|
/*
* Copyright (c) 1997, 2022, 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 "asm/macroAssembler.hpp"
#include "asm/macroAssembler.inline.hpp"
#include "ci/ciReplay.hpp"
#include "classfile/javaClasses.hpp"
#include "code/exceptionHandlerTable.hpp"
#include "code/nmethod.hpp"
#include "compiler/compileBroker.hpp"
#include "compiler/compileLog.hpp"
#include "compiler/disassembler.hpp"
#include "compiler/oopMap.hpp"
#include "gc/shared/barrierSet.hpp"
#include "gc/shared/c2/barrierSetC2.hpp"
#include "jfr/jfrEvents.hpp"
#include "jvm_io.h"
#include "memory/allocation.hpp"
#include "memory/resourceArea.hpp"
#include "opto/addnode.hpp"
#include "opto/block.hpp"
#include "opto/c2compiler.hpp"
#include "opto/callGenerator.hpp"
#include "opto/callnode.hpp"
#include "opto/castnode.hpp"
#include "opto/cfgnode.hpp"
#include "opto/chaitin.hpp"
#include "opto/compile.hpp"
#include "opto/connode.hpp"
#include "opto/convertnode.hpp"
#include "opto/divnode.hpp"
#include "opto/escape.hpp"
#include "opto/idealGraphPrinter.hpp"
#include "opto/loopnode.hpp"
#include "opto/machnode.hpp"
#include "opto/macro.hpp"
#include "opto/matcher.hpp"
#include "opto/mathexactnode.hpp"
#include "opto/memnode.hpp"
#include "opto/mulnode.hpp"
#include "opto/narrowptrnode.hpp"
#include "opto/node.hpp"
#include "opto/opcodes.hpp"
#include "opto/output.hpp"
#include "opto/parse.hpp"
#include "opto/phaseX.hpp"
#include "opto/rootnode.hpp"
#include "opto/runtime.hpp"
#include "opto/stringopts.hpp"
#include "opto/type.hpp"
#include "opto/vector.hpp"
#include "opto/vectornode.hpp"
#include "runtime/globals_extension.hpp"
#include "runtime/sharedRuntime.hpp"
#include "runtime/signature.hpp"
#include "runtime/stubRoutines.hpp"
#include "runtime/timer.hpp"
#include "utilities/align.hpp"
#include "utilities/copy.hpp"
#include "utilities/macros.hpp"
#include "utilities/resourceHash.hpp"
// -------------------- Compile::mach_constant_base_node -----------------------
// Constant table base node singleton.
MachConstantBaseNode* Compile::mach_constant_base_node() {
if (_mach_constant_base_node == NULL) {
_mach_constant_base_node = new MachConstantBaseNode();
_mach_constant_base_node->add_req(C->root());
}
return _mach_constant_base_node;
}
/// Support for intrinsics.
// Return the index at which m must be inserted (or already exists).
// The sort order is by the address of the ciMethod, with is_virtual as minor key.
class IntrinsicDescPair {
private:
ciMethod* _m;
bool _is_virtual;
public:
IntrinsicDescPair(ciMethod* m, bool is_virtual) : _m(m), _is_virtual(is_virtual) {}
static int compare(IntrinsicDescPair* const& key, CallGenerator* const& elt) {
ciMethod* m= elt->method();
ciMethod* key_m = key->_m;
if (key_m < m) return -1;
else if (key_m > m) return 1;
else {
bool is_virtual = elt->is_virtual();
bool key_virtual = key->_is_virtual;
if (key_virtual < is_virtual) return -1;
else if (key_virtual > is_virtual) return 1;
else return 0;
}
}
};
int Compile::intrinsic_insertion_index(ciMethod* m, bool is_virtual, bool& found) {
#ifdef ASSERT
for (int i = 1; i < _intrinsics.length(); i++) {
CallGenerator* cg1 = _intrinsics.at(i-1);
CallGenerator* cg2 = _intrinsics.at(i);
assert(cg1->method() != cg2->method()
? cg1->method() < cg2->method()
: cg1->is_virtual() < cg2->is_virtual(),
"compiler intrinsics list must stay sorted");
}
#endif
IntrinsicDescPair pair(m, is_virtual);
return _intrinsics.find_sorted<IntrinsicDescPair*, IntrinsicDescPair::compare>(&pair, found);
}
void Compile::register_intrinsic(CallGenerator* cg) {
bool found = false;
int index = intrinsic_insertion_index(cg->method(), cg->is_virtual(), found);
assert(!found, "registering twice");
_intrinsics.insert_before(index, cg);
assert(find_intrinsic(cg->method(), cg->is_virtual()) == cg, "registration worked");
}
CallGenerator* Compile::find_intrinsic(ciMethod* m, bool is_virtual) {
assert(m->is_loaded(), "don't try this on unloaded methods");
if (_intrinsics.length() > 0) {
bool found = false;
int index = intrinsic_insertion_index(m, is_virtual, found);
if (found) {
return _intrinsics.at(index);
}
}
// Lazily create intrinsics for intrinsic IDs well-known in the runtime.
if (m->intrinsic_id() != vmIntrinsics::_none &&
m->intrinsic_id() <= vmIntrinsics::LAST_COMPILER_INLINE) {
CallGenerator* cg = make_vm_intrinsic(m, is_virtual);
if (cg != NULL) {
// Save it for next time:
register_intrinsic(cg);
return cg;
} else {
gather_intrinsic_statistics(m->intrinsic_id(), is_virtual, _intrinsic_disabled);
}
}
return NULL;
}
// Compile::make_vm_intrinsic is defined in library_call.cpp.
#ifndef PRODUCT
// statistics gathering...
juint Compile::_intrinsic_hist_count[vmIntrinsics::number_of_intrinsics()] = {0};
jubyte Compile::_intrinsic_hist_flags[vmIntrinsics::number_of_intrinsics()] = {0};
inline int as_int(vmIntrinsics::ID id) {
return vmIntrinsics::as_int(id);
}
bool Compile::gather_intrinsic_statistics(vmIntrinsics::ID id, bool is_virtual, int flags) {
assert(id > vmIntrinsics::_none && id < vmIntrinsics::ID_LIMIT, "oob");
int oflags = _intrinsic_hist_flags[as_int(id)];
assert(flags != 0, "what happened?");
if (is_virtual) {
flags |= _intrinsic_virtual;
}
bool changed = (flags != oflags);
if ((flags & _intrinsic_worked) != 0) {
juint count = (_intrinsic_hist_count[as_int(id)] += 1);
if (count == 1) {
changed = true; // first time
}
// increment the overall count also:
_intrinsic_hist_count[as_int(vmIntrinsics::_none)] += 1;
}
if (changed) {
if (((oflags ^ flags) & _intrinsic_virtual) != 0) {
// Something changed about the intrinsic's virtuality.
if ((flags & _intrinsic_virtual) != 0) {
// This is the first use of this intrinsic as a virtual call.
if (oflags != 0) {
// We already saw it as a non-virtual, so note both cases.
