encodes (in a low bit) the need for a slow path.
jint layout_con = Klass::_lh_neutral_value;
Node* layout_val = get_layout_helper(klass_node, layout_con);
int layout_is_con = (layout_val == NULL);
if (extra_slow_test == NULL) extra_slow_test = intcon(0);
// Generate the initial go-slow test. It's either ALWAYS (return a
// Node for 1) or NEVER (return a NULL) or perhaps (in the reflective
// case) a computed value derived from the layout_helper.
Node* initial_slow_test = NULL;
if (layout_is_con) {
assert(!StressReflectiveCode, "stress mode does not use these paths");
bool must_go_slow = Klass::layout_helper_needs_slow_path(layout_con);
initial_slow_test = must_go_slow? intcon(1): extra_slow_test;
} else { // reflective case
// This reflective path is used by Unsafe.allocateInstance.
// (It may be stress-tested by specifying StressReflectiveCode.)
// Basically, we want to get into the VM is there's an illegal argument.
Node* bit = intcon(Klass::_lh_instance_slow_path_bit);
initial_slow_test = _gvn.transform( new AndINode(layout_val, bit) );
if (extra_slow_test != intcon(0)) {
initial_slow_test = _gvn.transform( new OrINode(initial_slow_test, extra_slow_test) );
}
// (Macro-expander will further convert this to a Bool, if necessary.)
}
// Find the size in bytes. This is easy; it's the layout_helper.
// The size value must be valid even if the slow path is taken.
Node* size = NULL;
if (layout_is_con) {
size = MakeConX(Klass::layout_helper_size_in_bytes(layout_con));
} else { // reflective case
// This reflective path is used by clone and Unsafe.allocateInstance.
size = ConvI2X(layout_val);
// Clear the low bits to extract layout_helper_size_in_bytes:
assert((int)Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
Node* mask = MakeConX(~ (intptr_t)right_n_bits(LogBytesPerLong));
size = _gvn.transform( new AndXNode(size, mask) );
}
if (return_size_val != NULL) {
(*return_size_val) = size;
}
// This is a precise notnull oop of the klass.
// (Actually, it need not be precise if this is a reflective allocation.)
// It's what we cast the result to.
const TypeKlassPtr* tklass = _gvn.type(klass_node)->isa_klassptr();
if (!tklass) tklass = TypeKlassPtr::OBJECT;
const TypeOopPtr* oop_type = tklass->as_instance_type();
// Now generate allocation code
// The entire memory state is needed for slow path of the allocation
// since GC and deoptimization can happened.
Node *mem = reset_memory();
set_all_memory(mem); // Create new memory state
AllocateNode* alloc = new AllocateNode(C, AllocateNode::alloc_type(Type::TOP),
control(), mem, i_o(),
size, klass_node,
initial_slow_test);
return set_output_for_allocation(alloc, oop_type, deoptimize_on_exception);
}
//-------------------------------new_array-------------------------------------
// helper for both newarray and anewarray
// The 'length' parameter is (obviously) the length of the array.
// See comments on new_instance for the meaning of the other arguments.
Node* GraphKit::new_array(Node* klass_node, // array klass (maybe variable)
Node* length, // number of array elements
int nargs, // number of arguments to push back for uncommon trap
Node* *return_size_val,
bool deoptimize_on_exception) {
jint layout_con = Klass::_lh_neutral_value;
Node* layout_val = get_layout_helper(klass_node, layout_con);
int layout_is_con = (layout_val == NULL);
if (!layout_is_con && !StressReflectiveCode &&
!too_many_traps(Deoptimization::Reason_class_check)) {
// This is a reflective array creation site.
// Optimistically assume that it is a subtype of Object[],
// so that we can fold up all the address arithmetic.
layout_con = Klass::array_layout_helper(T_OBJECT);
Node* cmp_lh = _gvn.transform( new CmpINode(layout_val, intcon(layout_con)) );
Node* bol_lh = _gvn.transform( new BoolNode(cmp_lh, BoolTest::eq) );
{ BuildCutout unless(this, bol_lh, PROB_MAX);
inc_sp(nargs);
uncommon_trap(Deoptimization::Reason_class_check,
Deoptimization::Action_maybe_recompile);
}
layout_val = NULL;
layout_is_con = true;
}
// Generate the initial go-slow test. Make sure we do not overflow
// if length is huge (near 2Gig) or negative! We do not need
// exact double-words here, just a close approximation of needed
// double-words. We can't add any offset or rounding bits, lest we
// take a size -1 of bytes and make it positive. Use an unsigned
// compare, so negative sizes look hugely positive.
int fast_size_limit = FastAllocateSizeLimit;
if (layout_is_con) {
assert(!StressReflectiveCode, "stress mode does not use these paths");
// Increase the size limit if we have exact knowledge of array type.
