if (depth() == 1) {
assert(map()->memory()->Opcode() == Op_Parm, "");
// Insert the memory aliasing node
set_all_memory(reset_memory());
}
assert(merged_memory(), "");
// Now add the locals which are initially bound to arguments:
uint arg_size = tf()->domain()->cnt();
ensure_stack(arg_size - TypeFunc::Parms); // OSR methods have funny args
for (i = TypeFunc::Parms; i < arg_size; i++) {
map()->init_req(i, inmap->argument(_caller, i - TypeFunc::Parms));
}
// Clear out the rest of the map (locals and stack)
for (i = arg_size; i < len; i++) {
map()->init_req(i, top());
}
SafePointNode* entry_map = stop();
return entry_map;
}
//-----------------------------do_method_entry--------------------------------
// Emit any code needed in the pseudo-block before BCI zero.
// The main thing to do is lock the receiver of a synchronized method.
void Parse::do_method_entry() {
set_parse_bci(InvocationEntryBci); // Pseudo-BCP
set_sp(0); // Java Stack Pointer
NOT_PRODUCT( count_compiled_calls(true/*at_method_entry*/, false/*is_inline*/); )
if (C->env()->dtrace_method_probes()) {
make_dtrace_method_entry(method());
}
// If the method is synchronized, we need to construct a lock node, attach
// it to the Start node, and pin it there.
if (method()->is_synchronized()) {
// Insert a FastLockNode right after the Start which takes as arguments
// the current thread pointer, the "this" pointer & the address of the
// stack slot pair used for the lock. The "this" pointer is a projection
// off the start node, but the locking spot has to be constructed by
// creating a ConLNode of 0, and boxing it with a BoxLockNode. The BoxLockNode
// becomes the second argument to the FastLockNode call. The
// FastLockNode becomes the new control parent to pin it to the start.
// Setup Object Pointer
Node *lock_obj = NULL;
if(method()->is_static()) {
ciInstance* mirror = _method->holder()->java_mirror();
const TypeInstPtr *t_lock = TypeInstPtr::make(mirror);
lock_obj = makecon(t_lock);
} else { // Else pass the "this" pointer,
lock_obj = local(0); // which is Parm0 from StartNode
}
// Clear out dead values from the debug info.
kill_dead_locals();
// Build the FastLockNode
_synch_lock = shared_lock(lock_obj);
}
// Feed profiling data for parameters to the type system so it can
// propagate it as speculative types
record_profiled_parameters_for_speculation();
if (depth() == 1) {
increment_and_test_invocation_counter(Tier2CompileThreshold);
}
}
//------------------------------init_blocks------------------------------------
// Initialize our parser map to contain the types/monitors at method entry.
void Parse::init_blocks() {
// Create the blocks.
_block_count = flow()->block_count();
_blocks = NEW_RESOURCE_ARRAY(Block, _block_count);
Copy::zero_to_bytes(_blocks, sizeof(Block)*_block_count);
int rpo;
// Initialize the structs.
for (rpo = 0; rpo < block_count(); rpo++) {
Block* block = rpo_at(rpo);
block->init_node(this, rpo);
}
// Collect predecessor and successor information.
for (rpo = 0; rpo < block_count(); rpo++) {
Block* block = rpo_at(rpo);
block->init_graph(this);
}
}
//-------------------------------init_node-------------------------------------
void Parse::Block::init_node(Parse* outer, int rpo) {
_flow = outer->flow()->rpo_at(rpo);
_pred_count = 0;
_preds_parsed = 0;
_count = 0;
assert(pred_count() == 0 && preds_parsed() == 0, "sanity");
assert(!(is_merged() || is_parsed() || is_handler() || has_merged_backedge()), "sanity");
assert(_live_locals.size() == 0, "sanity");
// entry point has additional predecessor
if (flow()->is_start()) _pred_count++;
assert(flow()->is_start() == (this == outer->start_block()), "");
}
//-------------------------------init_graph------------------------------------
void Parse::Block::init_graph(Parse* outer) {
// Create the successor list for this parser block.
GrowableArray<ciTypeFlow::Block*>* tfs = flow()->successors();
GrowableArray<ciTypeFlow::Block*>* tfe = flow()->exceptions();
int ns = tfs->length();
int ne = tfe->length();
_num_successors = ns;
_all_successors = ns+ne;
_successors = (ns+ne == 0) ? NULL : NEW_RESOURCE_ARRAY(Block*, ns+ne);
int p = 0;
for (int i = 0; i < ns+ne; i++) {
ciTypeFlow::Block* tf2 = (i < ns) ? tfs->at(i) : tfe->at(i-ns);
Block* block2 = outer->rpo_at(tf2->rpo());
_successors[i] = block2;
// Accumulate pred info for the other block, too.
if (i < ns) {
block2->_pred_count++;
} else {
block2->_is_handler = true;
}
#ifdef ASSERT
// A block's successors must be distinguishable by BCI.
// That is, no bytecode is allowed to branch to two different
// clones of the same code location.
for (int j = 0; j < i; j++) {
Block* block1 = _successors[j];
if (block1 == block2) continue; // duplicates are OK
assert(block1->start() != block2->start(), "successors have unique bcis");
}
#endif
}
// Note: We never call next_path_num along exception paths, so they
// never get processed as "ready". Also, the input phis of exception
// handlers get specially processed, so that
}
//---------------------------successor_for_bci---------------------------------
Parse::Block* Parse::Block::successor_for_bci(int bci) {
for (int i = 0; i < all_successors(); i++) {
Block* block2 = successor_at(i);
if (block2->start() == bci) return block2;
}
// We can actually reach here if ciTypeFlow traps out a block
// due to an unloaded class, and concurrently with compilation the
// class is then loaded, so that a later phase of the parser is
// able to see more of the bytecode CFG. Or, the flow pass and
// the parser can have a minor difference of opinion about executability
// of bytecodes. For example, "obj.field = null" is executable even
// if the field's type is an unloaded class; the flow pass used to
// make a trap for such code.
return NULL;
}
//-----------------------------stack_type_at-----------------------------------
const Type* Parse::Block::stack_type_at(int i) const {
return get_type(flow()->stack_type_at(i));
}
//-----------------------------local_type_at-----------------------------------
const Type* Parse::Block::local_type_at(int i) const {
// Make dead locals fall to bottom.
if (_live_locals.size() == 0) {
MethodLivenessResult live_locals = flow()->outer()->method()->liveness_at_bci(start());
// This bitmap can be zero length if we saw a breakpoint.
// In such cases, pretend they are all live.
((Block*)this)->_live_locals = live_locals;
}
if (_live_locals.size() > 0 && !_live_locals.at(i))
return Type::BOTTOM;
return get_type(flow()->local_type_at(i));
}
#ifndef PRODUCT
//----------------------------name_for_bc--------------------------------------
// helper method for BytecodeParseHistogram
static const char* name_for_bc(int i) {
return Bytecodes::is_defined(i) ? Bytecodes::name(Bytecodes::cast(i)) : "xxxunusedxxx";
}
//----------------------------BytecodeParseHistogram------------------------------------
Parse::BytecodeParseHistogram::BytecodeParseHistogram(Parse *p, Compile *c) {
_parser = p;
_compiler = c;
if( ! _initialized ) { _initialized = true; reset(); }
}
//----------------------------current_count------------------------------------
int Parse::BytecodeParseHistogram::current_count(BPHType bph_type) {
switch( bph_type ) {
case BPH_transforms: { return _parser->gvn().made_progress(); }
case BPH_values: { return _parser->gvn().made_new_values(); }
default: { ShouldNotReachHere(); return 0; }
}
}
//----------------------------initialized--------------------------------------
bool Parse::BytecodeParseHistogram::initialized() { return _initialized; }
//----------------------------reset--------------------------------------------
void Parse::BytecodeParseHistogram::reset() {
int i = Bytecodes::number_of_codes;
while (i-- > 0) { _bytecodes_parsed[i] = 0; _nodes_constructed[i] = 0; _nodes_transformed[i] = 0; _new_values[i] = 0; }
}
//----------------------------set_initial_state--------------------------------
// Record info when starting to parse one bytecode
void Parse::BytecodeParseHistogram::set_initial_state( Bytecodes::Code bc ) {
if( PrintParseStatistics && !_parser->is_osr_parse() ) {
_initial_bytecode = bc;
_initial_node_count = _compiler->unique();
_initial_transforms = current_count(BPH_transforms);
_initial_values = current_count(BPH_values);
}
}
//----------------------------record_change--------------------------------
// Record results of parsing one bytecode
void Parse::BytecodeParseHistogram::record_change() {
if( PrintParseStatistics && !_parser->is_osr_parse() ) {
++_bytecodes_parsed[_initial_bytecode];
_nodes_constructed [_initial_bytecode] += (_compiler->unique() - _initial_node_count);
_nodes_transformed [_initial_bytecode] += (current_count(BPH_transforms) - _initial_transforms);
_new_values [_initial_bytecode] += (current_count(BPH_values) - _initial_values);
}
}
//----------------------------print--------------------------------------------
void Parse::BytecodeParseHistogram::print(float cutoff) {
ResourceMark rm;
// print profile
int total = 0;
int i = 0;
for( i = 0; i < Bytecodes::number_of_codes; ++i ) { total += _bytecodes_parsed[i]; }
int abs_sum = 0;
tty->cr(); //0123456789012345678901234567890123456789012345678901234567890123456789
tty->print_cr("Histogram of %d parsed bytecodes:", total);
if( total == 0 ) { return; }
tty->cr();
tty->print_cr("absolute: count of compiled bytecodes of this type");
tty->print_cr("relative: percentage contribution to compiled nodes");
tty->print_cr("nodes : Average number of nodes constructed per bytecode");
tty->print_cr("rnodes : Significance towards total nodes constructed, (nodes*relative)");
tty->print_cr("transforms: Average amount of tranform progress per bytecode compiled");
tty->print_cr("values : Average number of node values improved per bytecode");
tty->print_cr("name : Bytecode name");
tty->cr();
tty->print_cr(" absolute relative nodes rnodes transforms values name");
tty->print_cr("----------------------------------------------------------------------");
while (--i > 0) {
int abs = _bytecodes_parsed[i];
float rel = abs * 100.0F / total;
float nodes = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _nodes_constructed[i])/_bytecodes_parsed[i];
float rnodes = _bytecodes_parsed[i] == 0 ? 0 : rel * nodes;
float xforms = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _nodes_transformed[i])/_bytecodes_parsed[i];
float values = _bytecodes_parsed[i] == 0 ? 0 : (1.0F * _new_values [i])/_bytecodes_parsed[i];
if (cutoff <= rel) {
tty->print_cr("%10d %7.2f%% %6.1f %6.2f %6.1f %6.1f %s", abs, rel, nodes, rnodes, xforms, values, name_for_bc(i));
abs_sum += abs;
}
}
tty->print_cr("----------------------------------------------------------------------");
float rel_sum = abs_sum * 100.0F / total;
tty->print_cr("%10d %7.2f%% (cutoff = %.2f%%)", abs_sum, rel_sum, cutoff);
tty->print_cr("----------------------------------------------------------------------");
tty->cr();
}
#endif
//----------------------------load_state_from----------------------------------
// Load block/map/sp. But not do not touch iter/bci.
void Parse::load_state_from(Block* block) {
set_block(block);
// load the block's JVM state:
set_map(block->start_map());
set_sp( block->start_sp());
}
//-----------------------------record_state------------------------------------
void Parse::Block::record_state(Parse* p) {
assert(!is_merged(), "can only record state once, on 1st inflow");
assert(start_sp() == p->sp(), "stack pointer must agree with ciTypeFlow");
set_start_map(p->stop());
}
//------------------------------do_one_block-----------------------------------
void Parse::do_one_block() {
if (TraceOptoParse) {
Block *b = block();
int ns = b->num_successors();
int nt = b->all_successors();
tty->print("Parsing block #%d at bci [%d,%d), successors: ",
block()->rpo(), block()->start(), block()->limit());
for (int i = 0; i < nt; i++) {
tty->print((( i < ns) ? " %d" : " %d(e)"), b->successor_at(i)->rpo());
}
if (b->is_loop_head()) tty->print(" lphd");
tty->cr();
}
assert(block()->is_merged(), "must be merged before being parsed");
block()->mark_parsed();
++_blocks_parsed;
// Set iterator to start of block.
iter().reset_to_bci(block()->start());
CompileLog* log = C->log();
// Parse bytecodes
while (!stopped() && !failing()) {
iter().next();
// Learn the current bci from the iterator:
set_parse_bci(iter().cur_bci());
if (bci() == block()->limit()) {
// Do not walk into the next block until directed by do_all_blocks.
merge(bci());
break;
}
assert(bci() < block()->limit(), "bci still in block");
if (log != NULL) {
// Output an optional context marker, to help place actions
// that occur during parsing of this BC. If there is no log
// output until the next context string, this context string
// will be silently ignored.
log->set_context("bc code='%d' bci='%d'", (int)bc(), bci());
}
if (block()->has_trap_at(bci())) {
// We must respect the flow pass's traps, because it will refuse
// to produce successors for trapping blocks.
int trap_index = block()->flow()->trap_index();
assert(trap_index != 0, "trap index must be valid");
uncommon_trap(trap_index);
break;
}
NOT_PRODUCT( parse_histogram()->set_initial_state(bc()); );
#ifdef ASSERT
int pre_bc_sp = sp();
int inputs, depth;
bool have_se = !stopped() && compute_stack_effects(inputs, depth);
assert(!have_se || pre_bc_sp >= inputs, err_msg_res("have enough stack to execute this BC: pre_bc_sp=%d, inputs=%d", pre_bc_sp, inputs));
#endif //ASSERT
do_one_bytecode();
assert(!have_se || stopped() || failing() || (sp() - pre_bc_sp) == depth,
err_msg_res("incorrect depth prediction: sp=%d, pre_bc_sp=%d, depth=%d", sp(), pre_bc_sp, depth));
do_exceptions();
NOT_PRODUCT( parse_histogram()->record_change(); );
if (log != NULL)
log->clear_context(); // skip marker if nothing was printed
// Fall into next bytecode. Each bytecode normally has 1 sequential
// successor which is typically made ready by visiting this bytecode.
// If the successor has several predecessors, then it is a merge
// point, starts a new basic block, and is handled like other basic blocks.
}
}
//------------------------------merge------------------------------------------
void Parse::set_parse_bci(int bci) {
set_bci(bci);
Node_Notes* nn = C->default_node_notes();
if (nn == NULL) return;
// Collect debug info for inlined calls unless -XX:-DebugInlinedCalls.
if (!DebugInlinedCalls && depth() > 1) {
return;
}
// Update the JVMS annotation, if present.
JVMState* jvms = nn->jvms();
if (jvms != NULL && jvms->bci() != bci) {
// Update the JVMS.
jvms = jvms->clone_shallow(C);
jvms->set_bci(bci);
nn->set_jvms(jvms);
}
}
//------------------------------merge------------------------------------------
// Merge the current mapping into the basic block starting at bci
void Parse::merge(int target_bci) {
Block* target = successor_for_bci(target_bci);
if (target == NULL) { handle_missing_successor(target_bci); return; }
assert(!target->is_ready(), "our arrival must be expected");
int pnum = target->next_path_num();
merge_common(target, pnum);
}
//-------------------------merge_new_path--------------------------------------
// Merge the current mapping into the basic block, using a new path
void Parse::merge_new_path(int target_bci) {
Block* target = successor_for_bci(target_bci);
if (target == NULL) { handle_missing_successor(target_bci); return; }
assert(!target->is_ready(), "new path into frozen graph");
int pnum = target->add_new_path();
merge_common(target, pnum);
}
//-------------------------merge_exception-------------------------------------
// Merge the current mapping into the basic block starting at bci
// The ex_oop must be pushed on the stack, unlike throw_to_exit.
void Parse::merge_exception(int target_bci) {
assert(sp() == 1, "must have only the throw exception on the stack");
Block* target = successor_for_bci(target_bci);
if (target == NULL) { handle_missing_successor(target_bci); return; }
assert(target->is_handler(), "exceptions are handled by special blocks");
int pnum = target->add_new_path();
merge_common(target, pnum);
}
//--------------------handle_missing_successor---------------------------------
void Parse::handle_missing_successor(int target_bci) {
#ifndef PRODUCT
Block* b = block();
int trap_bci = b->flow()->has_trap()? b->flow()->trap_bci(): -1;
tty->print_cr("### Missing successor at bci:%d for block #%d (trap_bci:%d)", target_bci, b->rpo(), trap_bci);
#endif
ShouldNotReachHere();
}
//--------------------------merge_common---------------------------------------
void Parse::merge_common(Parse::Block* target, int pnum) {
if (TraceOptoParse) {
tty->print("Merging state at block #%d bci:%d", target->rpo(), target->start());
}
// Zap extra stack slots to top
assert(sp() == target->start_sp(), "");
clean_stack(sp());
if (!target->is_merged()) { // No prior mapping at this bci
if (TraceOptoParse) { tty->print(" with empty state"); }
// If this path is dead, do not bother capturing it as a merge.
// It is "as if" we had 1 fewer predecessors from the beginning.
if (stopped()) {
if (TraceOptoParse) tty->print_cr(", but path is dead and doesn't count");
return;
}
// Record that a new block has been merged.
++_blocks_merged;
// Make a region if we know there are multiple or unpredictable inputs.
// (Also, if this is a plain fall-through, we might see another region,
// which must not be allowed into this block's map.)
if (pnum > PhiNode::Input // Known multiple inputs.
|| target->is_handler() // These have unpredictable inputs.
|| target->is_loop_head() // Known multiple inputs
|| control()->is_Region()) { // We must hide this guy.
int current_bci = bci();
set_parse_bci(target->start()); // Set target bci
if (target->is_SEL_head()) {
DEBUG_ONLY( target->mark_merged_backedge(block()); )
if (target->start() == 0) {
// Add loop predicate for the special case when
// there are backbranches to the method entry.
add_predicate();
}
}
// Add a Region to start the new basic block. Phis will be added
// later lazily.
int edges = target->pred_count();
if (edges < pnum) edges = pnum; // might be a new path!
RegionNode *r = new (C) RegionNode(edges+1);
gvn().set_type(r, Type::CONTROL);
record_for_igvn(r);
// zap all inputs to NULL for debugging (done in Node(uint) constructor)
// for (int j = 1; j < edges+1; j++) { r->init_req(j, NULL); }
r->init_req(pnum, control());
set_control(r);
set_parse_bci(current_bci); // Restore bci
}
// Convert the existing Parser mapping into a mapping at this bci.
store_state_to(target);
assert(target->is_merged(), "do not come here twice");
} else { // Prior mapping at this bci
if (TraceOptoParse) { tty->print(" with previous state"); }
#ifdef ASSERT
if (target->is_SEL_head()) {
target->mark_merged_backedge(block());
}
#endif
// We must not manufacture more phis if the target is already parsed.
bool nophi = target->is_parsed();
SafePointNode* newin = map();// Hang on to incoming mapping
Block* save_block = block(); // Hang on to incoming block;
load_state_from(target); // Get prior mapping
assert(newin->jvms()->locoff() == jvms()->locoff(), "JVMS layouts agree");
assert(newin->jvms()->stkoff() == jvms()->stkoff(), "JVMS layouts agree");
assert(newin->jvms()->monoff() == jvms()->monoff(), "JVMS layouts agree");
assert(newin->jvms()->endoff() == jvms()->endoff(), "JVMS layouts agree");
// Iterate over my current mapping and the old mapping.
// Where different, insert Phi functions.
// Use any existing Phi functions.
assert(control()->is_Region(), "must be merging to a region");
RegionNode* r = control()->as_Region();
// Compute where to merge into
// Merge incoming control path
r->init_req(pnum, newin->control());
if (pnum == 1) { // Last merge for this Region?
if (!block()->flow()->is_irreducible_entry()) {
Node* result = _gvn.transform_no_reclaim(r);
if (r != result && TraceOptoParse) {
tty->print_cr("Block #%d replace %d with %d", block()->rpo(), r->_idx, result->_idx);
}
}
record_for_igvn(r);
}
// Update all the non-control inputs to map:
assert(TypeFunc::Parms == newin->jvms()->locoff(), "parser map should contain only youngest jvms");
bool check_elide_phi = target->is_SEL_backedge(save_block);
for (uint j = 1; j < newin->req(); j++) {
Node* m = map()->in(j); // Current state of target.
Node* n = newin->in(j); // Incoming change to target state.
PhiNode* phi;
if (m->is_Phi() && m->as_Phi()->region() == r)
phi = m->as_Phi();
else
phi = NULL;
if (m != n) { // Different; must merge
switch (j) {
// Frame pointer and Return Address never changes
case TypeFunc::FramePtr:// Drop m, use the original value
case TypeFunc::ReturnAdr:
break;
case TypeFunc::Memory: // Merge inputs to the MergeMem node
assert(phi == NULL, "the merge contains phis, not vice versa");
merge_memory_edges(n->as_MergeMem(), pnum, nophi);
continue;
default: // All normal stuff
if (phi == NULL) {
const JVMState* jvms = map()->jvms();
if (EliminateNestedLocks &&
jvms->is_mon(j) && jvms->is_monitor_box(j)) {
// BoxLock nodes are not commoning.
// Use old BoxLock node as merged box.
assert(newin->jvms()->is_monitor_box(j), "sanity");
// This assert also tests that nodes are BoxLock.
assert(BoxLockNode::same_slot(n, m), "sanity");
C->gvn_replace_by(n, m);
} else if (!check_elide_phi || !target->can_elide_SEL_phi(j)) {
phi = ensure_phi(j, nophi);
}
}
break;
}
}
// At this point, n might be top if:
// - there is no phi (because TypeFlow detected a conflict), or
// - the corresponding control edges is top (a dead incoming path)
// It is a bug if we create a phi which sees a garbage value on a live path.
if (phi != NULL) {
assert(n != top() || r->in(pnum) == top(), "live value must not be garbage");
assert(phi->region() == r, "");
phi->set_req(pnum, n); // Then add 'n' to the merge
if (pnum == PhiNode::Input) {
// Last merge for this Phi.
// So far, Phis have had a reasonable type from ciTypeFlow.
// Now _gvn will join that with the meet of current inputs.
// BOTTOM is never permissible here, 'cause pessimistically
// Phis of pointers cannot lose the basic pointer type.
debug_only(const Type* bt1 = phi->bottom_type());
assert(bt1 != Type::BOTTOM, "should not be building conflict phis");
map()->set_req(j, _gvn.transform_no_reclaim(phi));
debug_only(const Type* bt2 = phi->bottom_type());
assert(bt2->higher_equal_speculative(bt1), "must be consistent with type-flow");
record_for_igvn(phi);
}
}
} // End of for all values to be merged
if (pnum == PhiNode::Input &&
!r->in(0)) { // The occasional useless Region
assert(control() == r, "");
set_control(r->nonnull_req());
}
map()->merge_replaced_nodes_with(newin);
// newin has been subsumed into the lazy merge, and is now dead.
set_block(save_block);
stop(); // done with this guy, for now
}
if (TraceOptoParse) {
tty->print_cr(" on path %d", pnum);
}
// Done with this parser state.
assert(stopped(), "");
}
//--------------------------merge_memory_edges---------------------------------
void Parse::merge_memory_edges(MergeMemNode* n, int pnum, bool nophi) {
// (nophi means we must not create phis, because we already parsed here)
assert(n != NULL, "");
// Merge the inputs to the MergeMems
MergeMemNode* m = merged_memory();
assert(control()->is_Region(), "must be merging to a region");
RegionNode* r = control()->as_Region();
PhiNode* base = NULL;
MergeMemNode* remerge = NULL;
for (MergeMemStream mms(m, n); mms.next_non_empty2(); ) {
Node *p = mms.force_memory();
Node *q = mms.memory2();
if (mms.is_empty() && nophi) {
// Trouble: No new splits allowed after a loop body is parsed.
// Instead, wire the new split into a MergeMem on the backedge.
// The optimizer will sort it out, slicing the phi.
if (remerge == NULL) {
assert(base != NULL, "");
assert(base->in(0) != NULL, "should not be xformed away");
remerge = MergeMemNode::make(C, base->in(pnum));
gvn().set_type(remerge, Type::MEMORY);
base->set_req(pnum, remerge);
}
remerge->set_memory_at(mms.alias_idx(), q);
continue;
}
assert(!q->is_MergeMem(), "");
PhiNode* phi;
if (p != q) {
phi = ensure_memory_phi(mms.alias_idx(), nophi);
} else {
if (p->is_Phi() && p->as_Phi()->region() == r)
phi = p->as_Phi();
else
phi = NULL;
}
// Insert q into local phi
if (phi != NULL) {
assert(phi->region() == r, "");
p = phi;
phi->set_req(pnum, q);
if (mms.at_base_memory()) {
base = phi; // delay transforming it
} else if (pnum == 1) {
record_for_igvn(phi);
p = _gvn.transform_no_reclaim(phi);
}
mms.set_memory(p);// store back through the iterator
}
}
// Transform base last, in case we must fiddle with remerging.
if (base != NULL && pnum == 1) {
record_for_igvn(base);
m->set_base_memory( _gvn.transform_no_reclaim(base) );
}
}
//------------------------ensure_phis_everywhere-------------------------------
void Parse::ensure_phis_everywhere() {
ensure_phi(TypeFunc::I_O);
// Ensure a phi on all currently known memories.
for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
ensure_memory_phi(mms.alias_idx());
debug_only(mms.set_memory()); // keep the iterator happy
}
// Note: This is our only chance to create phis for memory slices.
// If we miss a slice that crops up later, it will have to be
// merged into the base-memory phi that we are building here.
// Later, the optimizer will comb out the knot, and build separate
// phi-loops for each memory slice that matters.
// Monitors must nest nicely and not get confused amongst themselves.
// Phi-ify everything up to the monitors, though.
uint monoff = map()->jvms()->monoff();
uint nof_monitors = map()->jvms()->nof_monitors();
assert(TypeFunc::Parms == map()->jvms()->locoff(), "parser map should contain only youngest jvms");
bool check_elide_phi = block()->is_SEL_head();
for (uint i = TypeFunc::Parms; i < monoff; i++) {
if (!check_elide_phi || !block()->can_elide_SEL_phi(i)) {
ensure_phi(i);
}
}
// Even monitors need Phis, though they are well-structured.
// This is true for OSR methods, and also for the rare cases where
// a monitor object is the subject of a replace_in_map operation.
// See bugs 4426707 and 5043395.
for (uint m = 0; m < nof_monitors; m++) {
ensure_phi(map()->jvms()->monitor_obj_offset(m));
}
}
//-----------------------------add_new_path------------------------------------
// Add a previously unaccounted predecessor to this block.
int Parse::Block::add_new_path() {
// If there is no map, return the lowest unused path number.
if (!is_merged()) return pred_count()+1; // there will be a map shortly
SafePointNode* map = start_map();
if (!map->control()->is_Region())
return pred_count()+1; // there may be a region some day
RegionNode* r = map->control()->as_Region();
// Add new path to the region.
uint pnum = r->req();
r->add_req(NULL);
for (uint i = 1; i < map->req(); i++) {
Node* n = map->in(i);
if (i == TypeFunc::Memory) {
// Ensure a phi on all currently known memories.
for (MergeMemStream mms(n->as_MergeMem()); mms.next_non_empty(); ) {
Node* phi = mms.memory();
if (phi->is_Phi() && phi->as_Phi()->region() == r) {
assert(phi->req() == pnum, "must be same size as region");
phi->add_req(NULL);
}
}
} else {
if (n->is_Phi() && n->as_Phi()->region() == r) {
assert(n->req() == pnum, "must be same size as region");
n->add_req(NULL);
}
}
}
return pnum;
}
//------------------------------ensure_phi-------------------------------------
// Turn the idx'th entry of the current map into a Phi
PhiNode *Parse::ensure_phi(int idx, bool nocreate) {
SafePointNode* map = this->map();
Node* region = map->control();
assert(region->is_Region(), "");
Node* o = map->in(idx);
assert(o != NULL, "");
if (o == top()) return NULL; // TOP always merges into TOP
if (o->is_Phi() && o->as_Phi()->region() == region) {
return o->as_Phi();
}
// Now use a Phi here for merging
assert(!nocreate, "Cannot build a phi for a block already parsed.");
const JVMState* jvms = map->jvms();
const Type* t = NULL;
if (jvms->is_loc(idx)) {
t = block()->local_type_at(idx - jvms->locoff());
} else if (jvms->is_stk(idx)) {
t = block()->stack_type_at(idx - jvms->stkoff());
} else if (jvms->is_mon(idx)) {
assert(!jvms->is_monitor_box(idx), "no phis for boxes");
t = TypeInstPtr::BOTTOM; // this is sufficient for a lock object
} else if ((uint)idx < TypeFunc::Parms) {
t = o->bottom_type(); // Type::RETURN_ADDRESS or such-like.
} else {
assert(false, "no type information for this phi");
}
// If the type falls to bottom, then this must be a local that
// is mixing ints and oops or some such. Forcing it to top
// makes it go dead.
if (t == Type::BOTTOM) {
map->set_req(idx, top());
return NULL;
}
// Do not create phis for top either.