flags |= _intrinsic_both;
}
} else if ((oflags & _intrinsic_both) == 0) {
// This is the first use of this intrinsic as a non-virtual
flags |= _intrinsic_both;
}
}
_intrinsic_hist_flags[as_int(id)] = (jubyte) (oflags | flags);
}
// update the overall flags also:
_intrinsic_hist_flags[as_int(vmIntrinsics::_none)] |= (jubyte) flags;
return changed;
}
static char* format_flags(int flags, char* buf) {
buf[0] = 0;
if ((flags & Compile::_intrinsic_worked) != 0) strcat(buf, ",worked");
if ((flags & Compile::_intrinsic_failed) != 0) strcat(buf, ",failed");
if ((flags & Compile::_intrinsic_disabled) != 0) strcat(buf, ",disabled");
if ((flags & Compile::_intrinsic_virtual) != 0) strcat(buf, ",virtual");
if ((flags & Compile::_intrinsic_both) != 0) strcat(buf, ",nonvirtual");
if (buf[0] == 0) strcat(buf, ",");
assert(buf[0] == ',', "must be");
return &buf[1];
}
void Compile::print_intrinsic_statistics() {
char flagsbuf[100];
ttyLocker ttyl;
if (xtty != NULL) xtty->head("statistics type='intrinsic'");
tty->print_cr("Compiler intrinsic usage:");
juint total = _intrinsic_hist_count[as_int(vmIntrinsics::_none)];
if (total == 0) total = 1; // avoid div0 in case of no successes
#define PRINT_STAT_LINE(name, c, f) \
tty->print_cr(" %4d (%4.1f%%) %s (%s)", (int)(c), ((c) * 100.0) / total, name, f);
for (auto id : EnumRange<vmIntrinsicID>{}) {
int flags = _intrinsic_hist_flags[as_int(id)];
juint count = _intrinsic_hist_count[as_int(id)];
if ((flags | count) != 0) {
PRINT_STAT_LINE(vmIntrinsics::name_at(id), count, format_flags(flags, flagsbuf));
}
}
PRINT_STAT_LINE("total", total, format_flags(_intrinsic_hist_flags[as_int(vmIntrinsics::_none)], flagsbuf));
if (xtty != NULL) xtty->tail("statistics");
}
void Compile::print_statistics() {
{ ttyLocker ttyl;
if (xtty != NULL) xtty->head("statistics type='opto'");
Parse::print_statistics();
PhaseStringOpts::print_statistics();
PhaseCCP::print_statistics();
PhaseRegAlloc::print_statistics();
PhaseOutput::print_statistics();
PhasePeephole::print_statistics();
PhaseIdealLoop::print_statistics();
ConnectionGraph::print_statistics();
PhaseMacroExpand::print_statistics();
if (xtty != NULL) xtty->tail("statistics");
}
if (_intrinsic_hist_flags[as_int(vmIntrinsics::_none)] != 0) {
// put this under its own <statistics> element.
print_intrinsic_statistics();
}
}
#endif //PRODUCT
void Compile::gvn_replace_by(Node* n, Node* nn) {
for (DUIterator_Last imin, i = n->last_outs(imin); i >= imin; ) {
Node* use = n->last_out(i);
bool is_in_table = initial_gvn()->hash_delete(use);
uint uses_found = 0;
for (uint j = 0; j < use->len(); j++) {
if (use->in(j) == n) {
if (j < use->req())
use->set_req(j, nn);
else
use->set_prec(j, nn);
uses_found++;
}
}
if (is_in_table) {
// reinsert into table
initial_gvn()->hash_find_insert(use);
}
record_for_igvn(use);
i -= uses_found; // we deleted 1 or more copies of this edge
}
}
// Identify all nodes that are reachable from below, useful.
// Use breadth-first pass that records state in a Unique_Node_List,
// recursive traversal is slower.
void Compile::identify_useful_nodes(Unique_Node_List &useful) {
int estimated_worklist_size = live_nodes();
useful.map( estimated_worklist_size, NULL ); // preallocate space
// Initialize worklist
if (root() != NULL) { useful.push(root()); }
// If 'top' is cached, declare it useful to preserve cached node
if( cached_top_node() ) { useful.push(cached_top_node()); }
// Push all useful nodes onto the list, breadthfirst
for( uint next = 0; next < useful.size(); ++next ) {
assert( next < unique(), "Unique useful nodes < total nodes");
Node *n = useful.at(next);
uint max = n->len();
for( uint i = 0; i < max; ++i ) {
Node *m = n->in(i);
if (not_a_node(m)) continue;
useful.push(m);
}
}
}
// Update dead_node_list with any missing dead nodes using useful
// list. Consider all non-useful nodes to be useless i.e., dead nodes.
void Compile::update_dead_node_list(Unique_Node_List &useful) {
uint max_idx = unique();
VectorSet& useful_node_set = useful.member_set();
for (uint node_idx = 0; node_idx < max_idx; node_idx++) {
// If node with index node_idx is not in useful set,
// mark it as dead in dead node list.
if (!useful_node_set.test(node_idx)) {
record_dead_node(node_idx);
}
}
}
void Compile::remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful) {
int shift = 0;
for (int i = 0; i < inlines->length(); i++) {
CallGenerator* cg = inlines->at(i);
if (useful.member(cg->call_node())) {
if (shift > 0) {
inlines->at_put(i - shift, cg);
}
} else {
shift++; // skip over the dead element
}
}
if (shift > 0) {
inlines->trunc_to(inlines->length() - shift); // remove last elements from compacted array
}
}
void Compile::remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Node* dead) {
assert(dead != NULL && dead->is_Call(), "sanity");
int found = 0;
for (int i = 0; i < inlines->length(); i++) {
if (inlines->at(i)->call_node() == dead) {
inlines->remove_at(i);
found++;
NOT_DEBUG( break; ) // elements are unique, so exit early
}
}
assert(found <= 1, "not unique");
}
void Compile::remove_useless_nodes(GrowableArray<Node*>& node_list, Unique_Node_List& useful) {
for (int i = node_list.length() - 1; i >= 0; i--) {
Node* n = node_list.at(i);
if (!useful.member(n)) {
node_list.delete_at(i); // replaces i-th with last element which is known to be useful (already processed)
}
}
}
void Compile::remove_useless_node(Node* dead) {
remove_modified_node(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()) {
record_dead_node(dead->_idx);
}
if (dead->is_macro()) {
remove_macro_node(dead);
}
if (dead->is_expensive()) {
remove_expensive_node(dead);
}
if (dead->Opcode() == Op_Opaque4) {
remove_skeleton_predicate_opaq(dead);
}
if (dead->for_post_loop_opts_igvn()) {
remove_from_post_loop_opts_igvn(dead);
}
if (dead->is_Call()) {
remove_useless_late_inlines( &_late_inlines, dead);
remove_useless_late_inlines( &_string_late_inlines, dead);
remove_useless_late_inlines( &_boxing_late_inlines, dead);
remove_useless_late_inlines(&_vector_reboxing_late_inlines, dead);
if (dead->is_CallStaticJava()) {
remove_unstable_if_trap(dead->as_CallStaticJava(), false);
}
}
BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
bs->unregister_potential_barrier_node(dead);
}
// Disconnect all useless nodes by disconnecting those at the boundary.
void Compile::disconnect_useless_nodes(Unique_Node_List &useful, Unique_Node_List* worklist) {
uint next = 0;
while (next < useful.size()) {
Node *n = useful.at(next++);
if (n->is_SafePoint()) {
// We're done with a parsing phase. Replaced nodes are not valid
// beyond that point.