int log2_esize = Klass::layout_helper_log2_element_size(layout_con);
fast_size_limit <<= (LogBytesPerLong - log2_esize);
}
Node* initial_slow_cmp = _gvn.transform( new CmpUNode( length, intcon( fast_size_limit ) ) );
Node* initial_slow_test = _gvn.transform( new BoolNode( initial_slow_cmp, BoolTest::gt ) );
if (initial_slow_test->is_Bool()) {
// Hide it behind a CMoveI, or else PhaseIdealLoop::split_up will get sick.
initial_slow_test = initial_slow_test->as_Bool()->as_int_value(&_gvn);
}
// --- Size Computation ---
// array_size = round_to_heap(array_header + (length << elem_shift));
// where round_to_heap(x) == round_to(x, MinObjAlignmentInBytes)
// and round_to(x, y) == ((x + y-1) & ~(y-1))
// The rounding mask is strength-reduced, if possible.
int round_mask = MinObjAlignmentInBytes - 1;
Node* header_size = NULL;
int header_size_min = arrayOopDesc::base_offset_in_bytes(T_BYTE);
// (T_BYTE has the weakest alignment and size restrictions...)
if (layout_is_con) {
int hsize = Klass::layout_helper_header_size(layout_con);
int eshift = Klass::layout_helper_log2_element_size(layout_con);
BasicType etype = Klass::layout_helper_element_type(layout_con);
if ((round_mask & ~right_n_bits(eshift)) == 0)
round_mask = 0; // strength-reduce it if it goes away completely
assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
assert(header_size_min <= hsize, "generic minimum is smallest");
header_size_min = hsize;
header_size = intcon(hsize + round_mask);
} else {
Node* hss = intcon(Klass::_lh_header_size_shift);
Node* hsm = intcon(Klass::_lh_header_size_mask);
Node* hsize = _gvn.transform( new URShiftINode(layout_val, hss) );
hsize = _gvn.transform( new AndINode(hsize, hsm) );
Node* mask = intcon(round_mask);
header_size = _gvn.transform( new AddINode(hsize, mask) );
}
Node* elem_shift = NULL;
if (layout_is_con) {
int eshift = Klass::layout_helper_log2_element_size(layout_con);
if (eshift != 0)
elem_shift = intcon(eshift);
} else {
// There is no need to mask or shift this value.
// The semantics of LShiftINode include an implicit mask to 0x1F.
assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
elem_shift = layout_val;
}
// Transition to native address size for all offset calculations:
Node* lengthx = ConvI2X(length);
Node* headerx = ConvI2X(header_size);
#ifdef _LP64
{ const TypeLong* tllen = _gvn.find_long_type(lengthx);
if (tllen != NULL && tllen->_lo < 0) {
// Add a manual constraint to a positive range. Cf. array_element_address.
jlong size_max = arrayOopDesc::max_array_length(T_BYTE);
if (size_max > tllen->_hi) size_max = tllen->_hi;
const TypeLong* tlcon = TypeLong::make(CONST64(0), size_max, Type::WidenMin);
lengthx = _gvn.transform( new ConvI2LNode(length, tlcon));
}
}
#endif
// Combine header size (plus rounding) and body size. Then round down.
// This computation cannot overflow, because it is used only in two
// places, one where the length is sharply limited, and the other
// after a successful allocation.
Node* abody = lengthx;
if (elem_shift != NULL)
abody = _gvn.transform( new LShiftXNode(lengthx, elem_shift) );
Node* size = _gvn.transform( new AddXNode(headerx, abody) );
if (round_mask != 0) {
Node* mask = MakeConX(~round_mask);
size = _gvn.transform( new AndXNode(size, mask) );
}
// else if round_mask == 0, the size computation is self-rounding
if (return_size_val != NULL) {
// This is the size
(*return_size_val) = size;
}
// Now generate allocation code
// The entire memory state is needed for slow path of the allocation
// since GC and deoptimization can happened.
Node *mem = reset_memory();
set_all_memory(mem); // Create new memory state
// Create the AllocateArrayNode and its result projections
AllocateArrayNode* alloc
= new AllocateArrayNode(C, AllocateArrayNode::alloc_type(TypeInt::INT),
control(), mem, i_o(),
size, klass_node,
initial_slow_test,
length);
// Cast to correct type. Note that the klass_node may be constant or not,
// and in the latter case the actual array type will be inexact also.
// (This happens via a non-constant argument to inline_native_newArray.)