// A top on a non-null control flow must be an unused even after the.phi.
if (t == Type::TOP || t == Type::HALF) {
map->set_req(idx, top());
return NULL;
}
PhiNode* phi = PhiNode::make(region, o, t);
gvn().set_type(phi, t);
if (C->do_escape_analysis()) record_for_igvn(phi);
map->set_req(idx, phi);
return phi;
}
//--------------------------ensure_memory_phi----------------------------------
// Turn the idx'th slice of the current memory into a Phi
PhiNode *Parse::ensure_memory_phi(int idx, bool nocreate) {
MergeMemNode* mem = merged_memory();
Node* region = control();
assert(region->is_Region(), "");
Node *o = (idx == Compile::AliasIdxBot)? mem->base_memory(): mem->memory_at(idx);
assert(o != NULL && o != top(), "");
PhiNode* phi;
if (o->is_Phi() && o->as_Phi()->region() == region) {
phi = o->as_Phi();
if (phi == mem->base_memory() && idx >= Compile::AliasIdxRaw) {
// clone the shared base memory phi to make a new memory split
assert(!nocreate, "Cannot build a phi for a block already parsed.");
const Type* t = phi->bottom_type();
const TypePtr* adr_type = C->get_adr_type(idx);
phi = phi->slice_memory(adr_type);
gvn().set_type(phi, t);
}
return phi;
}
// Now use a Phi here for merging
assert(!nocreate, "Cannot build a phi for a block already parsed.");
const Type* t = o->bottom_type();
const TypePtr* adr_type = C->get_adr_type(idx);
phi = PhiNode::make(region, o, t, adr_type);
gvn().set_type(phi, t);
if (idx == Compile::AliasIdxBot)
mem->set_base_memory(phi);
else
mem->set_memory_at(idx, phi);
return phi;
}
//------------------------------call_register_finalizer-----------------------
// Check the klass of the receiver and call register_finalizer if the
// class need finalization.
void Parse::call_register_finalizer() {
Node* receiver = local(0);
assert(receiver != NULL && receiver->bottom_type()->isa_instptr() != NULL,
"must have non-null instance type");
const TypeInstPtr *tinst = receiver->bottom_type()->isa_instptr();
if (tinst != NULL && tinst->klass()->is_loaded() && !tinst->klass_is_exact()) {
// The type isn't known exactly so see if CHA tells us anything.
ciInstanceKlass* ik = tinst->klass()->as_instance_klass();
if (!Dependencies::has_finalizable_subclass(ik)) {
// No finalizable subclasses so skip the dynamic check.
C->dependencies()->assert_has_no_finalizable_subclasses(ik);
return;
}
}
// Insert a dynamic test for whether the instance needs
// finalization. In general this will fold up since the concrete
// class is often visible so the access flags are constant.
Node* klass_addr = basic_plus_adr( receiver, receiver, oopDesc::klass_offset_in_bytes() );
Node* klass = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), klass_addr, TypeInstPtr::KLASS));
Node* access_flags_addr = basic_plus_adr(klass, klass, in_bytes(Klass::access_flags_offset()));
Node* access_flags = make_load(NULL, access_flags_addr, TypeInt::INT, T_INT, MemNode::unordered);
Node* mask = _gvn.transform(new (C) AndINode(access_flags, intcon(JVM_ACC_HAS_FINALIZER)));
Node* check = _gvn.transform(new (C) CmpINode(mask, intcon(0)));
Node* test = _gvn.transform(new (C) BoolNode(check, BoolTest::ne));
IfNode* iff = create_and_map_if(control(), test, PROB_MAX, COUNT_UNKNOWN);
RegionNode* result_rgn = new (C) RegionNode(3);
record_for_igvn(result_rgn);
Node *skip_register = _gvn.transform(new (C) IfFalseNode(iff));
result_rgn->init_req(1, skip_register);
Node *needs_register = _gvn.transform(new (C) IfTrueNode(iff));
set_control(needs_register);
if (stopped()) {
// There is no slow path.
result_rgn->init_req(2, top());
} else {
Node *call = make_runtime_call(RC_NO_LEAF,
OptoRuntime::register_finalizer_Type(),
OptoRuntime::register_finalizer_Java(),
NULL, TypePtr::BOTTOM,
receiver);
make_slow_call_ex(call, env()->Throwable_klass(), true);
Node* fast_io = call->in(TypeFunc::I_O);
Node* fast_mem = call->in(TypeFunc::Memory);
// These two phis are pre-filled with copies of of the fast IO and Memory
Node* io_phi = PhiNode::make(result_rgn, fast_io, Type::ABIO);
Node* mem_phi = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
result_rgn->init_req(2, control());
io_phi ->init_req(2, i_o());
mem_phi ->init_req(2, reset_memory());
set_all_memory( _gvn.transform(mem_phi) );
set_i_o( _gvn.transform(io_phi) );
}
set_control( _gvn.transform(result_rgn) );
}
// Add check to deoptimize if RTM state is not ProfileRTM
void Parse::rtm_deopt() {
#if INCLUDE_RTM_OPT
if (C->profile_rtm()) {
assert(C->method() != NULL, "only for normal compilations");
assert(!C->method()->method_data()->is_empty(), "MDO is needed to record RTM state");
assert(depth() == 1, "generate check only for main compiled method");
// Set starting bci for uncommon trap.
set_parse_bci(is_osr_parse() ? osr_bci() : 0);
// Load the rtm_state from the MethodData.
const TypePtr* adr_type = TypeMetadataPtr::make(C->method()->method_data());
Node* mdo = makecon(adr_type);
int offset = MethodData::rtm_state_offset_in_bytes();
Node* adr_node = basic_plus_adr(mdo, mdo, offset);
Node* rtm_state = make_load(control(), adr_node, TypeInt::INT, T_INT, adr_type, MemNode::unordered);
// Separate Load from Cmp by Opaque.
// In expand_macro_nodes() it will be replaced either
// with this load when there are locks in the code
// or with ProfileRTM (cmp->in(2)) otherwise so that
// the check will fold.
Node* profile_state = makecon(TypeInt::make(ProfileRTM));
Node* opq = _gvn.transform( new (C) Opaque3Node(C, rtm_state, Opaque3Node::RTM_OPT) );
Node* chk = _gvn.transform( new (C) CmpINode(opq, profile_state) );
Node* tst = _gvn.transform( new (C) BoolNode(chk, BoolTest::eq) );
// Branch to failure if state was changed
{ BuildCutout unless(this, tst, PROB_ALWAYS);
uncommon_trap(Deoptimization::Reason_rtm_state_change,
Deoptimization::Action_make_not_entrant);
}
}
#endif
}
//------------------------------return_current---------------------------------
// Append current _map to _exit_return
void Parse::return_current(Node* value) {
if (RegisterFinalizersAtInit &&
method()->intrinsic_id() == vmIntrinsics::_Object_init) {
call_register_finalizer();
}
// Do not set_parse_bci, so that return goo is credited to the return insn.
set_bci(InvocationEntryBci);
if (method()->is_synchronized() && GenerateSynchronizationCode) {
shared_unlock(_synch_lock->box_node(), _synch_lock->obj_node());
}
if (C->env()->dtrace_method_probes()) {
make_dtrace_method_exit(method());
}
SafePointNode* exit_return = _exits.map();
exit_return->in( TypeFunc::Control )->add_req( control() );
exit_return->in( TypeFunc::I_O )->add_req( i_o () );
Node *mem = exit_return->in( TypeFunc::Memory );
for (MergeMemStream mms(mem->as_MergeMem(), merged_memory()); mms.next_non_empty2(); ) {
if (mms.is_empty()) {
// get a copy of the base memory, and patch just this one input
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), "");
gvn().set_type_bottom(phi);
phi->del_req(phi->req()-1); // prepare to re-patch
mms.set_memory(phi);
}
mms.memory()->add_req(mms.memory2());
}
// frame pointer is always same, already captured
if (value != NULL) {
// If returning oops to an interface-return, there is a silent free
// cast from oop to interface allowed by the Verifier. Make it explicit
// here.
Node* phi = _exits.argument(0);
const TypeInstPtr *tr = phi->bottom_type()->isa_instptr();
if (tr && tr->klass()->is_loaded() &&
tr->klass()->is_interface()) {
const TypeInstPtr *tp = value->bottom_type()->isa_instptr();
if (tp && tp->klass()->is_loaded() &&
!tp->klass()->is_interface()) {
// sharpen the type eagerly; this eases certain assert checking
if (tp->higher_equal(TypeInstPtr::NOTNULL))
tr = tr->join_speculative(TypeInstPtr::NOTNULL)->is_instptr();
value = _gvn.transform(new (C) CheckCastPPNode(0, value, tr));
}
} else {
// Also handle returns of oop-arrays to an arrays-of-interface return
const TypeInstPtr* phi_tip;
const TypeInstPtr* val_tip;
Type::get_arrays_base_elements(phi->bottom_type(), value->bottom_type(), &phi_tip, &val_tip);
if (phi_tip != NULL && phi_tip->is_loaded() && phi_tip->klass()->is_interface() &&
val_tip != NULL && val_tip->is_loaded() && !val_tip->klass()->is_interface()) {
value = _gvn.transform(new (C) CheckCastPPNode(0, value, phi->bottom_type()));
}
}
phi->add_req(value);
}
if (_first_return) {
_exits.map()->transfer_replaced_nodes_from(map(), _new_idx);
_first_return = false;
} else {
_exits.map()->merge_replaced_nodes_with(map());
}
stop_and_kill_map(); // This CFG path dies here
}
//------------------------------add_safepoint----------------------------------
void Parse::add_safepoint() {
// See if we can avoid this safepoint. No need for a SafePoint immediately
// after a Call (except Leaf Call) or another SafePoint.
Node *proj = control();
bool add_poll_param = SafePointNode::needs_polling_address_input();
uint parms = add_poll_param ? TypeFunc::Parms+1 : TypeFunc::Parms;
if( proj->is_Proj() ) {
Node *n0 = proj->in(0);
if( n0->is_Catch() ) {
n0 = n0->in(0)->in(0);
assert( n0->is_Call(), "expect a call here" );
}
if( n0->is_Call() ) {
if( n0->as_Call()->guaranteed_safepoint() )
return;
} else if( n0->is_SafePoint() && n0->req() >= parms ) {
return;
}
}
// Clear out dead values from the debug info.
kill_dead_locals();
// Clone the JVM State
SafePointNode *sfpnt = new (C) SafePointNode(parms, NULL);
// Capture memory state BEFORE a SafePoint. Since we can block at a
// SafePoint we need our GC state to be safe; i.e. we need all our current
// write barriers (card marks) to not float down after the SafePoint so we
// must read raw memory. Likewise we need all oop stores to match the card
// marks. If deopt can happen, we need ALL stores (we need the correct JVM
// state on a deopt).
// We do not need to WRITE the memory state after a SafePoint. The control
// edge will keep card-marks and oop-stores from floating up from below a
// SafePoint and our true dependency added here will keep them from floating
// down below a SafePoint.
// Clone the current memory state
Node* mem = MergeMemNode::make(C, map()->memory());
mem = _gvn.transform(mem);
// Pass control through the safepoint
sfpnt->init_req(TypeFunc::Control , control());
// Fix edges normally used by a call
sfpnt->init_req(TypeFunc::I_O , top() );
sfpnt->init_req(TypeFunc::Memory , mem );
sfpnt->init_req(TypeFunc::ReturnAdr, top() );
sfpnt->init_req(TypeFunc::FramePtr , top() );
// Create a node for the polling address
if( add_poll_param ) {
Node *polladr = ConPNode::make(C, (address)os::get_polling_page());
sfpnt->init_req(TypeFunc::Parms+0, _gvn.transform(polladr));
}
// Fix up the JVM State edges
add_safepoint_edges(sfpnt);
Node *transformed_sfpnt = _gvn.transform(sfpnt);
set_control(transformed_sfpnt);
// Provide an edge from root to safepoint. This makes the safepoint
// appear useful until the parse has completed.
if( OptoRemoveUseless && transformed_sfpnt->is_SafePoint() ) {
assert(C->root() != NULL, "Expect parse is still valid");
C->root()->add_prec(transformed_sfpnt);
}
}
#ifndef PRODUCT
//------------------------show_parse_info--------------------------------------
void Parse::show_parse_info() {
InlineTree* ilt = NULL;
if (C->ilt() != NULL) {
JVMState* caller_jvms = is_osr_parse() ? caller()->caller() : caller();
ilt = InlineTree::find_subtree_from_root(C->ilt(), caller_jvms, method());
}
if (PrintCompilation && Verbose) {
if (depth() == 1) {
if( ilt->count_inlines() ) {
tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
ilt->count_inline_bcs());
tty->cr();
}
} else {
if (method()->is_synchronized()) tty->print("s");
if (method()->has_exception_handlers()) tty->print("!");
// Check this is not the final compiled version
if (C->trap_can_recompile()) {
tty->print("-");
} else {
tty->print(" ");
}
method()->print_short_name();
if (is_osr_parse()) {
tty->print(" @ %d", osr_bci());
}
tty->print(" (%d bytes)",method()->code_size());
if (ilt->count_inlines()) {
tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
ilt->count_inline_bcs());
}
tty->cr();
}
}
if (PrintOpto && (depth() == 1 || PrintOptoInlining)) {
// Print that we succeeded; suppress this message on the first osr parse.
if (method()->is_synchronized()) tty->print("s");
if (method()->has_exception_handlers()) tty->print("!");
// Check this is not the final compiled version
if (C->trap_can_recompile() && depth() == 1) {
tty->print("-");
} else {
tty->print(" ");
}
if( depth() != 1 ) { tty->print(" "); } // missing compile count
for (int i = 1; i < depth(); ++i) { tty->print(" "); }
method()->print_short_name();
if (is_osr_parse()) {
tty->print(" @ %d", osr_bci());
}
if (ilt->caller_bci() != -1) {
tty->print(" @ %d", ilt->caller_bci());
}
tty->print(" (%d bytes)",method()->code_size());
if (ilt->count_inlines()) {
tty->print(" __inlined %d (%d bytes)", ilt->count_inlines(),
ilt->count_inline_bcs());
}
tty->cr();
}
}
//------------------------------dump-------------------------------------------
// Dump information associated with the bytecodes of current _method
void Parse::dump() {
if( method() != NULL ) {
// Iterate over bytecodes
ciBytecodeStream iter(method());
for( Bytecodes::Code bc = iter.next(); bc != ciBytecodeStream::EOBC() ; bc = iter.next() ) {
dump_bci( iter.cur_bci() );
tty->cr();
}
}
}
// Dump information associated with a byte code index, 'bci'
void Parse::dump_bci(int bci) {
// Output info on merge-points, cloning, and within _jsr..._ret
// NYI
tty->print(" bci:%d", bci);
}
#endif
C:\hotspot-69087d08d473\src\share\vm/opto/parse2.cpp
/*
* Copyright (c) 1998, 2016, 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 "ci/ciMethodData.hpp"
#include "classfile/systemDictionary.hpp"
#include "classfile/vmSymbols.hpp"
#include "compiler/compileLog.hpp"
#include "interpreter/linkResolver.hpp"
#include "memory/universe.inline.hpp"
#include "opto/addnode.hpp"
#include "opto/divnode.hpp"
#include "opto/idealGraphPrinter.hpp"
#include "opto/matcher.hpp"
#include "opto/memnode.hpp"
#include "opto/mulnode.hpp"
#include "opto/parse.hpp"
#include "opto/runtime.hpp"
#include "runtime/deoptimization.hpp"
#include "runtime/sharedRuntime.hpp"
extern int explicit_null_checks_inserted,
explicit_null_checks_elided;
//---------------------------------array_load----------------------------------
void Parse::array_load(BasicType elem_type) {
const Type* elem = Type::TOP;
Node* adr = array_addressing(elem_type, 0, &elem);
if (stopped()) return; // guaranteed null or range check
dec_sp(2); // Pop array and index
const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(elem_type);
Node* ld = make_load(control(), adr, elem, elem_type, adr_type, MemNode::unordered);
push(ld);
}
//--------------------------------array_store----------------------------------
void Parse::array_store(BasicType elem_type) {
const Type* elem = Type::TOP;
Node* adr = array_addressing(elem_type, 1, &elem);
if (stopped()) return; // guaranteed null or range check
Node* val = pop();
dec_sp(2); // Pop array and index
const TypeAryPtr* adr_type = TypeAryPtr::get_array_body_type(elem_type);
if (elem == TypeInt::BOOL) {
elem_type = T_BOOLEAN;
}
store_to_memory(control(), adr, val, elem_type, adr_type, StoreNode::release_if_reference(elem_type));
}
//------------------------------array_addressing-------------------------------
// Pull array and index from the stack. Compute pointer-to-element.
Node* Parse::array_addressing(BasicType type, int vals, const Type* *result2) {
Node *idx = peek(0+vals); // Get from stack without popping
Node *ary = peek(1+vals); // in case of exception
// Null check the array base, with correct stack contents
ary = null_check(ary, T_ARRAY);
// Compile-time detect of null-exception?
if (stopped()) return top();
const TypeAryPtr* arytype = _gvn.type(ary)->is_aryptr();
const TypeInt* sizetype = arytype->size();
const Type* elemtype = arytype->elem();
if (UseUniqueSubclasses && result2 != NULL) {
const Type* el = elemtype->make_ptr();
if (el && el->isa_instptr()) {
const TypeInstPtr* toop = el->is_instptr();
if (toop->klass()->as_instance_klass()->unique_concrete_subklass()) {
// If we load from "AbstractClass[]" we must see "ConcreteSubClass".
const Type* subklass = Type::get_const_type(toop->klass());
elemtype = subklass->join_speculative(el);
}
}
}
// Check for big class initializers with all constant offsets
// feeding into a known-size array.
const TypeInt* idxtype = _gvn.type(idx)->is_int();
// See if the highest idx value is less than the lowest array bound,
// and if the idx value cannot be negative:
bool need_range_check = true;
if (idxtype->_hi < sizetype->_lo && idxtype->_lo >= 0) {
need_range_check = false;
if (C->log() != NULL) C->log()->elem("observe that='!need_range_check'");
}
ciKlass * arytype_klass = arytype->klass();
if ((arytype_klass != NULL) && (!arytype_klass->is_loaded())) {
// Only fails for some -Xcomp runs
// The class is unloaded. We have to run this bytecode in the interpreter.
uncommon_trap(Deoptimization::Reason_unloaded,
Deoptimization::Action_reinterpret,
arytype->klass(), "!loaded array");
return top();
}
// Do the range check
if (GenerateRangeChecks && need_range_check) {
Node* tst;
if (sizetype->_hi <= 0) {
// The greatest array bound is negative, so we can conclude that we're
// compiling unreachable code, but the unsigned compare trick used below
// only works with non-negative lengths. Instead, hack "tst" to be zero so
// the uncommon_trap path will always be taken.
tst = _gvn.intcon(0);
} else {
// Range is constant in array-oop, so we can use the original state of mem
Node* len = load_array_length(ary);
// Test length vs index (standard trick using unsigned compare)
Node* chk = _gvn.transform( new (C) CmpUNode(idx, len) );
BoolTest::mask btest = BoolTest::lt;
tst = _gvn.transform( new (C) BoolNode(chk, btest) );
}
// Branch to failure if out of bounds
{ BuildCutout unless(this, tst, PROB_MAX);
if (C->allow_range_check_smearing()) {
// Do not use builtin_throw, since range checks are sometimes
// made more stringent by an optimistic transformation.
// This creates "tentative" range checks at this point,
// which are not guaranteed to throw exceptions.
// See IfNode::Ideal, is_range_check, adjust_check.
uncommon_trap(Deoptimization::Reason_range_check,
Deoptimization::Action_make_not_entrant,
NULL, "range_check");
} else {
// If we have already recompiled with the range-check-widening
// heroic optimization turned off, then we must really be throwing
// range check exceptions.
builtin_throw(Deoptimization::Reason_range_check, idx);
}
}
}
// Check for always knowing you are throwing a range-check exception
if (stopped()) return top();
// Make array address computation control dependent to prevent it
// from floating above the range check during loop optimizations.
Node* ptr = array_element_address(ary, idx, type, sizetype, control());
if (result2 != NULL) *result2 = elemtype;
assert(ptr != top(), "top should go hand-in-hand with stopped");
return ptr;
}
// returns IfNode
IfNode* Parse::jump_if_fork_int(Node* a, Node* b, BoolTest::mask mask) {
Node *cmp = _gvn.transform( new (C) CmpINode( a, b)); // two cases: shiftcount > 32 and shiftcount <= 32
Node *tst = _gvn.transform( new (C) BoolNode( cmp, mask));
IfNode *iff = create_and_map_if( control(), tst, ((mask == BoolTest::eq) ? PROB_STATIC_INFREQUENT : PROB_FAIR), COUNT_UNKNOWN );
return iff;
}
// return Region node
Node* Parse::jump_if_join(Node* iffalse, Node* iftrue) {
Node *region = new (C) RegionNode(3); // 2 results
record_for_igvn(region);
region->init_req(1, iffalse);
region->init_req(2, iftrue );
_gvn.set_type(region, Type::CONTROL);
region = _gvn.transform(region);
set_control (region);
return region;
}
//------------------------------helper for tableswitch-------------------------
void Parse::jump_if_true_fork(IfNode *iff, int dest_bci_if_true, int prof_table_index) {
// True branch, use existing map info
{ PreserveJVMState pjvms(this);
Node *iftrue = _gvn.transform( new (C) IfTrueNode (iff) );
set_control( iftrue );
profile_switch_case(prof_table_index);
merge_new_path(dest_bci_if_true);
}
// False branch
Node *iffalse = _gvn.transform( new (C) IfFalseNode(iff) );
set_control( iffalse );
}
void Parse::jump_if_false_fork(IfNode *iff, int dest_bci_if_true, int prof_table_index) {
// True branch, use existing map info
{ PreserveJVMState pjvms(this);
Node *iffalse = _gvn.transform( new (C) IfFalseNode (iff) );
set_control( iffalse );
profile_switch_case(prof_table_index);
merge_new_path(dest_bci_if_true);
}
// False branch
Node *iftrue = _gvn.transform( new (C) IfTrueNode(iff) );
set_control( iftrue );
}
void Parse::jump_if_always_fork(int dest_bci, int prof_table_index) {
// False branch, use existing map and control()
profile_switch_case(prof_table_index);
merge_new_path(dest_bci);
}
extern "C" {
static int jint_cmp(const void *i, const void *j) {
int a = *(jint *)i;
int b = *(jint *)j;
return a > b ? 1 : a < b ? -1 : 0;
}
}
// Default value for methodData switch indexing. Must be a negative value to avoid
// conflict with any legal switch index.
#define NullTableIndex -1
class SwitchRange : public StackObj {
// a range of integers coupled with a bci destination
jint _lo; // inclusive lower limit
jint _hi; // inclusive upper limit
int _dest;
int _table_index; // index into method data table
public:
jint lo() const { return _lo; }
jint hi() const { return _hi; }
int dest() const { return _dest; }
int table_index() const { return _table_index; }
bool is_singleton() const { return _lo == _hi; }
void setRange(jint lo, jint hi, int dest, int table_index) {
assert(lo <= hi, "must be a non-empty range");
_lo = lo, _hi = hi; _dest = dest; _table_index = table_index;
}
bool adjoinRange(jint lo, jint hi, int dest, int table_index) {
assert(lo <= hi, "must be a non-empty range");
if (lo == _hi+1 && dest == _dest && table_index == _table_index) {
_hi = hi;
return true;
}
return false;
}
void set (jint value, int dest, int table_index) {
setRange(value, value, dest, table_index);
}
bool adjoin(jint value, int dest, int table_index) {
return adjoinRange(value, value, dest, table_index);
}
void print() {
if (is_singleton())
tty->print(" {%d}=>%d", lo(), dest());
else if (lo() == min_jint)
tty->print(" {..%d}=>%d", hi(), dest());
else if (hi() == max_jint)
tty->print(" {%d..}=>%d", lo(), dest());
else
tty->print(" {%d..%d}=>%d", lo(), hi(), dest());
}
};
//-------------------------------do_tableswitch--------------------------------
void Parse::do_tableswitch() {
Node* lookup = pop();
// Get information about tableswitch
int default_dest = iter().get_dest_table(0);
int lo_index = iter().get_int_table(1);
int hi_index = iter().get_int_table(2);
int len = hi_index - lo_index + 1;
if (len < 1) {
// If this is a backward branch, add safepoint
maybe_add_safepoint(default_dest);
merge(default_dest);
return;
}
// generate decision tree, using trichotomy when possible
int rnum = len+2;
bool makes_backward_branch = false;
SwitchRange* ranges = NEW_RESOURCE_ARRAY(SwitchRange, rnum);
int rp = -1;
if (lo_index != min_jint) {
ranges[++rp].setRange(min_jint, lo_index-1, default_dest, NullTableIndex);
}
for (int j = 0; j < len; j++) {
jint match_int = lo_index+j;
int dest = iter().get_dest_table(j+3);
makes_backward_branch |= (dest <= bci());
int table_index = method_data_update() ? j : NullTableIndex;
if (rp < 0 || !ranges[rp].adjoin(match_int, dest, table_index)) {
ranges[++rp].set(match_int, dest, table_index);
}
}
jint highest = lo_index+(len-1);
assert(ranges[rp].hi() == highest, "");
if (highest != max_jint
&& !ranges[rp].adjoinRange(highest+1, max_jint, default_dest, NullTableIndex)) {
ranges[++rp].setRange(highest+1, max_jint, default_dest, NullTableIndex);
}
assert(rp < len+2, "not too many ranges");
// Safepoint in case if backward branch observed
if( makes_backward_branch && UseLoopSafepoints )
add_safepoint();
jump_switch_ranges(lookup, &ranges[0], &ranges[rp]);
}
//------------------------------do_lookupswitch--------------------------------
void Parse::do_lookupswitch() {
Node *lookup = pop(); // lookup value
// Get information about lookupswitch
int default_dest = iter().get_dest_table(0);
int len = iter().get_int_table(1);
if (len < 1) { // If this is a backward branch, add safepoint
maybe_add_safepoint(default_dest);
merge(default_dest);
return;
}
// generate decision tree, using trichotomy when possible
jint* table = NEW_RESOURCE_ARRAY(jint, len*2);
{
for( int j = 0; j < len; j++ ) {
table[j+j+0] = iter().get_int_table(2+j+j);
table[j+j+1] = iter().get_dest_table(2+j+j+1);
}
qsort( table, len, 2*sizeof(table[0]), jint_cmp );
}
int rnum = len*2+1;
bool makes_backward_branch = false;
SwitchRange* ranges = NEW_RESOURCE_ARRAY(SwitchRange, rnum);
int rp = -1;
for( int j = 0; j < len; j++ ) {
jint match_int = table[j+j+0];
int dest = table[j+j+1];
int next_lo = rp < 0 ? min_jint : ranges[rp].hi()+1;
int table_index = method_data_update() ? j : NullTableIndex;
makes_backward_branch |= (dest <= bci());
if( match_int != next_lo ) {
ranges[++rp].setRange(next_lo, match_int-1, default_dest, NullTableIndex);
}
if( rp < 0 || !ranges[rp].adjoin(match_int, dest, table_index) ) {
ranges[++rp].set(match_int, dest, table_index);
}
}
jint highest = table[2*(len-1)];
assert(ranges[rp].hi() == highest, "");
if( highest != max_jint
&& !ranges[rp].adjoinRange(highest+1, max_jint, default_dest, NullTableIndex) ) {
ranges[++rp].setRange(highest+1, max_jint, default_dest, NullTableIndex);
}
assert(rp < rnum, "not too many ranges");
// Safepoint in case backward branch observed
if( makes_backward_branch && UseLoopSafepoints )
add_safepoint();
jump_switch_ranges(lookup, &ranges[0], &ranges[rp]);
}
//----------------------------create_jump_tables-------------------------------
bool Parse::create_jump_tables(Node* key_val, SwitchRange* lo, SwitchRange* hi) {
// Are jumptables enabled
if (!UseJumpTables) return false;
// Are jumptables supported
if (!Matcher::has_match_rule(Op_Jump)) return false;
// Don't make jump table if profiling
if (method_data_update()) return false;
// Decide if a guard is needed to lop off big ranges at either (or
// both) end(s) of the input set. We'll call this the default target
// even though we can't be sure that it is the true "default".
bool needs_guard = false;
int default_dest;
int64 total_outlier_size = 0;
int64 hi_size = ((int64)hi->hi()) - ((int64)hi->lo()) + 1;
int64 lo_size = ((int64)lo->hi()) - ((int64)lo->lo()) + 1;
if (lo->dest() == hi->dest()) {
total_outlier_size = hi_size + lo_size;
default_dest = lo->dest();
} else if (lo_size > hi_size) {
total_outlier_size = lo_size;
default_dest = lo->dest();
} else {
total_outlier_size = hi_size;
default_dest = hi->dest();
}
// If a guard test will eliminate very sparse end ranges, then
// it is worth the cost of an extra jump.
if (total_outlier_size > (MaxJumpTableSparseness * 4)) {
needs_guard = true;
if (default_dest == lo->dest()) lo++;
if (default_dest == hi->dest()) hi--;
}
// Find the total number of cases and ranges
int64 num_cases = ((int64)hi->hi()) - ((int64)lo->lo()) + 1;
int num_range = hi - lo + 1;
// Don't create table if: too large, too small, or too sparse.
if (num_cases < MinJumpTableSize || num_cases > MaxJumpTableSize)
return false;
if (num_cases > (MaxJumpTableSparseness * num_range))
return false;
// Normalize table lookups to zero
int lowval = lo->lo();
key_val = _gvn.transform( new (C) SubINode(key_val, _gvn.intcon(lowval)) );
// Generate a guard to protect against input keyvals that aren't
// in the switch domain.
if (needs_guard) {
Node* size = _gvn.intcon(num_cases);
Node* cmp = _gvn.transform( new (C) CmpUNode(key_val, size) );
Node* tst = _gvn.transform( new (C) BoolNode(cmp, BoolTest::ge) );
IfNode* iff = create_and_map_if( control(), tst, PROB_FAIR, COUNT_UNKNOWN);
jump_if_true_fork(iff, default_dest, NullTableIndex);
}
// Create an ideal node JumpTable that has projections
// of all possible ranges for a switch statement
// The key_val input must be converted to a pointer offset and scaled.
// Compare Parse::array_addressing above.
#ifdef _LP64
// Clean the 32-bit int into a real 64-bit offset.