n->as_SafePoint()->delete_replaced_nodes();
}
// Use raw traversal of out edges since this code removes out edges
int max = n->outcnt();
for (int j = 0; j < max; ++j) {
Node* child = n->raw_out(j);
if (!useful.member(child)) {
assert(!child->is_top() || child != top(),
"If top is cached in Compile object it is in useful list");
// Only need to remove this out-edge to the useless node
n->raw_del_out(j);
--j;
--max;
}
}
if (n->outcnt() == 1 && n->has_special_unique_user()) {
worklist->push(n->unique_out());
}
}
remove_useless_nodes(_macro_nodes, useful); // remove useless macro nodes
remove_useless_nodes(_predicate_opaqs, useful); // remove useless predicate opaque nodes
remove_useless_nodes(_skeleton_predicate_opaqs, useful);
remove_useless_nodes(_expensive_nodes, useful); // remove useless expensive nodes
remove_useless_nodes(_for_post_loop_igvn, useful); // remove useless node recorded for post loop opts IGVN pass
remove_useless_unstable_if_traps(useful); // remove useless unstable_if traps
remove_useless_coarsened_locks(useful); // remove useless coarsened locks nodes
#ifdef ASSERT
if (_modified_nodes != NULL) {
_modified_nodes->remove_useless_nodes(useful.member_set());
}
#endif
BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
bs->eliminate_useless_gc_barriers(useful, this);
// clean up the late inline lists
remove_useless_late_inlines( &_late_inlines, useful);
remove_useless_late_inlines( &_string_late_inlines, useful);
remove_useless_late_inlines( &_boxing_late_inlines, useful);
remove_useless_late_inlines(&_vector_reboxing_late_inlines, useful);
debug_only(verify_graph_edges(true/*check for no_dead_code*/);)
}
// ============================================================================
//------------------------------CompileWrapper---------------------------------
class CompileWrapper : public StackObj {
Compile *const _compile;
public:
CompileWrapper(Compile* compile);
~CompileWrapper();
};
CompileWrapper::CompileWrapper(Compile* compile) : _compile(compile) {
// the Compile* pointer is stored in the current ciEnv:
ciEnv* env = compile->env();
assert(env == ciEnv::current(), "must already be a ciEnv active");
assert(env->compiler_data() == NULL, "compile already active?");
env->set_compiler_data(compile);
assert(compile == Compile::current(), "sanity");
compile->set_type_dict(NULL);
compile->set_clone_map(new Dict(cmpkey, hashkey, _compile->comp_arena()));
compile->clone_map().set_clone_idx(0);
compile->set_type_last_size(0);
compile->set_last_tf(NULL, NULL);
compile->set_indexSet_arena(NULL);
compile->set_indexSet_free_block_list(NULL);
compile->init_type_arena();
Type::Initialize(compile);
_compile->begin_method();
_compile->clone_map().set_debug(_compile->has_method() && _compile->directive()->CloneMapDebugOption);
}
CompileWrapper::~CompileWrapper() {
// simulate crash during compilation
assert(CICrashAt < 0 || _compile->compile_id() != CICrashAt, "just as planned");
_compile->end_method();
_compile->env()->set_compiler_data(NULL);
}
//----------------------------print_compile_messages---------------------------
void Compile::print_compile_messages() {
#ifndef PRODUCT
// Check if recompiling
if (!subsume_loads() && PrintOpto) {
// Recompiling without allowing machine instructions to subsume loads
tty->print_cr("*********************************************************");
tty->print_cr("** Bailout: Recompile without subsuming loads **");
tty->print_cr("*********************************************************");
}
if ((do_escape_analysis() != DoEscapeAnalysis) && PrintOpto) {
// Recompiling without escape analysis
tty->print_cr("*********************************************************");
tty->print_cr("** Bailout: Recompile without escape analysis **");
tty->print_cr("*********************************************************");
}
if (do_iterative_escape_analysis() != DoEscapeAnalysis && PrintOpto) {
// Recompiling without iterative escape analysis
tty->print_cr("*********************************************************");
tty->print_cr("** Bailout: Recompile without iterative escape analysis**");
tty->print_cr("*********************************************************");
}
if ((eliminate_boxing() != EliminateAutoBox) && PrintOpto) {
// Recompiling without boxing elimination
tty->print_cr("*********************************************************");
tty->print_cr("** Bailout: Recompile without boxing elimination **");
tty->print_cr("*********************************************************");
}
if ((do_locks_coarsening() != EliminateLocks) && PrintOpto) {
// Recompiling without locks coarsening
tty->print_cr("*********************************************************");
tty->print_cr("** Bailout: Recompile without locks coarsening **");
tty->print_cr("*********************************************************");
}
if (env()->break_at_compile()) {
// Open the debugger when compiling this method.
tty->print("### Breaking when compiling: ");
method()->print_short_name();
tty->cr();
BREAKPOINT;
}
if( PrintOpto ) {
if (is_osr_compilation()) {
tty->print("[OSR]%3d", _compile_id);
} else {
tty->print("%3d", _compile_id);
}
}
#endif
}
#ifndef PRODUCT
void Compile::print_ideal_ir(const char* phase_name) {
ttyLocker ttyl;
// keep the following output all in one block
// This output goes directly to the tty, not the compiler log.
// To enable tools to match it up with the compilation activity,
// be sure to tag this tty output with the compile ID.
if (xtty != NULL) {
xtty->head("ideal compile_id='%d'%s compile_phase='%s'",
compile_id(),
is_osr_compilation() ? " compile_kind='osr'" : "",
phase_name);
}
if (_output == nullptr) {
tty->print_cr("AFTER: %s", phase_name);
// Print out all nodes in ascending order of index.
root()->dump_bfs(MaxNodeLimit, nullptr, "+S$");
} else {
// Dump the node blockwise if we have a scheduling
_output->print_scheduling();
}
if (xtty != NULL) {
xtty->tail("ideal");
}
}
#endif
// ============================================================================
//------------------------------Compile standard-------------------------------
debug_only( int Compile::_debug_idx = 100000; )
// Compile a method. entry_bci is -1 for normal compilations and indicates
// the continuation bci for on stack replacement.
Compile::Compile( ciEnv* ci_env, ciMethod* target, int osr_bci,
Options options, DirectiveSet* directive)
: Phase(Compiler),
_compile_id(ci_env->compile_id()),
_options(options),
_method(target),
_entry_bci(osr_bci),
_ilt(NULL),
_stub_function(NULL),
_stub_name(NULL),
_stub_entry_point(NULL),
_max_node_limit(MaxNodeLimit),
_post_loop_opts_phase(false),
_inlining_progress(false),
_inlining_incrementally(false),
_do_cleanup(false),
_has_reserved_stack_access(target->has_reserved_stack_access()),
#ifndef PRODUCT
_igv_idx(0),
_trace_opto_output(directive->TraceOptoOutputOption),
#endif
_has_method_handle_invokes(false),
_clinit_barrier_on_entry(false),
_stress_seed(0),
_comp_arena(mtCompiler),
_barrier_set_state(BarrierSet::barrier_set()->barrier_set_c2()->create_barrier_state(comp_arena())),
_env(ci_env),
_directive(directive),
_log(ci_env->log()),
_failure_reason(NULL),
_intrinsics (comp_arena(), 0, 0, NULL),
_macro_nodes (comp_arena(), 8, 0, NULL),
_predicate_opaqs (comp_arena(), 8, 0, NULL),
_skeleton_predicate_opaqs (comp_arena(), 8, 0, NULL),
_expensive_nodes (comp_arena(), 8, 0, NULL),
_for_post_loop_igvn(comp_arena(), 8, 0, NULL),
_unstable_if_traps (comp_arena(), 8, 0, NULL),
_coarsened_locks (comp_arena(), 8, 0, NULL),
_congraph(NULL),
NOT_PRODUCT(_igv_printer(NULL) COMMA)
_dead_node_list(comp_arena()),
_dead_node_count(0),
_node_arena(mtCompiler),
_old_arena(mtCompiler),
_mach_constant_base_node(NULL),
_Compile_types(mtCompiler),
_initial_gvn(NULL),
_for_igvn(NULL),
_late_inlines(comp_arena(), 2, 0, NULL),
_string_late_inlines(comp_arena(), 2, 0, NULL),
_boxing_late_inlines(comp_arena(), 2, 0, NULL),
_vector_reboxing_late_inlines(comp_arena(), 2, 0, NULL),
_late_inlines_pos(0),
_number_of_mh_late_inlines(0),
_print_inlining_stream(new (mtCompiler) stringStream()),
_print_inlining_list(NULL),
_print_inlining_idx(0),
_print_inlining_output(NULL),
_replay_inline_data(NULL),
_java_calls(0),
_inner_loops(0),
_interpreter_frame_size(0),
_output(NULL)
#ifndef PRODUCT
, _in_dump_cnt(0)
#endif
{
C = this;
CompileWrapper cw(this);
if (CITimeVerbose) {
tty->print(" ");
target->holder()->name()->print();
tty->print(".");
target->print_short_name();
tty->print(" ");
}
TraceTime t1("Total compilation time", &_t_totalCompilation, CITime, CITimeVerbose);
TraceTime t2(NULL, &_t_methodCompilation, CITime, false);
#if defined(SUPPORT_ASSEMBLY) || defined(SUPPORT_ABSTRACT_ASSEMBLY)
bool print_opto_assembly = directive->PrintOptoAssemblyOption;
// We can always print a disassembly, either abstract (hex dump) or
// with the help of a suitable hsdis library. Thus, we should not
// couple print_assembly and print_opto_assembly controls.
// But: always print opto and regular assembly on compile command 'print'.
bool print_assembly = directive->PrintAssemblyOption;
set_print_assembly(print_opto_assembly || print_assembly);
#else
set_print_assembly(false); // must initialize.