// In any case, the value of klass_node provides the desired array type.
const TypeInt* length_type = _gvn.find_int_type(length);
const TypeOopPtr* ary_type = _gvn.type(klass_node)->is_klassptr()->as_instance_type();
if (ary_type->isa_aryptr() && length_type != NULL) {
// Try to get a better type than POS for the size
ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
}
Node* javaoop = set_output_for_allocation(alloc, ary_type, deoptimize_on_exception);
// Cast length on remaining path to be as narrow as possible
if (map()->find_edge(length) >= 0) {
Node* ccast = alloc->make_ideal_length(ary_type, &_gvn);
if (ccast != length) {
_gvn.set_type_bottom(ccast);
record_for_igvn(ccast);
replace_in_map(length, ccast);
}
}
return javaoop;
}
// The following "Ideal_foo" functions are placed here because they recognize
// the graph shapes created by the functions immediately above.
//---------------------------Ideal_allocation----------------------------------
// Given an oop pointer or raw pointer, see if it feeds from an AllocateNode.
AllocateNode* AllocateNode::Ideal_allocation(Node* ptr, PhaseTransform* phase) {
if (ptr == NULL) { // reduce dumb test in callers
return NULL;
}
if (ptr->is_CheckCastPP()) { // strip only one raw-to-oop cast
ptr = ptr->in(1);
if (ptr == NULL) return NULL;
}
// Return NULL for allocations with several casts:
// j.l.reflect.Array.newInstance(jobject, jint)
// Object.clone()
// to keep more precise type from last cast.
if (ptr->is_Proj()) {
Node* allo = ptr->in(0);
if (allo != NULL && allo->is_Allocate()) {
return allo->as_Allocate();
}
}
// Report failure to match.
return NULL;
}
// Fancy version which also strips off an offset (and reports it to caller).
AllocateNode* AllocateNode::Ideal_allocation(Node* ptr, PhaseTransform* phase,
intptr_t& offset) {
Node* base = AddPNode::Ideal_base_and_offset(ptr, phase, offset);
if (base == NULL) return NULL;
return Ideal_allocation(base, phase);
}
// Trace Initialize <- Proj[Parm] <- Allocate
AllocateNode* InitializeNode::allocation() {
Node* rawoop = in(InitializeNode::RawAddress);
if (rawoop->is_Proj()) {
Node* alloc = rawoop->in(0);
if (alloc->is_Allocate()) {
return alloc->as_Allocate();
}
}
return NULL;
}
// Trace Allocate -> Proj[Parm] -> Initialize
InitializeNode* AllocateNode::initialization() {
ProjNode* rawoop = proj_out(AllocateNode::RawAddress);
if (rawoop == NULL) return NULL;
for (DUIterator_Fast imax, i = rawoop->fast_outs(imax); i < imax; i++) {
Node* init = rawoop->fast_out(i);
if (init->is_Initialize()) {
assert(init->as_Initialize()->allocation() == this, "2-way link");
return init->as_Initialize();
}
}
return NULL;
}
//----------------------------- loop predicates ---------------------------
//------------------------------add_predicate_impl----------------------------
void GraphKit::add_predicate_impl(Deoptimization::DeoptReason reason, int nargs) {
// Too many traps seen?
if (too_many_traps(reason)) {
#ifdef ASSERT
if (TraceLoopPredicate) {
int tc = C->trap_count(reason);
tty->print("too many traps=%s tcount=%d in ",
Deoptimization::trap_reason_name(reason), tc);
method()->print(); // which method has too many predicate traps
tty->cr();
}
#endif
// We cannot afford to take more traps here,
// do not generate predicate.
return;
}
Node *cont = _gvn.intcon(1);
Node* opq = _gvn.transform(new Opaque1Node(C, cont));
Node *bol = _gvn.transform(new Conv2BNode(opq));
IfNode* iff = create_and_map_if(control(), bol, PROB_MAX, COUNT_UNKNOWN);
Node* iffalse = _gvn.transform(new IfFalseNode(iff));
C->add_predicate_opaq(opq);
{
PreserveJVMState pjvms(this);
set_control(iffalse);
inc_sp(nargs);
uncommon_trap(reason, Deoptimization::Action_maybe_recompile);
}
Node* iftrue = _gvn.transform(new IfTrueNode(iff));
set_control(iftrue);
}
//------------------------------add_predicate---------------------------------
void GraphKit::add_predicate(int nargs) {
if (UseLoopPredicate) {
add_predicate_impl(Deoptimization::Reason_predicate, nargs);
}
// loop's limit check predicate should be near the loop.
if (LoopLimitCheck) {
add_predicate_impl(Deoptimization::Reason_loop_limit_check, nargs);
}
}
//----------------------------- store barriers ----------------------------
#define __ ideal.
void GraphKit::sync_kit(IdealKit& ideal) {
set_all_memory(__ merged_memory());
set_i_o(__ i_o());
set_control(__ ctrl());
}
void GraphKit::final_sync(IdealKit& ideal) {
// Final sync IdealKit and graphKit.