// Otherwise, the jint value 0 might turn into an offset of 0x0800000000.
const TypeInt* ikeytype = TypeInt::make(0, num_cases-1, Type::WidenMin);
// Make I2L conversion control dependent to prevent it from
// floating above the range check during loop optimizations.
key_val = C->constrained_convI2L(&_gvn, key_val, ikeytype, control());
#endif
// Shift the value by wordsize so we have an index into the table, rather
// than a switch value
Node *shiftWord = _gvn.MakeConX(wordSize);
key_val = _gvn.transform( new (C) MulXNode( key_val, shiftWord));
// Create the JumpNode
Node* jtn = _gvn.transform( new (C) JumpNode(control(), key_val, num_cases) );
// These are the switch destinations hanging off the jumpnode
int i = 0;
for (SwitchRange* r = lo; r <= hi; r++) {
for (int64 j = r->lo(); j <= r->hi(); j++, i++) {
Node* input = _gvn.transform(new (C) JumpProjNode(jtn, i, r->dest(), (int)(j - lowval)));
{
PreserveJVMState pjvms(this);
set_control(input);
jump_if_always_fork(r->dest(), r->table_index());
}
}
}
assert(i == num_cases, "miscount of cases");
stop_and_kill_map(); // no more uses for this JVMS
return true;
}
//----------------------------jump_switch_ranges-------------------------------
void Parse::jump_switch_ranges(Node* key_val, SwitchRange *lo, SwitchRange *hi, int switch_depth) {
Block* switch_block = block();
if (switch_depth == 0) {
// Do special processing for the top-level call.
assert(lo->lo() == min_jint, "initial range must exhaust Type::INT");
assert(hi->hi() == max_jint, "initial range must exhaust Type::INT");
// Decrement pred-numbers for the unique set of nodes.
#ifdef ASSERT
// Ensure that the block's successors are a (duplicate-free) set.
int successors_counted = 0; // block occurrences in [hi..lo]
int unique_successors = switch_block->num_successors();
for (int i = 0; i < unique_successors; i++) {
Block* target = switch_block->successor_at(i);
// Check that the set of successors is the same in both places.
int successors_found = 0;
for (SwitchRange* p = lo; p <= hi; p++) {
if (p->dest() == target->start()) successors_found++;
}
assert(successors_found > 0, "successor must be known");
successors_counted += successors_found;
}
assert(successors_counted == (hi-lo)+1, "no unexpected successors");
#endif
// Maybe prune the inputs, based on the type of key_val.
jint min_val = min_jint;
jint max_val = max_jint;
const TypeInt* ti = key_val->bottom_type()->isa_int();
if (ti != NULL) {
min_val = ti->_lo;
max_val = ti->_hi;
assert(min_val <= max_val, "invalid int type");
}
while (lo->hi() < min_val) lo++;
if (lo->lo() < min_val) lo->setRange(min_val, lo->hi(), lo->dest(), lo->table_index());
while (hi->lo() > max_val) hi--;
if (hi->hi() > max_val) hi->setRange(hi->lo(), max_val, hi->dest(), hi->table_index());
}
#ifndef PRODUCT
if (switch_depth == 0) {
_max_switch_depth = 0;
_est_switch_depth = log2_intptr((hi-lo+1)-1)+1;
}
#endif
assert(lo <= hi, "must be a non-empty set of ranges");
if (lo == hi) {
jump_if_always_fork(lo->dest(), lo->table_index());
} else {
assert(lo->hi() == (lo+1)->lo()-1, "contiguous ranges");
assert(hi->lo() == (hi-1)->hi()+1, "contiguous ranges");
if (create_jump_tables(key_val, lo, hi)) return;
int nr = hi - lo + 1;
SwitchRange* mid = lo + nr/2;
// if there is an easy choice, pivot at a singleton:
if (nr > 3 && !mid->is_singleton() && (mid-1)->is_singleton()) mid--;
assert(lo < mid && mid <= hi, "good pivot choice");
assert(nr != 2 || mid == hi, "should pick higher of 2");
assert(nr != 3 || mid == hi-1, "should pick middle of 3");
Node *test_val = _gvn.intcon(mid->lo());
if (mid->is_singleton()) {
IfNode *iff_ne = jump_if_fork_int(key_val, test_val, BoolTest::ne);
jump_if_false_fork(iff_ne, mid->dest(), mid->table_index());
// Special Case: If there are exactly three ranges, and the high
// and low range each go to the same place, omit the "gt" test,
// since it will not discriminate anything.
bool eq_test_only = (hi == lo+2 && hi->dest() == lo->dest());
if (eq_test_only) {
assert(mid == hi-1, "");
}
// if there is a higher range, test for it and process it:
if (mid < hi && !eq_test_only) {
// two comparisons of same values--should enable 1 test for 2 branches
// Use BoolTest::le instead of BoolTest::gt
IfNode *iff_le = jump_if_fork_int(key_val, test_val, BoolTest::le);
Node *iftrue = _gvn.transform( new (C) IfTrueNode(iff_le) );
Node *iffalse = _gvn.transform( new (C) IfFalseNode(iff_le) );
{ PreserveJVMState pjvms(this);
set_control(iffalse);
jump_switch_ranges(key_val, mid+1, hi, switch_depth+1);
}
set_control(iftrue);
}
} else {
// mid is a range, not a singleton, so treat mid..hi as a unit
IfNode *iff_ge = jump_if_fork_int(key_val, test_val, BoolTest::ge);
// if there is a higher range, test for it and process it:
if (mid == hi) {
jump_if_true_fork(iff_ge, mid->dest(), mid->table_index());
} else {
Node *iftrue = _gvn.transform( new (C) IfTrueNode(iff_ge) );
Node *iffalse = _gvn.transform( new (C) IfFalseNode(iff_ge) );
{ PreserveJVMState pjvms(this);
set_control(iftrue);
jump_switch_ranges(key_val, mid, hi, switch_depth+1);
}
set_control(iffalse);
}
}
// in any case, process the lower range
jump_switch_ranges(key_val, lo, mid-1, switch_depth+1);
}
// Decrease pred_count for each successor after all is done.
if (switch_depth == 0) {
int unique_successors = switch_block->num_successors();
for (int i = 0; i < unique_successors; i++) {
Block* target = switch_block->successor_at(i);
// Throw away the pre-allocated path for each unique successor.
target->next_path_num();
}
}
#ifndef PRODUCT
_max_switch_depth = MAX2(switch_depth, _max_switch_depth);
if (TraceOptoParse && Verbose && WizardMode && switch_depth == 0) {
SwitchRange* r;
int nsing = 0;
for( r = lo; r <= hi; r++ ) {
if( r->is_singleton() ) nsing++;
}
tty->print(">>> ");
_method->print_short_name();
tty->print_cr(" switch decision tree");
tty->print_cr(" %d ranges (%d singletons), max_depth=%d, est_depth=%d",
(int) (hi-lo+1), nsing, _max_switch_depth, _est_switch_depth);
if (_max_switch_depth > _est_switch_depth) {
tty->print_cr("******** BAD SWITCH DEPTH ********");
}
tty->print(" ");
for( r = lo; r <= hi; r++ ) {
r->print();
}
tty->cr();
}
#endif
}
void Parse::modf() {
Node *f2 = pop();
Node *f1 = pop();
Node* c = make_runtime_call(RC_LEAF, OptoRuntime::modf_Type(),
CAST_FROM_FN_PTR(address, SharedRuntime::frem),
"frem", NULL, //no memory effects
f1, f2);
Node* res = _gvn.transform(new (C) ProjNode(c, TypeFunc::Parms + 0));
push(res);
}
void Parse::modd() {
Node *d2 = pop_pair();
Node *d1 = pop_pair();
Node* c = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(),
CAST_FROM_FN_PTR(address, SharedRuntime::drem),
"drem", NULL, //no memory effects
d1, top(), d2, top());
Node* res_d = _gvn.transform(new (C) ProjNode(c, TypeFunc::Parms + 0));
#ifdef ASSERT
Node* res_top = _gvn.transform(new (C) ProjNode(c, TypeFunc::Parms + 1));
assert(res_top == top(), "second value must be top");
#endif
push_pair(res_d);
}
void Parse::l2f() {
Node* f2 = pop();
Node* f1 = pop();
Node* c = make_runtime_call(RC_LEAF, OptoRuntime::l2f_Type(),
CAST_FROM_FN_PTR(address, SharedRuntime::l2f),
"l2f", NULL, //no memory effects
f1, f2);
Node* res = _gvn.transform(new (C) ProjNode(c, TypeFunc::Parms + 0));
push(res);
}
void Parse::do_irem() {
// Must keep both values on the expression-stack during null-check
zero_check_int(peek());
// Compile-time detect of null-exception?
if (stopped()) return;
Node* b = pop();
Node* a = pop();
const Type *t = _gvn.type(b);
if (t != Type::TOP) {
const TypeInt *ti = t->is_int();
if (ti->is_con()) {
int divisor = ti->get_con();
// check for positive power of 2
if (divisor > 0 &&
(divisor & ~(divisor-1)) == divisor) {
// yes !
Node *mask = _gvn.intcon((divisor - 1));
// Sigh, must handle negative dividends
Node *zero = _gvn.intcon(0);
IfNode *ifff = jump_if_fork_int(a, zero, BoolTest::lt);
Node *iff = _gvn.transform( new (C) IfFalseNode(ifff) );
Node *ift = _gvn.transform( new (C) IfTrueNode (ifff) );
Node *reg = jump_if_join(ift, iff);
Node *phi = PhiNode::make(reg, NULL, TypeInt::INT);
// Negative path; negate/and/negate
Node *neg = _gvn.transform( new (C) SubINode(zero, a) );
Node *andn= _gvn.transform( new (C) AndINode(neg, mask) );
Node *negn= _gvn.transform( new (C) SubINode(zero, andn) );
phi->init_req(1, negn);
// Fast positive case
Node *andx = _gvn.transform( new (C) AndINode(a, mask) );
phi->init_req(2, andx);
// Push the merge
push( _gvn.transform(phi) );
return;
}
}
}
// Default case
push( _gvn.transform( new (C) ModINode(control(),a,b) ) );
}
// Handle jsr and jsr_w bytecode
void Parse::do_jsr() {
assert(bc() == Bytecodes::_jsr || bc() == Bytecodes::_jsr_w, "wrong bytecode");
// Store information about current state, tagged with new _jsr_bci
int return_bci = iter().next_bci();
int jsr_bci = (bc() == Bytecodes::_jsr) ? iter().get_dest() : iter().get_far_dest();
// Update method data
profile_taken_branch(jsr_bci);
// The way we do things now, there is only one successor block
// for the jsr, because the target code is cloned by ciTypeFlow.
Block* target = successor_for_bci(jsr_bci);
// What got pushed?
const Type* ret_addr = target->peek();
assert(ret_addr->singleton(), "must be a constant (cloned jsr body)");
// Effect on jsr on stack
push(_gvn.makecon(ret_addr));
// Flow to the jsr.
merge(jsr_bci);
}
// Handle ret bytecode
void Parse::do_ret() {
// Find to whom we return.
assert(block()->num_successors() == 1, "a ret can only go one place now");
Block* target = block()->successor_at(0);
assert(!target->is_ready(), "our arrival must be expected");
profile_ret(target->flow()->start());
int pnum = target->next_path_num();
merge_common(target, pnum);
}
static bool has_injected_profile(BoolTest::mask btest, Node* test, int& taken, int& not_taken) {
if (btest != BoolTest::eq && btest != BoolTest::ne) {
// Only ::eq and ::ne are supported for profile injection.
return false;
}
if (test->is_Cmp() &&
test->in(1)->Opcode() == Op_ProfileBoolean) {
ProfileBooleanNode* profile = (ProfileBooleanNode*)test->in(1);
int false_cnt = profile->false_count();
int true_cnt = profile->true_count();
// Counts matching depends on the actual test operation (::eq or ::ne).
// No need to scale the counts because profile injection was designed
// to feed exact counts into VM.
taken = (btest == BoolTest::eq) ? false_cnt : true_cnt;
not_taken = (btest == BoolTest::eq) ? true_cnt : false_cnt;
profile->consume();
return true;
}
return false;
}
//--------------------------dynamic_branch_prediction--------------------------
// Try to gather dynamic branch prediction behavior. Return a probability
// of the branch being taken and set the "cnt" field. Returns a -1.0
// if we need to use static prediction for some reason.
float Parse::dynamic_branch_prediction(float &cnt, BoolTest::mask btest, Node* test) {
ResourceMark rm;
cnt = COUNT_UNKNOWN;
int taken = 0;
int not_taken = 0;
bool use_mdo = !has_injected_profile(btest, test, taken, not_taken);
if (use_mdo) {
// Use MethodData information if it is available
// FIXME: free the ProfileData structure
ciMethodData* methodData = method()->method_data();
if (!methodData->is_mature()) return PROB_UNKNOWN;
ciProfileData* data = methodData->bci_to_data(bci());
if (data == NULL) {
return PROB_UNKNOWN;
}
if (!data->is_JumpData()) return PROB_UNKNOWN;
// get taken and not taken values
taken = data->as_JumpData()->taken();
not_taken = 0;
if (data->is_BranchData()) {
not_taken = data->as_BranchData()->not_taken();
}
// scale the counts to be commensurate with invocation counts:
taken = method()->scale_count(taken);
not_taken = method()->scale_count(not_taken);
}
// Give up if too few (or too many, in which case the sum will overflow) counts to be meaningful.
// We also check that individual counters are positive first, otherwise the sum can become positive.
if (taken < 0 || not_taken < 0 || taken + not_taken < 40) {
if (C->log() != NULL) {
C->log()->elem("branch target_bci='%d' taken='%d' not_taken='%d'", iter().get_dest(), taken, not_taken);
}
return PROB_UNKNOWN;
}
// Compute frequency that we arrive here
float sum = taken + not_taken;
// Adjust, if this block is a cloned private block but the
// Jump counts are shared. Taken the private counts for
// just this path instead of the shared counts.
if( block()->count() > 0 )
sum = block()->count();
cnt = sum / FreqCountInvocations;
// Pin probability to sane limits
float prob;
if( !taken )
prob = (0+PROB_MIN) / 2;
else if( !not_taken )
prob = (1+PROB_MAX) / 2;
else { // Compute probability of true path
prob = (float)taken / (float)(taken + not_taken);
if (prob > PROB_MAX) prob = PROB_MAX;
if (prob < PROB_MIN) prob = PROB_MIN;
}
assert((cnt > 0.0f) && (prob > 0.0f),
"Bad frequency assignment in if");
if (C->log() != NULL) {
const char* prob_str = NULL;
if (prob >= PROB_MAX) prob_str = (prob == PROB_MAX) ? "max" : "always";
if (prob <= PROB_MIN) prob_str = (prob == PROB_MIN) ? "min" : "never";
char prob_str_buf[30];
if (prob_str == NULL) {
sprintf(prob_str_buf, "%g", prob);
prob_str = prob_str_buf;
}
C->log()->elem("branch target_bci='%d' taken='%d' not_taken='%d' cnt='%g' prob='%s'",
iter().get_dest(), taken, not_taken, cnt, prob_str);
}
return prob;
}
//-----------------------------branch_prediction-------------------------------
float Parse::branch_prediction(float& cnt,
BoolTest::mask btest,
int target_bci,
Node* test) {
float prob = dynamic_branch_prediction(cnt, btest, test);
// If prob is unknown, switch to static prediction
if (prob != PROB_UNKNOWN) return prob;
prob = PROB_FAIR; // Set default value
if (btest == BoolTest::eq) // Exactly equal test?
prob = PROB_STATIC_INFREQUENT; // Assume its relatively infrequent
else if (btest == BoolTest::ne)
prob = PROB_STATIC_FREQUENT; // Assume its relatively frequent
// If this is a conditional test guarding a backwards branch,
// assume its a loop-back edge. Make it a likely taken branch.
if (target_bci < bci()) {
if (is_osr_parse()) { // Could be a hot OSR'd loop; force deopt
// Since it's an OSR, we probably have profile data, but since
// branch_prediction returned PROB_UNKNOWN, the counts are too small.
// Let's make a special check here for completely zero counts.
ciMethodData* methodData = method()->method_data();
if (!methodData->is_empty()) {
ciProfileData* data = methodData->bci_to_data(bci());
// Only stop for truly zero counts, which mean an unknown part
// of the OSR-ed method, and we want to deopt to gather more stats.
// If you have ANY counts, then this loop is simply 'cold' relative
// to the OSR loop.
if (data == NULL ||
(data->as_BranchData()->taken() + data->as_BranchData()->not_taken() == 0)) {
// This is the only way to return PROB_UNKNOWN:
return PROB_UNKNOWN;
}
}
}
prob = PROB_STATIC_FREQUENT; // Likely to take backwards branch
}
assert(prob != PROB_UNKNOWN, "must have some guess at this point");
return prob;
}
// The magic constants are chosen so as to match the output of
// branch_prediction() when the profile reports a zero taken count.
// It is important to distinguish zero counts unambiguously, because
// some branches (e.g., _213_javac.Assembler.eliminate) validly produce
// very small but nonzero probabilities, which if confused with zero
// counts would keep the program recompiling indefinitely.
bool Parse::seems_never_taken(float prob) const {
return prob < PROB_MIN;
}
// True if the comparison seems to be the kind that will not change its
// statistics from true to false. See comments in adjust_map_after_if.
// This question is only asked along paths which are already
// classifed as untaken (by seems_never_taken), so really,
// if a path is never taken, its controlling comparison is
// already acting in a stable fashion. If the comparison
// seems stable, we will put an expensive uncommon trap
// on the untaken path.
bool Parse::seems_stable_comparison() const {
if (C->too_many_traps(method(), bci(), Deoptimization::Reason_unstable_if)) {
return false;
}
return true;
}
//-------------------------------repush_if_args--------------------------------
// Push arguments of an "if" bytecode back onto the stack by adjusting _sp.
inline int Parse::repush_if_args() {
#ifndef PRODUCT
if (PrintOpto && WizardMode) {
tty->print("defending against excessive implicit null exceptions on %s @%d in ",
Bytecodes::name(iter().cur_bc()), iter().cur_bci());
method()->print_name(); tty->cr();
}
#endif
int bc_depth = - Bytecodes::depth(iter().cur_bc());
assert(bc_depth == 1 || bc_depth == 2, "only two kinds of branches");
DEBUG_ONLY(sync_jvms()); // argument(n) requires a synced jvms
assert(argument(0) != NULL, "must exist");
assert(bc_depth == 1 || argument(1) != NULL, "two must exist");
inc_sp(bc_depth);
return bc_depth;
}
//----------------------------------do_ifnull----------------------------------
void Parse::do_ifnull(BoolTest::mask btest, Node *c) {
int target_bci = iter().get_dest();
Block* branch_block = successor_for_bci(target_bci);
Block* next_block = successor_for_bci(iter().next_bci());
float cnt;
float prob = branch_prediction(cnt, btest, target_bci, c);
if (prob == PROB_UNKNOWN) {
// (An earlier version of do_ifnull omitted this trap for OSR methods.)
#ifndef PRODUCT
if (PrintOpto && Verbose)
tty->print_cr("Never-taken edge stops compilation at bci %d",bci());
#endif
repush_if_args(); // to gather stats on loop
// We need to mark this branch as taken so that if we recompile we will
// see that it is possible. In the tiered system the interpreter doesn't
// do profiling and by the time we get to the lower tier from the interpreter
// the path may be cold again. Make sure it doesn't look untaken
profile_taken_branch(target_bci, !ProfileInterpreter);
uncommon_trap(Deoptimization::Reason_unreached,
Deoptimization::Action_reinterpret,
NULL, "cold");
if (C->eliminate_boxing()) {
// Mark the successor blocks as parsed
branch_block->next_path_num();
next_block->next_path_num();
}
return;
}
explicit_null_checks_inserted++;
// Generate real control flow
Node *tst = _gvn.transform( new (C) BoolNode( c, btest ) );
// Sanity check the probability value
assert(prob > 0.0f,"Bad probability in Parser");
// Need xform to put node in hash table
IfNode *iff = create_and_xform_if( control(), tst, prob, cnt );
assert(iff->_prob > 0.0f,"Optimizer made bad probability in parser");
// True branch
{ PreserveJVMState pjvms(this);
Node* iftrue = _gvn.transform( new (C) IfTrueNode (iff) );
set_control(iftrue);
if (stopped()) { // Path is dead?
explicit_null_checks_elided++;
if (C->eliminate_boxing()) {
// Mark the successor block as parsed
branch_block->next_path_num();
}
} else { // Path is live.
// Update method data
profile_taken_branch(target_bci);
adjust_map_after_if(btest, c, prob, branch_block, next_block);
if (!stopped()) {
merge(target_bci);
}
}
}
// False branch
Node* iffalse = _gvn.transform( new (C) IfFalseNode(iff) );
set_control(iffalse);
if (stopped()) { // Path is dead?
explicit_null_checks_elided++;
if (C->eliminate_boxing()) {
// Mark the successor block as parsed
next_block->next_path_num();
}
} else { // Path is live.
// Update method data
profile_not_taken_branch();
adjust_map_after_if(BoolTest(btest).negate(), c, 1.0-prob,
next_block, branch_block);
}
}
//------------------------------------do_if------------------------------------
void Parse::do_if(BoolTest::mask btest, Node* c) {
int target_bci = iter().get_dest();
Block* branch_block = successor_for_bci(target_bci);
Block* next_block = successor_for_bci(iter().next_bci());
float cnt;
float prob = branch_prediction(cnt, btest, target_bci, c);
float untaken_prob = 1.0 - prob;
if (prob == PROB_UNKNOWN) {
#ifndef PRODUCT
if (PrintOpto && Verbose)
tty->print_cr("Never-taken edge stops compilation at bci %d",bci());
#endif
repush_if_args(); // to gather stats on loop
// We need to mark this branch as taken so that if we recompile we will
// see that it is possible. In the tiered system the interpreter doesn't
// do profiling and by the time we get to the lower tier from the interpreter
// the path may be cold again. Make sure it doesn't look untaken
profile_taken_branch(target_bci, !ProfileInterpreter);
uncommon_trap(Deoptimization::Reason_unreached,
Deoptimization::Action_reinterpret,
NULL, "cold");
if (C->eliminate_boxing()) {
// Mark the successor blocks as parsed
branch_block->next_path_num();
next_block->next_path_num();
}
return;
}
// Sanity check the probability value
assert(0.0f < prob && prob < 1.0f,"Bad probability in Parser");
bool taken_if_true = true;
// Convert BoolTest to canonical form:
if (!BoolTest(btest).is_canonical()) {
btest = BoolTest(btest).negate();
taken_if_true = false;
// prob is NOT updated here; it remains the probability of the taken
// path (as opposed to the prob of the path guarded by an 'IfTrueNode').
}
assert(btest != BoolTest::eq, "!= is the only canonical exact test");
Node* tst0 = new (C) BoolNode(c, btest);
Node* tst = _gvn.transform(tst0);
BoolTest::mask taken_btest = BoolTest::illegal;
BoolTest::mask untaken_btest = BoolTest::illegal;
if (tst->is_Bool()) {
// Refresh c from the transformed bool node, since it may be
// simpler than the original c. Also re-canonicalize btest.
// This wins when (Bool ne (Conv2B p) 0) => (Bool ne (CmpP p NULL)).
// That can arise from statements like: if (x instanceof C) ...
if (tst != tst0) {
// Canonicalize one more time since transform can change it.
btest = tst->as_Bool()->_test._test;
if (!BoolTest(btest).is_canonical()) {
// Reverse edges one more time...
tst = _gvn.transform( tst->as_Bool()->negate(&_gvn) );
btest = tst->as_Bool()->_test._test;
assert(BoolTest(btest).is_canonical(), "sanity");
taken_if_true = !taken_if_true;
}
c = tst->in(1);
}
BoolTest::mask neg_btest = BoolTest(btest).negate();
taken_btest = taken_if_true ? btest : neg_btest;
untaken_btest = taken_if_true ? neg_btest : btest;
}
// Generate real control flow
float true_prob = (taken_if_true ? prob : untaken_prob);
IfNode* iff = create_and_map_if(control(), tst, true_prob, cnt);
assert(iff->_prob > 0.0f,"Optimizer made bad probability in parser");
Node* taken_branch = new (C) IfTrueNode(iff);
Node* untaken_branch = new (C) IfFalseNode(iff);
if (!taken_if_true) { // Finish conversion to canonical form
Node* tmp = taken_branch;
taken_branch = untaken_branch;
untaken_branch = tmp;
}
// Branch is taken:
{ PreserveJVMState pjvms(this);
taken_branch = _gvn.transform(taken_branch);
set_control(taken_branch);
if (stopped()) {
if (C->eliminate_boxing()) {
// Mark the successor block as parsed
branch_block->next_path_num();
}
} else {
// Update method data
profile_taken_branch(target_bci);
adjust_map_after_if(taken_btest, c, prob, branch_block, next_block);
if (!stopped()) {
merge(target_bci);
}
}
}
untaken_branch = _gvn.transform(untaken_branch);
set_control(untaken_branch);
// Branch not taken.
if (stopped()) {
if (C->eliminate_boxing()) {
// Mark the successor block as parsed
next_block->next_path_num();
}
} else {
// Update method data
profile_not_taken_branch();
adjust_map_after_if(untaken_btest, c, untaken_prob,
next_block, branch_block);
}
}
bool Parse::path_is_suitable_for_uncommon_trap(float prob) const {
// Don't want to speculate on uncommon traps when running with -Xcomp
if (!UseInterpreter) {
return false;
}
return (seems_never_taken(prob) && seems_stable_comparison());
}
//----------------------------adjust_map_after_if------------------------------
// Adjust the JVM state to reflect the result of taking this path.
// Basically, it means inspecting the CmpNode controlling this
// branch, seeing how it constrains a tested value, and then
// deciding if it's worth our while to encode this constraint
// as graph nodes in the current abstract interpretation map.
void Parse::adjust_map_after_if(BoolTest::mask btest, Node* c, float prob,
Block* path, Block* other_path) {
if (stopped() || !c->is_Cmp() || btest == BoolTest::illegal)
return; // nothing to do
bool is_fallthrough = (path == successor_for_bci(iter().next_bci()));
if (path_is_suitable_for_uncommon_trap(prob)) {
repush_if_args();
uncommon_trap(Deoptimization::Reason_unstable_if,
Deoptimization::Action_reinterpret,
NULL,
(is_fallthrough ? "taken always" : "taken never"));
return;
}
Node* val = c->in(1);
Node* con = c->in(2);
const Type* tcon = _gvn.type(con);
const Type* tval = _gvn.type(val);
bool have_con = tcon->singleton();
if (tval->singleton()) {
if (!have_con) {
// Swap, so constant is in con.
con = val;
tcon = tval;
val = c->in(2);
tval = _gvn.type(val);
btest = BoolTest(btest).commute();
have_con = true;
} else {
// Do we have two constants? Then leave well enough alone.
have_con = false;
}
}
if (!have_con) // remaining adjustments need a con
return;
sharpen_type_after_if(btest, con, tcon, val, tval);
}
static Node* extract_obj_from_klass_load(PhaseGVN* gvn, Node* n) {
Node* ldk;
if (n->is_DecodeNKlass()) {
if (n->in(1)->Opcode() != Op_LoadNKlass) {
return NULL;
} else {
ldk = n->in(1);
}
} else if (n->Opcode() != Op_LoadKlass) {
return NULL;
} else {
ldk = n;
}
assert(ldk != NULL && ldk->is_Load(), "should have found a LoadKlass or LoadNKlass node");
Node* adr = ldk->in(MemNode::Address);
intptr_t off = 0;
Node* obj = AddPNode::Ideal_base_and_offset(adr, gvn, off);
if (obj == NULL || off != oopDesc::klass_offset_in_bytes()) // loading oopDesc::_klass?
return NULL;
const TypePtr* tp = gvn->type(obj)->is_ptr();
if (tp == NULL || !(tp->isa_instptr() || tp->isa_aryptr())) // is obj a Java object ptr?
return NULL;
return obj;
}
void Parse::sharpen_type_after_if(BoolTest::mask btest,
Node* con, const Type* tcon,
Node* val, const Type* tval) {
// Look for opportunities to sharpen the type of a node
// whose klass is compared with a constant klass.
if (btest == BoolTest::eq && tcon->isa_klassptr()) {
Node* obj = extract_obj_from_klass_load(&_gvn, val);
const TypeOopPtr* con_type = tcon->isa_klassptr()->as_instance_type();
if (obj != NULL && (con_type->isa_instptr() || con_type->isa_aryptr())) {
// Found:
// Bool(CmpP(LoadKlass(obj._klass), ConP(Foo.klass)), [eq])
// or the narrowOop equivalent.
const Type* obj_type = _gvn.type(obj);
const TypeOopPtr* tboth = obj_type->join_speculative(con_type)->isa_oopptr();
if (tboth != NULL && tboth->klass_is_exact() && tboth != obj_type &&
tboth->higher_equal(obj_type)) {
// obj has to be of the exact type Foo if the CmpP succeeds.
int obj_in_map = map()->find_edge(obj);
JVMState* jvms = this->jvms();
if (obj_in_map >= 0 &&
(jvms->is_loc(obj_in_map) || jvms->is_stk(obj_in_map))) {
TypeNode* ccast = new (C) CheckCastPPNode(control(), obj, tboth);
const Type* tcc = ccast->as_Type()->type();
assert(tcc != obj_type && tcc->higher_equal_speculative(obj_type), "must improve");
// Delay transform() call to allow recovery of pre-cast value
// at the control merge.
_gvn.set_type_bottom(ccast);
record_for_igvn(ccast);
// Here's the payoff.
replace_in_map(obj, ccast);
}
}
}
}
int val_in_map = map()->find_edge(val);
if (val_in_map < 0) return; // replace_in_map would be useless
{
JVMState* jvms = this->jvms();
if (!(jvms->is_loc(val_in_map) ||
jvms->is_stk(val_in_map)))
return; // again, it would be useless
}
// Check for a comparison to a constant, and "know" that the compared
// value is constrained on this path.
assert(tcon->singleton(), "");
ConstraintCastNode* ccast = NULL;
Node* cast = NULL;
switch (btest) {
case BoolTest::eq: // Constant test?
{
const Type* tboth = tcon->join_speculative(tval);
if (tboth == tval) break; // Nothing to gain.
if (tcon->isa_int()) {
ccast = new (C) CastIINode(val, tboth);
} else if (tcon == TypePtr::NULL_PTR) {
// Cast to null, but keep the pointer identity temporarily live.
ccast = new (C) CastPPNode(val, tboth);
} else {
const TypeF* tf = tcon->isa_float_constant();
const TypeD* td = tcon->isa_double_constant();
// Exclude tests vs float/double 0 as these could be
// either +0 or -0. Just because you are equal to +0
// doesn't mean you ARE +0!