#endif
#ifndef PRODUCT
set_parsed_irreducible_loop(false);
if (directive->ReplayInlineOption) {
_replay_inline_data = ciReplay::load_inline_data(method(), entry_bci(), ci_env->comp_level());
}
#endif
set_print_inlining(directive->PrintInliningOption || PrintOptoInlining);
set_print_intrinsics(directive->PrintIntrinsicsOption);
set_has_irreducible_loop(true); // conservative until build_loop_tree() reset it
if (ProfileTraps RTM_OPT_ONLY( || UseRTMLocking )) {
// Make sure the method being compiled gets its own MDO,
// so we can at least track the decompile_count().
// Need MDO to record RTM code generation state.
method()->ensure_method_data();
}
Init(/*do_aliasing=*/ true);
print_compile_messages();
_ilt = InlineTree::build_inline_tree_root();
// Even if NO memory addresses are used, MergeMem nodes must have at least 1 slice
assert(num_alias_types() >= AliasIdxRaw, "");
#define MINIMUM_NODE_HASH 1023
// Node list that Iterative GVN will start with
Unique_Node_List for_igvn(comp_arena());
set_for_igvn(&for_igvn);
// GVN that will be run immediately on new nodes
uint estimated_size = method()->code_size()*4+64;
estimated_size = (estimated_size < MINIMUM_NODE_HASH ? MINIMUM_NODE_HASH : estimated_size);
PhaseGVN gvn(node_arena(), estimated_size);
set_initial_gvn(&gvn);
print_inlining_init();
{ // Scope for timing the parser
TracePhase tp("parse", &timers[_t_parser]);
// Put top into the hash table ASAP.
initial_gvn()->transform_no_reclaim(top());
// Set up tf(), start(), and find a CallGenerator.
CallGenerator* cg = NULL;
if (is_osr_compilation()) {
const TypeTuple *domain = StartOSRNode::osr_domain();
const TypeTuple *range = TypeTuple::make_range(method()->signature());
init_tf(TypeFunc::make(domain, range));
StartNode* s = new StartOSRNode(root(), domain);
initial_gvn()->set_type_bottom(s);
init_start(s);
cg = CallGenerator::for_osr(method(), entry_bci());
} else {
// Normal case.
init_tf(TypeFunc::make(method()));
StartNode* s = new StartNode(root(), tf()->domain());
initial_gvn()->set_type_bottom(s);
init_start(s);
if (method()->intrinsic_id() == vmIntrinsics::_Reference_get) {
// With java.lang.ref.reference.get() we must go through the
// intrinsic - even when get() is the root
// method of the compile - so that, if necessary, the value in
// the referent field of the reference object gets recorded by
// the pre-barrier code.
cg = find_intrinsic(method(), false);
}
if (cg == NULL) {
float past_uses = method()->interpreter_invocation_count();
float expected_uses = past_uses;
cg = CallGenerator::for_inline(method(), expected_uses);
}
}
if (failing()) return;
if (cg == NULL) {
record_method_not_compilable("cannot parse method");
return;
}
JVMState* jvms = build_start_state(start(), tf());
if ((jvms = cg->generate(jvms)) == NULL) {
if (!failure_reason_is(C2Compiler::retry_class_loading_during_parsing())) {
record_method_not_compilable("method parse failed");
}
return;
}
GraphKit kit(jvms);
if (!kit.stopped()) {
// Accept return values, and transfer control we know not where.
// This is done by a special, unique ReturnNode bound to root.
return_values(kit.jvms());
}
if (kit.has_exceptions()) {
// Any exceptions that escape from this call must be rethrown
// to whatever caller is dynamically above us on the stack.
// This is done by a special, unique RethrowNode bound to root.
rethrow_exceptions(kit.transfer_exceptions_into_jvms());
}
assert(IncrementalInline || (_late_inlines.length() == 0 && !has_mh_late_inlines()), "incremental inlining is off");
if (_late_inlines.length() == 0 && !has_mh_late_inlines() && !failing() && has_stringbuilder()) {
inline_string_calls(true);
}
if (failing()) return;
print_method(PHASE_BEFORE_REMOVEUSELESS, 3);
// Remove clutter produced by parsing.
if (!failing()) {
ResourceMark rm;
PhaseRemoveUseless pru(initial_gvn(), &for_igvn);
}
}
// Note: Large methods are capped off in do_one_bytecode().
if (failing()) return;
// After parsing, node notes are no longer automagic.
// They must be propagated by register_new_node_with_optimizer(),
// clone(), or the like.
set_default_node_notes(NULL);
#ifndef PRODUCT
if (should_print_igv(1)) {
_igv_printer->print_inlining();
}
#endif
if (failing()) return;
NOT_PRODUCT( verify_graph_edges(); )
// If any phase is randomized for stress testing, seed random number
// generation and log the seed for repeatability.
if (StressLCM || StressGCM || StressIGVN || StressCCP) {
if (FLAG_IS_DEFAULT(StressSeed) || (FLAG_IS_ERGO(StressSeed) && RepeatCompilation)) {
_stress_seed = static_cast<uint>(Ticks::now().nanoseconds());
FLAG_SET_ERGO(StressSeed, _stress_seed);
} else {
_stress_seed = StressSeed;
}
if (_log != NULL) {
_log->elem("stress_test seed='%u'", _stress_seed);
}
}
// Now optimize
Optimize();
if (failing()) return;
NOT_PRODUCT( verify_graph_edges(); )
#ifndef PRODUCT
if (should_print_ideal()) {
print_ideal_ir("print_ideal");
}
#endif
#ifdef ASSERT
BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
bs->verify_gc_barriers(this, BarrierSetC2::BeforeCodeGen);
#endif
// Dump compilation data to replay it.
if (directive->DumpReplayOption) {
env()->dump_replay_data(_compile_id);
}
if (directive->DumpInlineOption && (ilt() != NULL)) {
env()->dump_inline_data(_compile_id);
}
// Now that we know the size of all the monitors we can add a fixed slot
// for the original deopt pc.
int next_slot = fixed_slots() + (sizeof(address) / VMRegImpl::stack_slot_size);
set_fixed_slots(next_slot);
// Compute when to use implicit null checks. Used by matching trap based
// nodes and NullCheck optimization.
set_allowed_deopt_reasons();
// Now generate code
Code_Gen();
}
//------------------------------Compile----------------------------------------
// Compile a runtime stub
Compile::Compile( ciEnv* ci_env,
TypeFunc_generator generator,
address stub_function,
const char *stub_name,
int is_fancy_jump,
bool pass_tls,
bool return_pc,
DirectiveSet* directive)
: Phase(Compiler),
_compile_id(0),
_options(Options::for_runtime_stub()),
_method(NULL),
_entry_bci(InvocationEntryBci),
_stub_function(stub_function),
_stub_name(stub_name),
_stub_entry_point(NULL),
_max_node_limit(MaxNodeLimit),
_post_loop_opts_phase(false),
_inlining_progress(false),
_inlining_incrementally(false),
_has_reserved_stack_access(false),
#ifndef PRODUCT
_igv_idx(0),
_trace_opto_output(directive->TraceOptoOutputOption),
#endif
_has_method_handle_invokes(false),
_clinit_barrier_on_entry(false),
_stress_seed(0),
_comp_arena(mtCompiler),
_barrier_set_state(BarrierSet::barrier_set()->barrier_set_c2()->create_barrier_state(comp_arena())),
_env(ci_env),
_directive(directive),
_log(ci_env->log()),
_failure_reason(NULL),
_congraph(NULL),
NOT_PRODUCT(_igv_printer(NULL) COMMA)
_dead_node_list(comp_arena()),
_dead_node_count(0),
_node_arena(mtCompiler),
_old_arena(mtCompiler),
_mach_constant_base_node(NULL),
_Compile_types(mtCompiler),
_initial_gvn(NULL),
_for_igvn(NULL),
_number_of_mh_late_inlines(0),
_print_inlining_stream(new (mtCompiler) stringStream()),
_print_inlining_list(NULL),
_print_inlining_idx(0),
_print_inlining_output(NULL),
_replay_inline_data(NULL),
_java_calls(0),
_inner_loops(0),
_interpreter_frame_size(0),
_output(NULL),
#ifndef PRODUCT
_in_dump_cnt(0),
#endif
_allowed_reasons(0) {
C = this;
TraceTime t1(NULL, &_t_totalCompilation, CITime, false);
TraceTime t2(NULL, &_t_stubCompilation, CITime, false);
#ifndef PRODUCT
set_print_assembly(PrintFrameConverterAssembly);
set_parsed_irreducible_loop(false);
#else
set_print_assembly(false); // Must initialize.