sync_kit(ideal);
}
// vanilla/CMS post barrier
// Insert a write-barrier store. This is to let generational GC work; we have
// to flag all oop-stores before the next GC point.
void GraphKit::write_barrier_post(Node* oop_store,
Node* obj,
Node* adr,
uint adr_idx,
Node* val,
bool use_precise) {
// No store check needed if we're storing a NULL or an old object
// (latter case is probably a string constant). The concurrent
// mark sweep garbage collector, however, needs to have all nonNull
// oop updates flagged via card-marks.
if (val != NULL && val->is_Con()) {
// must be either an oop or NULL
const Type* t = val->bottom_type();
if (t == TypePtr::NULL_PTR || t == Type::TOP)
// stores of null never (?) need barriers
return;
}
if (use_ReduceInitialCardMarks()
&& obj == just_allocated_object(control())) {
// We can skip marks on a freshly-allocated object in Eden.
// Keep this code in sync with new_store_pre_barrier() in runtime.cpp.
// That routine informs GC to take appropriate compensating steps,
// upon a slow-path allocation, so as to make this card-mark
// elision safe.
return;
}
if (!use_precise) {
// All card marks for a (non-array) instance are in one place:
adr = obj;
}
// (Else it's an array (or unknown), and we want more precise card marks.)
assert(adr != NULL, "");
IdealKit ideal(this, true);
// Convert the pointer to an int prior to doing math on it
Node* cast = __ CastPX(__ ctrl(), adr);
// Divide by card size
assert(Universe::heap()->barrier_set()->kind() == BarrierSet::CardTableModRef,
"Only one we handle so far.");
Node* card_offset = __ URShiftX( cast, __ ConI(CardTableModRefBS::card_shift) );
// Combine card table base and card offset
Node* card_adr = __ AddP(__ top(), byte_map_base_node(), card_offset );
// Get the alias_index for raw card-mark memory
int adr_type = Compile::AliasIdxRaw;
Node* zero = __ ConI(0); // Dirty card value
BasicType bt = T_BYTE;
if (UseCondCardMark) {
// The classic GC reference write barrier is typically implemented
// as a store into the global card mark table. Unfortunately
// unconditional stores can result in false sharing and excessive
// coherence traffic as well as false transactional aborts.
// UseCondCardMark enables MP "polite" conditional card mark
// stores. In theory we could relax the load from ctrl() to
// no_ctrl, but that doesn't buy much latitude.
Node* card_val = __ load( __ ctrl(), card_adr, TypeInt::BYTE, bt, adr_type);
__ if_then(card_val, BoolTest::ne, zero);
}
// Smash zero into card
if( !UseConcMarkSweepGC ) {
__ store(__ ctrl(), card_adr, zero, bt, adr_type, MemNode::release);
} else {
// Specialized path for CM store barrier
__ storeCM(__ ctrl(), card_adr, zero, oop_store, adr_idx, bt, adr_type);
}
if (UseCondCardMark) {
__ end_if();
}
// Final sync IdealKit and GraphKit.
final_sync(ideal);
}
/*
* Determine if the G1 pre-barrier can be removed. The pre-barrier is
* required by SATB to make sure all objects live at the start of the
* marking are kept alive, all reference updates need to any previous
* reference stored before writing.
*
* If the previous value is NULL there is no need to save the old value.
* References that are NULL are filtered during runtime by the barrier
* code to avoid unnecessary queuing.
*
* However in the case of newly allocated objects it might be possible to
* prove that the reference about to be overwritten is NULL during compile
* time and avoid adding the barrier code completely.
*
* The compiler needs to determine that the object in which a field is about
* to be written is newly allocated, and that no prior store to the same field
* has happened since the allocation.
*
* Returns true if the pre-barrier can be removed
*/
bool GraphKit::g1_can_remove_pre_barrier(PhaseTransform* phase, Node* adr,
BasicType bt, uint adr_idx) {
intptr_t offset = 0;
Node* base = AddPNode::Ideal_base_and_offset(adr, phase, offset);
AllocateNode* alloc = AllocateNode::Ideal_allocation(base, phase);
if (offset == Type::OffsetBot) {
return false; // cannot unalias unless there are precise offsets
}
if (alloc == NULL) {
return false; // No allocation found
}
intptr_t size_in_bytes = type2aelembytes(bt);
Node* mem = memory(adr_idx); // start searching here...
for (int cnt = 0; cnt < 50; cnt++) {
if (mem->is_Store()) {
Node* st_adr = mem->in(MemNode::Address);
intptr_t st_offset = 0;
Node* st_base = AddPNode::Ideal_base_and_offset(st_adr, phase, st_offset);
if (st_base == NULL) {
break; // inscrutable pointer
}
// Break we have found a store with same base and offset as ours so break
if (st_base == base && st_offset == offset) {
break;
}
if (st_offset != offset && st_offset != Type::OffsetBot) {
const int MAX_STORE = BytesPerLong;
if (st_offset >= offset + size_in_bytes ||
st_offset <= offset - MAX_STORE ||
st_offset <= offset - mem->as_Store()->memory_size()) {
// Success: The offsets are provably independent.
