Node* LibraryCallKit::string_indexOf(Node* string_object, ciTypeArray* target_array, jint targetOffset_i,
jint cache_i, jint md2_i) {
Node* no_ctrl = NULL;
float likely = PROB_LIKELY(0.9);
float unlikely = PROB_UNLIKELY(0.9);
const int nargs = 0; // no arguments to push back for uncommon trap in predicate
Node* source = load_String_value(no_ctrl, string_object);
Node* sourceOffset = load_String_offset(no_ctrl, string_object);
Node* sourceCount = load_String_length(no_ctrl, string_object);
Node* target = _gvn.transform( makecon(TypeOopPtr::make_from_constant(target_array, true)));
jint target_length = target_array->length();
const TypeAry* target_array_type = TypeAry::make(TypeInt::CHAR, TypeInt::make(0, target_length, Type::WidenMin));
const TypeAryPtr* target_type = TypeAryPtr::make(TypePtr::BotPTR, target_array_type, target_array->klass(), true, Type::OffsetBot);
if (UseImplicitStableValues) {
target = cast_array_to_stable(target, target_type);
}
IdealKit kit(this, false, true);
#define __ kit.
Node* zero = __ ConI(0);
Node* one = __ ConI(1);
Node* cache = __ ConI(cache_i);
Node* md2 = __ ConI(md2_i);
Node* lastChar = __ ConI(target_array->char_at(target_length - 1));
Node* targetCount = __ ConI(target_length);
Node* targetCountLess1 = __ ConI(target_length - 1);
Node* targetOffset = __ ConI(targetOffset_i);
Node* sourceEnd = __ SubI(__ AddI(sourceOffset, sourceCount), targetCountLess1);
IdealVariable rtn(kit), i(kit), j(kit); __ declarations_done();
Node* outer_loop = __ make_label(2 /* goto */);
Node* return_ = __ make_label(1);
__ set(rtn,__ ConI(-1));
__ loop(this, nargs, i, sourceOffset, BoolTest::lt, sourceEnd); {
Node* i2 = __ AddI(__ value(i), targetCountLess1);
Node* src = load_array_element(__ ctrl(), source, i2, TypeAryPtr::CHARS);
__ if_then(src, BoolTest::eq, lastChar, unlikely); {
__ loop(this, nargs, j, zero, BoolTest::lt, targetCountLess1); {
Node* tpj = __ AddI(targetOffset, __ value(j));
Node* targ = load_array_element(no_ctrl, target, tpj, target_type);
Node* ipj = __ AddI(__ value(i), __ value(j));
Node* src2 = load_array_element(no_ctrl, source, ipj, TypeAryPtr::CHARS);
__ if_then(targ, BoolTest::ne, src2); {
__ if_then(__ AndI(cache, __ LShiftI(one, src2)), BoolTest::eq, zero); {
__ if_then(md2, BoolTest::lt, __ AddI(__ value(j), one)); {
__ increment(i, __ AddI(__ value(j), one));
__ goto_(outer_loop);
} __ end_if(); __ dead(j);
}__ end_if(); __ dead(j);
__ increment(i, md2);
__ goto_(outer_loop);
}__ end_if();
__ increment(j, one);
}__ end_loop(); __ dead(j);
__ set(rtn, __ SubI(__ value(i), sourceOffset)); __ dead(i);
__ goto_(return_);
}__ end_if();
__ if_then(__ AndI(cache, __ LShiftI(one, src)), BoolTest::eq, zero, likely); {
__ increment(i, targetCountLess1);
}__ end_if();
__ increment(i, one);
__ bind(outer_loop);
}__ end_loop(); __ dead(i);
__ bind(return_);
final_sync(kit);
Node* result = __ value(rtn);
#undef __
C->set_has_loops(true);
return result;
}
bool LibraryCallKit::inline_string_indexOf() {
Node* receiver = argument(0);
Node* arg = argument(1);
Node* result;
if (Matcher::has_match_rule(Op_StrIndexOf) &&
UseSSE42Intrinsics) {
receiver = null_check(receiver);
arg = null_check(arg);
if (stopped()) {
return true;
}
ciInstanceKlass* str_klass = env()->String_klass();
const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(str_klass);
RegionNode* result_rgn = new (C) RegionNode(4);
Node* result_phi = new (C) PhiNode(result_rgn, TypeInt::INT);
Node* no_ctrl = NULL;
Node* source = load_String_value(no_ctrl, receiver);
Node* source_offset = load_String_offset(no_ctrl, receiver);
Node* source_start = array_element_address(source, source_offset, T_CHAR);
Node* source_cnt = load_String_length(no_ctrl, receiver);
Node* substr = load_String_value(no_ctrl, arg);
Node* substr_offset = load_String_offset(no_ctrl, arg);
Node* substr_start = array_element_address(substr, substr_offset, T_CHAR);