#endif
set_has_irreducible_loop(false); // no loops
CompileWrapper cw(this);
Init(/*do_aliasing=*/ false);
init_tf((*generator)());
{
// The following is a dummy for the sake of GraphKit::gen_stub
Unique_Node_List for_igvn(comp_arena());
set_for_igvn(&for_igvn); // not used, but some GraphKit guys push on this
PhaseGVN gvn(Thread::current()->resource_area(),255);
set_initial_gvn(&gvn); // not significant, but GraphKit guys use it pervasively
gvn.transform_no_reclaim(top());
GraphKit kit;
kit.gen_stub(stub_function, stub_name, is_fancy_jump, pass_tls, return_pc);
}
NOT_PRODUCT( verify_graph_edges(); )
Code_Gen();
}
//------------------------------Init-------------------------------------------
// Prepare for a single compilation
void Compile::Init(bool aliasing) {
_do_aliasing = aliasing;
_unique = 0;
_regalloc = NULL;
_tf = NULL; // filled in later
_top = NULL; // cached later
_matcher = NULL; // filled in later
_cfg = NULL; // filled in later
IA32_ONLY( set_24_bit_selection_and_mode(true, false); )
_node_note_array = NULL;
_default_node_notes = NULL;
DEBUG_ONLY( _modified_nodes = NULL; ) // Used in Optimize()
_immutable_memory = NULL; // filled in at first inquiry
// Globally visible Nodes
// First set TOP to NULL to give safe behavior during creation of RootNode
set_cached_top_node(NULL);
set_root(new RootNode());
// Now that you have a Root to point to, create the real TOP
set_cached_top_node( new ConNode(Type::TOP) );
set_recent_alloc(NULL, NULL);
// Create Debug Information Recorder to record scopes, oopmaps, etc.
env()->set_oop_recorder(new OopRecorder(env()->arena()));
env()->set_debug_info(new DebugInformationRecorder(env()->oop_recorder()));
env()->set_dependencies(new Dependencies(env()));
_fixed_slots = 0;
set_has_split_ifs(false);
set_has_loops(false); // first approximation
set_has_stringbuilder(false);
set_has_boxed_value(false);
_trap_can_recompile = false; // no traps emitted yet
_major_progress = true; // start out assuming good things will happen
set_has_unsafe_access(false);
set_max_vector_size(0);
set_clear_upper_avx(false); //false as default for clear upper bits of ymm registers
Copy::zero_to_bytes(_trap_hist, sizeof(_trap_hist));
set_decompile_count(0);
set_do_freq_based_layout(_directive->BlockLayoutByFrequencyOption);
_loop_opts_cnt = LoopOptsCount;
set_do_inlining(Inline);
set_max_inline_size(MaxInlineSize);
set_freq_inline_size(FreqInlineSize);
set_do_scheduling(OptoScheduling);
set_do_vector_loop(false);
set_has_monitors(false);
if (AllowVectorizeOnDemand) {
if (has_method() && (_directive->VectorizeOption || _directive->VectorizeDebugOption)) {
set_do_vector_loop(true);
NOT_PRODUCT(if (do_vector_loop() && Verbose) {tty->print("Compile::Init: do vectorized loops (SIMD like) for method %s\n", method()->name()->as_quoted_ascii());})
} else if (has_method() && method()->name() != 0 &&
method()->intrinsic_id() == vmIntrinsics::_forEachRemaining) {
set_do_vector_loop(true);
}
}
set_use_cmove(UseCMoveUnconditionally /* || do_vector_loop()*/); //TODO: consider do_vector_loop() mandate use_cmove unconditionally
NOT_PRODUCT(if (use_cmove() && Verbose && has_method()) {tty->print("Compile::Init: use CMove without profitability tests for method %s\n", method()->name()->as_quoted_ascii());})
set_rtm_state(NoRTM); // No RTM lock eliding by default
_max_node_limit = _directive->MaxNodeLimitOption;
#if INCLUDE_RTM_OPT
if (UseRTMLocking && has_method() && (method()->method_data_or_null() != NULL)) {
int rtm_state = method()->method_data()->rtm_state();
if (method_has_option(CompileCommand::NoRTMLockEliding) || ((rtm_state & NoRTM) != 0)) {
// Don't generate RTM lock eliding code.
set_rtm_state(NoRTM);
} else if (method_has_option(CompileCommand::UseRTMLockEliding) || ((rtm_state & UseRTM) != 0) || !UseRTMDeopt) {
// Generate RTM lock eliding code without abort ratio calculation code.
set_rtm_state(UseRTM);
} else if (UseRTMDeopt) {
// Generate RTM lock eliding code and include abort ratio calculation
// code if UseRTMDeopt is on.
set_rtm_state(ProfileRTM);
}
}
#endif
if (VM_Version::supports_fast_class_init_checks() && has_method() && !is_osr_compilation() && method()->needs_clinit_barrier()) {
set_clinit_barrier_on_entry(true);
}
if (debug_info()->recording_non_safepoints()) {
set_node_note_array(new(comp_arena()) GrowableArray<Node_Notes*>
(comp_arena(), 8, 0, NULL));
set_default_node_notes(Node_Notes::make(this));
}
const int grow_ats = 16;
_max_alias_types = grow_ats;
_alias_types = NEW_ARENA_ARRAY(comp_arena(), AliasType*, grow_ats);
AliasType* ats = NEW_ARENA_ARRAY(comp_arena(), AliasType, grow_ats);
Copy::zero_to_bytes(ats, sizeof(AliasType)*grow_ats);
{
for (int i = 0; i < grow_ats; i++) _alias_types[i] = &ats[i];
}
// Initialize the first few types.
_alias_types[AliasIdxTop]->Init(AliasIdxTop, NULL);
_alias_types[AliasIdxBot]->Init(AliasIdxBot, TypePtr::BOTTOM);
_alias_types[AliasIdxRaw]->Init(AliasIdxRaw, TypeRawPtr::BOTTOM);
_num_alias_types = AliasIdxRaw+1;
// Zero out the alias type cache.
Copy::zero_to_bytes(_alias_cache, sizeof(_alias_cache));
// A NULL adr_type hits in the cache right away. Preload the right answer.
probe_alias_cache(NULL)->_index = AliasIdxTop;
#ifdef ASSERT
_type_verify_symmetry = true;
_phase_optimize_finished = false;
_exception_backedge = false;
#endif
}
//---------------------------init_start----------------------------------------
// Install the StartNode on this compile object.
void Compile::init_start(StartNode* s) {
if (failing())
return; // already failing
assert(s == start(), "");
}
/**
* Return the 'StartNode'. We must not have a pending failure, since the ideal graph
* can be in an inconsistent state, i.e., we can get segmentation faults when traversing
* the ideal graph.
*/
StartNode* Compile::start() const {
assert (!failing(), "Must not have pending failure. Reason is: %s", failure_reason());
for (DUIterator_Fast imax, i = root()->fast_outs(imax); i < imax; i++) {
Node* start = root()->fast_out(i);
if (start->is_Start()) {
return start->as_Start();
}
}
fatal("Did not find Start node!");
return NULL;
}
//-------------------------------immutable_memory-------------------------------------
// Access immutable memory
Node* Compile::immutable_memory() {
if (_immutable_memory != NULL) {
return _immutable_memory;
}
StartNode* s = start();
for (DUIterator_Fast imax, i = s->fast_outs(imax); true; i++) {
Node *p = s->fast_out(i);
if (p != s && p->as_Proj()->_con == TypeFunc::Memory) {
_immutable_memory = p;
return _immutable_memory;
}
}
ShouldNotReachHere();
return NULL;
}
//----------------------set_cached_top_node------------------------------------
// Install the cached top node, and make sure Node::is_top works correctly.
void Compile::set_cached_top_node(Node* tn) {
if (tn != NULL) verify_top(tn);
Node* old_top = _top;
_top = tn;
// Calling Node::setup_is_top allows the nodes the chance to adjust
// their _out arrays.
if (_top != NULL) _top->setup_is_top();
if (old_top != NULL) old_top->setup_is_top();
assert(_top == NULL || top()->is_top(), "");
}
#ifdef ASSERT
uint Compile::count_live_nodes_by_graph_walk() {
Unique_Node_List useful(comp_arena());
// Get useful node list by walking the graph.
identify_useful_nodes(useful);
return useful.size();
}
void Compile::print_missing_nodes() {
// Return if CompileLog is NULL and PrintIdealNodeCount is false.
if ((_log == NULL) && (! PrintIdealNodeCount)) {
return;
}
// This is an expensive function. It is executed only when the user
// specifies VerifyIdealNodeCount option or otherwise knows the
// additional work that needs to be done to identify reachable nodes
// by walking the flow graph and find the missing ones using
// _dead_node_list.