// Note, following code also replaces Long and Oop values.
if ((!tf || tf->_f != 0.0) &&
(!td || td->_d != 0.0))
cast = con; // Replace non-constant val by con.
}
}
break;
case BoolTest::ne:
if (tcon == TypePtr::NULL_PTR) {
cast = cast_not_null(val, false);
}
break;
default:
// (At this point we could record int range types with CastII.)
break;
}
if (ccast != NULL) {
const Type* tcc = ccast->as_Type()->type();
assert(tcc != tval && tcc->higher_equal_speculative(tval), "must improve");
// Delay transform() call to allow recovery of pre-cast value
// at the control merge.
ccast->set_req(0, control());
_gvn.set_type_bottom(ccast);
record_for_igvn(ccast);
cast = ccast;
}
if (cast != NULL) { // Here's the payoff.
replace_in_map(val, cast);
}
}
/**
* Use speculative type to optimize CmpP node: if comparison is
* against the low level class, cast the object to the speculative
* type if any. CmpP should then go away.
*
* @param c expected CmpP node
* @return result of CmpP on object casted to speculative type
*
*/
Node* Parse::optimize_cmp_with_klass(Node* c) {
// If this is transformed by the _gvn to a comparison with the low
// level klass then we may be able to use speculation
if (c->Opcode() == Op_CmpP &&
(c->in(1)->Opcode() == Op_LoadKlass || c->in(1)->Opcode() == Op_DecodeNKlass) &&
c->in(2)->is_Con()) {
Node* load_klass = NULL;
Node* decode = NULL;
if (c->in(1)->Opcode() == Op_DecodeNKlass) {
decode = c->in(1);
load_klass = c->in(1)->in(1);
} else {
load_klass = c->in(1);
}
if (load_klass->in(2)->is_AddP()) {
Node* addp = load_klass->in(2);
Node* obj = addp->in(AddPNode::Address);
const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
if (obj_type->speculative_type() != NULL) {
ciKlass* k = obj_type->speculative_type();
inc_sp(2);
obj = maybe_cast_profiled_obj(obj, k);
dec_sp(2);
// Make the CmpP use the casted obj
addp = basic_plus_adr(obj, addp->in(AddPNode::Offset));
load_klass = load_klass->clone();
load_klass->set_req(2, addp);
load_klass = _gvn.transform(load_klass);
if (decode != NULL) {
decode = decode->clone();
decode->set_req(1, load_klass);
load_klass = _gvn.transform(decode);
}
c = c->clone();
c->set_req(1, load_klass);
c = _gvn.transform(c);
}
}
}
return c;
}
//------------------------------do_one_bytecode--------------------------------
// Parse this bytecode, and alter the Parsers JVM->Node mapping
void Parse::do_one_bytecode() {
Node *a, *b, *c, *d; // Handy temps
BoolTest::mask btest;
int i;
assert(!has_exceptions(), "bytecode entry state must be clear of throws");
if (C->check_node_count(NodeLimitFudgeFactor * 5,
"out of nodes parsing method")) {
return;
}
#ifdef ASSERT
// for setting breakpoints
if (TraceOptoParse) {
tty->print(" @");
dump_bci(bci());
tty->cr();
}
#endif
switch (bc()) {
case Bytecodes::_nop:
// do nothing
break;
case Bytecodes::_lconst_0:
push_pair(longcon(0));
break;
case Bytecodes::_lconst_1:
push_pair(longcon(1));
break;
case Bytecodes::_fconst_0:
push(zerocon(T_FLOAT));
break;
case Bytecodes::_fconst_1:
push(makecon(TypeF::ONE));
break;
case Bytecodes::_fconst_2:
push(makecon(TypeF::make(2.0f)));
break;
case Bytecodes::_dconst_0:
push_pair(zerocon(T_DOUBLE));
break;
case Bytecodes::_dconst_1:
push_pair(makecon(TypeD::ONE));
break;
case Bytecodes::_iconst_m1:push(intcon(-1)); break;
case Bytecodes::_iconst_0: push(intcon( 0)); break;
case Bytecodes::_iconst_1: push(intcon( 1)); break;
case Bytecodes::_iconst_2: push(intcon( 2)); break;
case Bytecodes::_iconst_3: push(intcon( 3)); break;
case Bytecodes::_iconst_4: push(intcon( 4)); break;
case Bytecodes::_iconst_5: push(intcon( 5)); break;
case Bytecodes::_bipush: push(intcon(iter().get_constant_u1())); break;
case Bytecodes::_sipush: push(intcon(iter().get_constant_u2())); break;
case Bytecodes::_aconst_null: push(null()); break;
case Bytecodes::_ldc:
case Bytecodes::_ldc_w:
case Bytecodes::_ldc2_w:
// If the constant is unresolved, run this BC once in the interpreter.
{
ciConstant constant = iter().get_constant();
if (constant.basic_type() == T_OBJECT &&
!constant.as_object()->is_loaded()) {
int index = iter().get_constant_pool_index();
constantTag tag = iter().get_constant_pool_tag(index);
uncommon_trap(Deoptimization::make_trap_request
(Deoptimization::Reason_unloaded,
Deoptimization::Action_reinterpret,
index),
NULL, tag.internal_name());
break;
}
assert(constant.basic_type() != T_OBJECT || constant.as_object()->is_instance(),
"must be java_mirror of klass");
bool pushed = push_constant(constant, true);
guarantee(pushed, "must be possible to push this constant");
}
break;
case Bytecodes::_aload_0:
push( local(0) );
break;
case Bytecodes::_aload_1:
push( local(1) );
break;
case Bytecodes::_aload_2:
push( local(2) );
break;
case Bytecodes::_aload_3:
push( local(3) );
break;
case Bytecodes::_aload:
push( local(iter().get_index()) );
break;
case Bytecodes::_fload_0:
case Bytecodes::_iload_0:
push( local(0) );
break;
case Bytecodes::_fload_1:
case Bytecodes::_iload_1:
push( local(1) );
break;
case Bytecodes::_fload_2:
case Bytecodes::_iload_2:
push( local(2) );
break;
case Bytecodes::_fload_3:
case Bytecodes::_iload_3:
push( local(3) );
break;
case Bytecodes::_fload:
case Bytecodes::_iload:
push( local(iter().get_index()) );
break;
case Bytecodes::_lload_0:
push_pair_local( 0 );
break;
case Bytecodes::_lload_1:
push_pair_local( 1 );
break;
case Bytecodes::_lload_2:
push_pair_local( 2 );
break;
case Bytecodes::_lload_3:
push_pair_local( 3 );
break;
case Bytecodes::_lload:
push_pair_local( iter().get_index() );
break;
case Bytecodes::_dload_0:
push_pair_local(0);
break;
case Bytecodes::_dload_1:
push_pair_local(1);
break;
case Bytecodes::_dload_2:
push_pair_local(2);
break;
case Bytecodes::_dload_3:
push_pair_local(3);
break;
case Bytecodes::_dload:
push_pair_local(iter().get_index());
break;
case Bytecodes::_fstore_0:
case Bytecodes::_istore_0:
case Bytecodes::_astore_0:
set_local( 0, pop() );
break;
case Bytecodes::_fstore_1:
case Bytecodes::_istore_1:
case Bytecodes::_astore_1:
set_local( 1, pop() );
break;
case Bytecodes::_fstore_2:
case Bytecodes::_istore_2:
case Bytecodes::_astore_2:
set_local( 2, pop() );
break;
case Bytecodes::_fstore_3:
case Bytecodes::_istore_3:
case Bytecodes::_astore_3:
set_local( 3, pop() );
break;
case Bytecodes::_fstore:
case Bytecodes::_istore:
case Bytecodes::_astore:
set_local( iter().get_index(), pop() );
break;
// long stores
case Bytecodes::_lstore_0:
set_pair_local( 0, pop_pair() );
break;
case Bytecodes::_lstore_1:
set_pair_local( 1, pop_pair() );
break;
case Bytecodes::_lstore_2:
set_pair_local( 2, pop_pair() );
break;
case Bytecodes::_lstore_3:
set_pair_local( 3, pop_pair() );
break;
case Bytecodes::_lstore:
set_pair_local( iter().get_index(), pop_pair() );
break;
// double stores
case Bytecodes::_dstore_0:
set_pair_local( 0, dstore_rounding(pop_pair()) );
break;
case Bytecodes::_dstore_1:
set_pair_local( 1, dstore_rounding(pop_pair()) );
break;
case Bytecodes::_dstore_2:
set_pair_local( 2, dstore_rounding(pop_pair()) );
break;
case Bytecodes::_dstore_3:
set_pair_local( 3, dstore_rounding(pop_pair()) );
break;
case Bytecodes::_dstore:
set_pair_local( iter().get_index(), dstore_rounding(pop_pair()) );
break;
case Bytecodes::_pop: dec_sp(1); break;
case Bytecodes::_pop2: dec_sp(2); break;
case Bytecodes::_swap:
a = pop();
b = pop();
push(a);
push(b);
break;
case Bytecodes::_dup:
a = pop();
push(a);
push(a);
break;
case Bytecodes::_dup_x1:
a = pop();
b = pop();
push( a );
push( b );
push( a );
break;
case Bytecodes::_dup_x2:
a = pop();
b = pop();
c = pop();
push( a );
push( c );
push( b );
push( a );
break;
case Bytecodes::_dup2:
a = pop();
b = pop();
push( b );
push( a );
push( b );
push( a );
break;
case Bytecodes::_dup2_x1:
// before: .. c, b, a
// after: .. b, a, c, b, a
// not tested
a = pop();
b = pop();
c = pop();
push( b );
push( a );
push( c );
push( b );
push( a );
break;
case Bytecodes::_dup2_x2:
// before: .. d, c, b, a
// after: .. b, a, d, c, b, a
// not tested
a = pop();
b = pop();
c = pop();
d = pop();
push( b );
push( a );
push( d );
push( c );
push( b );
push( a );
break;
case Bytecodes::_arraylength: {
// Must do null-check with value on expression stack
Node *ary = null_check(peek(), T_ARRAY);
// Compile-time detect of null-exception?
if (stopped()) return;
a = pop();
push(load_array_length(a));
break;
}
case Bytecodes::_baload: array_load(T_BYTE); break;
case Bytecodes::_caload: array_load(T_CHAR); break;
case Bytecodes::_iaload: array_load(T_INT); break;
case Bytecodes::_saload: array_load(T_SHORT); break;
case Bytecodes::_faload: array_load(T_FLOAT); break;
case Bytecodes::_aaload: array_load(T_OBJECT); break;
case Bytecodes::_laload: {
a = array_addressing(T_LONG, 0);
if (stopped()) return; // guaranteed null or range check
dec_sp(2); // Pop array and index
push_pair(make_load(control(), a, TypeLong::LONG, T_LONG, TypeAryPtr::LONGS, MemNode::unordered));
break;
}
case Bytecodes::_daload: {
a = array_addressing(T_DOUBLE, 0);
if (stopped()) return; // guaranteed null or range check
dec_sp(2); // Pop array and index
push_pair(make_load(control(), a, Type::DOUBLE, T_DOUBLE, TypeAryPtr::DOUBLES, MemNode::unordered));
break;
}
case Bytecodes::_bastore: array_store(T_BYTE); break;
case Bytecodes::_castore: array_store(T_CHAR); break;
case Bytecodes::_iastore: array_store(T_INT); break;
case Bytecodes::_sastore: array_store(T_SHORT); break;
case Bytecodes::_fastore: array_store(T_FLOAT); break;
case Bytecodes::_aastore: {
d = array_addressing(T_OBJECT, 1);
if (stopped()) return; // guaranteed null or range check
array_store_check();
c = pop(); // Oop to store
b = pop(); // index (already used)
a = pop(); // the array itself
const TypeOopPtr* elemtype = _gvn.type(a)->is_aryptr()->elem()->make_oopptr();
const TypeAryPtr* adr_type = TypeAryPtr::OOPS;
Node* store = store_oop_to_array(control(), a, d, adr_type, c, elemtype, T_OBJECT, MemNode::release);
break;
}
case Bytecodes::_lastore: {
a = array_addressing(T_LONG, 2);
if (stopped()) return; // guaranteed null or range check
c = pop_pair();
dec_sp(2); // Pop array and index
store_to_memory(control(), a, c, T_LONG, TypeAryPtr::LONGS, MemNode::unordered);
break;
}
case Bytecodes::_dastore: {
a = array_addressing(T_DOUBLE, 2);
if (stopped()) return; // guaranteed null or range check
c = pop_pair();
dec_sp(2); // Pop array and index
c = dstore_rounding(c);
store_to_memory(control(), a, c, T_DOUBLE, TypeAryPtr::DOUBLES, MemNode::unordered);
break;
}
case Bytecodes::_getfield:
do_getfield();
break;
case Bytecodes::_getstatic:
do_getstatic();
break;
case Bytecodes::_putfield:
do_putfield();
break;
case Bytecodes::_putstatic:
do_putstatic();
break;
case Bytecodes::_irem:
do_irem();
break;
case Bytecodes::_idiv:
// Must keep both values on the expression-stack during null-check
zero_check_int(peek());
// Compile-time detect of null-exception?
if (stopped()) return;
b = pop();
a = pop();
push( _gvn.transform( new (C) DivINode(control(),a,b) ) );
break;
case Bytecodes::_imul:
b = pop(); a = pop();
push( _gvn.transform( new (C) MulINode(a,b) ) );
break;
case Bytecodes::_iadd:
b = pop(); a = pop();
push( _gvn.transform( new (C) AddINode(a,b) ) );
break;
case Bytecodes::_ineg:
a = pop();
push( _gvn.transform( new (C) SubINode(_gvn.intcon(0),a)) );
break;
case Bytecodes::_isub:
b = pop(); a = pop();
push( _gvn.transform( new (C) SubINode(a,b) ) );
break;
case Bytecodes::_iand:
b = pop(); a = pop();
push( _gvn.transform( new (C) AndINode(a,b) ) );
break;
case Bytecodes::_ior:
b = pop(); a = pop();
push( _gvn.transform( new (C) OrINode(a,b) ) );
break;
case Bytecodes::_ixor:
b = pop(); a = pop();
push( _gvn.transform( new (C) XorINode(a,b) ) );
break;
case Bytecodes::_ishl:
b = pop(); a = pop();
push( _gvn.transform( new (C) LShiftINode(a,b) ) );
break;
case Bytecodes::_ishr:
b = pop(); a = pop();
push( _gvn.transform( new (C) RShiftINode(a,b) ) );
break;
case Bytecodes::_iushr:
b = pop(); a = pop();
push( _gvn.transform( new (C) URShiftINode(a,b) ) );
break;
case Bytecodes::_fneg:
a = pop();
b = _gvn.transform(new (C) NegFNode (a));
push(b);
break;
case Bytecodes::_fsub:
b = pop();
a = pop();
c = _gvn.transform( new (C) SubFNode(a,b) );
d = precision_rounding(c);
push( d );
break;
case Bytecodes::_fadd:
b = pop();
a = pop();
c = _gvn.transform( new (C) AddFNode(a,b) );
d = precision_rounding(c);
push( d );
break;
case Bytecodes::_fmul:
b = pop();
a = pop();
c = _gvn.transform( new (C) MulFNode(a,b) );
d = precision_rounding(c);
push( d );
break;
case Bytecodes::_fdiv:
b = pop();
a = pop();
c = _gvn.transform( new (C) DivFNode(0,a,b) );
d = precision_rounding(c);
push( d );
break;
case Bytecodes::_frem:
if (Matcher::has_match_rule(Op_ModF)) {
// Generate a ModF node.
b = pop();
a = pop();
c = _gvn.transform( new (C) ModFNode(0,a,b) );
d = precision_rounding(c);
push( d );
}
else {
// Generate a call.
modf();
}
break;
case Bytecodes::_fcmpl:
b = pop();
a = pop();
c = _gvn.transform( new (C) CmpF3Node( a, b));
push(c);
break;
case Bytecodes::_fcmpg:
b = pop();
a = pop();
// Same as fcmpl but need to flip the unordered case. Swap the inputs,
// which negates the result sign except for unordered. Flip the unordered
// as well by using CmpF3 which implements unordered-lesser instead of
// unordered-greater semantics. Finally, commute the result bits. Result
// is same as using a CmpF3Greater except we did it with CmpF3 alone.
c = _gvn.transform( new (C) CmpF3Node( b, a));
c = _gvn.transform( new (C) SubINode(_gvn.intcon(0),c) );
push(c);
break;
case Bytecodes::_f2i:
a = pop();
push(_gvn.transform(new (C) ConvF2INode(a)));
break;
case Bytecodes::_d2i:
a = pop_pair();
b = _gvn.transform(new (C) ConvD2INode(a));
push( b );
break;
case Bytecodes::_f2d:
a = pop();
b = _gvn.transform( new (C) ConvF2DNode(a));
push_pair( b );
break;
case Bytecodes::_d2f:
a = pop_pair();
b = _gvn.transform( new (C) ConvD2FNode(a));
// This breaks _227_mtrt (speed & correctness) and _222_mpegaudio (speed)
//b = _gvn.transform(new (C) RoundFloatNode(0, b) );
push( b );
break;
case Bytecodes::_l2f:
if (Matcher::convL2FSupported()) {
a = pop_pair();
b = _gvn.transform( new (C) ConvL2FNode(a));
// For i486.ad, FILD doesn't restrict precision to 24 or 53 bits.
// Rather than storing the result into an FP register then pushing
// out to memory to round, the machine instruction that implements
// ConvL2D is responsible for rounding.
// c = precision_rounding(b);
c = _gvn.transform(b);
push(c);
} else {
l2f();
}
break;
case Bytecodes::_l2d:
a = pop_pair();
b = _gvn.transform( new (C) ConvL2DNode(a));
// For i486.ad, rounding is always necessary (see _l2f above).
// c = dprecision_rounding(b);
c = _gvn.transform(b);
push_pair(c);
break;
case Bytecodes::_f2l:
a = pop();
b = _gvn.transform( new (C) ConvF2LNode(a));
push_pair(b);
break;
case Bytecodes::_d2l:
a = pop_pair();
b = _gvn.transform( new (C) ConvD2LNode(a));
push_pair(b);
break;
case Bytecodes::_dsub:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) SubDNode(a,b) );
d = dprecision_rounding(c);
push_pair( d );
break;
case Bytecodes::_dadd:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) AddDNode(a,b) );
d = dprecision_rounding(c);
push_pair( d );
break;
case Bytecodes::_dmul:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) MulDNode(a,b) );
d = dprecision_rounding(c);
push_pair( d );
break;
case Bytecodes::_ddiv:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) DivDNode(0,a,b) );
d = dprecision_rounding(c);
push_pair( d );
break;
case Bytecodes::_dneg:
a = pop_pair();
b = _gvn.transform(new (C) NegDNode (a));
push_pair(b);
break;
case Bytecodes::_drem:
if (Matcher::has_match_rule(Op_ModD)) {
// Generate a ModD node.
b = pop_pair();
a = pop_pair();
// a % b
c = _gvn.transform( new (C) ModDNode(0,a,b) );
d = dprecision_rounding(c);
push_pair( d );
}
else {
// Generate a call.
modd();
}
break;
case Bytecodes::_dcmpl:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) CmpD3Node( a, b));
push(c);
break;
case Bytecodes::_dcmpg:
b = pop_pair();
a = pop_pair();
// Same as dcmpl but need to flip the unordered case.
// Commute the inputs, which negates the result sign except for unordered.
// Flip the unordered as well by using CmpD3 which implements
// unordered-lesser instead of unordered-greater semantics.
// Finally, negate the result bits. Result is same as using a
// CmpD3Greater except we did it with CmpD3 alone.
c = _gvn.transform( new (C) CmpD3Node( b, a));
c = _gvn.transform( new (C) SubINode(_gvn.intcon(0),c) );
push(c);
break;
// Note for longs -> lo word is on TOS, hi word is on TOS - 1
case Bytecodes::_land:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) AndLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lor:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) OrLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lxor:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) XorLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lshl:
b = pop(); // the shift count
a = pop_pair(); // value to be shifted
c = _gvn.transform( new (C) LShiftLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lshr:
b = pop(); // the shift count
a = pop_pair(); // value to be shifted
c = _gvn.transform( new (C) RShiftLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lushr:
b = pop(); // the shift count
a = pop_pair(); // value to be shifted
c = _gvn.transform( new (C) URShiftLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lmul:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) MulLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lrem:
// Must keep both values on the expression-stack during null-check
assert(peek(0) == top(), "long word order");
zero_check_long(peek(1));
// Compile-time detect of null-exception?
if (stopped()) return;
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) ModLNode(control(),a,b) );
push_pair(c);
break;
case Bytecodes::_ldiv:
// Must keep both values on the expression-stack during null-check
assert(peek(0) == top(), "long word order");
zero_check_long(peek(1));
// Compile-time detect of null-exception?
if (stopped()) return;
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) DivLNode(control(),a,b) );
push_pair(c);
break;
case Bytecodes::_ladd:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) AddLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lsub:
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) SubLNode(a,b) );
push_pair(c);
break;
case Bytecodes::_lcmp:
// Safepoints are now inserted _before_ branches. The long-compare
// bytecode painfully produces a 3-way value (-1,0,+1) which requires a
// slew of control flow. These are usually followed by a CmpI vs zero and
// a branch; this pattern then optimizes to the obvious long-compare and
// branch. However, if the branch is backwards there's a Safepoint
// inserted. The inserted Safepoint captures the JVM state at the
// pre-branch point, i.e. it captures the 3-way value. Thus if a
// long-compare is used to control a loop the debug info will force
// computation of the 3-way value, even though the generated code uses a
// long-compare and branch. We try to rectify the situation by inserting
// a SafePoint here and have it dominate and kill the safepoint added at a
// following backwards branch. At this point the JVM state merely holds 2
// longs but not the 3-way value.
if( UseLoopSafepoints ) {
switch( iter().next_bc() ) {
case Bytecodes::_ifgt:
case Bytecodes::_iflt:
case Bytecodes::_ifge:
case Bytecodes::_ifle:
case Bytecodes::_ifne:
case Bytecodes::_ifeq:
// If this is a backwards branch in the bytecodes, add Safepoint
maybe_add_safepoint(iter().next_get_dest());
}
}
b = pop_pair();
a = pop_pair();
c = _gvn.transform( new (C) CmpL3Node( a, b ));
push(c);
break;
case Bytecodes::_lneg:
a = pop_pair();
b = _gvn.transform( new (C) SubLNode(longcon(0),a));
push_pair(b);
break;
case Bytecodes::_l2i:
a = pop_pair();
push( _gvn.transform( new (C) ConvL2INode(a)));
break;
case Bytecodes::_i2l:
a = pop();
b = _gvn.transform( new (C) ConvI2LNode(a));
push_pair(b);
break;
case Bytecodes::_i2b:
// Sign extend
a = pop();
a = _gvn.transform( new (C) LShiftINode(a,_gvn.intcon(24)) );
a = _gvn.transform( new (C) RShiftINode(a,_gvn.intcon(24)) );
push( a );
break;
case Bytecodes::_i2s:
a = pop();
a = _gvn.transform( new (C) LShiftINode(a,_gvn.intcon(16)) );
a = _gvn.transform( new (C) RShiftINode(a,_gvn.intcon(16)) );
push( a );
break;
case Bytecodes::_i2c:
a = pop();
push( _gvn.transform( new (C) AndINode(a,_gvn.intcon(0xFFFF)) ) );
break;
case Bytecodes::_i2f:
a = pop();
b = _gvn.transform( new (C) ConvI2FNode(a) ) ;
c = precision_rounding(b);
push (b);
break;
case Bytecodes::_i2d:
a = pop();
b = _gvn.transform( new (C) ConvI2DNode(a));
push_pair(b);
break;
case Bytecodes::_iinc: // Increment local
i = iter().get_index(); // Get local index
set_local( i, _gvn.transform( new (C) AddINode( _gvn.intcon(iter().get_iinc_con()), local(i) ) ) );
break;
// Exit points of synchronized methods must have an unlock node
case Bytecodes::_return:
return_current(NULL);
break;
case Bytecodes::_ireturn:
case Bytecodes::_areturn:
case Bytecodes::_freturn:
return_current(pop());
break;
case Bytecodes::_lreturn:
return_current(pop_pair());
break;
case Bytecodes::_dreturn:
return_current(pop_pair());
break;
case Bytecodes::_athrow:
// null exception oop throws NULL pointer exception
null_check(peek());
if (stopped()) return;
// Hook the thrown exception directly to subsequent handlers.
if (BailoutToInterpreterForThrows) {
// Keep method interpreted from now on.
uncommon_trap(Deoptimization::Reason_unhandled,
Deoptimization::Action_make_not_compilable);
return;
}
if (env()->jvmti_can_post_on_exceptions()) {
// check if we must post exception events, take uncommon trap if so (with must_throw = false)
uncommon_trap_if_should_post_on_exceptions(Deoptimization::Reason_unhandled, false);
}
// Here if either can_post_on_exceptions or should_post_on_exceptions is false
add_exception_state(make_exception_state(peek()));
break;
case Bytecodes::_goto: // fall through
case Bytecodes::_goto_w: {
int target_bci = (bc() == Bytecodes::_goto) ? iter().get_dest() : iter().get_far_dest();
// If this is a backwards branch in the bytecodes, add Safepoint
maybe_add_safepoint(target_bci);
// Update method data
profile_taken_branch(target_bci);
// Merge the current control into the target basic block
merge(target_bci);
// See if we can get some profile data and hand it off to the next block
Block *target_block = block()->successor_for_bci(target_bci);
if (target_block->pred_count() != 1) break;
ciMethodData* methodData = method()->method_data();
if (!methodData->is_mature()) break;
ciProfileData* data = methodData->bci_to_data(bci());
assert( data->is_JumpData(), "" );
int taken = ((ciJumpData*)data)->taken();
taken = method()->scale_count(taken);
target_block->set_count(taken);
break;
}
case Bytecodes::_ifnull: btest = BoolTest::eq; goto handle_if_null;
case Bytecodes::_ifnonnull: btest = BoolTest::ne; goto handle_if_null;
handle_if_null:
// If this is a backwards branch in the bytecodes, add Safepoint
maybe_add_safepoint(iter().get_dest());
a = null();
b = pop();
c = _gvn.transform( new (C) CmpPNode(b, a) );
do_ifnull(btest, c);
break;
case Bytecodes::_if_acmpeq: btest = BoolTest::eq; goto handle_if_acmp;
case Bytecodes::_if_acmpne: btest = BoolTest::ne; goto handle_if_acmp;
handle_if_acmp:
// If this is a backwards branch in the bytecodes, add Safepoint
maybe_add_safepoint(iter().get_dest());
a = pop();
b = pop();
c = _gvn.transform( new (C) CmpPNode(b, a) );
c = optimize_cmp_with_klass(c);
do_if(btest, c);
break;
case Bytecodes::_ifeq: btest = BoolTest::eq; goto handle_ifxx;
case Bytecodes::_ifne: btest = BoolTest::ne; goto handle_ifxx;
case Bytecodes::_iflt: btest = BoolTest::lt; goto handle_ifxx;
case Bytecodes::_ifle: btest = BoolTest::le; goto handle_ifxx;
case Bytecodes::_ifgt: btest = BoolTest::gt; goto handle_ifxx;
case Bytecodes::_ifge: btest = BoolTest::ge; goto handle_ifxx;
handle_ifxx:
// If this is a backwards branch in the bytecodes, add Safepoint
maybe_add_safepoint(iter().get_dest());
a = _gvn.intcon(0);
b = pop();
c = _gvn.transform( new (C) CmpINode(b, a) );
do_if(btest, c);
break;
case Bytecodes::_if_icmpeq: btest = BoolTest::eq; goto handle_if_icmp;
case Bytecodes::_if_icmpne: btest = BoolTest::ne; goto handle_if_icmp;
case Bytecodes::_if_icmplt: btest = BoolTest::lt; goto handle_if_icmp;
case Bytecodes::_if_icmple: btest = BoolTest::le; goto handle_if_icmp;
case Bytecodes::_if_icmpgt: btest = BoolTest::gt; goto handle_if_icmp;
case Bytecodes::_if_icmpge: btest = BoolTest::ge; goto handle_if_icmp;
handle_if_icmp:
// If this is a backwards branch in the bytecodes, add Safepoint
maybe_add_safepoint(iter().get_dest());
a = pop();
b = pop();
c = _gvn.transform( new (C) CmpINode( b, a ) );
do_if(btest, c);
break;
case Bytecodes::_tableswitch:
do_tableswitch();
break;
case Bytecodes::_lookupswitch:
do_lookupswitch();
break;
case Bytecodes::_invokestatic:
case Bytecodes::_invokedynamic:
case Bytecodes::_invokespecial:
case Bytecodes::_invokevirtual:
case Bytecodes::_invokeinterface:
do_call();
break;
case Bytecodes::_checkcast:
do_checkcast();
break;
case Bytecodes::_instanceof:
do_instanceof();
break;
case Bytecodes::_anewarray:
do_anewarray();
break;
case Bytecodes::_newarray:
do_newarray((BasicType)iter().get_index());
break;
case Bytecodes::_multianewarray:
do_multianewarray();
break;
case Bytecodes::_new:
do_new();
break;
case Bytecodes::_jsr:
case Bytecodes::_jsr_w:
do_jsr();
break;
case Bytecodes::_ret:
do_ret();
break;
case Bytecodes::_monitorenter:
do_monitor_enter();
break;
case Bytecodes::_monitorexit:
do_monitor_exit();
break;
case Bytecodes::_breakpoint:
// Breakpoint set concurrently to compile
// %%% use an uncommon trap?