// (You may ask, why not just test st_offset != offset and be done?
// The answer is that stores of different sizes can co-exist
// in the same sequence of RawMem effects. We sometimes initialize
// a whole 'tile' of array elements with a single jint or jlong.)
mem = mem->in(MemNode::Memory);
continue; // advance through independent store memory
}
}
if (st_base != base
&& MemNode::detect_ptr_independence(base, alloc, st_base,
AllocateNode::Ideal_allocation(st_base, phase),
phase)) {
// Success: The bases are provably independent.
mem = mem->in(MemNode::Memory);
continue; // advance through independent store memory
}
} else if (mem->is_Proj() && mem->in(0)->is_Initialize()) {
InitializeNode* st_init = mem->in(0)->as_Initialize();
AllocateNode* st_alloc = st_init->allocation();
// Make sure that we are looking at the same allocation site.
// The alloc variable is guaranteed to not be null here from earlier check.
if (alloc == st_alloc) {
// Check that the initialization is storing NULL so that no previous store
// has been moved up and directly write a reference
Node* captured_store = st_init->find_captured_store(offset,
type2aelembytes(T_OBJECT),
phase);
if (captured_store == NULL || captured_store == st_init->zero_memory()) {
return true;
}
}
}
// Unless there is an explicit 'continue', we must bail out here,
// because 'mem' is an inscrutable memory state (e.g., a call).
break;
}
return false;
}
// G1 pre/post barriers
void GraphKit::g1_write_barrier_pre(bool do_load,
Node* obj,
Node* adr,
uint alias_idx,
Node* val,
const TypeOopPtr* val_type,
Node* pre_val,
BasicType bt) {
// Some sanity checks
// Note: val is unused in this routine.
if (do_load) {
// We need to generate the load of the previous value
assert(obj != NULL, "must have a base");
assert(adr != NULL, "where are loading from?");
assert(pre_val == NULL, "loaded already?");
assert(val_type != NULL, "need a type");
if (use_ReduceInitialCardMarks()
&& g1_can_remove_pre_barrier(&_gvn, adr, bt, alias_idx)) {
return;
}
} else {
// In this case both val_type and alias_idx are unused.
assert(pre_val != NULL, "must be loaded already");
// Nothing to be done if pre_val is null.
if (pre_val->bottom_type() == TypePtr::NULL_PTR) return;
assert(pre_val->bottom_type()->basic_type() == T_OBJECT, "or we shouldn't be here");
}
assert(bt == T_OBJECT, "or we shouldn't be here");
IdealKit ideal(this, true);
Node* tls = __ thread(); // ThreadLocalStorage
Node* no_ctrl = NULL;
Node* no_base = __ top();
Node* zero = __ ConI(0);
Node* zeroX = __ ConX(0);
float likely = PROB_LIKELY(0.999);
float unlikely = PROB_UNLIKELY(0.999);
BasicType active_type = in_bytes(PtrQueue::byte_width_of_active()) == 4 ? T_INT : T_BYTE;
assert(in_bytes(PtrQueue::byte_width_of_active()) == 4 || in_bytes(PtrQueue::byte_width_of_active()) == 1, "flag width");
// Offsets into the thread
const int marking_offset = in_bytes(JavaThread::satb_mark_queue_offset() + // 648
PtrQueue::byte_offset_of_active());
const int index_offset = in_bytes(JavaThread::satb_mark_queue_offset() + // 656
PtrQueue::byte_offset_of_index());
const int buffer_offset = in_bytes(JavaThread::satb_mark_queue_offset() + // 652
PtrQueue::byte_offset_of_buf());
// Now the actual pointers into the thread
Node* marking_adr = __ AddP(no_base, tls, __ ConX(marking_offset));
Node* buffer_adr = __ AddP(no_base, tls, __ ConX(buffer_offset));
Node* index_adr = __ AddP(no_base, tls, __ ConX(index_offset));
// Now some of the values
Node* marking = __ load(__ ctrl(), marking_adr, TypeInt::INT, active_type, Compile::AliasIdxRaw);
// if (!marking)
__ if_then(marking, BoolTest::ne, zero, unlikely); {
BasicType index_bt = TypeX_X->basic_type();
assert(sizeof(size_t) == type2aelembytes(index_bt), "Loading G1 PtrQueue::_index with wrong size.");
Node* index = __ load(__ ctrl(), index_adr, TypeX_X, index_bt, Compile::AliasIdxRaw);
if (do_load) {
// load original value
// alias_idx correct??
pre_val = __ load(__ ctrl(), adr, val_type, bt, alias_idx);
}
// if (pre_val != NULL)
__ if_then(pre_val, BoolTest::ne, null()); {
Node* buffer = __ load(__ ctrl(), buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw);
// is the queue for this thread full?