Unique_Node_List useful(comp_arena());
// Get useful node list by walking the graph.
identify_useful_nodes(useful);
uint l_nodes = C->live_nodes();
uint l_nodes_by_walk = useful.size();
if (l_nodes != l_nodes_by_walk) {
if (_log != NULL) {
_log->begin_head("mismatched_nodes count='%d'", abs((int) (l_nodes - l_nodes_by_walk)));
_log->stamp();
_log->end_head();
}
VectorSet& useful_member_set = useful.member_set();
int last_idx = l_nodes_by_walk;
for (int i = 0; i < last_idx; i++) {
if (useful_member_set.test(i)) {
if (_dead_node_list.test(i)) {
if (_log != NULL) {
_log->elem("mismatched_node_info node_idx='%d' type='both live and dead'", i);
}
if (PrintIdealNodeCount) {
// Print the log message to tty
tty->print_cr("mismatched_node idx='%d' both live and dead'", i);
useful.at(i)->dump();
}
}
}
else if (! _dead_node_list.test(i)) {
if (_log != NULL) {
_log->elem("mismatched_node_info node_idx='%d' type='neither live nor dead'", i);
}
if (PrintIdealNodeCount) {
// Print the log message to tty
tty->print_cr("mismatched_node idx='%d' type='neither live nor dead'", i);
}
}
}
if (_log != NULL) {
_log->tail("mismatched_nodes");
}
}
}
void Compile::record_modified_node(Node* n) {
if (_modified_nodes != NULL && !_inlining_incrementally && !n->is_Con()) {
_modified_nodes->push(n);
}
}
void Compile::remove_modified_node(Node* n) {
if (_modified_nodes != NULL) {
_modified_nodes->remove(n);
}
}
#endif
#ifndef PRODUCT
void Compile::verify_top(Node* tn) const {
if (tn != NULL) {
assert(tn->is_Con(), "top node must be a constant");
assert(((ConNode*)tn)->type() == Type::TOP, "top node must have correct type");
assert(tn->in(0) != NULL, "must have live top node");
}
}
#endif
///-------------------Managing Per-Node Debug & Profile Info-------------------
void Compile::grow_node_notes(GrowableArray<Node_Notes*>* arr, int grow_by) {
guarantee(arr != NULL, "");
int num_blocks = arr->length();
if (grow_by < num_blocks) grow_by = num_blocks;
int num_notes = grow_by * _node_notes_block_size;
Node_Notes* notes = NEW_ARENA_ARRAY(node_arena(), Node_Notes, num_notes);
Copy::zero_to_bytes(notes, num_notes * sizeof(Node_Notes));
while (num_notes > 0) {
arr->append(notes);
notes += _node_notes_block_size;
num_notes -= _node_notes_block_size;
}
assert(num_notes == 0, "exact multiple, please");
}
bool Compile::copy_node_notes_to(Node* dest, Node* source) {
if (source == NULL || dest == NULL) return false;
if (dest->is_Con())
return false; // Do not push debug info onto constants.
#ifdef ASSERT
// Leave a bread crumb trail pointing to the original node:
if (dest != NULL && dest != source && dest->debug_orig() == NULL) {
dest->set_debug_orig(source);
}
#endif
if (node_note_array() == NULL)
return false; // Not collecting any notes now.
// This is a copy onto a pre-existing node, which may already have notes.
// If both nodes have notes, do not overwrite any pre-existing notes.
Node_Notes* source_notes = node_notes_at(source->_idx);
if (source_notes == NULL || source_notes->is_clear()) return false;
Node_Notes* dest_notes = node_notes_at(dest->_idx);
if (dest_notes == NULL || dest_notes->is_clear()) {
return set_node_notes_at(dest->_idx, source_notes);
}
Node_Notes merged_notes = (*source_notes);
// The order of operations here ensures that dest notes will win...
merged_notes.update_from(dest_notes);
return set_node_notes_at(dest->_idx, &merged_notes);
}
//--------------------------allow_range_check_smearing-------------------------
// Gating condition for coalescing similar range checks.
// Sometimes we try 'speculatively' replacing a series of a range checks by a
// single covering check that is at least as strong as any of them.
// If the optimization succeeds, the simplified (strengthened) range check
// will always succeed. If it fails, we will deopt, and then give up
// on the optimization.
bool Compile::allow_range_check_smearing() const {
// If this method has already thrown a range-check,
// assume it was because we already tried range smearing
// and it failed.
uint already_trapped = trap_count(Deoptimization::Reason_range_check);
return !already_trapped;
}
//------------------------------flatten_alias_type-----------------------------
const TypePtr *Compile::flatten_alias_type( const TypePtr *tj ) const {
assert(do_aliasing(), "Aliasing should be enabled");
int offset = tj->offset();
TypePtr::PTR ptr = tj->ptr();
// Known instance (scalarizable allocation) alias only with itself.
bool is_known_inst = tj->isa_oopptr() != NULL &&
tj->is_oopptr()->is_known_instance();
// Process weird unsafe references.
if (offset == Type::OffsetBot && (tj->isa_instptr() /*|| tj->isa_klassptr()*/)) {
assert(InlineUnsafeOps || StressReflectiveCode, "indeterminate pointers come only from unsafe ops");
assert(!is_known_inst, "scalarizable allocation should not have unsafe references");
tj = TypeOopPtr::BOTTOM;
ptr = tj->ptr();
offset = tj->offset();
}
// Array pointers need some flattening
const TypeAryPtr* ta = tj->isa_aryptr();
if (ta && ta->is_stable()) {
// Erase stability property for alias analysis.
tj = ta = ta->cast_to_stable(false);
}
if( ta && is_known_inst ) {
if ( offset != Type::OffsetBot &&
offset > arrayOopDesc::length_offset_in_bytes() ) {
offset = Type::OffsetBot; // Flatten constant access into array body only
tj = ta = ta->
remove_speculative()->
cast_to_ptr_type(ptr)->
with_offset(offset);
}
} else if (ta) {
// For arrays indexed by constant indices, we flatten the alias
// space to include all of the array body. Only the header, klass
// and array length can be accessed un-aliased.
if( offset != Type::OffsetBot ) {
if( ta->const_oop() ) { // MethodData* or Method*
offset = Type::OffsetBot; // Flatten constant access into array body
tj = ta = ta->
remove_speculative()->
cast_to_ptr_type(ptr)->
cast_to_exactness(false)->
with_offset(offset);
} else if( offset == arrayOopDesc::length_offset_in_bytes() ) {
// range is OK as-is.
tj = ta = TypeAryPtr::RANGE;
} else if( offset == oopDesc::klass_offset_in_bytes() ) {
tj = TypeInstPtr::KLASS; // all klass loads look alike
ta = TypeAryPtr::RANGE; // generic ignored junk
ptr = TypePtr::BotPTR;
} else if( offset == oopDesc::mark_offset_in_bytes() ) {
tj = TypeInstPtr::MARK;
ta = TypeAryPtr::RANGE; // generic ignored junk
ptr = TypePtr::BotPTR;
} else { // Random constant offset into array body
offset = Type::OffsetBot; // Flatten constant access into array body
tj = ta = ta->
remove_speculative()->
cast_to_ptr_type(ptr)->
cast_to_exactness(false)->
with_offset(offset);
}
}
// Arrays of fixed size alias with arrays of unknown size.