C->record_failure("breakpoint in method");
return;
default:
#ifndef PRODUCT
map()->dump(99);
#endif
tty->print("\nUnhandled bytecode %s\n", Bytecodes::name(bc()) );
ShouldNotReachHere();
}
#ifndef PRODUCT
IdealGraphPrinter *printer = IdealGraphPrinter::printer();
if(printer) {
char buffer[256];
sprintf(buffer, "Bytecode %d: %s", bci(), Bytecodes::name(bc()));
bool old = printer->traverse_outs();
printer->set_traverse_outs(true);
printer->print_method(C, buffer, 4);
printer->set_traverse_outs(old);
}
#endif
}
C:\hotspot-69087d08d473\src\share\vm/opto/parse3.cpp
/*
* Copyright (c) 1998, 2013, 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 "interpreter/linkResolver.hpp"
#include "memory/universe.inline.hpp"
#include "oops/objArrayKlass.hpp"
#include "opto/addnode.hpp"
#include "opto/memnode.hpp"
#include "opto/parse.hpp"
#include "opto/rootnode.hpp"
#include "opto/runtime.hpp"
#include "opto/subnode.hpp"
#include "runtime/deoptimization.hpp"
#include "runtime/handles.inline.hpp"
//=============================================================================
// Helper methods for _get* and _put* bytecodes
//=============================================================================
bool Parse::static_field_ok_in_clinit(ciField *field, ciMethod *method) {
// Could be the field_holder's <clinit> method, or <clinit> for a subklass.
// Better to check now than to Deoptimize as soon as we execute
assert( field->is_static(), "Only check if field is static");
// is_being_initialized() is too generous. It allows access to statics
// by threads that are not running the <clinit> before the <clinit> finishes.
// return field->holder()->is_being_initialized();
// The following restriction is correct but conservative.
// It is also desirable to allow compilation of methods called from <clinit>
// but this generated code will need to be made safe for execution by
// other threads, or the transition from interpreted to compiled code would
// need to be guarded.
ciInstanceKlass *field_holder = field->holder();
bool access_OK = false;
if (method->holder()->is_subclass_of(field_holder)) {
if (method->is_static()) {
if (method->name() == ciSymbol::class_initializer_name()) {
// OK to access static fields inside initializer
access_OK = true;
}
} else {
if (method->name() == ciSymbol::object_initializer_name()) {
// It's also OK to access static fields inside a constructor,
// because any thread calling the constructor must first have
// synchronized on the class by executing a '_new' bytecode.
access_OK = true;
}
}
}
return access_OK;
}
void Parse::do_field_access(bool is_get, bool is_field) {
bool will_link;
ciField* field = iter().get_field(will_link);
assert(will_link, "getfield: typeflow responsibility");
ciInstanceKlass* field_holder = field->holder();
if (is_field == field->is_static()) {
// Interpreter will throw java_lang_IncompatibleClassChangeError
// Check this before allowing <clinit> methods to access static fields
uncommon_trap(Deoptimization::Reason_unhandled,
Deoptimization::Action_none);
return;
}
if (!is_field && !field_holder->is_initialized()) {
if (!static_field_ok_in_clinit(field, method())) {
uncommon_trap(Deoptimization::Reason_uninitialized,
Deoptimization::Action_reinterpret,
NULL, "!static_field_ok_in_clinit");
return;
}
}
// Deoptimize on putfield writes to call site target field.
if (!is_get && field->is_call_site_target()) {
uncommon_trap(Deoptimization::Reason_unhandled,
Deoptimization::Action_reinterpret,
NULL, "put to call site target field");
return;
}
assert(field->will_link(method()->holder(), bc()), "getfield: typeflow responsibility");
// Note: We do not check for an unloaded field type here any more.
// Generate code for the object pointer.
Node* obj;
if (is_field) {
int obj_depth = is_get ? 0 : field->type()->size();
obj = null_check(peek(obj_depth));
// Compile-time detect of null-exception?
if (stopped()) return;
#ifdef ASSERT
const TypeInstPtr *tjp = TypeInstPtr::make(TypePtr::NotNull, iter().get_declared_field_holder());
assert(_gvn.type(obj)->higher_equal(tjp), "cast_up is no longer needed");
#endif
if (is_get) {
(void) pop(); // pop receiver before getting
do_get_xxx(obj, field, is_field);
} else {
do_put_xxx(obj, field, is_field);
(void) pop(); // pop receiver after putting
}
} else {
const TypeInstPtr* tip = TypeInstPtr::make(field_holder->java_mirror());
obj = _gvn.makecon(tip);
if (is_get) {
do_get_xxx(obj, field, is_field);
} else {
do_put_xxx(obj, field, is_field);
}
}
}
void Parse::do_get_xxx(Node* obj, ciField* field, bool is_field) {
// Does this field have a constant value? If so, just push the value.
if (field->is_constant()) {
// final or stable field
const Type* stable_type = NULL;
if (FoldStableValues && field->is_stable()) {
stable_type = Type::get_const_type(field->type());
if (field->type()->is_array_klass()) {
int stable_dimension = field->type()->as_array_klass()->dimension();
stable_type = stable_type->is_aryptr()->cast_to_stable(true, stable_dimension);
}
}
if (field->is_static()) {
// final static field
if (C->eliminate_boxing()) {
// The pointers in the autobox arrays are always non-null.
ciSymbol* klass_name = field->holder()->name();
if (field->name() == ciSymbol::cache_field_name() &&
field->holder()->uses_default_loader() &&
(klass_name == ciSymbol::java_lang_Character_CharacterCache() ||
klass_name == ciSymbol::java_lang_Byte_ByteCache() ||
klass_name == ciSymbol::java_lang_Short_ShortCache() ||
klass_name == ciSymbol::java_lang_Integer_IntegerCache() ||
klass_name == ciSymbol::java_lang_Long_LongCache())) {
bool require_const = true;
bool autobox_cache = true;
if (push_constant(field->constant_value(), require_const, autobox_cache)) {
return;
}
}
}
if (push_constant(field->constant_value(), false, false, stable_type))
return;
} else {
// final or stable non-static field
// Treat final non-static fields of trusted classes (classes in
// java.lang.invoke and sun.invoke packages and subpackages) as
// compile time constants.
if (obj->is_Con()) {
const TypeOopPtr* oop_ptr = obj->bottom_type()->isa_oopptr();
ciObject* constant_oop = oop_ptr->const_oop();
ciConstant constant = field->constant_value_of(constant_oop);
if (FoldStableValues && field->is_stable() && constant.is_null_or_zero()) {
// fall through to field load; the field is not yet initialized
} else {
if (push_constant(constant, true, false, stable_type))
return;
}
}
}
}
Node* leading_membar = NULL;
ciType* field_klass = field->type();
bool is_vol = field->is_volatile();
// Compute address and memory type.
int offset = field->offset_in_bytes();
const TypePtr* adr_type = C->alias_type(field)->adr_type();
Node *adr = basic_plus_adr(obj, obj, offset);
BasicType bt = field->layout_type();
// Build the resultant type of the load
const Type *type;
bool must_assert_null = false;
if( bt == T_OBJECT ) {
if (!field->type()->is_loaded()) {
type = TypeInstPtr::BOTTOM;
must_assert_null = true;
} else if (field->is_constant() && field->is_static()) {
// This can happen if the constant oop is non-perm.
ciObject* con = field->constant_value().as_object();
// Do not "join" in the previous type; it doesn't add value,
// and may yield a vacuous result if the field is of interface type.
type = TypeOopPtr::make_from_constant(con)->isa_oopptr();
assert(type != NULL, "field singleton type must be consistent");
} else {
type = TypeOopPtr::make_from_klass(field_klass->as_klass());
}
} else {
type = Type::get_const_basic_type(bt);
}
if (support_IRIW_for_not_multiple_copy_atomic_cpu && field->is_volatile()) {
leading_membar = insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
}
// Build the load.
//
MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
Node* ld = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
// Adjust Java stack
if (type2size[bt] == 1)
push(ld);
else
push_pair(ld);
if (must_assert_null) {
// Do not take a trap here. It's possible that the program
// will never load the field's class, and will happily see
// null values in this field forever. Don't stumble into a
// trap for such a program, or we might get a long series
// of useless recompilations. (Or, we might load a class
// which should not be loaded.) If we ever see a non-null
// value, we will then trap and recompile. (The trap will
// not need to mention the class index, since the class will
// already have been loaded if we ever see a non-null value.)
// uncommon_trap(iter().get_field_signature_index());
#ifndef PRODUCT
if (PrintOpto && (Verbose || WizardMode)) {
method()->print_name(); tty->print_cr(" asserting nullness of field at bci: %d", bci());
}
#endif
if (C->log() != NULL) {
C->log()->elem("assert_null reason='field' klass='%d'",
C->log()->identify(field->type()));
}
// If there is going to be a trap, put it at the next bytecode:
set_bci(iter().next_bci());
null_assert(peek());
set_bci(iter().cur_bci()); // put it back
}
// If reference is volatile, prevent following memory ops from
// floating up past the volatile read. Also prevents commoning
// another volatile read.
if (field->is_volatile()) {
// Memory barrier includes bogus read of value to force load BEFORE membar
assert(leading_membar == NULL || support_IRIW_for_not_multiple_copy_atomic_cpu, "no leading membar expected");
Node* mb = insert_mem_bar(Op_MemBarAcquire, ld);
mb->as_MemBar()->set_trailing_load();
}
}
void Parse::do_put_xxx(Node* obj, ciField* field, bool is_field) {
Node* leading_membar = NULL;
bool is_vol = field->is_volatile();
// If reference is volatile, prevent following memory ops from
// floating down past the volatile write. Also prevents commoning
// another volatile read.
if (is_vol) {
leading_membar = insert_mem_bar(Op_MemBarRelease);
}
// Compute address and memory type.
int offset = field->offset_in_bytes();
const TypePtr* adr_type = C->alias_type(field)->adr_type();
Node* adr = basic_plus_adr(obj, obj, offset);
BasicType bt = field->layout_type();
// Value to be stored
Node* val = type2size[bt] == 1 ? pop() : pop_pair();
// Round doubles before storing
if (bt == T_DOUBLE) val = dstore_rounding(val);
// Conservatively release stores of object references.
const MemNode::MemOrd mo =
is_vol ?
// Volatile fields need releasing stores.
MemNode::release :
// Non-volatile fields also need releasing stores if they hold an
// object reference, because the object reference might point to
// a freshly created object.
StoreNode::release_if_reference(bt);
// Store the value.
Node* store;
if (bt == T_OBJECT) {
const TypeOopPtr* field_type;
if (!field->type()->is_loaded()) {
field_type = TypeInstPtr::BOTTOM;
} else {
field_type = TypeOopPtr::make_from_klass(field->type()->as_klass());
}
store = store_oop_to_object(control(), obj, adr, adr_type, val, field_type, bt, mo);
} else {
store = store_to_memory(control(), adr, val, bt, adr_type, mo, is_vol);
}
// If reference is volatile, prevent following volatiles ops from
// floating up before the volatile write.
if (is_vol) {
// If not multiple copy atomic, we do the MemBarVolatile before the load.
if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
Node* mb = insert_mem_bar(Op_MemBarVolatile, store); // Use fat membar
MemBarNode::set_store_pair(leading_membar->as_MemBar(), mb->as_MemBar());
}
// Remember we wrote a volatile field.
// For not multiple copy atomic cpu (ppc64) a barrier should be issued
// in constructors which have such stores. See do_exits() in parse1.cpp.
if (is_field) {
set_wrote_volatile(true);
}
}
// If the field is final, the rules of Java say we are in <init> or <clinit>.
// Note the presence of writes to final non-static fields, so that we
// can insert a memory barrier later on to keep the writes from floating
// out of the constructor.
// Any method can write a @Stable field; insert memory barriers after those also.
if (is_field && (field->is_final() || field->is_stable())) {
set_wrote_final(true);
// Preserve allocation ptr to create precedent edge to it in membar
// generated on exit from constructor.
if (C->eliminate_boxing() &&
adr_type->isa_oopptr() && adr_type->is_oopptr()->is_ptr_to_boxed_value() &&
AllocateNode::Ideal_allocation(obj, &_gvn) != NULL) {
set_alloc_with_final(obj);
}
}
}
bool Parse::push_constant(ciConstant constant, bool require_constant, bool is_autobox_cache, const Type* stable_type) {
const Type* con_type = Type::make_from_constant(constant, require_constant, is_autobox_cache);
switch (constant.basic_type()) {
case T_ARRAY:
case T_OBJECT:
// cases:
// can_be_constant = (oop not scavengable || ScavengeRootsInCode != 0)
// should_be_constant = (oop not scavengable || ScavengeRootsInCode >= 2)
// An oop is not scavengable if it is in the perm gen.
if (stable_type != NULL && con_type != NULL && con_type->isa_oopptr())
con_type = con_type->join_speculative(stable_type);
break;
case T_ILLEGAL:
// Invalid ciConstant returned due to OutOfMemoryError in the CI
assert(C->env()->failing(), "otherwise should not see this");
// These always occur because of object types; we are going to
// bail out anyway, so make the stack depths match up
push( zerocon(T_OBJECT) );
return false;
}
if (con_type == NULL)
// we cannot inline the oop, but we can use it later to narrow a type
return false;
push_node(constant.basic_type(), makecon(con_type));
return true;
}
//=============================================================================
void Parse::do_anewarray() {
bool will_link;
ciKlass* klass = iter().get_klass(will_link);
// Uncommon Trap when class that array contains is not loaded
// we need the loaded class for the rest of graph; do not
// initialize the container class (see Java spec)!!!
assert(will_link, "anewarray: typeflow responsibility");
ciObjArrayKlass* array_klass = ciObjArrayKlass::make(klass);
// Check that array_klass object is loaded
if (!array_klass->is_loaded()) {
// Generate uncommon_trap for unloaded array_class
uncommon_trap(Deoptimization::Reason_unloaded,
Deoptimization::Action_reinterpret,
array_klass);
return;
}
kill_dead_locals();
const TypeKlassPtr* array_klass_type = TypeKlassPtr::make(array_klass);
Node* count_val = pop();
Node* obj = new_array(makecon(array_klass_type), count_val, 1);
push(obj);
}
void Parse::do_newarray(BasicType elem_type) {
kill_dead_locals();
Node* count_val = pop();
const TypeKlassPtr* array_klass = TypeKlassPtr::make(ciTypeArrayKlass::make(elem_type));
Node* obj = new_array(makecon(array_klass), count_val, 1);
// Push resultant oop onto stack
push(obj);
}
// Expand simple expressions like new int[3][5] and new Object[2][nonConLen].
// Also handle the degenerate 1-dimensional case of anewarray.
Node* Parse::expand_multianewarray(ciArrayKlass* array_klass, Node* *lengths, int ndimensions, int nargs) {
Node* length = lengths[0];
assert(length != NULL, "");
Node* array = new_array(makecon(TypeKlassPtr::make(array_klass)), length, nargs);
if (ndimensions > 1) {
jint length_con = find_int_con(length, -1);
guarantee(length_con >= 0, "non-constant multianewarray");
ciArrayKlass* array_klass_1 = array_klass->as_obj_array_klass()->element_klass()->as_array_klass();
const TypePtr* adr_type = TypeAryPtr::OOPS;
const TypeOopPtr* elemtype = _gvn.type(array)->is_aryptr()->elem()->make_oopptr();
const intptr_t header = arrayOopDesc::base_offset_in_bytes(T_OBJECT);
for (jint i = 0; i < length_con; i++) {
Node* elem = expand_multianewarray(array_klass_1, &lengths[1], ndimensions-1, nargs);
intptr_t offset = header + ((intptr_t)i << LogBytesPerHeapOop);
Node* eaddr = basic_plus_adr(array, offset);
store_oop_to_array(control(), array, eaddr, adr_type, elem, elemtype, T_OBJECT, MemNode::unordered);
}
}
return array;
}
void Parse::do_multianewarray() {
int ndimensions = iter().get_dimensions();
// the m-dimensional array
bool will_link;
ciArrayKlass* array_klass = iter().get_klass(will_link)->as_array_klass();
assert(will_link, "multianewarray: typeflow responsibility");
// Note: Array classes are always initialized; no is_initialized check.
kill_dead_locals();
// get the lengths from the stack (first dimension is on top)
Node** length = NEW_RESOURCE_ARRAY(Node*, ndimensions + 1);
length[ndimensions] = NULL; // terminating null for make_runtime_call
int j;
for (j = ndimensions-1; j >= 0 ; j--) length[j] = pop();
// The original expression was of this form: new T[length0][length1]...
// It is often the case that the lengths are small (except the last).
// If that happens, use the fast 1-d creator a constant number of times.
const jint expand_limit = MIN2((juint)MultiArrayExpandLimit, (juint)100);
jint expand_count = 1; // count of allocations in the expansion
jint expand_fanout = 1; // running total fanout
for (j = 0; j < ndimensions-1; j++) {
jint dim_con = find_int_con(length[j], -1);
expand_fanout *= dim_con;
expand_count += expand_fanout; // count the level-J sub-arrays
if (dim_con <= 0
|| dim_con > expand_limit
|| expand_count > expand_limit) {
expand_count = 0;
break;
}
}
// Can use multianewarray instead of [a]newarray if only one dimension,
// or if all non-final dimensions are small constants.
if (ndimensions == 1 || (1 <= expand_count && expand_count <= expand_limit)) {
Node* obj = NULL;
// Set the original stack and the reexecute bit for the interpreter
// to reexecute the multianewarray bytecode if deoptimization happens.
// Do it unconditionally even for one dimension multianewarray.
// Note: the reexecute bit will be set in GraphKit::add_safepoint_edges()
// when AllocateArray node for newarray is created.
{ PreserveReexecuteState preexecs(this);
inc_sp(ndimensions);
// Pass 0 as nargs since uncommon trap code does not need to restore stack.
obj = expand_multianewarray(array_klass, &length[0], ndimensions, 0);
} //original reexecute and sp are set back here
push(obj);
return;
}
address fun = NULL;
switch (ndimensions) {
case 1: ShouldNotReachHere(); break;
case 2: fun = OptoRuntime::multianewarray2_Java(); break;
case 3: fun = OptoRuntime::multianewarray3_Java(); break;
case 4: fun = OptoRuntime::multianewarray4_Java(); break;
case 5: fun = OptoRuntime::multianewarray5_Java(); break;
};
Node* c = NULL;
if (fun != NULL) {
c = make_runtime_call(RC_NO_LEAF | RC_NO_IO,
OptoRuntime::multianewarray_Type(ndimensions),
fun, NULL, TypeRawPtr::BOTTOM,
makecon(TypeKlassPtr::make(array_klass)),
length[0], length[1], length[2],
(ndimensions > 2) ? length[3] : NULL,
(ndimensions > 3) ? length[4] : NULL);
} else {
// Create a java array for dimension sizes
Node* dims = NULL;
{ PreserveReexecuteState preexecs(this);
inc_sp(ndimensions);
Node* dims_array_klass = makecon(TypeKlassPtr::make(ciArrayKlass::make(ciType::make(T_INT))));
dims = new_array(dims_array_klass, intcon(ndimensions), 0);
// Fill-in it with values
for (j = 0; j < ndimensions; j++) {
Node *dims_elem = array_element_address(dims, intcon(j), T_INT);
store_to_memory(control(), dims_elem, length[j], T_INT, TypeAryPtr::INTS, MemNode::unordered);
}
}
c = make_runtime_call(RC_NO_LEAF | RC_NO_IO,
OptoRuntime::multianewarrayN_Type(),
OptoRuntime::multianewarrayN_Java(), NULL, TypeRawPtr::BOTTOM,
makecon(TypeKlassPtr::make(array_klass)),
dims);
}
make_slow_call_ex(c, env()->Throwable_klass(), false);
Node* res = _gvn.transform(new (C) ProjNode(c, TypeFunc::Parms));
const Type* type = TypeOopPtr::make_from_klass_raw(array_klass);
// Improve the type: We know it's not null, exact, and of a given length.
type = type->is_ptr()->cast_to_ptr_type(TypePtr::NotNull);
type = type->is_aryptr()->cast_to_exactness(true);
const TypeInt* ltype = _gvn.find_int_type(length[0]);
if (ltype != NULL)
type = type->is_aryptr()->cast_to_size(ltype);
// We cannot sharpen the nested sub-arrays, since the top level is mutable.
Node* cast = _gvn.transform( new (C) CheckCastPPNode(control(), res, type) );
push(cast);
// Possible improvements:
// - Make a fast path for small multi-arrays. (W/ implicit init. loops.)
// - Issue CastII against length[*] values, to TypeInt::POS.
}
C:\hotspot-69087d08d473\src\share\vm/opto/parseHelper.cpp
/*
* Copyright (c) 1998, 2013, 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 "classfile/systemDictionary.hpp"
#include "compiler/compileLog.hpp"
#include "oops/objArrayKlass.hpp"
#include "opto/addnode.hpp"
#include "opto/memnode.hpp"
#include "opto/mulnode.hpp"
#include "opto/parse.hpp"
#include "opto/rootnode.hpp"
#include "opto/runtime.hpp"
#include "runtime/sharedRuntime.hpp"
//------------------------------make_dtrace_method_entry_exit ----------------
// Dtrace -- record entry or exit of a method if compiled with dtrace support
void GraphKit::make_dtrace_method_entry_exit(ciMethod* method, bool is_entry) {
const TypeFunc *call_type = OptoRuntime::dtrace_method_entry_exit_Type();
address call_address = is_entry ? CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry) :
CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit);
const char *call_name = is_entry ? "dtrace_method_entry" : "dtrace_method_exit";
// Get base of thread-local storage area
Node* thread = _gvn.transform( new (C) ThreadLocalNode() );
// Get method
const TypePtr* method_type = TypeMetadataPtr::make(method);
Node *method_node = _gvn.transform( ConNode::make(C, method_type) );
kill_dead_locals();
// For some reason, this call reads only raw memory.
const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
make_runtime_call(RC_LEAF | RC_NARROW_MEM,
call_type, call_address,
call_name, raw_adr_type,
thread, method_node);
}
//=============================================================================
//------------------------------do_checkcast-----------------------------------
void Parse::do_checkcast() {
bool will_link;
ciKlass* klass = iter().get_klass(will_link);
Node *obj = peek();
// Throw uncommon trap if class is not loaded or the value we are casting
// _from_ is not loaded, and value is not null. If the value _is_ NULL,
// then the checkcast does nothing.
const TypeOopPtr *tp = _gvn.type(obj)->isa_oopptr();
if (!will_link || (tp && tp->klass() && !tp->klass()->is_loaded())) {
if (C->log() != NULL) {
if (!will_link) {
C->log()->elem("assert_null reason='checkcast' klass='%d'",
C->log()->identify(klass));
}
if (tp && tp->klass() && !tp->klass()->is_loaded()) {
// %%% Cannot happen?
C->log()->elem("assert_null reason='checkcast source' klass='%d'",
C->log()->identify(tp->klass()));
}
}
null_assert(obj);
assert( stopped() || _gvn.type(peek())->higher_equal(TypePtr::NULL_PTR), "what's left behind is null" );
if (!stopped()) {
profile_null_checkcast();
}
return;
}
Node *res = gen_checkcast(obj, makecon(TypeKlassPtr::make(klass)) );
// Pop from stack AFTER gen_checkcast because it can uncommon trap and
// the debug info has to be correct.
pop();
push(res);
}
//------------------------------do_instanceof----------------------------------
void Parse::do_instanceof() {
if (stopped()) return;
// We would like to return false if class is not loaded, emitting a
// dependency, but Java requires instanceof to load its operand.
// Throw uncommon trap if class is not loaded
bool will_link;
ciKlass* klass = iter().get_klass(will_link);
if (!will_link) {
if (C->log() != NULL) {
C->log()->elem("assert_null reason='instanceof' klass='%d'",
C->log()->identify(klass));
}
null_assert(peek());
assert( stopped() || _gvn.type(peek())->higher_equal(TypePtr::NULL_PTR), "what's left behind is null" );
if (!stopped()) {
// The object is now known to be null.
// Shortcut the effect of gen_instanceof and return "false" directly.
pop(); // pop the null
push(_gvn.intcon(0)); // push false answer
}
return;
}
// Push the bool result back on stack
Node* res = gen_instanceof(peek(), makecon(TypeKlassPtr::make(klass)), true);
// Pop from stack AFTER gen_instanceof because it can uncommon trap.
pop();
push(res);
}
//------------------------------array_store_check------------------------------
// pull array from stack and check that the store is valid
void Parse::array_store_check() {
// Shorthand access to array store elements without popping them.
Node *obj = peek(0);
Node *idx = peek(1);
Node *ary = peek(2);
if (_gvn.type(obj) == TypePtr::NULL_PTR) {
// There's never a type check on null values.
// This cutout lets us avoid the uncommon_trap(Reason_array_check)
// below, which turns into a performance liability if the
// gen_checkcast folds up completely.
return;
}
// Extract the array klass type
int klass_offset = oopDesc::klass_offset_in_bytes();
Node* p = basic_plus_adr( ary, ary, klass_offset );
// p's type is array-of-OOPS plus klass_offset
Node* array_klass = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeInstPtr::KLASS));
// Get the array klass
const TypeKlassPtr *tak = _gvn.type(array_klass)->is_klassptr();
// The type of array_klass is usually INexact array-of-oop. Heroically
// cast array_klass to EXACT array and uncommon-trap if the cast fails.
// Make constant out of the inexact array klass, but use it only if the cast
// succeeds.
bool always_see_exact_class = false;
if (MonomorphicArrayCheck
&& !too_many_traps(Deoptimization::Reason_array_check)
&& !tak->klass_is_exact()
&& tak != TypeKlassPtr::OBJECT) {
// Regarding the fourth condition in the if-statement from above:
//
// If the compiler has determined that the type of array 'ary' (represented
// by 'array_klass') is java/lang/Object, the compiler must not assume that
// the array 'ary' is monomorphic.
//
// If 'ary' were of type java/lang/Object, this arraystore would have to fail,
// because it is not possible to perform a arraystore into an object that is not
// a "proper" array.
//
// Therefore, let's obtain at runtime the type of 'ary' and check if we can still
// successfully perform the store.
//
// The implementation reasons for the condition are the following:
//
// java/lang/Object is the superclass of all arrays, but it is represented by the VM
// as an InstanceKlass. The checks generated by gen_checkcast() (see below) expect
// 'array_klass' to be ObjArrayKlass, which can result in invalid memory accesses.
//
// See issue JDK-8057622 for details.
always_see_exact_class = true;
// (If no MDO at all, hope for the best, until a trap actually occurs.)
// Make a constant out of the inexact array klass
const TypeKlassPtr *extak = tak->cast_to_exactness(true)->is_klassptr();
Node* con = makecon(extak);
Node* cmp = _gvn.transform(new (C) CmpPNode( array_klass, con ));
Node* bol = _gvn.transform(new (C) BoolNode( cmp, BoolTest::eq ));
Node* ctrl= control();
{ BuildCutout unless(this, bol, PROB_MAX);
uncommon_trap(Deoptimization::Reason_array_check,
Deoptimization::Action_maybe_recompile,
tak->klass());
}
if (stopped()) { // MUST uncommon-trap?
set_control(ctrl); // Then Don't Do It, just fall into the normal checking
} else { // Cast array klass to exactness:
// Use the exact constant value we know it is.
replace_in_map(array_klass,con);
CompileLog* log = C->log();
if (log != NULL) {
log->elem("cast_up reason='monomorphic_array' from='%d' to='(exact)'",
log->identify(tak->klass()));
}
array_klass = con; // Use cast value moving forward
}
}
// Come here for polymorphic array klasses
// Extract the array element class
int element_klass_offset = in_bytes(ObjArrayKlass::element_klass_offset());
Node *p2 = basic_plus_adr(array_klass, array_klass, element_klass_offset);
// We are allowed to use the constant type only if cast succeeded. If always_see_exact_class is true,
// we must set a control edge from the IfTrue node created by the uncommon_trap above to the
// LoadKlassNode.
Node* a_e_klass = _gvn.transform(LoadKlassNode::make(_gvn, always_see_exact_class ? control() : NULL,
immutable_memory(), p2, tak));
// Check (the hard way) and throw if not a subklass.
// Result is ignored, we just need the CFG effects.
gen_checkcast(obj, a_e_klass);
}
void Parse::emit_guard_for_new(ciInstanceKlass* klass) {
// Emit guarded new
// if (klass->_init_thread != current_thread ||
// klass->_init_state != being_initialized)
// uncommon_trap
Node* cur_thread = _gvn.transform( new (C) ThreadLocalNode() );
Node* merge = new (C) RegionNode(3);
_gvn.set_type(merge, Type::CONTROL);
Node* kls = makecon(TypeKlassPtr::make(klass));
Node* init_thread_offset = _gvn.MakeConX(in_bytes(InstanceKlass::init_thread_offset()));
Node* adr_node = basic_plus_adr(kls, kls, init_thread_offset);
Node* init_thread = make_load(NULL, adr_node, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
Node *tst = Bool( CmpP( init_thread, cur_thread), BoolTest::eq);
IfNode* iff = create_and_map_if(control(), tst, PROB_ALWAYS, COUNT_UNKNOWN);
set_control(IfTrue(iff));
merge->set_req(1, IfFalse(iff));
Node* init_state_offset = _gvn.MakeConX(in_bytes(InstanceKlass::init_state_offset()));
adr_node = basic_plus_adr(kls, kls, init_state_offset);
// Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
// can generate code to load it as unsigned byte.