__ if_then(index, BoolTest::ne, zeroX, likely); {
// decrement the index
Node* next_index = _gvn.transform(new SubXNode(index, __ ConX(sizeof(intptr_t))));
// Now get the buffer location we will log the previous value into and store it
Node *log_addr = __ AddP(no_base, buffer, next_index);
__ store(__ ctrl(), log_addr, pre_val, T_OBJECT, Compile::AliasIdxRaw, MemNode::unordered);
// update the index
__ store(__ ctrl(), index_adr, next_index, index_bt, Compile::AliasIdxRaw, MemNode::unordered);
} __ else_(); {
// logging buffer is full, call the runtime
const TypeFunc *tf = OptoRuntime::g1_wb_pre_Type();
__ make_leaf_call(tf, CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), "g1_wb_pre", pre_val, tls);
} __ end_if(); // (!index)
} __ end_if(); // (pre_val != NULL)
} __ end_if(); // (!marking)
// Final sync IdealKit and GraphKit.
final_sync(ideal);
}
/*
* G1 similar to any GC with a Young Generation requires a way to keep track of
* references from Old Generation to Young Generation to make sure all live
* objects are found. G1 also requires to keep track of object references
* between different regions to enable evacuation of old regions, which is done
* as part of mixed collections. References are tracked in remembered sets and
* is continuously updated as reference are written to with the help of the
* post-barrier.
*
* To reduce the number of updates to the remembered set the post-barrier
* filters updates to fields in objects located in the Young Generation,
* the same region as the reference, when the NULL is being written or
* if the card is already marked as dirty by an earlier write.
*
* Under certain circumstances it is possible to avoid generating the
* post-barrier completely if it is possible during compile time to prove
* the object is newly allocated and that no safepoint exists between the
* allocation and the store.
*
* In the case of slow allocation the allocation code must handle the barrier
* as part of the allocation in the case the allocated object is not located
* in the nursery, this would happen for humongous objects. This is similar to
* how CMS is required to handle this case, see the comments for the method
* CollectedHeap::new_store_pre_barrier and OptoRuntime::new_store_pre_barrier.
* A deferred card mark is required for these objects and handled in the above
* mentioned methods.
*
* Returns true if the post barrier can be removed
*/
bool GraphKit::g1_can_remove_post_barrier(PhaseTransform* phase, Node* store,
Node* adr) {
intptr_t offset = 0;
Node* base = AddPNode::Ideal_base_and_offset(adr, phase, offset);
AllocateNode* alloc = AllocateNode::Ideal_allocation(base, phase);
if (offset == Type::OffsetBot) {
return false; // cannot unalias unless there are precise offsets
}
if (alloc == NULL) {
return false; // No allocation found
}
// Start search from Store node
Node* mem = store->in(MemNode::Control);
if (mem->is_Proj() && mem->in(0)->is_Initialize()) {
InitializeNode* st_init = mem->in(0)->as_Initialize();
AllocateNode* st_alloc = st_init->allocation();
// Make sure we are looking at the same allocation
if (alloc == st_alloc) {
return true;
}
}
return false;
}
//
// Update the card table and add card address to the queue
//
void GraphKit::g1_mark_card(IdealKit& ideal,
Node* card_adr,
Node* oop_store,
uint oop_alias_idx,
Node* index,
Node* index_adr,
Node* buffer,
const TypeFunc* tf) {
Node* zero = __ ConI(0);
Node* zeroX = __ ConX(0);
Node* no_base = __ top();
BasicType card_bt = T_BYTE;
// Smash zero into card. MUST BE ORDERED WRT TO STORE
__ storeCM(__ ctrl(), card_adr, zero, oop_store, oop_alias_idx, card_bt, Compile::AliasIdxRaw);
// Now do the queue work
__ if_then(index, BoolTest::ne, zeroX); {
Node* next_index = _gvn.transform(new SubXNode(index, __ ConX(sizeof(intptr_t))));
Node* log_addr = __ AddP(no_base, buffer, next_index);
// Order, see storeCM.
