6589834: deoptimization problem with -XX:+DeoptimizeALot
Summary: Relocate the stack pointer adjustment to where uncommon_trap is actually inserted for new_array.
Reviewed-by: kvn, jrose
/*
* Copyright 1998-2008 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
* CA 95054 USA or visit www.sun.com if you need additional information or
* have any questions.
*
*/
#include "incls/_precompiled.incl"
#include "incls/_parse3.cpp.incl"
//=============================================================================
// 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;
}
}
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 = do_null_check(peek(obj_depth), T_OBJECT);
// Compile-time detect of null-exception?
if (stopped()) return;
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");
if (is_get) {
--_sp; // pop receiver before getting
do_get_xxx(tjp, obj, field, is_field);
} else {
do_put_xxx(tjp, obj, field, is_field);
--_sp; // pop receiver after putting
}
} else {
const TypeKlassPtr* tkp = TypeKlassPtr::make(field_holder);
obj = _gvn.makecon(tkp);
if (is_get) {
do_get_xxx(tkp, obj, field, is_field);
} else {
do_put_xxx(tkp, obj, field, is_field);
}
}
}
void Parse::do_get_xxx(const TypePtr* obj_type, Node* obj, ciField* field, bool is_field) {
// Does this field have a constant value? If so, just push the value.
if (field->is_constant() && push_constant(field->constant_value())) return;
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()) {
// 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);
}
// Build the load.
Node* ld = make_load(NULL, adr, type, bt, adr_type, 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());
do_null_assert(peek(), T_OBJECT);
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
insert_mem_bar(Op_MemBarAcquire, ld);
}
}
void Parse::do_put_xxx(const TypePtr* obj_type, Node* obj, ciField* field, bool is_field) {
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) 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);
// Store the value.
Node* store;
if (bt == T_OBJECT) {
const TypePtr* 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);
} else {
store = store_to_memory( control(), adr, val, bt, adr_type, is_vol );
}
// If reference is volatile, prevent following volatiles ops from
// floating up before the volatile write.
if (is_vol) {
// First place the specific membar for THIS volatile index. This first
// membar is dependent on the store, keeping any other membars generated
// below from floating up past the store.
int adr_idx = C->get_alias_index(adr_type);
insert_mem_bar_volatile(Op_MemBarVolatile, adr_idx);
// Now place a membar for AliasIdxBot for the unknown yet-to-be-parsed
// volatile alias indices. Skip this if the membar is redundant.
if (adr_idx != Compile::AliasIdxBot) {
insert_mem_bar_volatile(Op_MemBarVolatile, Compile::AliasIdxBot);
}
// Finally, place alias-index-specific membars for each volatile index
// that isn't the adr_idx membar. Typically there's only 1 or 2.
for( int i = Compile::AliasIdxRaw; i < C->num_alias_types(); i++ ) {
if (i != adr_idx && C->alias_type(i)->is_volatile()) {
insert_mem_bar_volatile(Op_MemBarVolatile, i);
}
}
}
// 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.
if (is_field && field->is_final()) {
set_wrote_final(true);
}
}
bool Parse::push_constant(ciConstant constant) {
switch (constant.basic_type()) {
case T_BOOLEAN: push( intcon(constant.as_boolean()) ); break;
case T_INT: push( intcon(constant.as_int()) ); break;
case T_CHAR: push( intcon(constant.as_char()) ); break;
case T_BYTE: push( intcon(constant.as_byte()) ); break;
case T_SHORT: push( intcon(constant.as_short()) ); break;
case T_FLOAT: push( makecon(TypeF::make(constant.as_float())) ); break;
case T_DOUBLE: push_pair( makecon(TypeD::make(constant.as_double())) ); break;
case T_LONG: push_pair( longcon(constant.as_long()) ); break;
case T_ARRAY:
case T_OBJECT: {
// the oop is in perm space if the ciObject "has_encoding"
ciObject* oop_constant = constant.as_object();
if (oop_constant->is_null_object()) {
push( zerocon(T_OBJECT) );
break;
} else if (oop_constant->has_encoding()) {
push( makecon(TypeOopPtr::make_from_constant(oop_constant)) );
break;
} else {
// we cannot inline the oop, but we can use it later to narrow a type
return false;
}
}
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;
}
default:
ShouldNotReachHere();
return false;
}
// success
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 Type* elemtype = _gvn.type(array)->is_aryptr()->elem();
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);
}
}
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.
enum { MAX_DIMENSION = 5 };
if (ndimensions > MAX_DIMENSION || ndimensions <= 0) {
uncommon_trap(Deoptimization::Reason_unhandled,
Deoptimization::Action_none);
return;
}
kill_dead_locals();
// get the lengths from the stack (first dimension is on top)
Node* length[MAX_DIMENSION+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 (expand_count == 1 || (1 <= expand_count && expand_count <= expand_limit)) {
Node* obj = expand_multianewarray(array_klass, &length[0], ndimensions, ndimensions);
push(obj);
return;
}
address fun = NULL;
switch (ndimensions) {
//case 1: Actually, there is no case 1. It's handled by new_array.
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;
default: ShouldNotReachHere();
};
Node* 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],
length[3], length[4]);
Node* res = _gvn.transform(new (C, 1) 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, 2) 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.
}