Node* init_state = make_load(NULL, adr_node, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
Node* being_init = _gvn.intcon(InstanceKlass::being_initialized);
tst = Bool( CmpI( init_state, being_init), BoolTest::eq);
iff = create_and_map_if(control(), tst, PROB_ALWAYS, COUNT_UNKNOWN);
set_control(IfTrue(iff));
merge->set_req(2, IfFalse(iff));
PreserveJVMState pjvms(this);
record_for_igvn(merge);
set_control(merge);
uncommon_trap(Deoptimization::Reason_uninitialized,
Deoptimization::Action_reinterpret,
klass);
}
//------------------------------do_new-----------------------------------------
void Parse::do_new() {
kill_dead_locals();
bool will_link;
ciInstanceKlass* klass = iter().get_klass(will_link)->as_instance_klass();
assert(will_link, "_new: typeflow responsibility");
// Should initialize, or throw an InstantiationError?
if (!klass->is_initialized() && !klass->is_being_initialized() ||
klass->is_abstract() || klass->is_interface() ||
klass->name() == ciSymbol::java_lang_Class() ||
iter().is_unresolved_klass()) {
uncommon_trap(Deoptimization::Reason_uninitialized,
Deoptimization::Action_reinterpret,
klass);
return;
}
if (klass->is_being_initialized()) {
emit_guard_for_new(klass);
}
Node* kls = makecon(TypeKlassPtr::make(klass));
Node* obj = new_instance(kls);
// Push resultant oop onto stack
push(obj);
// Keep track of whether opportunities exist for StringBuilder
// optimizations.
if (OptimizeStringConcat &&
(klass == C->env()->StringBuilder_klass() ||
klass == C->env()->StringBuffer_klass())) {
C->set_has_stringbuilder(true);
}
// Keep track of boxed values for EliminateAutoBox optimizations.
if (C->eliminate_boxing() && klass->is_box_klass()) {
C->set_has_boxed_value(true);
}
}
#ifndef PRODUCT
//------------------------------dump_map_adr_mem-------------------------------
// Debug dump of the mapping from address types to MergeMemNode indices.
void Parse::dump_map_adr_mem() const {
tty->print_cr("--- Mapping from address types to memory Nodes ---");
MergeMemNode *mem = map() == NULL ? NULL : (map()->memory()->is_MergeMem() ?
map()->memory()->as_MergeMem() : NULL);
for (uint i = 0; i < (uint)C->num_alias_types(); i++) {
C->alias_type(i)->print_on(tty);
tty->print("\t");
// Node mapping, if any
if (mem && i < mem->req() && mem->in(i) && mem->in(i) != mem->empty_memory()) {
mem->in(i)->dump();
} else {
tty->cr();
}
}
}
#endif
//=============================================================================
//
// parser methods for profiling
//----------------------test_counter_against_threshold ------------------------
void Parse::test_counter_against_threshold(Node* cnt, int limit) {
// Test the counter against the limit and uncommon trap if greater.
// This code is largely copied from the range check code in
// array_addressing()
// Test invocation count vs threshold
Node *threshold = makecon(TypeInt::make(limit));
Node *chk = _gvn.transform( new (C) CmpUNode( cnt, threshold) );
BoolTest::mask btest = BoolTest::lt;
Node *tst = _gvn.transform( new (C) BoolNode( chk, btest) );
// Branch to failure if threshold exceeded
{ BuildCutout unless(this, tst, PROB_ALWAYS);
uncommon_trap(Deoptimization::Reason_age,
Deoptimization::Action_maybe_recompile);
}
}
//----------------------increment_and_test_invocation_counter-------------------
void Parse::increment_and_test_invocation_counter(int limit) {
if (!count_invocations()) return;
// Get the Method* node.
ciMethod* m = method();
MethodCounters* counters_adr = m->ensure_method_counters();
if (counters_adr == NULL) {
C->record_failure("method counters allocation failed");
return;
}
Node* ctrl = control();
const TypePtr* adr_type = TypeRawPtr::make((address) counters_adr);
Node *counters_node = makecon(adr_type);
Node* adr_iic_node = basic_plus_adr(counters_node, counters_node,
MethodCounters::interpreter_invocation_counter_offset_in_bytes());
Node* cnt = make_load(ctrl, adr_iic_node, TypeInt::INT, T_INT, adr_type, MemNode::unordered);
test_counter_against_threshold(cnt, limit);
// Add one to the counter and store
Node* incr = _gvn.transform(new (C) AddINode(cnt, _gvn.intcon(1)));
store_to_memory(ctrl, adr_iic_node, incr, T_INT, adr_type, MemNode::unordered);
}
//----------------------------method_data_addressing---------------------------
Node* Parse::method_data_addressing(ciMethodData* md, ciProfileData* data, ByteSize counter_offset, Node* idx, uint stride) {
// Get offset within MethodData* of the data array
ByteSize data_offset = MethodData::data_offset();
// Get cell offset of the ProfileData within data array
int cell_offset = md->dp_to_di(data->dp());
// Add in counter_offset, the # of bytes into the ProfileData of counter or flag
int offset = in_bytes(data_offset) + cell_offset + in_bytes(counter_offset);
const TypePtr* adr_type = TypeMetadataPtr::make(md);
Node* mdo = makecon(adr_type);
Node* ptr = basic_plus_adr(mdo, mdo, offset);
if (stride != 0) {
Node* str = _gvn.MakeConX(stride);
Node* scale = _gvn.transform( new (C) MulXNode( idx, str ) );
ptr = _gvn.transform( new (C) AddPNode( mdo, ptr, scale ) );
}
return ptr;
}
//--------------------------increment_md_counter_at----------------------------
void Parse::increment_md_counter_at(ciMethodData* md, ciProfileData* data, ByteSize counter_offset, Node* idx, uint stride) {
Node* adr_node = method_data_addressing(md, data, counter_offset, idx, stride);
const TypePtr* adr_type = _gvn.type(adr_node)->is_ptr();
Node* cnt = make_load(NULL, adr_node, TypeInt::INT, T_INT, adr_type, MemNode::unordered);
Node* incr = _gvn.transform(new (C) AddINode(cnt, _gvn.intcon(DataLayout::counter_increment)));
store_to_memory(NULL, adr_node, incr, T_INT, adr_type, MemNode::unordered);
}
//--------------------------test_for_osr_md_counter_at-------------------------
void Parse::test_for_osr_md_counter_at(ciMethodData* md, ciProfileData* data, ByteSize counter_offset, int limit) {
Node* adr_node = method_data_addressing(md, data, counter_offset);
const TypePtr* adr_type = _gvn.type(adr_node)->is_ptr();
Node* cnt = make_load(NULL, adr_node, TypeInt::INT, T_INT, adr_type, MemNode::unordered);
test_counter_against_threshold(cnt, limit);
}
//-------------------------------set_md_flag_at--------------------------------
void Parse::set_md_flag_at(ciMethodData* md, ciProfileData* data, int flag_constant) {
Node* adr_node = method_data_addressing(md, data, DataLayout::flags_offset());
const TypePtr* adr_type = _gvn.type(adr_node)->is_ptr();
Node* flags = make_load(NULL, adr_node, TypeInt::BYTE, T_BYTE, adr_type, MemNode::unordered);
Node* incr = _gvn.transform(new (C) OrINode(flags, _gvn.intcon(flag_constant)));
store_to_memory(NULL, adr_node, incr, T_BYTE, adr_type, MemNode::unordered);
}
//----------------------------profile_taken_branch-----------------------------
void Parse::profile_taken_branch(int target_bci, bool force_update) {
// This is a potential osr_site if we have a backedge.
int cur_bci = bci();
bool osr_site =
(target_bci <= cur_bci) && count_invocations() && UseOnStackReplacement;
// If we are going to OSR, restart at the target bytecode.
set_bci(target_bci);
// To do: factor out the the limit calculations below. These duplicate
// the similar limit calculations in the interpreter.
if (method_data_update() || force_update) {
ciMethodData* md = method()->method_data();
assert(md != NULL, "expected valid ciMethodData");
ciProfileData* data = md->bci_to_data(cur_bci);
assert(data->is_JumpData(), "need JumpData for taken branch");
increment_md_counter_at(md, data, JumpData::taken_offset());
}
// In the new tiered system this is all we need to do. In the old
// (c2 based) tiered sytem we must do the code below.
#ifndef TIERED
if (method_data_update()) {
ciMethodData* md = method()->method_data();
if (osr_site) {
ciProfileData* data = md->bci_to_data(cur_bci);
int limit = (CompileThreshold
* (OnStackReplacePercentage - InterpreterProfilePercentage)) / 100;
test_for_osr_md_counter_at(md, data, JumpData::taken_offset(), limit);
}
} else {
// With method data update off, use the invocation counter to trigger an
// OSR compilation, as done in the interpreter.
if (osr_site) {
int limit = (CompileThreshold * OnStackReplacePercentage) / 100;
increment_and_test_invocation_counter(limit);
}
}
#endif // TIERED
// Restore the original bytecode.
set_bci(cur_bci);
}
//--------------------------profile_not_taken_branch---------------------------
void Parse::profile_not_taken_branch(bool force_update) {
if (method_data_update() || force_update) {
ciMethodData* md = method()->method_data();
assert(md != NULL, "expected valid ciMethodData");
ciProfileData* data = md->bci_to_data(bci());
assert(data->is_BranchData(), "need BranchData for not taken branch");
increment_md_counter_at(md, data, BranchData::not_taken_offset());
}
}
//---------------------------------profile_call--------------------------------
void Parse::profile_call(Node* receiver) {
if (!method_data_update()) return;
switch (bc()) {
case Bytecodes::_invokevirtual:
case Bytecodes::_invokeinterface:
profile_receiver_type(receiver);
break;
case Bytecodes::_invokestatic:
case Bytecodes::_invokedynamic:
case Bytecodes::_invokespecial:
profile_generic_call();
break;
default: fatal("unexpected call bytecode");
}
}
//------------------------------profile_generic_call---------------------------
void Parse::profile_generic_call() {
assert(method_data_update(), "must be generating profile code");
ciMethodData* md = method()->method_data();
assert(md != NULL, "expected valid ciMethodData");
ciProfileData* data = md->bci_to_data(bci());
assert(data->is_CounterData(), "need CounterData for not taken branch");
increment_md_counter_at(md, data, CounterData::count_offset());
}
//-----------------------------profile_receiver_type---------------------------
void Parse::profile_receiver_type(Node* receiver) {
assert(method_data_update(), "must be generating profile code");
ciMethodData* md = method()->method_data();
assert(md != NULL, "expected valid ciMethodData");
ciProfileData* data = md->bci_to_data(bci());
assert(data->is_ReceiverTypeData(), "need ReceiverTypeData here");
// Skip if we aren't tracking receivers
if (TypeProfileWidth < 1) {
increment_md_counter_at(md, data, CounterData::count_offset());
return;
}
ciReceiverTypeData* rdata = (ciReceiverTypeData*)data->as_ReceiverTypeData();
Node* method_data = method_data_addressing(md, rdata, in_ByteSize(0));
// Using an adr_type of TypePtr::BOTTOM to work around anti-dep problems.
// A better solution might be to use TypeRawPtr::BOTTOM with RC_NARROW_MEM.
make_runtime_call(RC_LEAF, OptoRuntime::profile_receiver_type_Type(),
CAST_FROM_FN_PTR(address,
OptoRuntime::profile_receiver_type_C),
"profile_receiver_type_C",
TypePtr::BOTTOM,
method_data, receiver);
}
//---------------------------------profile_ret---------------------------------
void Parse::profile_ret(int target_bci) {
if (!method_data_update()) return;
// Skip if we aren't tracking ret targets
if (TypeProfileWidth < 1) return;
ciMethodData* md = method()->method_data();
assert(md != NULL, "expected valid ciMethodData");
ciProfileData* data = md->bci_to_data(bci());
assert(data->is_RetData(), "need RetData for ret");
ciRetData* ret_data = (ciRetData*)data->as_RetData();
// Look for the target_bci is already in the table
uint row;
bool table_full = true;
for (row = 0; row < ret_data->row_limit(); row++) {
int key = ret_data->bci(row);
table_full &= (key != RetData::no_bci);
if (key == target_bci) break;
}
if (row >= ret_data->row_limit()) {
// The target_bci was not found in the table.
if (!table_full) {
// XXX: Make slow call to update RetData
}
return;
}
// the target_bci is already in the table
increment_md_counter_at(md, data, RetData::bci_count_offset(row));
}
//--------------------------profile_null_checkcast----------------------------
void Parse::profile_null_checkcast() {
// Set the null-seen flag, done in conjunction with the usual null check. We
// never unset the flag, so this is a one-way switch.
if (!method_data_update()) return;
ciMethodData* md = method()->method_data();
assert(md != NULL, "expected valid ciMethodData");
ciProfileData* data = md->bci_to_data(bci());
assert(data->is_BitData(), "need BitData for checkcast");
set_md_flag_at(md, data, BitData::null_seen_byte_constant());
}
//-----------------------------profile_switch_case-----------------------------
void Parse::profile_switch_case(int table_index) {
if (!method_data_update()) return;
ciMethodData* md = method()->method_data();
assert(md != NULL, "expected valid ciMethodData");
ciProfileData* data = md->bci_to_data(bci());
assert(data->is_MultiBranchData(), "need MultiBranchData for switch case");
if (table_index >= 0) {
increment_md_counter_at(md, data, MultiBranchData::case_count_offset(table_index));
} else {
increment_md_counter_at(md, data, MultiBranchData::default_count_offset());
}
}
C:\hotspot-69087d08d473\src\share\vm/opto/phase.cpp
/*
* Copyright (c) 1997, 2013, 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 "code/nmethod.hpp"
#include "compiler/compileBroker.hpp"
#include "opto/compile.hpp"
#include "opto/matcher.hpp"
#include "opto/node.hpp"
#include "opto/phase.hpp"
#ifndef PRODUCT
int Phase::_total_bytes_compiled = 0;
elapsedTimer Phase::_t_totalCompilation;
elapsedTimer Phase::_t_methodCompilation;
elapsedTimer Phase::_t_stubCompilation;
#endif
// The next timers used for LogCompilation
elapsedTimer Phase::_t_parser;
elapsedTimer Phase::_t_optimizer;
elapsedTimer Phase::_t_escapeAnalysis;
elapsedTimer Phase::_t_connectionGraph;
elapsedTimer Phase::_t_idealLoop;
elapsedTimer Phase::_t_ccp;
elapsedTimer Phase::_t_matcher;
elapsedTimer Phase::_t_registerAllocation;
elapsedTimer Phase::_t_output;
#ifndef PRODUCT
elapsedTimer Phase::_t_graphReshaping;
elapsedTimer Phase::_t_scheduler;
elapsedTimer Phase::_t_blockOrdering;
elapsedTimer Phase::_t_macroEliminate;
elapsedTimer Phase::_t_macroExpand;
elapsedTimer Phase::_t_peephole;
elapsedTimer Phase::_t_postalloc_expand;
elapsedTimer Phase::_t_codeGeneration;
elapsedTimer Phase::_t_registerMethod;
elapsedTimer Phase::_t_temporaryTimer1;
elapsedTimer Phase::_t_temporaryTimer2;
elapsedTimer Phase::_t_idealLoopVerify;
// Subtimers for _t_optimizer
elapsedTimer Phase::_t_iterGVN;
elapsedTimer Phase::_t_iterGVN2;
elapsedTimer Phase::_t_incrInline;
elapsedTimer Phase::_t_renumberLive;
// Subtimers for _t_registerAllocation
elapsedTimer Phase::_t_ctorChaitin;
elapsedTimer Phase::_t_buildIFGphysical;
elapsedTimer Phase::_t_computeLive;
elapsedTimer Phase::_t_regAllocSplit;
elapsedTimer Phase::_t_postAllocCopyRemoval;
elapsedTimer Phase::_t_mergeMultidefs;
elapsedTimer Phase::_t_fixupSpills;
// Subtimers for _t_output
elapsedTimer Phase::_t_instrSched;
elapsedTimer Phase::_t_buildOopMaps;
#endif
//------------------------------Phase------------------------------------------
Phase::Phase( PhaseNumber pnum ) : _pnum(pnum), C( pnum == Compiler ? NULL : Compile::current()) {
// Poll for requests from shutdown mechanism to quiesce compiler (4448539, 4448544).
// This is an effective place to poll, since the compiler is full of phases.
// In particular, every inlining site uses a recursively created Parse phase.
CompileBroker::maybe_block();
}
#ifndef PRODUCT
static const double minimum_reported_time = 0.0001; // seconds
static const double expected_method_compile_coverage = 0.97; // %
static const double minimum_meaningful_method_compile = 2.00; // seconds
void Phase::print_timers() {
tty->print_cr ("Accumulated compiler times:");
tty->print_cr ("---------------------------");
tty->print_cr (" Total compilation: %3.3f sec.", Phase::_t_totalCompilation.seconds());
tty->print (" method compilation : %3.3f sec", Phase::_t_methodCompilation.seconds());
tty->print ("/%d bytes",_total_bytes_compiled);
tty->print_cr (" (%3.0f bytes per sec) ", Phase::_total_bytes_compiled / Phase::_t_methodCompilation.seconds());
tty->print_cr (" stub compilation : %3.3f sec.", Phase::_t_stubCompilation.seconds());
tty->print_cr (" Phases:");
tty->print_cr (" parse : %3.3f sec", Phase::_t_parser.seconds());
tty->print_cr (" optimizer : %3.3f sec", Phase::_t_optimizer.seconds());
if( Verbose || WizardMode ) {
if (DoEscapeAnalysis) {
// EA is part of Optimizer.
tty->print_cr (" escape analysis: %3.3f sec", Phase::_t_escapeAnalysis.seconds());
tty->print_cr (" connection graph: %3.3f sec", Phase::_t_connectionGraph.seconds());
tty->print_cr (" macroEliminate : %3.3f sec", Phase::_t_macroEliminate.seconds());
}
tty->print_cr (" iterGVN : %3.3f sec", Phase::_t_iterGVN.seconds());
tty->print_cr (" incrInline : %3.3f sec", Phase::_t_incrInline.seconds());
tty->print_cr (" renumberLive : %3.3f sec", Phase::_t_renumberLive.seconds());
tty->print_cr (" idealLoop : %3.3f sec", Phase::_t_idealLoop.seconds());
tty->print_cr (" idealLoopVerify: %3.3f sec", Phase::_t_idealLoopVerify.seconds());
tty->print_cr (" ccp : %3.3f sec", Phase::_t_ccp.seconds());
tty->print_cr (" iterGVN2 : %3.3f sec", Phase::_t_iterGVN2.seconds());
tty->print_cr (" macroExpand : %3.3f sec", Phase::_t_macroExpand.seconds());
tty->print_cr (" graphReshape : %3.3f sec", Phase::_t_graphReshaping.seconds());
double optimizer_subtotal = Phase::_t_iterGVN.seconds() + Phase::_t_iterGVN2.seconds() + Phase::_t_renumberLive.seconds() +
Phase::_t_escapeAnalysis.seconds() + Phase::_t_macroEliminate.seconds() +
Phase::_t_idealLoop.seconds() + Phase::_t_ccp.seconds() +
Phase::_t_macroExpand.seconds() + Phase::_t_graphReshaping.seconds();
double percent_of_optimizer = ((optimizer_subtotal == 0.0) ? 0.0 : (optimizer_subtotal / Phase::_t_optimizer.seconds() * 100.0));
tty->print_cr (" subtotal : %3.3f sec, %3.2f %%", optimizer_subtotal, percent_of_optimizer);
}
tty->print_cr (" matcher : %3.3f sec", Phase::_t_matcher.seconds());
tty->print_cr (" scheduler : %3.3f sec", Phase::_t_scheduler.seconds());
tty->print_cr (" regalloc : %3.3f sec", Phase::_t_registerAllocation.seconds());
if( Verbose || WizardMode ) {
tty->print_cr (" ctorChaitin : %3.3f sec", Phase::_t_ctorChaitin.seconds());
tty->print_cr (" buildIFG : %3.3f sec", Phase::_t_buildIFGphysical.seconds());
tty->print_cr (" computeLive : %3.3f sec", Phase::_t_computeLive.seconds());
tty->print_cr (" regAllocSplit : %3.3f sec", Phase::_t_regAllocSplit.seconds());
tty->print_cr (" postAllocCopyRemoval: %3.3f sec", Phase::_t_postAllocCopyRemoval.seconds());
tty->print_cr (" mergeMultidefs: %3.3f sec", Phase::_t_mergeMultidefs.seconds());
tty->print_cr (" fixupSpills : %3.3f sec", Phase::_t_fixupSpills.seconds());
double regalloc_subtotal = Phase::_t_ctorChaitin.seconds() +
Phase::_t_buildIFGphysical.seconds() + Phase::_t_computeLive.seconds() +
Phase::_t_regAllocSplit.seconds() + Phase::_t_fixupSpills.seconds() +
Phase::_t_postAllocCopyRemoval.seconds() + Phase::_t_mergeMultidefs.seconds();
double percent_of_regalloc = ((regalloc_subtotal == 0.0) ? 0.0 : (regalloc_subtotal / Phase::_t_registerAllocation.seconds() * 100.0));
tty->print_cr (" subtotal : %3.3f sec, %3.2f %%", regalloc_subtotal, percent_of_regalloc);
}
tty->print_cr (" blockOrdering : %3.3f sec", Phase::_t_blockOrdering.seconds());
tty->print_cr (" peephole : %3.3f sec", Phase::_t_peephole.seconds());
if (Matcher::require_postalloc_expand) {
tty->print_cr (" postalloc_expand: %3.3f sec", Phase::_t_postalloc_expand.seconds());
}
tty->print_cr (" codeGen : %3.3f sec", Phase::_t_codeGeneration.seconds());
tty->print_cr (" install_code : %3.3f sec", Phase::_t_registerMethod.seconds());
tty->print_cr (" -------------- : ----------");
double phase_subtotal = Phase::_t_parser.seconds() +
Phase::_t_optimizer.seconds() + Phase::_t_graphReshaping.seconds() +
Phase::_t_matcher.seconds() + Phase::_t_scheduler.seconds() +
Phase::_t_registerAllocation.seconds() + Phase::_t_blockOrdering.seconds() +
Phase::_t_codeGeneration.seconds() + Phase::_t_registerMethod.seconds();
double percent_of_method_compile = ((phase_subtotal == 0.0) ? 0.0 : phase_subtotal / Phase::_t_methodCompilation.seconds()) * 100.0;
// counters inside Compile::CodeGen include time for adapters and stubs
// so phase-total can be greater than 100%
tty->print_cr (" total : %3.3f sec, %3.2f %%", phase_subtotal, percent_of_method_compile);
assert( percent_of_method_compile > expected_method_compile_coverage ||
phase_subtotal < minimum_meaningful_method_compile,
"Must account for method compilation");
if( Phase::_t_temporaryTimer1.seconds() > minimum_reported_time ) {
tty->cr();
tty->print_cr (" temporaryTimer1: %3.3f sec", Phase::_t_temporaryTimer1.seconds());
}
if( Phase::_t_temporaryTimer2.seconds() > minimum_reported_time ) {
tty->cr();
tty->print_cr (" temporaryTimer2: %3.3f sec", Phase::_t_temporaryTimer2.seconds());
}
tty->print_cr (" output : %3.3f sec", Phase::_t_output.seconds());
tty->print_cr (" isched : %3.3f sec", Phase::_t_instrSched.seconds());
tty->print_cr (" bldOopMaps : %3.3f sec", Phase::_t_buildOopMaps.seconds());
}
#endif
C:\hotspot-69087d08d473\src\share\vm/opto/phase.hpp
/*
* Copyright (c) 1997, 2013, 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.
*
*/
#ifndef SHARE_VM_OPTO_PHASE_HPP
#define SHARE_VM_OPTO_PHASE_HPP
#include "libadt/port.hpp"
#include "runtime/timer.hpp"
class Compile;
//------------------------------Phase------------------------------------------
// Most optimizations are done in Phases. Creating a phase does any long
// running analysis required, and caches the analysis in internal data
// structures. Later the analysis is queried using transform() calls to
// guide transforming the program. When the Phase is deleted, so is any
// cached analysis info. This basic Phase class mostly contains timing and
// memory management code.
class Phase : public StackObj {
public:
enum PhaseNumber {
Compiler, // Top-level compiler phase
Parser, // Parse bytecodes
Remove_Useless, // Remove useless nodes
Remove_Useless_And_Renumber_Live, // First, remove useless nodes from the graph. Then, renumber live nodes.
Optimistic, // Optimistic analysis phase
GVN, // Pessimistic global value numbering phase
Ins_Select, // Instruction selection phase
CFG, // Build a CFG
BlockLayout, // Linear ordering of blocks
Register_Allocation, // Register allocation, duh
LIVE, // Dragon-book LIVE range problem
StringOpts, // StringBuilder related optimizations
Interference_Graph, // Building the IFG
Coalesce, // Coalescing copies
Ideal_Loop, // Find idealized trip-counted loops
Macro_Expand, // Expand macro nodes
Peephole, // Apply peephole optimizations
last_phase
};
protected:
enum PhaseNumber _pnum; // Phase number (for stat gathering)
#ifndef PRODUCT
static int _total_bytes_compiled;
// accumulated timers
static elapsedTimer _t_totalCompilation;
static elapsedTimer _t_methodCompilation;
static elapsedTimer _t_stubCompilation;
#endif
// The next timers used for LogCompilation
static elapsedTimer _t_parser;
static elapsedTimer _t_optimizer;
public:
// ConnectionGraph can't be Phase since it is used after EA done.
static elapsedTimer _t_escapeAnalysis;
static elapsedTimer _t_connectionGraph;
protected:
static elapsedTimer _t_idealLoop;
static elapsedTimer _t_ccp;
static elapsedTimer _t_matcher;
static elapsedTimer _t_registerAllocation;
static elapsedTimer _t_output;
#ifndef PRODUCT
static elapsedTimer _t_graphReshaping;
static elapsedTimer _t_scheduler;
static elapsedTimer _t_blockOrdering;
static elapsedTimer _t_macroEliminate;
static elapsedTimer _t_macroExpand;
static elapsedTimer _t_peephole;
static elapsedTimer _t_postalloc_expand;
static elapsedTimer _t_codeGeneration;
static elapsedTimer _t_registerMethod;
static elapsedTimer _t_temporaryTimer1;
static elapsedTimer _t_temporaryTimer2;
static elapsedTimer _t_idealLoopVerify;
// Subtimers for _t_optimizer
static elapsedTimer _t_iterGVN;
static elapsedTimer _t_iterGVN2;
static elapsedTimer _t_incrInline;
static elapsedTimer _t_renumberLive;
// Subtimers for _t_registerAllocation
static elapsedTimer _t_ctorChaitin;
static elapsedTimer _t_buildIFGphysical;
static elapsedTimer _t_computeLive;
static elapsedTimer _t_regAllocSplit;
static elapsedTimer _t_postAllocCopyRemoval;
static elapsedTimer _t_mergeMultidefs;
static elapsedTimer _t_fixupSpills;
// Subtimers for _t_output
static elapsedTimer _t_instrSched;
static elapsedTimer _t_buildOopMaps;
#endif
public:
Compile * C;
Phase( PhaseNumber pnum );
#ifndef PRODUCT
static void print_timers();
#endif
};
#endif // SHARE_VM_OPTO_PHASE_HPP
C:\hotspot-69087d08d473\src\share\vm/opto/phasetype.hpp
/*
* Copyright (c) 2012, 2013, 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.
*
*/
#ifndef SHARE_VM_OPTO_PHASETYPE_HPP
#define SHARE_VM_OPTO_PHASETYPE_HPP
enum CompilerPhaseType {
PHASE_BEFORE_STRINGOPTS,
PHASE_AFTER_STRINGOPTS,
PHASE_BEFORE_REMOVEUSELESS,
PHASE_AFTER_PARSING,
PHASE_ITER_GVN1,
PHASE_PHASEIDEAL_BEFORE_EA,
PHASE_ITER_GVN_AFTER_EA,
PHASE_ITER_GVN_AFTER_ELIMINATION,
PHASE_PHASEIDEALLOOP1,
PHASE_PHASEIDEALLOOP2,
PHASE_PHASEIDEALLOOP3,
PHASE_CPP1,
PHASE_ITER_GVN2,
PHASE_PHASEIDEALLOOP_ITERATIONS,
PHASE_OPTIMIZE_FINISHED,
PHASE_GLOBAL_CODE_MOTION,
PHASE_FINAL_CODE,
PHASE_AFTER_EA,
PHASE_BEFORE_CLOOPS,
PHASE_AFTER_CLOOPS,
PHASE_BEFORE_BEAUTIFY_LOOPS,
PHASE_AFTER_BEAUTIFY_LOOPS,
PHASE_BEFORE_MATCHING,
PHASE_INCREMENTAL_INLINE,
PHASE_INCREMENTAL_BOXING_INLINE,
PHASE_END,
PHASE_FAILURE,
PHASE_NUM_TYPES
};
class CompilerPhaseTypeHelper {
public:
static const char* to_string(CompilerPhaseType cpt) {
switch (cpt) {
case PHASE_BEFORE_STRINGOPTS: return "Before StringOpts";
case PHASE_AFTER_STRINGOPTS: return "After StringOpts";
case PHASE_BEFORE_REMOVEUSELESS: return "Before RemoveUseless";
case PHASE_AFTER_PARSING: return "After Parsing";
case PHASE_ITER_GVN1: return "Iter GVN 1";
case PHASE_PHASEIDEAL_BEFORE_EA: return "PhaseIdealLoop before EA";
case PHASE_ITER_GVN_AFTER_EA: return "Iter GVN after EA";
case PHASE_ITER_GVN_AFTER_ELIMINATION: return "Iter GVN after eliminating allocations and locks";
case PHASE_PHASEIDEALLOOP1: return "PhaseIdealLoop 1";
case PHASE_PHASEIDEALLOOP2: return "PhaseIdealLoop 2";
case PHASE_PHASEIDEALLOOP3: return "PhaseIdealLoop 3";
case PHASE_CPP1: return "PhaseCPP 1";
case PHASE_ITER_GVN2: return "Iter GVN 2";
case PHASE_PHASEIDEALLOOP_ITERATIONS: return "PhaseIdealLoop iterations";
case PHASE_OPTIMIZE_FINISHED: return "Optimize finished";
case PHASE_GLOBAL_CODE_MOTION: return "Global code motion";
case PHASE_FINAL_CODE: return "Final Code";
case PHASE_AFTER_EA: return "After Escape Analysis";
case PHASE_BEFORE_CLOOPS: return "Before CountedLoop";
case PHASE_AFTER_CLOOPS: return "After CountedLoop";
case PHASE_BEFORE_BEAUTIFY_LOOPS: return "Before beautify loops";
case PHASE_AFTER_BEAUTIFY_LOOPS: return "After beautify loops";
case PHASE_BEFORE_MATCHING: return "Before Matching";
case PHASE_INCREMENTAL_INLINE: return "Incremental Inline";
case PHASE_INCREMENTAL_BOXING_INLINE: return "Incremental Boxing Inline";
case PHASE_END: return "End";
case PHASE_FAILURE: return "Failure";
default:
ShouldNotReachHere();
return NULL;
}
}
};
#endif //SHARE_VM_OPTO_PHASETYPE_HPP
C:\hotspot-69087d08d473\src\share\vm/opto/phaseX.cpp
/*
* Copyright (c) 1997, 2017, 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 "memory/allocation.inline.hpp"
#include "opto/block.hpp"
#include "opto/callnode.hpp"
#include "opto/cfgnode.hpp"
#include "opto/connode.hpp"
#include "opto/idealGraphPrinter.hpp"
#include "opto/loopnode.hpp"
#include "opto/machnode.hpp"
#include "opto/opcodes.hpp"
#include "opto/phaseX.hpp"
#include "opto/regalloc.hpp"
#include "opto/rootnode.hpp"
//=============================================================================
#define NODE_HASH_MINIMUM_SIZE 255
//------------------------------NodeHash---------------------------------------
NodeHash::NodeHash(uint est_max_size) :
_max( round_up(est_max_size < NODE_HASH_MINIMUM_SIZE ? NODE_HASH_MINIMUM_SIZE : est_max_size) ),
_a(Thread::current()->resource_area()),
_table( NEW_ARENA_ARRAY( _a , Node* , _max ) ), // (Node**)_a->Amalloc(_max * sizeof(Node*)) ),
_inserts(0), _insert_limit( insert_limit() ),
_look_probes(0), _lookup_hits(0), _lookup_misses(0),
_total_insert_probes(0), _total_inserts(0),
_insert_probes(0), _grows(0) {
// _sentinel must be in the current node space
_sentinel = new (Compile::current()) ProjNode(NULL, TypeFunc::Control);
memset(_table,0,sizeof(Node*)*_max);
}
//------------------------------NodeHash---------------------------------------
NodeHash::NodeHash(Arena *arena, uint est_max_size) :
_max( round_up(est_max_size < NODE_HASH_MINIMUM_SIZE ? NODE_HASH_MINIMUM_SIZE : est_max_size) ),
_a(arena),
_table( NEW_ARENA_ARRAY( _a , Node* , _max ) ),
_inserts(0), _insert_limit( insert_limit() ),
_look_probes(0), _lookup_hits(0), _lookup_misses(0),
_delete_probes(0), _delete_hits(0), _delete_misses(0),
_total_insert_probes(0), _total_inserts(0),
_insert_probes(0), _grows(0) {
// _sentinel must be in the current node space
_sentinel = new (Compile::current()) ProjNode(NULL, TypeFunc::Control);
memset(_table,0,sizeof(Node*)*_max);
}
//------------------------------NodeHash---------------------------------------
NodeHash::NodeHash(NodeHash *nh) {
debug_only(_table = (Node**)badAddress); // interact correctly w/ operator=
// just copy in all the fields
*this = *nh;
// nh->_sentinel must be in the current node space
}
void NodeHash::replace_with(NodeHash *nh) {
debug_only(_table = (Node**)badAddress); // interact correctly w/ operator=
// just copy in all the fields
*this = *nh;
// nh->_sentinel must be in the current node space
}
//------------------------------hash_find--------------------------------------
// Find in hash table
Node *NodeHash::hash_find( const Node *n ) {
// ((Node*)n)->set_hash( n->hash() );
uint hash = n->hash();
if (hash == Node::NO_HASH) {
debug_only( _lookup_misses++ );
return NULL;
}
uint key = hash & (_max-1);
uint stride = key | 0x01;
debug_only( _look_probes++ );
Node *k = _table[key]; // Get hashed value
if( !k ) { // ?Miss?
debug_only( _lookup_misses++ );
return NULL; // Miss!