__ store(__ ctrl(), log_addr, card_adr, T_ADDRESS, Compile::AliasIdxRaw, MemNode::unordered);
__ store(__ ctrl(), index_adr, next_index, TypeX_X->basic_type(), Compile::AliasIdxRaw, MemNode::unordered);
} __ else_(); {
__ make_leaf_call(tf, CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), "g1_wb_post", card_adr, __ thread());
} __ end_if();
}
void GraphKit::g1_write_barrier_post(Node* oop_store,
Node* obj,
Node* adr,
uint alias_idx,
Node* val,
BasicType bt,
bool use_precise) {
// If we are writing a NULL then we need no post barrier
if (val != NULL && val->is_Con() && val->bottom_type() == TypePtr::NULL_PTR) {
// Must be NULL
const Type* t = val->bottom_type();
assert(t == Type::TOP || t == TypePtr::NULL_PTR, "must be NULL");
// No post barrier if writing NULLx
return;
}
if (use_ReduceInitialCardMarks() && obj == just_allocated_object(control())) {
// We can skip marks on a freshly-allocated object in Eden.
// Keep this code in sync with new_store_pre_barrier() in runtime.cpp.
// That routine informs GC to take appropriate compensating steps,
// upon a slow-path allocation, so as to make this card-mark
// elision safe.
return;
}
if (use_ReduceInitialCardMarks()
&& g1_can_remove_post_barrier(&_gvn, oop_store, adr)) {
return;
}
if (!use_precise) {
// All card marks for a (non-array) instance are in one place:
adr = obj;
}
// (Else it's an array (or unknown), and we want more precise card marks.)
assert(adr != NULL, "");
IdealKit ideal(this, true);
Node* tls = __ thread(); // ThreadLocalStorage
Node* no_base = __ top();
float likely = PROB_LIKELY(0.999);
float unlikely = PROB_UNLIKELY(0.999);
Node* young_card = __ ConI((jint)G1SATBCardTableModRefBS::g1_young_card_val());
Node* dirty_card = __ ConI((jint)CardTableModRefBS::dirty_card_val());
Node* zeroX = __ ConX(0);
// Get the alias_index for raw card-mark memory
const TypePtr* card_type = TypeRawPtr::BOTTOM;
const TypeFunc *tf = OptoRuntime::g1_wb_post_Type();
// Offsets into the thread
const int index_offset = in_bytes(JavaThread::dirty_card_queue_offset() +
PtrQueue::byte_offset_of_index());
const int buffer_offset = in_bytes(JavaThread::dirty_card_queue_offset() +
PtrQueue::byte_offset_of_buf());
// Pointers into the thread
Node* buffer_adr = __ AddP(no_base, tls, __ ConX(buffer_offset));
Node* index_adr = __ AddP(no_base, tls, __ ConX(index_offset));
// Now some values
// Use ctrl to avoid hoisting these values past a safepoint, which could
// potentially reset these fields in the JavaThread.
Node* index = __ load(__ ctrl(), index_adr, TypeX_X, TypeX_X->basic_type(), Compile::AliasIdxRaw);
Node* buffer = __ load(__ ctrl(), buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw);
// Convert the store obj pointer to an int prior to doing math on it
// Must use ctrl to prevent "integerized oop" existing across safepoint
Node* cast = __ CastPX(__ ctrl(), adr);
// Divide pointer by card size
Node* card_offset = __ URShiftX( cast, __ ConI(CardTableModRefBS::card_shift) );
// Combine card table base and card offset
Node* card_adr = __ AddP(no_base, byte_map_base_node(), card_offset );
// If we know the value being stored does it cross regions?
if (val != NULL) {
// Does the store cause us to cross regions?
// Should be able to do an unsigned compare of region_size instead of
// and extra shift. Do we have an unsigned compare??
// Node* region_size = __ ConI(1 << HeapRegion::LogOfHRGrainBytes);
Node* xor_res = __ URShiftX ( __ XorX( cast, __ CastPX(__ ctrl(), val)), __ ConI(HeapRegion::LogOfHRGrainBytes));
// if (xor_res == 0) same region so skip
__ if_then(xor_res, BoolTest::ne, zeroX); {
// No barrier if we are storing a NULL
__ if_then(val, BoolTest::ne, null(), unlikely); {
// Ok must mark the card if not already dirty
// load the original value of the card
Node* card_val = __ load(__ ctrl(), card_adr, TypeInt::INT, T_BYTE, Compile::AliasIdxRaw);
__ if_then(card_val, BoolTest::ne, young_card); {
sync_kit(ideal);
// Use Op_MemBarVolatile to achieve the effect of a StoreLoad barrier.
insert_mem_bar(Op_MemBarVolatile, oop_store);
__ sync_kit(this);
Node* card_val_reload = __ load(__ ctrl(), card_adr, TypeInt::INT, T_BYTE, Compile::AliasIdxRaw);
__ if_then(card_val_reload, BoolTest::ne, dirty_card); {
g1_mark_card(ideal, card_adr, oop_store, alias_idx, index, index_adr, buffer, tf);
} __ end_if();
} __ end_if();
} __ end_if();
} __ end_if();
} else {
// Object.clone() instrinsic uses this path.