if (ta->size() != TypeInt::POS) {
const TypeAry *tary = TypeAry::make(ta->elem(), TypeInt::POS);
tj = ta = ta->
remove_speculative()->
cast_to_ptr_type(ptr)->
with_ary(tary)->
cast_to_exactness(false);
}
// Arrays of known objects become arrays of unknown objects.
if (ta->elem()->isa_narrowoop() && ta->elem() != TypeNarrowOop::BOTTOM) {
const TypeAry *tary = TypeAry::make(TypeNarrowOop::BOTTOM, ta->size());
tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,NULL,false,offset);
}
if (ta->elem()->isa_oopptr() && ta->elem() != TypeInstPtr::BOTTOM) {
const TypeAry *tary = TypeAry::make(TypeInstPtr::BOTTOM, ta->size());
tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,NULL,false,offset);
}
// Arrays of bytes and of booleans both use 'bastore' and 'baload' so
// cannot be distinguished by bytecode alone.
if (ta->elem() == TypeInt::BOOL) {
const TypeAry *tary = TypeAry::make(TypeInt::BYTE, ta->size());
ciKlass* aklass = ciTypeArrayKlass::make(T_BYTE);
tj = ta = TypeAryPtr::make(ptr,ta->const_oop(),tary,aklass,false,offset);
}
// During the 2nd round of IterGVN, NotNull castings are removed.
// Make sure the Bottom and NotNull variants alias the same.
// Also, make sure exact and non-exact variants alias the same.
if (ptr == TypePtr::NotNull || ta->klass_is_exact() || ta->speculative() != NULL) {
tj = ta = ta->
remove_speculative()->
cast_to_ptr_type(TypePtr::BotPTR)->
cast_to_exactness(false)->
with_offset(offset);
}
}
// Oop pointers need some flattening
const TypeInstPtr *to = tj->isa_instptr();
if (to && to != TypeOopPtr::BOTTOM) {
ciInstanceKlass* ik = to->instance_klass();
if( ptr == TypePtr::Constant ) {
if (ik != ciEnv::current()->Class_klass() ||
offset < ik->layout_helper_size_in_bytes()) {
// No constant oop pointers (such as Strings); they alias with
// unknown strings.
assert(!is_known_inst, "not scalarizable allocation");
tj = to = to->
cast_to_instance_id(TypeOopPtr::InstanceBot)->
remove_speculative()->
cast_to_ptr_type(TypePtr::BotPTR)->
cast_to_exactness(false);
}
} else if( is_known_inst ) {
tj = to; // Keep NotNull and klass_is_exact for instance type
} else if( ptr == TypePtr::NotNull || to->klass_is_exact() ) {
// During the 2nd round of IterGVN, NotNull castings are removed.
// Make sure the Bottom and NotNull variants alias the same.
// Also, make sure exact and non-exact variants alias the same.
tj = to = to->
remove_speculative()->
cast_to_instance_id(TypeOopPtr::InstanceBot)->
cast_to_ptr_type(TypePtr::BotPTR)->
cast_to_exactness(false);
}
if (to->speculative() != NULL) {
tj = to = to->remove_speculative();
}
// Canonicalize the holder of this field
if (offset >= 0 && offset < instanceOopDesc::base_offset_in_bytes()) {
// First handle header references such as a LoadKlassNode, even if the
// object's klass is unloaded at compile time (4965979).
if (!is_known_inst) { // Do it only for non-instance types
tj = to = TypeInstPtr::make(TypePtr::BotPTR, env()->Object_klass(), false, NULL, offset);
}
} else if (offset < 0 || offset >= ik->layout_helper_size_in_bytes()) {
// Static fields are in the space above the normal instance
// fields in the java.lang.Class instance.
if (ik != ciEnv::current()->Class_klass()) {
to = NULL;
tj = TypeOopPtr::BOTTOM;
offset = tj->offset();
}
} else {
ciInstanceKlass *canonical_holder = ik->get_canonical_holder(offset);
assert(offset < canonical_holder->layout_helper_size_in_bytes(), "");
if (!ik->equals(canonical_holder) || tj->offset() != offset) {
if( is_known_inst ) {
tj = to = TypeInstPtr::make(to->ptr(), canonical_holder, true, NULL, offset, to->instance_id());
} else {
tj = to = TypeInstPtr::make(to->ptr(), canonical_holder, false, NULL, offset);
}
}
}
}
// Klass pointers to object array klasses need some flattening
const TypeKlassPtr *tk = tj->isa_klassptr();
if( tk ) {
// If we are referencing a field within a Klass, we need
// to assume the worst case of an Object. Both exact and
// inexact types must flatten to the same alias class so
// use NotNull as the PTR.
if ( offset == Type::OffsetBot || (offset >= 0 && (size_t)offset < sizeof(Klass)) ) {
tj = tk = TypeInstKlassPtr::make(TypePtr::NotNull,
env()->Object_klass(),
offset);
}
if (tk->isa_aryklassptr() && tk->is_aryklassptr()->elem()->isa_klassptr()) {
ciKlass* k = ciObjArrayKlass::make(env()->Object_klass());
if (!k || !k->is_loaded()) { // Only fails for some -Xcomp runs
tj = tk = TypeInstKlassPtr::make(TypePtr::NotNull, env()->Object_klass(), offset);
} else {
tj = tk = TypeAryKlassPtr::make(TypePtr::NotNull, tk->is_aryklassptr()->elem(), k, offset);
}
}
// Check for precise loads from the primary supertype array and force them
// to the supertype cache alias index. Check for generic array loads from
// the primary supertype array and also force them to the supertype cache
// alias index. Since the same load can reach both, we need to merge
// these 2 disparate memories into the same alias class. Since the
// primary supertype array is read-only, there's no chance of confusion
// where we bypass an array load and an array store.
int primary_supers_offset = in_bytes(Klass::primary_supers_offset());
if (offset == Type::OffsetBot ||
(offset >= primary_supers_offset &&
offset < (int)(primary_supers_offset + Klass::primary_super_limit() * wordSize)) ||
offset == (int)in_bytes(Klass::secondary_super_cache_offset())) {
offset = in_bytes(Klass::secondary_super_cache_offset());
tj = tk = tk->with_offset(offset);
}
}
// Flatten all Raw pointers together.
if (tj->base() == Type::RawPtr)
tj = TypeRawPtr::BOTTOM;
if (tj->base() == Type::AnyPtr)
tj = TypePtr::BOTTOM; // An error, which the caller must check for.