}
int op = n->Opcode();
uint req = n->req();
while( 1 ) { // While probing hash table
if( k->req() == req && // Same count of inputs
k->Opcode() == op ) { // Same Opcode
for( uint i=0; i<req; i++ )
if( n->in(i)!=k->in(i)) // Different inputs?
goto collision; // "goto" is a speed hack...
if( n->cmp(*k) ) { // Check for any special bits
debug_only( _lookup_hits++ );
return k; // Hit!
}
}
collision:
debug_only( _look_probes++ );
key = (key + stride/*7*/) & (_max-1); // Stride through table with relative prime
k = _table[key]; // Get hashed value
if( !k ) { // ?Miss?
debug_only( _lookup_misses++ );
return NULL; // Miss!
}
}
ShouldNotReachHere();
return NULL;
}
//------------------------------hash_find_insert-------------------------------
// Find in hash table, insert if not already present
// Used to preserve unique entries in hash table
Node *NodeHash::hash_find_insert( Node *n ) {
// n->set_hash( );
uint hash = n->hash();
if (hash == Node::NO_HASH) {
debug_only( _lookup_misses++ );
return NULL;
}
uint key = hash & (_max-1);
uint stride = key | 0x01; // stride must be relatively prime to table siz
uint first_sentinel = 0; // replace a sentinel if seen.
debug_only( _look_probes++ );
Node *k = _table[key]; // Get hashed value
if( !k ) { // ?Miss?
debug_only( _lookup_misses++ );
_table[key] = n; // Insert into table!
debug_only(n->enter_hash_lock()); // Lock down the node while in the table.
check_grow(); // Grow table if insert hit limit
return NULL; // Miss!
}
else if( k == _sentinel ) {
first_sentinel = key; // Can insert here
}
int op = n->Opcode();
uint req = n->req();
while( 1 ) { // While probing hash table
if( k->req() == req && // Same count of inputs
k->Opcode() == op ) { // Same Opcode
for( uint i=0; i<req; i++ )
if( n->in(i)!=k->in(i)) // Different inputs?
goto collision; // "goto" is a speed hack...
if( n->cmp(*k) ) { // Check for any special bits
debug_only( _lookup_hits++ );
return k; // Hit!
}
}
collision:
debug_only( _look_probes++ );
key = (key + stride) & (_max-1); // Stride through table w/ relative prime
k = _table[key]; // Get hashed value
if( !k ) { // ?Miss?
debug_only( _lookup_misses++ );
key = (first_sentinel == 0) ? key : first_sentinel; // ?saw sentinel?
_table[key] = n; // Insert into table!
debug_only(n->enter_hash_lock()); // Lock down the node while in the table.
check_grow(); // Grow table if insert hit limit
return NULL; // Miss!
}
else if( first_sentinel == 0 && k == _sentinel ) {
first_sentinel = key; // Can insert here
}
}
ShouldNotReachHere();
return NULL;
}
//------------------------------hash_insert------------------------------------
// Insert into hash table
void NodeHash::hash_insert( Node *n ) {
// // "conflict" comments -- print nodes that conflict
// bool conflict = false;
// n->set_hash();
uint hash = n->hash();
if (hash == Node::NO_HASH) {
return;
}
check_grow();
uint key = hash & (_max-1);
uint stride = key | 0x01;
while( 1 ) { // While probing hash table
debug_only( _insert_probes++ );
Node *k = _table[key]; // Get hashed value
if( !k || (k == _sentinel) ) break; // Found a slot
assert( k != n, "already inserted" );
// if( PrintCompilation && PrintOptoStatistics && Verbose ) { tty->print(" conflict: "); k->dump(); conflict = true; }
key = (key + stride) & (_max-1); // Stride through table w/ relative prime
}
_table[key] = n; // Insert into table!
debug_only(n->enter_hash_lock()); // Lock down the node while in the table.
// if( conflict ) { n->dump(); }
}
//------------------------------hash_delete------------------------------------
// Replace in hash table with sentinel
bool NodeHash::hash_delete( const Node *n ) {
Node *k;
uint hash = n->hash();
if (hash == Node::NO_HASH) {
debug_only( _delete_misses++ );
return false;
}
uint key = hash & (_max-1);
uint stride = key | 0x01;
debug_only( uint counter = 0; );
for( ; /* (k != NULL) && (k != _sentinel) */; ) {
debug_only( counter++ );
debug_only( _delete_probes++ );
k = _table[key]; // Get hashed value
if( !k ) { // Miss?
debug_only( _delete_misses++ );
#ifdef ASSERT
if( VerifyOpto ) {
for( uint i=0; i < _max; i++ )
assert( _table[i] != n, "changed edges with rehashing" );
}
#endif
return false; // Miss! Not in chain
}
else if( n == k ) {
debug_only( _delete_hits++ );
_table[key] = _sentinel; // Hit! Label as deleted entry
debug_only(((Node*)n)->exit_hash_lock()); // Unlock the node upon removal from table.
return true;
}
else {
// collision: move through table with prime offset
key = (key + stride/*7*/) & (_max-1);
assert( counter <= _insert_limit, "Cycle in hash-table");
}
}
ShouldNotReachHere();
return false;
}
//------------------------------round_up---------------------------------------
// Round up to nearest power of 2
uint NodeHash::round_up( uint x ) {
x += (x>>2); // Add 25% slop
if( x <16 ) return 16; // Small stuff
uint i=16;
while( i < x ) i <<= 1; // Double to fit
return i; // Return hash table size
}
//------------------------------grow-------------------------------------------
// Grow _table to next power of 2 and insert old entries
void NodeHash::grow() {
// Record old state
uint old_max = _max;
Node **old_table = _table;
// Construct new table with twice the space
_grows++;
_total_inserts += _inserts;
_total_insert_probes += _insert_probes;
_inserts = 0;
_insert_probes = 0;
_max = _max << 1;
_table = NEW_ARENA_ARRAY( _a , Node* , _max ); // (Node**)_a->Amalloc( _max * sizeof(Node*) );
memset(_table,0,sizeof(Node*)*_max);
_insert_limit = insert_limit();
// Insert old entries into the new table
for( uint i = 0; i < old_max; i++ ) {
Node *m = *old_table++;
if( !m || m == _sentinel ) continue;
debug_only(m->exit_hash_lock()); // Unlock the node upon removal from old table.
hash_insert(m);
}
}
//------------------------------clear------------------------------------------
// Clear all entries in _table to NULL but keep storage
void NodeHash::clear() {
#ifdef ASSERT
// Unlock all nodes upon removal from table.
for (uint i = 0; i < _max; i++) {
Node* n = _table[i];
if (!n || n == _sentinel) continue;
n->exit_hash_lock();
}
#endif
memset( _table, 0, _max * sizeof(Node*) );
}
//-----------------------remove_useless_nodes----------------------------------
// Remove useless nodes from value table,
// implementation does not depend on hash function
void NodeHash::remove_useless_nodes(VectorSet &useful) {
// Dead nodes in the hash table inherited from GVN should not replace
// existing nodes, remove dead nodes.
uint max = size();
Node *sentinel_node = sentinel();
for( uint i = 0; i < max; ++i ) {
Node *n = at(i);
if(n != NULL && n != sentinel_node && !useful.test(n->_idx)) {
debug_only(n->exit_hash_lock()); // Unlock the node when removed
_table[i] = sentinel_node; // Replace with placeholder
}
}
}
void NodeHash::check_no_speculative_types() {
#ifdef ASSERT
uint max = size();
Node *sentinel_node = sentinel();
for (uint i = 0; i < max; ++i) {
Node *n = at(i);
if(n != NULL && n != sentinel_node && n->is_Type()) {
TypeNode* tn = n->as_Type();
const Type* t = tn->type();
const Type* t_no_spec = t->remove_speculative();
assert(t == t_no_spec, "dead node in hash table or missed node during speculative cleanup");
}
}
#endif
}
#ifndef PRODUCT
//------------------------------dump-------------------------------------------
// Dump statistics for the hash table
void NodeHash::dump() {
_total_inserts += _inserts;
_total_insert_probes += _insert_probes;
if (PrintCompilation && PrintOptoStatistics && Verbose && (_inserts > 0)) {
if (WizardMode) {
for (uint i=0; i<_max; i++) {
if (_table[i])
tty->print("%d/%d/%d ",i,_table[i]->hash()&(_max-1),_table[i]->_idx);
}
}
tty->print("\nGVN Hash stats: %d grows to %d max_size\n", _grows, _max);
tty->print(" %d/%d (%8.1f%% full)\n", _inserts, _max, (double)_inserts/_max*100.0);
tty->print(" %dp/(%dh+%dm) (%8.2f probes/lookup)\n", _look_probes, _lookup_hits, _lookup_misses, (double)_look_probes/(_lookup_hits+_lookup_misses));
tty->print(" %dp/%di (%8.2f probes/insert)\n", _total_insert_probes, _total_inserts, (double)_total_insert_probes/_total_inserts);
// sentinels increase lookup cost, but not insert cost
assert((_lookup_misses+_lookup_hits)*4+100 >= _look_probes, "bad hash function");
assert( _inserts+(_inserts>>3) < _max, "table too full" );
assert( _inserts*3+100 >= _insert_probes, "bad hash function" );
}
}
Node *NodeHash::find_index(uint idx) { // For debugging
// Find an entry by its index value
for( uint i = 0; i < _max; i++ ) {
Node *m = _table[i];
if( !m || m == _sentinel ) continue;
if( m->_idx == (uint)idx ) return m;
}
return NULL;
}
#endif
#ifdef ASSERT
NodeHash::~NodeHash() {
// Unlock all nodes upon destruction of table.
if (_table != (Node**)badAddress) clear();
}
void NodeHash::operator=(const NodeHash& nh) {
// Unlock all nodes upon replacement of table.
if (&nh == this) return;
if (_table != (Node**)badAddress) clear();
memcpy(this, &nh, sizeof(*this));
// Do not increment hash_lock counts again.
// Instead, be sure we never again use the source table.
((NodeHash*)&nh)->_table = (Node**)badAddress;
}
#endif
//=============================================================================
//------------------------------PhaseRemoveUseless-----------------------------
// 1) Use a breadthfirst walk to collect useful nodes reachable from root.
PhaseRemoveUseless::PhaseRemoveUseless(PhaseGVN *gvn, Unique_Node_List *worklist, PhaseNumber phase_num) : Phase(phase_num),
_useful(Thread::current()->resource_area()) {
// Implementation requires 'UseLoopSafepoints == true' and an edge from root
// to each SafePointNode at a backward branch. Inserted in add_safepoint().
if( !UseLoopSafepoints || !OptoRemoveUseless ) return;
// Identify nodes that are reachable from below, useful.
C->identify_useful_nodes(_useful);
// Update dead node list
C->update_dead_node_list(_useful);
// Remove all useless nodes from PhaseValues' recorded types
// Must be done before disconnecting nodes to preserve hash-table-invariant
gvn->remove_useless_nodes(_useful.member_set());
// Remove all useless nodes from future worklist
worklist->remove_useless_nodes(_useful.member_set());
// Disconnect 'useless' nodes that are adjacent to useful nodes
C->remove_useless_nodes(_useful);
}
//=============================================================================
//------------------------------PhaseRenumberLive------------------------------
// First, remove useless nodes (equivalent to identifying live nodes).
// Then, renumber live nodes.
//
// The set of live nodes is returned by PhaseRemoveUseless in the _useful structure.
// If the number of live nodes is 'x' (where 'x' == _useful.size()), then the
// PhaseRenumberLive updates the node ID of each node (the _idx field) with a unique
// value in the range [0, x).
//
// At the end of the PhaseRenumberLive phase, the compiler's count of unique nodes is
// updated to 'x' and the list of dead nodes is reset (as there are no dead nodes).
//
// The PhaseRenumberLive phase updates two data structures with the new node IDs.
// (1) The worklist is used by the PhaseIterGVN phase to identify nodes that must be
// processed. A new worklist (with the updated node IDs) is returned in 'new_worklist'.
// (2) Type information (the field PhaseGVN::_types) maps type information to each
// node ID. The mapping is updated to use the new node IDs as well. Updated type
// information is returned in PhaseGVN::_types.
//
// The PhaseRenumberLive phase does not preserve the order of elements in the worklist.
//
// Other data structures used by the compiler are not updated. The hash table for value
// numbering (the field PhaseGVN::_table) is not updated because computing the hash
// values is not based on node IDs. The field PhaseGVN::_nodes is not updated either
// because it is empty wherever PhaseRenumberLive is used.
PhaseRenumberLive::PhaseRenumberLive(PhaseGVN* gvn,
Unique_Node_List* worklist, Unique_Node_List* new_worklist,
PhaseNumber phase_num) :
PhaseRemoveUseless(gvn, worklist, Remove_Useless_And_Renumber_Live) {
assert(RenumberLiveNodes, "RenumberLiveNodes must be set to true for node renumbering to take place");
assert(C->live_nodes() == _useful.size(), "the number of live nodes must match the number of useful nodes");
assert(gvn->nodes_size() == 0, "GVN must not contain any nodes at this point");
uint old_unique_count = C->unique();
uint live_node_count = C->live_nodes();
uint worklist_size = worklist->size();
// Storage for the updated type information.
Type_Array new_type_array(C->comp_arena());
// Iterate over the set of live nodes.
uint current_idx = 0; // The current new node ID. Incremented after every assignment.
for (uint i = 0; i < _useful.size(); i++) {
Node* n = _useful.at(i);
// Sanity check that fails if we ever decide to execute this phase after EA
assert(!n->is_Phi() || n->as_Phi()->inst_mem_id() == -1, "should not be linked to data Phi");
const Type* type = gvn->type_or_null(n);
new_type_array.map(current_idx, type);
bool in_worklist = false;
if (worklist->member(n)) {
in_worklist = true;
}
n->set_idx(current_idx); // Update node ID.
if (in_worklist) {
new_worklist->push(n);
}
current_idx++;
}
assert(worklist_size == new_worklist->size(), "the new worklist must have the same size as the original worklist");
assert(live_node_count == current_idx, "all live nodes must be processed");
// Replace the compiler's type information with the updated type information.
gvn->replace_types(new_type_array);
// Update the unique node count of the compilation to the number of currently live nodes.
C->set_unique(live_node_count);
// Set the dead node count to 0 and reset dead node list.
C->reset_dead_node_list();
}
//=============================================================================
//------------------------------PhaseTransform---------------------------------
PhaseTransform::PhaseTransform( PhaseNumber pnum ) : Phase(pnum),
_arena(Thread::current()->resource_area()),
_nodes(_arena),
_types(_arena)
{
init_con_caches();
#ifndef PRODUCT
clear_progress();
clear_transforms();
set_allow_progress(true);
#endif
// Force allocation for currently existing nodes
_types.map(C->unique(), NULL);
}
//------------------------------PhaseTransform---------------------------------
PhaseTransform::PhaseTransform( Arena *arena, PhaseNumber pnum ) : Phase(pnum),
_arena(arena),
_nodes(arena),
_types(arena)
{
init_con_caches();
#ifndef PRODUCT
clear_progress();
clear_transforms();
set_allow_progress(true);
#endif
// Force allocation for currently existing nodes
_types.map(C->unique(), NULL);
}
//------------------------------PhaseTransform---------------------------------
// Initialize with previously generated type information
PhaseTransform::PhaseTransform( PhaseTransform *pt, PhaseNumber pnum ) : Phase(pnum),
_arena(pt->_arena),
_nodes(pt->_nodes),
_types(pt->_types)
{
init_con_caches();
#ifndef PRODUCT
clear_progress();
clear_transforms();
set_allow_progress(true);
#endif
}
void PhaseTransform::init_con_caches() {
memset(_icons,0,sizeof(_icons));
memset(_lcons,0,sizeof(_lcons));
memset(_zcons,0,sizeof(_zcons));
}
//--------------------------------find_int_type--------------------------------
const TypeInt* PhaseTransform::find_int_type(Node* n) {
if (n == NULL) return NULL;
// Call type_or_null(n) to determine node's type since we might be in
// parse phase and call n->Value() may return wrong type.
// (For example, a phi node at the beginning of loop parsing is not ready.)
const Type* t = type_or_null(n);
if (t == NULL) return NULL;
return t->isa_int();
}
//-------------------------------find_long_type--------------------------------
const TypeLong* PhaseTransform::find_long_type(Node* n) {
if (n == NULL) return NULL;
// (See comment above on type_or_null.)
const Type* t = type_or_null(n);
if (t == NULL) return NULL;
return t->isa_long();
}
#ifndef PRODUCT
void PhaseTransform::dump_old2new_map() const {
_nodes.dump();
}
void PhaseTransform::dump_new( uint nidx ) const {
for( uint i=0; i<_nodes.Size(); i++ )
if( _nodes[i] && _nodes[i]->_idx == nidx ) {
_nodes[i]->dump();
tty->cr();
tty->print_cr("Old index= %d",i);
return;
}
tty->print_cr("Node %d not found in the new indices", nidx);
}
//------------------------------dump_types-------------------------------------
void PhaseTransform::dump_types( ) const {
_types.dump();
}
//------------------------------dump_nodes_and_types---------------------------
void PhaseTransform::dump_nodes_and_types(const Node *root, uint depth, bool only_ctrl) {
VectorSet visited(Thread::current()->resource_area());
dump_nodes_and_types_recur( root, depth, only_ctrl, visited );
}
//------------------------------dump_nodes_and_types_recur---------------------
void PhaseTransform::dump_nodes_and_types_recur( const Node *n, uint depth, bool only_ctrl, VectorSet &visited) {
if( !n ) return;
if( depth == 0 ) return;
if( visited.test_set(n->_idx) ) return;
for( uint i=0; i<n->len(); i++ ) {
if( only_ctrl && !(n->is_Region()) && i != TypeFunc::Control ) continue;
dump_nodes_and_types_recur( n->in(i), depth-1, only_ctrl, visited );
}
n->dump();
if (type_or_null(n) != NULL) {
tty->print(" "); type(n)->dump(); tty->cr();
}
}
#endif
//=============================================================================
//------------------------------PhaseValues------------------------------------
// Set minimum table size to "255"
PhaseValues::PhaseValues( Arena *arena, uint est_max_size ) : PhaseTransform(arena, GVN), _table(arena, est_max_size) {
NOT_PRODUCT( clear_new_values(); )
}
//------------------------------PhaseValues------------------------------------
// Set minimum table size to "255"
PhaseValues::PhaseValues( PhaseValues *ptv ) : PhaseTransform( ptv, GVN ),
_table(&ptv->_table) {
NOT_PRODUCT( clear_new_values(); )
}
//------------------------------PhaseValues------------------------------------
// Used by +VerifyOpto. Clear out hash table but copy _types array.
PhaseValues::PhaseValues( PhaseValues *ptv, const char *dummy ) : PhaseTransform( ptv, GVN ),
_table(ptv->arena(),ptv->_table.size()) {
NOT_PRODUCT( clear_new_values(); )
}
//------------------------------~PhaseValues-----------------------------------
#ifndef PRODUCT
PhaseValues::~PhaseValues() {
_table.dump();
// Statistics for value progress and efficiency
if( PrintCompilation && Verbose && WizardMode ) {
tty->print("\n%sValues: %d nodes ---> %d/%d (%d)",
is_IterGVN() ? "Iter" : " ", C->unique(), made_progress(), made_transforms(), made_new_values());
if( made_transforms() != 0 ) {
tty->print_cr(" ratio %f", made_progress()/(float)made_transforms() );
} else {
tty->cr();
}
}
}
#endif
//------------------------------makecon----------------------------------------
ConNode* PhaseTransform::makecon(const Type *t) {
assert(t->singleton(), "must be a constant");
assert(!t->empty() || t == Type::TOP, "must not be vacuous range");
switch (t->base()) { // fast paths
case Type::Half:
case Type::Top: return (ConNode*) C->top();
case Type::Int: return intcon( t->is_int()->get_con() );
case Type::Long: return longcon( t->is_long()->get_con() );
}
if (t->is_zero_type())
return zerocon(t->basic_type());
return uncached_makecon(t);
}
//--------------------------uncached_makecon-----------------------------------
// Make an idealized constant - one of ConINode, ConPNode, etc.
ConNode* PhaseValues::uncached_makecon(const Type *t) {
assert(t->singleton(), "must be a constant");
ConNode* x = ConNode::make(C, t);
ConNode* k = (ConNode*)hash_find_insert(x); // Value numbering
if (k == NULL) {
set_type(x, t); // Missed, provide type mapping
GrowableArray<Node_Notes*>* nna = C->node_note_array();
if (nna != NULL) {
Node_Notes* loc = C->locate_node_notes(nna, x->_idx, true);
loc->clear(); // do not put debug info on constants
}
} else {
x->destruct(); // Hit, destroy duplicate constant
x = k; // use existing constant
}
return x;
}
//------------------------------intcon-----------------------------------------
// Fast integer constant. Same as "transform(new ConINode(TypeInt::make(i)))"
ConINode* PhaseTransform::intcon(int i) {
// Small integer? Check cache! Check that cached node is not dead
if (i >= _icon_min && i <= _icon_max) {
ConINode* icon = _icons[i-_icon_min];
if (icon != NULL && icon->in(TypeFunc::Control) != NULL)
return icon;
}
ConINode* icon = (ConINode*) uncached_makecon(TypeInt::make(i));
assert(icon->is_Con(), "");
if (i >= _icon_min && i <= _icon_max)
_icons[i-_icon_min] = icon; // Cache small integers
return icon;
}
//------------------------------longcon----------------------------------------
// Fast long constant.
ConLNode* PhaseTransform::longcon(jlong l) {
// Small integer? Check cache! Check that cached node is not dead
if (l >= _lcon_min && l <= _lcon_max) {
ConLNode* lcon = _lcons[l-_lcon_min];
if (lcon != NULL && lcon->in(TypeFunc::Control) != NULL)
return lcon;
}
ConLNode* lcon = (ConLNode*) uncached_makecon(TypeLong::make(l));
assert(lcon->is_Con(), "");
if (l >= _lcon_min && l <= _lcon_max)
_lcons[l-_lcon_min] = lcon; // Cache small integers
return lcon;
}
//------------------------------zerocon-----------------------------------------
// Fast zero or null constant. Same as "transform(ConNode::make(Type::get_zero_type(bt)))"
ConNode* PhaseTransform::zerocon(BasicType bt) {
assert((uint)bt <= _zcon_max, "domain check");
ConNode* zcon = _zcons[bt];
if (zcon != NULL && zcon->in(TypeFunc::Control) != NULL)
return zcon;
zcon = (ConNode*) uncached_makecon(Type::get_zero_type(bt));
_zcons[bt] = zcon;
return zcon;
}
//=============================================================================
//------------------------------transform--------------------------------------
// Return a node which computes the same function as this node, but in a
// faster or cheaper fashion.
Node *PhaseGVN::transform( Node *n ) {
return transform_no_reclaim(n);
}
//------------------------------transform--------------------------------------
// Return a node which computes the same function as this node, but
// in a faster or cheaper fashion.
Node *PhaseGVN::transform_no_reclaim( Node *n ) {
NOT_PRODUCT( set_transforms(); )
// Apply the Ideal call in a loop until it no longer applies
Node *k = n;
NOT_PRODUCT( uint loop_count = 0; )
while( 1 ) {
Node *i = k->Ideal(this, /*can_reshape=*/false);
if( !i ) break;
assert( i->_idx >= k->_idx, "Idealize should return new nodes, use Identity to return old nodes" );
k = i;
assert(loop_count++ < K, "infinite loop in PhaseGVN::transform");
}
NOT_PRODUCT( if( loop_count != 0 ) { set_progress(); } )
// If brand new node, make space in type array.
ensure_type_or_null(k);
// Since I just called 'Value' to compute the set of run-time values
// for this Node, and 'Value' is non-local (and therefore expensive) I'll
// cache Value. Later requests for the local phase->type of this Node can
// use the cached Value instead of suffering with 'bottom_type'.
const Type *t = k->Value(this); // Get runtime Value set
assert(t != NULL, "value sanity");
if (type_or_null(k) != t) {
#ifndef PRODUCT
// Do not count initial visit to node as a transformation
if (type_or_null(k) == NULL) {
inc_new_values();
set_progress();
}
#endif
set_type(k, t);
// If k is a TypeNode, capture any more-precise type permanently into Node
k->raise_bottom_type(t);
}
if( t->singleton() && !k->is_Con() ) {
NOT_PRODUCT( set_progress(); )
return makecon(t); // Turn into a constant
}
// Now check for Identities
Node *i = k->Identity(this); // Look for a nearby replacement
if( i != k ) { // Found? Return replacement!
NOT_PRODUCT( set_progress(); )
return i;
}
// Global Value Numbering
i = hash_find_insert(k); // Insert if new
if( i && (i != k) ) {
// Return the pre-existing node
NOT_PRODUCT( set_progress(); )
return i;
}
// Return Idealized original
return k;
}
#ifdef ASSERT
//------------------------------dead_loop_check--------------------------------
// Check for a simple dead loop when a data node references itself directly
// or through an other data node excluding cons and phis.
void PhaseGVN::dead_loop_check( Node *n ) {
// Phi may reference itself in a loop
if (n != NULL && !n->is_dead_loop_safe() && !n->is_CFG()) {
// Do 2 levels check and only data inputs.
bool no_dead_loop = true;
uint cnt = n->req();
for (uint i = 1; i < cnt && no_dead_loop; i++) {
Node *in = n->in(i);
if (in == n) {
no_dead_loop = false;
} else if (in != NULL && !in->is_dead_loop_safe()) {
uint icnt = in->req();
for (uint j = 1; j < icnt && no_dead_loop; j++) {
if (in->in(j) == n || in->in(j) == in)
no_dead_loop = false;
}
}
}
if (!no_dead_loop) n->dump(3);
assert(no_dead_loop, "dead loop detected");
}
}
#endif
//=============================================================================
//------------------------------PhaseIterGVN-----------------------------------
// Initialize hash table to fresh and clean for +VerifyOpto
PhaseIterGVN::PhaseIterGVN( PhaseIterGVN *igvn, const char *dummy ) : PhaseGVN(igvn,dummy), _worklist( ),
_stack(C->live_nodes() >> 1),
_delay_transform(false) {
}
//------------------------------PhaseIterGVN-----------------------------------
// Initialize with previous PhaseIterGVN info; used by PhaseCCP
PhaseIterGVN::PhaseIterGVN( PhaseIterGVN *igvn ) : PhaseGVN(igvn),
_worklist( igvn->_worklist ),
_stack( igvn->_stack ),
_delay_transform(igvn->_delay_transform)
{
}
//------------------------------PhaseIterGVN-----------------------------------
// Initialize with previous PhaseGVN info from Parser
PhaseIterGVN::PhaseIterGVN( PhaseGVN *gvn ) : PhaseGVN(gvn),
_worklist(*C->for_igvn()),
// TODO: Before incremental inlining it was allocated only once and it was fine. Now that
// the constructor is used in incremental inlining, this consumes too much memory:
// _stack(C->live_nodes() >> 1),
// So, as a band-aid, we replace this by:
_stack(C->comp_arena(), 32),
_delay_transform(false)
{
uint max;
// Dead nodes in the hash table inherited from GVN were not treated as
// roots during def-use info creation; hence they represent an invisible
// use. Clear them out.
max = _table.size();
for( uint i = 0; i < max; ++i ) {
Node *n = _table.at(i);
if(n != NULL && n != _table.sentinel() && n->outcnt() == 0) {
if( n->is_top() ) continue;
assert( false, "Parse::remove_useless_nodes missed this node");
hash_delete(n);
}
}
// Any Phis or Regions on the worklist probably had uses that could not
// make more progress because the uses were made while the Phis and Regions
// were in half-built states. Put all uses of Phis and Regions on worklist.
max = _worklist.size();
for( uint j = 0; j < max; j++ ) {
Node *n = _worklist.at(j);
uint uop = n->Opcode();
if( uop == Op_Phi || uop == Op_Region ||
n->is_Type() ||
n->is_Mem() )
add_users_to_worklist(n);
}
}
#ifndef PRODUCT
void PhaseIterGVN::verify_step(Node* n) {
_verify_window[_verify_counter % _verify_window_size] = n;
++_verify_counter;
ResourceMark rm;
ResourceArea *area = Thread::current()->resource_area();
VectorSet old_space(area), new_space(area);
if (C->unique() < 1000 ||
0 == _verify_counter % (C->unique() < 10000 ? 10 : 100)) {
++_verify_full_passes;
Node::verify_recur(C->root(), -1, old_space, new_space);
}
const int verify_depth = 4;
for ( int i = 0; i < _verify_window_size; i++ ) {
Node* n = _verify_window[i];
if ( n == NULL ) continue;
if( n->in(0) == NodeSentinel ) { // xform_idom
_verify_window[i] = n->in(1);
--i; continue;
}
// Typical fanout is 1-2, so this call visits about 6 nodes.