g1_mark_card(ideal, card_adr, oop_store, alias_idx, index, index_adr, buffer, tf);
}
// Final sync IdealKit and GraphKit.
final_sync(ideal);
}
#undef __
Node* GraphKit::load_String_offset(Node* ctrl, Node* str) {
if (java_lang_String::has_offset_field()) {
int offset_offset = java_lang_String::offset_offset_in_bytes();
const TypeInstPtr* string_type = TypeInstPtr::make(TypePtr::NotNull, C->env()->String_klass(),
false, NULL, 0);
const TypePtr* offset_field_type = string_type->add_offset(offset_offset);
int offset_field_idx = C->get_alias_index(offset_field_type);
return make_load(ctrl,
basic_plus_adr(str, str, offset_offset),
TypeInt::INT, T_INT, offset_field_idx, MemNode::unordered);
} else {
return intcon(0);
}
}
Node* GraphKit::load_String_length(Node* ctrl, Node* str) {
if (java_lang_String::has_count_field()) {
int count_offset = java_lang_String::count_offset_in_bytes();
const TypeInstPtr* string_type = TypeInstPtr::make(TypePtr::NotNull, C->env()->String_klass(),
false, NULL, 0);
const TypePtr* count_field_type = string_type->add_offset(count_offset);
int count_field_idx = C->get_alias_index(count_field_type);
return make_load(ctrl,
basic_plus_adr(str, str, count_offset),
TypeInt::INT, T_INT, count_field_idx, MemNode::unordered);
} else {
return load_array_length(load_String_value(ctrl, str));
}
}
Node* GraphKit::load_String_value(Node* ctrl, Node* str) {
int value_offset = java_lang_String::value_offset_in_bytes();
const TypeInstPtr* string_type = TypeInstPtr::make(TypePtr::NotNull, C->env()->String_klass(),
false, NULL, 0);
const TypePtr* value_field_type = string_type->add_offset(value_offset);
const TypeAryPtr* value_type = TypeAryPtr::make(TypePtr::NotNull,
TypeAry::make(TypeInt::CHAR,TypeInt::POS),
ciTypeArrayKlass::make(T_CHAR), true, 0);
int value_field_idx = C->get_alias_index(value_field_type);
Node* load = make_load(ctrl, basic_plus_adr(str, str, value_offset),
value_type, T_OBJECT, value_field_idx, MemNode::unordered);
// String.value field is known to be @Stable.
if (UseImplicitStableValues) {
load = cast_array_to_stable(load, value_type);
}
return load;
}
void GraphKit::store_String_offset(Node* ctrl, Node* str, Node* value) {
int offset_offset = java_lang_String::offset_offset_in_bytes();
const TypeInstPtr* string_type = TypeInstPtr::make(TypePtr::NotNull, C->env()->String_klass(),
false, NULL, 0);
const TypePtr* offset_field_type = string_type->add_offset(offset_offset);
int offset_field_idx = C->get_alias_index(offset_field_type);
store_to_memory(ctrl, basic_plus_adr(str, offset_offset),
value, T_INT, offset_field_idx, MemNode::unordered);
}
void GraphKit::store_String_value(Node* ctrl, Node* str, Node* value) {
int value_offset = java_lang_String::value_offset_in_bytes();
const TypeInstPtr* string_type = TypeInstPtr::make(TypePtr::NotNull, C->env()->String_klass(),
false, NULL, 0);
const TypePtr* value_field_type = string_type->add_offset(value_offset);
store_oop_to_object(ctrl, str, basic_plus_adr(str, value_offset), value_field_type,
value, TypeAryPtr::CHARS, T_OBJECT, MemNode::unordered);
}
void GraphKit::store_String_length(Node* ctrl, Node* str, Node* value) {
int count_offset = java_lang_String::count_offset_in_bytes();
const TypeInstPtr* string_type = TypeInstPtr::make(TypePtr::NotNull, C->env()->String_klass(),
false, NULL, 0);
const TypePtr* count_field_type = string_type->add_offset(count_offset);
int count_field_idx = C->get_alias_index(count_field_type);
store_to_memory(ctrl, basic_plus_adr(str, count_offset),
value, T_INT, count_field_idx, MemNode::unordered);
}
Node* GraphKit::cast_array_to_stable(Node* ary, const TypeAryPtr* ary_type) {
// Reify the property as a CastPP node in Ideal graph to comply with monotonicity
// assumption of CCP analysis.
return _gvn.transform(new CastPPNode(ary, ary_type->cast_to_stable(true)));
}