offset = tj->offset();
assert( offset != Type::OffsetTop, "Offset has fallen from constant" );
assert( (offset != Type::OffsetBot && tj->base() != Type::AryPtr) ||
(offset == Type::OffsetBot && tj->base() == Type::AryPtr) ||
(offset == Type::OffsetBot && tj == TypeOopPtr::BOTTOM) ||
(offset == Type::OffsetBot && tj == TypePtr::BOTTOM) ||
(offset == oopDesc::mark_offset_in_bytes() && tj->base() == Type::AryPtr) ||
(offset == oopDesc::klass_offset_in_bytes() && tj->base() == Type::AryPtr) ||
(offset == arrayOopDesc::length_offset_in_bytes() && tj->base() == Type::AryPtr),
"For oops, klasses, raw offset must be constant; for arrays the offset is never known" );
assert( tj->ptr() != TypePtr::TopPTR &&
tj->ptr() != TypePtr::AnyNull &&
tj->ptr() != TypePtr::Null, "No imprecise addresses" );
// assert( tj->ptr() != TypePtr::Constant ||
// tj->base() == Type::RawPtr ||
// tj->base() == Type::KlassPtr, "No constant oop addresses" );
return tj;
}
void Compile::AliasType::Init(int i, const TypePtr* at) {
assert(AliasIdxTop <= i && i < Compile::current()->_max_alias_types, "Invalid alias index");
_index = i;
_adr_type = at;
_field = NULL;
_element = NULL;
_is_rewritable = true; // default
const TypeOopPtr *atoop = (at != NULL) ? at->isa_oopptr() : NULL;
if (atoop != NULL && atoop->is_known_instance()) {
const TypeOopPtr *gt = atoop->cast_to_instance_id(TypeOopPtr::InstanceBot);
_general_index = Compile::current()->get_alias_index(gt);
} else {
_general_index = 0;
}
}
BasicType Compile::AliasType::basic_type() const {
if (element() != NULL) {
const Type* element = adr_type()->is_aryptr()->elem();
return element->isa_narrowoop() ? T_OBJECT : element->array_element_basic_type();
} if (field() != NULL) {
return field()->layout_type();
} else {
return T_ILLEGAL; // unknown
}
}
//---------------------------------print_on------------------------------------
#ifndef PRODUCT
void Compile::AliasType::print_on(outputStream* st) {
if (index() < 10)
st->print("@ <%d> ", index());
else st->print("@ <%d>", index());
st->print(is_rewritable() ? " " : " RO");
int offset = adr_type()->offset();
if (offset == Type::OffsetBot)
st->print(" +any");
else st->print(" +%-3d", offset);
st->print(" in ");
adr_type()->dump_on(st);
const TypeOopPtr* tjp = adr_type()->isa_oopptr();
if (field() != NULL && tjp) {
if (tjp->is_instptr()->instance_klass() != field()->holder() ||
tjp->offset() != field()->offset_in_bytes()) {
st->print(" != ");
field()->print();
st->print(" ***");
}
}
}
void print_alias_types() {
Compile* C = Compile::current();
tty->print_cr("--- Alias types, AliasIdxBot .. %d", C->num_alias_types()-1);
for (int idx = Compile::AliasIdxBot; idx < C->num_alias_types(); idx++) {
C->alias_type(idx)->print_on(tty);
tty->cr();
}
}
#endif
//----------------------------probe_alias_cache--------------------------------
Compile::AliasCacheEntry* Compile::probe_alias_cache(const TypePtr* adr_type) {
intptr_t key = (intptr_t) adr_type;
key ^= key >> logAliasCacheSize;
return &_alias_cache[key & right_n_bits(logAliasCacheSize)];
}
//-----------------------------grow_alias_types--------------------------------
void Compile::grow_alias_types() {
const int old_ats = _max_alias_types; // how many before?
const int new_ats = old_ats; // how many more?
const int grow_ats = old_ats+new_ats; // how many now?
_max_alias_types = grow_ats;
_alias_types = REALLOC_ARENA_ARRAY(comp_arena(), AliasType*, _alias_types, old_ats, grow_ats);
AliasType* ats = NEW_ARENA_ARRAY(comp_arena(), AliasType, new_ats);
Copy::zero_to_bytes(ats, sizeof(AliasType)*new_ats);
for (int i = 0; i < new_ats; i++) _alias_types[old_ats+i] = &ats[i];
}
//--------------------------------find_alias_type------------------------------
Compile::AliasType* Compile::find_alias_type(const TypePtr* adr_type, bool no_create, ciField* original_field) {
if (!do_aliasing()) {
return alias_type(AliasIdxBot);
}
AliasCacheEntry* ace = probe_alias_cache(adr_type);
if (ace->_adr_type == adr_type) {
return alias_type(ace->_index);
}
// Handle special cases.
if (adr_type == NULL) return alias_type(AliasIdxTop);
if (adr_type == TypePtr::BOTTOM) return alias_type(AliasIdxBot);
// Do it the slow way.
const TypePtr* flat = flatten_alias_type(adr_type);
#ifdef ASSERT
{
ResourceMark rm;
assert(flat == flatten_alias_type(flat), "not idempotent: adr_type = %s; flat = %s => %s",
Type::str(adr_type), Type::str(flat), Type::str(flatten_alias_type(flat)));
assert(flat != TypePtr::BOTTOM, "cannot alias-analyze an untyped ptr: adr_type = %s",
Type::str(adr_type));
if (flat->isa_oopptr() && !flat->isa_klassptr()) {
const TypeOopPtr* foop = flat->is_oopptr();
// Scalarizable allocations have exact klass always.
bool exact = !foop->klass_is_exact() || foop->is_known_instance();
const TypePtr* xoop = foop->cast_to_exactness(exact)->is_ptr();
assert(foop == flatten_alias_type(xoop), "exactness must not affect alias type: foop = %s; xoop = %s",
Type::str(foop), Type::str(xoop));
}
}
#endif
int idx = AliasIdxTop;
for (int i = 0; i < num_alias_types(); i++) {
if (alias_type(i)->adr_type() == flat) {
idx = i;
break;
}
}
if (idx == AliasIdxTop) {
if (no_create) return NULL;
// Grow the array if necessary.
if (_num_alias_types == _max_alias_types) grow_alias_types();
// Add a new alias type.
idx = _num_alias_types++;
_alias_types[idx]->Init(idx, flat);
if (flat == TypeInstPtr::KLASS) alias_type(idx)->set_rewritable(false);
if (flat == TypeAryPtr::RANGE) alias_type(idx)->set_rewritable(false);
if (flat->isa_instptr()) {
if (flat->offset() == java_lang_Class::klass_offset()
&& flat->is_instptr()->instance_klass() == env()->Class_klass())
alias_type(idx)->set_rewritable(false);
}
if (flat->isa_aryptr()) {
#ifdef ASSERT
const int header_size_min = arrayOopDesc::base_offset_in_bytes(T_BYTE);
// (T_BYTE has the weakest alignment and size restrictions...)
assert(flat->offset() < header_size_min, "array body reference must be OffsetBot");
#endif
if (flat->offset() == TypePtr::OffsetBot) {
alias_type(idx)->set_element(flat->is_aryptr()->elem());
}
}
if (flat->isa_klassptr()) {
if (flat->offset() == in_bytes(Klass::super_check_offset_offset()))
alias_type(idx)->set_rewritable(false);
if (flat->offset() == in_bytes(Klass::modifier_flags_offset()))
alias_type(idx)->set_rewritable(false);
if (flat->offset() == in_bytes(Klass::access_flags_offset()))
alias_type(idx)->set_rewritable(false);
if (flat->offset() == in_bytes(Klass::java_mirror_offset()))
alias_type(idx)->set_rewritable(false);
if (flat->offset() == in_bytes(Klass::secondary_super_cache_offset()))
alias_type(idx)->set_rewritable(false);
}
// %%% (We would like to finalize JavaThread::threadObj_offset(),
// but the base pointer type is not distinctive enough to identify
// references into JavaThread.)
// Check for final fields.
const TypeInstPtr* tinst = flat->isa_instptr();
if (tinst && tinst->offset() >= instanceOopDesc::base_offset_in_bytes()) {
ciField* field;
if (tinst->const_oop() != NULL &&
tinst->instance_klass() == ciEnv::current()->Class_klass() &&
tinst->offset() >= (tinst->instance_klass()->layout_helper_size_in_bytes())) {
// static field
ciInstanceKlass* k = tinst->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
field = k->get_field_by_offset(tinst->offset(), true);
} else {
ciInstanceKlass *k = tinst->instance_klass();
field = k->get_field_by_offset(tinst->offset(), false);
}
assert(field == NULL ||
original_field == NULL ||
(field->holder() == original_field->holder() &&
field->offset() == original_field->offset() &&
field->is_static() == original_field->is_static()), "wrong field?");
// Set field() and is_rewritable() attributes.
if (field != NULL) alias_type(idx)->set_field(field);
}
}
// Fill the cache for next time.
ace->_adr_type = adr_type;
ace->_index = idx;
assert(alias_type(adr_type) == alias_type(idx), "type must be installed");
// Might as well try to fill the cache for the flattened version, too.
AliasCacheEntry* face = probe_alias_cache(flat);
if (face->_adr_type == NULL) {
face->_adr_type = flat;
face->_index = idx;
assert(alias_type(flat) == alias_type(idx), "flat type must work too");
}
return alias_type(idx);
}
Compile::AliasType* Compile::alias_type(ciField* field) {
const TypeOopPtr* t;
if (field->is_static())
t = TypeInstPtr::make(field->holder()->java_mirror());
else
--> --------------------
--> maximum size reached
--> --------------------
¤ Dauer der Verarbeitung: 0.72 Sekunden
(vorverarbeitet)
¤
|
Haftungshinweis
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung ist noch experimentell.
|