Node::verify_recur(n, verify_depth, old_space, new_space);
}
}
#endif
//------------------------------init_worklist----------------------------------
// Initialize worklist for each node.
void PhaseIterGVN::init_worklist( Node *n ) {
if( _worklist.member(n) ) return;
_worklist.push(n);
uint cnt = n->req();
for( uint i =0 ; i < cnt; i++ ) {
Node *m = n->in(i);
if( m ) init_worklist(m);
}
}
//------------------------------optimize---------------------------------------
void PhaseIterGVN::optimize() {
debug_only(uint num_processed = 0;);
#ifndef PRODUCT
{
_verify_counter = 0;
_verify_full_passes = 0;
for ( int i = 0; i < _verify_window_size; i++ ) {
_verify_window[i] = NULL;
}
}
#endif
#ifdef ASSERT
Node* prev = NULL;
uint rep_cnt = 0;
#endif
uint loop_count = 0;
// Pull from worklist; transform node;
// If node has changed: update edge info and put uses on worklist.
while( _worklist.size() ) {
if (C->check_node_count(NodeLimitFudgeFactor * 2,
"out of nodes optimizing method")) {
return;
}
Node *n = _worklist.pop();
if (++loop_count >= K * C->live_nodes()) {
debug_only(n->dump(4);)
assert(false, "infinite loop in PhaseIterGVN::optimize");
C->record_method_not_compilable("infinite loop in PhaseIterGVN::optimize");
return;
}
#ifdef ASSERT
if (n == prev) {
if (++rep_cnt > 3) {
n->dump(4);
assert(false, "loop in Ideal transformation");
}
} else {
rep_cnt = 0;
}
prev = n;
#endif
if (TraceIterativeGVN && Verbose) {
tty->print(" Pop ");
NOT_PRODUCT( n->dump(); )
debug_only(if( (num_processed++ % 100) == 0 ) _worklist.print_set();)
}
if (n->outcnt() != 0) {
#ifndef PRODUCT
uint wlsize = _worklist.size();
const Type* oldtype = type_or_null(n);
#endif //PRODUCT
Node *nn = transform_old(n);
#ifndef PRODUCT
if (TraceIterativeGVN) {
const Type* newtype = type_or_null(n);
if (nn != n) {
// print old node
tty->print("< ");
if (oldtype != newtype && oldtype != NULL) {
oldtype->dump();
}
do { tty->print("\t"); } while (tty->position() < 16);
tty->print("<");
n->dump();
}
if (oldtype != newtype || nn != n) {
// print new node and/or new type
if (oldtype == NULL) {
tty->print("* ");
} else if (nn != n) {
tty->print("> ");
} else {
tty->print("= ");
}
if (newtype == NULL) {
tty->print("null");
} else {
newtype->dump();
}
do { tty->print("\t"); } while (tty->position() < 16);
nn->dump();
}
if (Verbose && wlsize < _worklist.size()) {
tty->print(" Push {");
while (wlsize != _worklist.size()) {
Node* pushed = _worklist.at(wlsize++);
tty->print(" %d", pushed->_idx);
}
tty->print_cr(" }");
}
}
if( VerifyIterativeGVN && nn != n ) {
verify_step((Node*) NULL); // ignore n, it might be subsumed
}
#endif
} else if (!n->is_top()) {
remove_dead_node(n);
}
}
#ifndef PRODUCT
C->verify_graph_edges();
if( VerifyOpto && allow_progress() ) {
// Must turn off allow_progress to enable assert and break recursion
C->root()->verify();
{ // Check if any progress was missed using IterGVN
// Def-Use info enables transformations not attempted in wash-pass
// e.g. Region/Phi cleanup, ...
// Null-check elision -- may not have reached fixpoint
// do not propagate to dominated nodes
ResourceMark rm;
PhaseIterGVN igvn2(this,"Verify"); // Fresh and clean!
// Fill worklist completely
igvn2.init_worklist(C->root());
igvn2.set_allow_progress(false);
igvn2.optimize();
igvn2.set_allow_progress(true);
}
}
if ( VerifyIterativeGVN && PrintOpto ) {
if ( _verify_counter == _verify_full_passes )
tty->print_cr("VerifyIterativeGVN: %d transforms and verify passes",
(int) _verify_full_passes);
else
tty->print_cr("VerifyIterativeGVN: %d transforms, %d full verify passes",
(int) _verify_counter, (int) _verify_full_passes);
}
#endif
}
//------------------register_new_node_with_optimizer---------------------------
// Register a new node with the optimizer. Update the types array, the def-use
// info. Put on worklist.
Node* PhaseIterGVN::register_new_node_with_optimizer(Node* n, Node* orig) {
set_type_bottom(n);
_worklist.push(n);
if (orig != NULL) C->copy_node_notes_to(n, orig);
return n;
}
//------------------------------transform--------------------------------------
// Non-recursive: idealize Node 'n' with respect to its inputs and its value
Node *PhaseIterGVN::transform( Node *n ) {
if (_delay_transform) {
// Register the node but don't optimize for now
register_new_node_with_optimizer(n);
return n;
}
// If brand new node, make space in type array, and give it a type.
ensure_type_or_null(n);
if (type_or_null(n) == NULL) {
set_type_bottom(n);
}
return transform_old(n);
}
//------------------------------transform_old----------------------------------
Node *PhaseIterGVN::transform_old( Node *n ) {
#ifndef PRODUCT
debug_only(uint loop_count = 0;);
set_transforms();
#endif
// Remove 'n' from hash table in case it gets modified
_table.hash_delete(n);
if( VerifyIterativeGVN ) {
assert( !_table.find_index(n->_idx), "found duplicate entry in table");
}
// Apply the Ideal call in a loop until it no longer applies
Node *k = n;
DEBUG_ONLY(dead_loop_check(k);)
DEBUG_ONLY(bool is_new = (k->outcnt() == 0);)
Node *i = k->Ideal(this, /*can_reshape=*/true);
assert(i != k || is_new || i->outcnt() > 0, "don't return dead nodes");
#ifndef PRODUCT
if( VerifyIterativeGVN )
verify_step(k);
if( i && VerifyOpto ) {
if( !allow_progress() ) {
if (i->is_Add() && i->outcnt() == 1) {
// Switched input to left side because this is the only use
} else if( i->is_If() && (i->in(0) == NULL) ) {
// This IF is dead because it is dominated by an equivalent IF When
// dominating if changed, info is not propagated sparsely to 'this'
// Propagating this info further will spuriously identify other
// progress.
return i;
} else
set_progress();
} else
set_progress();
}
#endif
while( i ) {
#ifndef PRODUCT
debug_only( if( loop_count >= K ) i->dump(4); )
assert(loop_count < K, "infinite loop in PhaseIterGVN::transform");
debug_only( loop_count++; )
#endif
assert((i->_idx >= k->_idx) || i->is_top(), "Idealize should return new nodes, use Identity to return old nodes");
// Made a change; put users of original Node on worklist
add_users_to_worklist( k );
// Replacing root of transform tree?
if( k != i ) {
// Make users of old Node now use new.
subsume_node( k, i );
k = i;
}
DEBUG_ONLY(dead_loop_check(k);)
// Try idealizing again
DEBUG_ONLY(is_new = (k->outcnt() == 0);)
i = k->Ideal(this, /*can_reshape=*/true);
assert(i != k || is_new || i->outcnt() > 0, "don't return dead nodes");
#ifndef PRODUCT
if( VerifyIterativeGVN )
verify_step(k);
if( i && VerifyOpto ) set_progress();
#endif
}
// If brand new node, make space in type array.
ensure_type_or_null(k);
// See what kind of values 'k' takes on at runtime
const Type *t = k->Value(this);
assert(t != NULL, "value sanity");
// Since I just called 'Value' to compute the set of run-time values
// for this Node, and 'Value' is non-local (and therefore expensive) I'll
// cache Value. Later requests for the local phase->type of this Node can
// use the cached Value instead of suffering with 'bottom_type'.
if (t != type_or_null(k)) {
NOT_PRODUCT( set_progress(); )
NOT_PRODUCT( inc_new_values();)
set_type(k, t);
// If k is a TypeNode, capture any more-precise type permanently into Node
k->raise_bottom_type(t);
// Move users of node to worklist
add_users_to_worklist( k );
}
// If 'k' computes a constant, replace it with a constant
if( t->singleton() && !k->is_Con() ) {
NOT_PRODUCT( set_progress(); )
Node *con = makecon(t); // Make a constant
add_users_to_worklist( k );
subsume_node( k, con ); // Everybody using k now uses con
return con;
}
// Now check for Identities
i = k->Identity(this); // Look for a nearby replacement
if( i != k ) { // Found? Return replacement!
NOT_PRODUCT( set_progress(); )
add_users_to_worklist( k );
subsume_node( k, i ); // Everybody using k now uses i
return i;
}
// Global Value Numbering
i = hash_find_insert(k); // Check for pre-existing node
if( i && (i != k) ) {
// Return the pre-existing node if it isn't dead
NOT_PRODUCT( set_progress(); )
add_users_to_worklist( k );
subsume_node( k, i ); // Everybody using k now uses i
return i;
}
// Return Idealized original
return k;
}
//---------------------------------saturate------------------------------------
const Type* PhaseIterGVN::saturate(const Type* new_type, const Type* old_type,
const Type* limit_type) const {
return new_type->narrow(old_type);
}
//------------------------------remove_globally_dead_node----------------------
// Kill a globally dead Node. All uses are also globally dead and are
// aggressively trimmed.
void PhaseIterGVN::remove_globally_dead_node( Node *dead ) {
enum DeleteProgress {
PROCESS_INPUTS,
PROCESS_OUTPUTS
};
assert(_stack.is_empty(), "not empty");
_stack.push(dead, PROCESS_INPUTS);
while (_stack.is_nonempty()) {
dead = _stack.node();
if (dead->Opcode() == Op_SafePoint) {
dead->as_SafePoint()->disconnect_from_root(this);
}
uint progress_state = _stack.index();
assert(dead != C->root(), "killing root, eh?");
assert(!dead->is_top(), "add check for top when pushing");
NOT_PRODUCT( set_progress(); )
if (progress_state == PROCESS_INPUTS) {
// After following inputs, continue to outputs
_stack.set_index(PROCESS_OUTPUTS);
if (!dead->is_Con()) { // Don't kill cons but uses
bool recurse = false;
// Remove from hash table
_table.hash_delete( dead );
// Smash all inputs to 'dead', isolating him completely
for (uint i = 0; i < dead->req(); i++) {
Node *in = dead->in(i);
if (in != NULL && in != C->top()) { // Points to something?
int nrep = dead->replace_edge(in, NULL); // Kill edges
assert((nrep > 0), "sanity");
if (in->outcnt() == 0) { // Made input go dead?
_stack.push(in, PROCESS_INPUTS); // Recursively remove
recurse = true;
} else if (in->outcnt() == 1 &&
in->has_special_unique_user()) {
_worklist.push(in->unique_out());
} else if (in->outcnt() <= 2 && dead->is_Phi()) {
if (in->Opcode() == Op_Region) {
_worklist.push(in);
} else if (in->is_Store()) {
DUIterator_Fast imax, i = in->fast_outs(imax);
_worklist.push(in->fast_out(i));
i++;
if (in->outcnt() == 2) {
_worklist.push(in->fast_out(i));
i++;
}
assert(!(i < imax), "sanity");
}
}
if (ReduceFieldZeroing && dead->is_Load() && i == MemNode::Memory &&
in->is_Proj() && in->in(0) != NULL && in->in(0)->is_Initialize()) {
// A Load that directly follows an InitializeNode is
// going away. The Stores that follow are candidates
// again to be captured by the InitializeNode.
for (DUIterator_Fast jmax, j = in->fast_outs(jmax); j < jmax; j++) {
Node *n = in->fast_out(j);
if (n->is_Store()) {
_worklist.push(n);
}
}
}
} // if (in != NULL && in != C->top())
} // for (uint i = 0; i < dead->req(); i++)
if (recurse) {
continue;
}
} // if (!dead->is_Con())
} // if (progress_state == PROCESS_INPUTS)
// Aggressively kill globally dead uses
// (Rather than pushing all the outs at once, we push one at a time,
// plus the parent to resume later, because of the indefinite number
// of edge deletions per loop trip.)
if (dead->outcnt() > 0) {
// Recursively remove output edges
_stack.push(dead->raw_out(0), PROCESS_INPUTS);
} else {
// Finished disconnecting all input and output edges.
_stack.pop();
// Remove dead node from iterative worklist
_worklist.remove(dead);
// Constant node that has no out-edges and has only one in-edge from
// root is usually dead. However, sometimes reshaping walk makes
// it reachable by adding use edges. So, we will NOT count Con nodes
// as dead to be conservative about the dead node count at any
// given time.
if (!dead->is_Con()) {
C->record_dead_node(dead->_idx);
}
if (dead->is_macro()) {
C->remove_macro_node(dead);
}
if (dead->is_expensive()) {
C->remove_expensive_node(dead);
}
CastIINode* cast = dead->isa_CastII();
if (cast != NULL && cast->has_range_check()) {
C->remove_range_check_cast(cast);
}
}
} // while (_stack.is_nonempty())
}
//------------------------------subsume_node-----------------------------------
// Remove users from node 'old' and add them to node 'nn'.
void PhaseIterGVN::subsume_node( Node *old, Node *nn ) {
if (old->Opcode() == Op_SafePoint) {
old->as_SafePoint()->disconnect_from_root(this);
}
assert( old != hash_find(old), "should already been removed" );
assert( old != C->top(), "cannot subsume top node");
// Copy debug or profile information to the new version:
C->copy_node_notes_to(nn, old);
// Move users of node 'old' to node 'nn'
for (DUIterator_Last imin, i = old->last_outs(imin); i >= imin; ) {
Node* use = old->last_out(i); // for each use...
// use might need re-hashing (but it won't if it's a new node)
bool is_in_table = _table.hash_delete( use );
// Update use-def info as well
// We remove all occurrences of old within use->in,
// so as to avoid rehashing any node more than once.
// The hash table probe swamps any outer loop overhead.
uint num_edges = 0;
for (uint jmax = use->len(), j = 0; j < jmax; j++) {
if (use->in(j) == old) {
use->set_req(j, nn);
++num_edges;
}
}
// Insert into GVN hash table if unique
// If a duplicate, 'use' will be cleaned up when pulled off worklist
if( is_in_table ) {
hash_find_insert(use);
}
i -= num_edges; // we deleted 1 or more copies of this edge
}
// Search for instance field data PhiNodes in the same region pointing to the old
// memory PhiNode and update their instance memory ids to point to the new node.
if (old->is_Phi() && old->as_Phi()->type()->has_memory() && old->in(0) != NULL) {
Node* region = old->in(0);
for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
PhiNode* phi = region->fast_out(i)->isa_Phi();
if (phi != NULL && phi->inst_mem_id() == (int)old->_idx) {
phi->set_inst_mem_id((int)nn->_idx);
}
}
}
// Smash all inputs to 'old', isolating him completely
Node *temp = new (C) Node(1);
temp->init_req(0,nn); // Add a use to nn to prevent him from dying
remove_dead_node( old );
temp->del_req(0); // Yank bogus edge
#ifndef PRODUCT
if( VerifyIterativeGVN ) {
for ( int i = 0; i < _verify_window_size; i++ ) {
if ( _verify_window[i] == old )
_verify_window[i] = nn;
}
}
#endif
_worklist.remove(temp); // this can be necessary
temp->destruct(); // reuse the _idx of this little guy
}
//------------------------------add_users_to_worklist--------------------------
void PhaseIterGVN::add_users_to_worklist0( Node *n ) {
for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
_worklist.push(n->fast_out(i)); // Push on worklist
}
}
// Return counted loop Phi if as a counted loop exit condition, cmp
// compares the the induction variable with n
static PhiNode* countedloop_phi_from_cmp(CmpINode* cmp, Node* n) {
for (DUIterator_Fast imax, i = cmp->fast_outs(imax); i < imax; i++) {
Node* bol = cmp->fast_out(i);
for (DUIterator_Fast i2max, i2 = bol->fast_outs(i2max); i2 < i2max; i2++) {
Node* iff = bol->fast_out(i2);
if (iff->is_CountedLoopEnd()) {
CountedLoopEndNode* cle = iff->as_CountedLoopEnd();
if (cle->limit() == n) {
PhiNode* phi = cle->phi();
if (phi != NULL) {
return phi;
}
}
}
}
}
return NULL;
}
void PhaseIterGVN::add_users_to_worklist( Node *n ) {
add_users_to_worklist0(n);
// Move users of node to worklist
for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
Node* use = n->fast_out(i); // Get use
if( use->is_Multi() || // Multi-definer? Push projs on worklist
use->is_Store() ) // Enable store/load same address
add_users_to_worklist0(use);
// If we changed the receiver type to a call, we need to revisit
// the Catch following the call. It's looking for a non-NULL
// receiver to know when to enable the regular fall-through path
// in addition to the NullPtrException path.
if (use->is_CallDynamicJava() && n == use->in(TypeFunc::Parms)) {
Node* p = use->as_CallDynamicJava()->proj_out(TypeFunc::Control);
if (p != NULL) {
add_users_to_worklist0(p);
}
}
uint use_op = use->Opcode();
if(use->is_Cmp()) { // Enable CMP/BOOL optimization
add_users_to_worklist(use); // Put Bool on worklist
if (use->outcnt() > 0) {
Node* bol = use->raw_out(0);
if (bol->outcnt() > 0) {
Node* iff = bol->raw_out(0);
if (iff->outcnt() == 2) {
// Look for the 'is_x2logic' pattern: "x ? : 0 : 1" and put the
// phi merging either 0 or 1 onto the worklist
Node* ifproj0 = iff->raw_out(0);
Node* ifproj1 = iff->raw_out(1);
if (ifproj0->outcnt() > 0 && ifproj1->outcnt() > 0) {
Node* region0 = ifproj0->raw_out(0);
Node* region1 = ifproj1->raw_out(0);
if( region0 == region1 )
add_users_to_worklist0(region0);
}
}
}
}
if (use_op == Op_CmpI) {
Node* phi = countedloop_phi_from_cmp((CmpINode*)use, n);
if (phi != NULL) {
// If an opaque node feeds into the limit condition of a
// CountedLoop, we need to process the Phi node for the
// induction variable when the opaque node is removed:
// the range of values taken by the Phi is now known and
// so its type is also known.
_worklist.push(phi);
}
Node* in1 = use->in(1);
for (uint i = 0; i < in1->outcnt(); i++) {
if (in1->raw_out(i)->Opcode() == Op_CastII) {
Node* castii = in1->raw_out(i);
if (castii->in(0) != NULL && castii->in(0)->in(0) != NULL && castii->in(0)->in(0)->is_If()) {
Node* ifnode = castii->in(0)->in(0);
if (ifnode->in(1) != NULL && ifnode->in(1)->is_Bool() && ifnode->in(1)->in(1) == use) {
// Reprocess a CastII node that may depend on an
// opaque node value when the opaque node is
// removed. In case it carries a dependency we can do
// a better job of computing its type.
_worklist.push(castii);
}
}
}
}
}
}
// If changed Cast input, check Phi users for simple cycles
if( use->is_ConstraintCast() || use->is_CheckCastPP() ) {
for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
Node* u = use->fast_out(i2);
if (u->is_Phi())
_worklist.push(u);
}
}
// If changed LShift inputs, check RShift users for useless sign-ext
if( use_op == Op_LShiftI ) {
for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
Node* u = use->fast_out(i2);
if (u->Opcode() == Op_RShiftI)
_worklist.push(u);
}
}
// If changed AddI/SubI inputs, check CmpU for range check optimization.
if (use_op == Op_AddI || use_op == Op_SubI) {
for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
Node* u = use->fast_out(i2);
if (u->is_Cmp() && (u->Opcode() == Op_CmpU)) {
_worklist.push(u);
}
}
}
// If changed AddP inputs, check Stores for loop invariant
if( use_op == Op_AddP ) {
for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
Node* u = use->fast_out(i2);
if (u->is_Mem())
_worklist.push(u);
}
}
// If changed initialization activity, check dependent Stores
if (use_op == Op_Allocate || use_op == Op_AllocateArray) {
InitializeNode* init = use->as_Allocate()->initialization();
if (init != NULL) {
Node* imem = init->proj_out(TypeFunc::Memory);
if (imem != NULL) add_users_to_worklist0(imem);
}
}
if (use_op == Op_Initialize) {
Node* imem = use->as_Initialize()->proj_out(TypeFunc::Memory);
if (imem != NULL) add_users_to_worklist0(imem);
}
}
}
/**
* Remove the speculative part of all types that we know of
*/
void PhaseIterGVN::remove_speculative_types() {
assert(UseTypeSpeculation, "speculation is off");
for (uint i = 0; i < _types.Size(); i++) {
const Type* t = _types.fast_lookup(i);
if (t != NULL) {
_types.map(i, t->remove_speculative());
}
}
_table.check_no_speculative_types();
}
//=============================================================================
#ifndef PRODUCT
uint PhaseCCP::_total_invokes = 0;
uint PhaseCCP::_total_constants = 0;
#endif
//------------------------------PhaseCCP---------------------------------------
// Conditional Constant Propagation, ala Wegman & Zadeck
PhaseCCP::PhaseCCP( PhaseIterGVN *igvn ) : PhaseIterGVN(igvn) {
NOT_PRODUCT( clear_constants(); )
assert( _worklist.size() == 0, "" );
// Clear out _nodes from IterGVN. Must be clear to transform call.
_nodes.clear(); // Clear out from IterGVN
analyze();
}
#ifndef PRODUCT
//------------------------------~PhaseCCP--------------------------------------
PhaseCCP::~PhaseCCP() {
inc_invokes();
_total_constants += count_constants();
}
#endif
#ifdef ASSERT
static bool ccp_type_widens(const Type* t, const Type* t0) {
assert(t->meet(t0) == t, "Not monotonic");
switch (t->base() == t0->base() ? t->base() : Type::Top) {
case Type::Int:
assert(t0->isa_int()->_widen <= t->isa_int()->_widen, "widen increases");
break;
case Type::Long:
assert(t0->isa_long()->_widen <= t->isa_long()->_widen, "widen increases");
break;
}
return true;
}
#endif //ASSERT
//------------------------------analyze----------------------------------------
void PhaseCCP::analyze() {
// Initialize all types to TOP, optimistic analysis
for (int i = C->unique() - 1; i >= 0; i--) {
_types.map(i,Type::TOP);
}
// Push root onto worklist
Unique_Node_List worklist;
worklist.push(C->root());
// Pull from worklist; compute new value; push changes out.
// This loop is the meat of CCP.
while( worklist.size() ) {
Node *n = worklist.pop();
const Type *t = n->Value(this);
if (t != type(n)) {
assert(ccp_type_widens(t, type(n)), "ccp type must widen");
#ifndef PRODUCT
if( TracePhaseCCP ) {
t->dump();
do { tty->print("\t"); } while (tty->position() < 16);
n->dump();
}
#endif
set_type(n, t);
for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
Node* m = n->fast_out(i); // Get user
if (m->is_Region()) { // New path to Region? Must recheck Phis too
for (DUIterator_Fast i2max, i2 = m->fast_outs(i2max); i2 < i2max; i2++) {
Node* p = m->fast_out(i2); // Propagate changes to uses
if (p->bottom_type() != type(p)) { // If not already bottomed out
worklist.push(p); // Propagate change to user
}
}
}
// If we changed the receiver type to a call, we need to revisit
// the Catch following the call. It's looking for a non-NULL
// receiver to know when to enable the regular fall-through path
// in addition to the NullPtrException path
if (m->is_Call()) {
for (DUIterator_Fast i2max, i2 = m->fast_outs(i2max); i2 < i2max; i2++) {
Node* p = m->fast_out(i2); // Propagate changes to uses
if (p->is_Proj() && p->as_Proj()->_con == TypeFunc::Control) {
Node* catch_node = p->find_out_with(Op_Catch);
if (catch_node != NULL) {
worklist.push(catch_node);
}
}
}
}
if (m->bottom_type() != type(m)) { // If not already bottomed out
worklist.push(m); // Propagate change to user
}
// CmpU nodes can get their type information from two nodes up in the
// graph (instead of from the nodes immediately above). Make sure they
// are added to the worklist if nodes they depend on are updated, since
// they could be missed and get wrong types otherwise.
uint m_op = m->Opcode();
if (m_op == Op_AddI || m_op == Op_SubI) {
for (DUIterator_Fast i2max, i2 = m->fast_outs(i2max); i2 < i2max; i2++) {
Node* p = m->fast_out(i2); // Propagate changes to uses
if (p->Opcode() == Op_CmpU) {
// Got a CmpU which might need the new type information from node n.
if(p->bottom_type() != type(p)) { // If not already bottomed out
worklist.push(p); // Propagate change to user
}
}
}
}
// If n is used in a counted loop exit condition then the type
// of the counted loop's Phi depends on the type of n. See
// PhiNode::Value().
if (m_op == Op_CmpI) {
PhiNode* phi = countedloop_phi_from_cmp((CmpINode*)m, n);
if (phi != NULL) {
worklist.push(phi);
}
}
}
}
}
}
//------------------------------do_transform-----------------------------------
// Top level driver for the recursive transformer
void PhaseCCP::do_transform() {
// Correct leaves of new-space Nodes; they point to old-space.
C->set_root( transform(C->root())->as_Root() );
assert( C->top(), "missing TOP node" );
assert( C->root(), "missing root" );
}
//------------------------------transform--------------------------------------
// Given a Node in old-space, clone him into new-space.
// Convert any of his old-space children into new-space children.
Node *PhaseCCP::transform( Node *n ) {
Node *new_node = _nodes[n->_idx]; // Check for transformed node
if( new_node != NULL )
return new_node; // Been there, done that, return old answer
new_node = transform_once(n); // Check for constant
_nodes.map( n->_idx, new_node ); // Flag as having been cloned
// Allocate stack of size _nodes.Size()/2 to avoid frequent realloc
GrowableArray <Node *> trstack(C->live_nodes() >> 1);
trstack.push(new_node); // Process children of cloned node
while ( trstack.is_nonempty() ) {
Node *clone = trstack.pop();
uint cnt = clone->req();
for( uint i = 0; i < cnt; i++ ) { // For all inputs do
Node *input = clone->in(i);
if( input != NULL ) { // Ignore NULLs
Node *new_input = _nodes[input->_idx]; // Check for cloned input node
if( new_input == NULL ) {
new_input = transform_once(input); // Check for constant
_nodes.map( input->_idx, new_input );// Flag as having been cloned
trstack.push(new_input);
}
assert( new_input == clone->in(i), "insanity check");
}
}
}
return new_node;
}
//------------------------------transform_once---------------------------------
// For PhaseCCP, transformation is IDENTITY unless Node computed a constant.
Node *PhaseCCP::transform_once( Node *n ) {
const Type *t = type(n);
// Constant? Use constant Node instead
if( t->singleton() ) {
Node *nn = n; // Default is to return the original constant
if( t == Type::TOP ) {
// cache my top node on the Compile instance
if( C->cached_top_node() == NULL || C->cached_top_node()->in(0) == NULL ) {
C->set_cached_top_node( ConNode::make(C, Type::TOP) );
set_type(C->top(), Type::TOP);
}
nn = C->top();
}
if( !n->is_Con() ) {
if( t != Type::TOP ) {
nn = makecon(t); // ConNode::make(t);
NOT_PRODUCT( inc_constants(); )
} else if( n->is_Region() ) { // Unreachable region
// Note: nn == C->top()
n->set_req(0, NULL); // Cut selfreference
// Eagerly remove dead phis to avoid phis copies creation.
for (DUIterator i = n->outs(); n->has_out(i); i++) {
Node* m = n->out(i);
if( m->is_Phi() ) {
assert(type(m) == Type::TOP, "Unreachable region should not have live phis.");
replace_node(m, nn);
--i; // deleted this phi; rescan starting with next position
}
}
}
replace_node(n,nn); // Update DefUse edges for new constant
}
return nn;
}
06-09
5640
5640
07-25
05-10
1055
1055
05-07
619
619
“相关推荐”对你有帮助么？
-
非常没帮助 -
没帮助 -
一般 -
有帮助 -
非常有帮助
提交
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