Node* substr_cnt = load_String_length(no_ctrl, arg);
Node* cmp = _gvn.transform(new(C) CmpINode(substr_cnt, source_cnt));
Node* bol = _gvn.transform(new(C) BoolNode(cmp, BoolTest::gt));
Node* if_gt = generate_slow_guard(bol, NULL);
if (if_gt != NULL) {
result_phi->init_req(2, intcon(-1));
result_rgn->init_req(2, if_gt);
}
if (!stopped()) {
cmp = _gvn.transform(new(C) CmpINode(substr_cnt, intcon(0)));
bol = _gvn.transform(new(C) BoolNode(cmp, BoolTest::eq));
Node* if_zero = generate_slow_guard(bol, NULL);
if (if_zero != NULL) {
result_phi->init_req(3, intcon(0));
result_rgn->init_req(3, if_zero);
}
}
if (!stopped()) {
result = make_string_method_node(Op_StrIndexOf, source_start, source_cnt, substr_start, substr_cnt);
result_phi->init_req(1, result);
result_rgn->init_req(1, control());
}
set_control(_gvn.transform(result_rgn));
record_for_igvn(result_rgn);
result = _gvn.transform(result_phi);
} else { // Use LibraryCallKit::string_indexOf
if (!arg->is_Con()) {
return false;
}
const TypeOopPtr* str_type = _gvn.type(arg)->isa_oopptr();
if (str_type == NULL) {
return false;
}
ciInstanceKlass* klass = env()->String_klass();
ciObject* str_const = str_type->const_oop();
if (str_const == NULL || str_const->klass() != klass) {
return false;
}
ciInstance* str = str_const->as_instance();
assert(str != NULL, "must be instance");
ciObject* v = str->field_value_by_offset(java_lang_String::value_offset_in_bytes()).as_object();
ciTypeArray* pat = v->as_type_array(); // pattern (argument) character array
int o;
int c;
if (java_lang_String::has_offset_field()) {
o = str->field_value_by_offset(java_lang_String::offset_offset_in_bytes()).as_int();
c = str->field_value_by_offset(java_lang_String::count_offset_in_bytes()).as_int();
} else {
o = 0;
c = pat->length();
}
if (o != 0 || c != pat->length()) {
return false;
}
receiver = null_check(receiver, T_OBJECT);
if (stopped()) {
return true;
}
if (c == 0) {
set_result(intcon(0));
return true;
}
jchar lastChar = pat->char_at(o + (c - 1));
int cache = 0;
int i;
for (i = 0; i < c - 1; i++) {
assert(i < pat->length(), "out of range");
cache |= (1 << (pat->char_at(o + i) & (sizeof(cache) * BitsPerByte - 1)));
}
int md2 = c;
for (i = 0; i < c - 1; i++) {
assert(i < pat->length(), "out of range");
if (pat->char_at(o + i) == lastChar) {
md2 = (c - 1) - i;
}
}
result = string_indexOf(receiver, pat, o, cache, md2);
}
set_result(result);
return true;
}
Node* LibraryCallKit::round_double_node(Node* n) {
if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
n = _gvn.transform(new (C) RoundDoubleNode(0, n));
return n;
}
bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
Node* arg = round_double_node(argument(0));
Node* n = NULL;
switch (id) {
case vmIntrinsics::_dabs: n = new (C) AbsDNode( arg); break;
case vmIntrinsics::_dsqrt: n = new (C) SqrtDNode(C, control(), arg); break;
case vmIntrinsics::_dlog: n = new (C) LogDNode(C, control(), arg); break;
case vmIntrinsics::_dlog10: n = new (C) Log10DNode(C, control(), arg); break;
default: fatal_unexpected_iid(id); break;
}
set_result(_gvn.transform(n));
return true;
}
bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
Node* arg = round_double_node(argument(0));
Node* n = NULL;
switch (id) {
case vmIntrinsics::_dsin: n = new (C) SinDNode(C, control(), arg); break;
case vmIntrinsics::_dcos: n = new (C) CosDNode(C, control(), arg); break;
case vmIntrinsics::_dtan: n = new (C) TanDNode(C, control(), arg); break;
default: fatal_unexpected_iid(id); break;
}
n = _gvn.transform(n);
if (Matcher::strict_fp_requires_explicit_rounding) {
static const double pi_4 = 0.7853981633974483;
static const double neg_pi_4 = -0.7853981633974483;
RegionNode* r = new (C) RegionNode(3);
Node* phi = new (C) PhiNode(r, Type::DOUBLE);
Node *abs = _gvn.transform(new (C) AbsDNode(arg));
Node *pi4 = makecon(TypeD::make(pi_4));
Node *cmp = _gvn.transform(new (C) CmpDNode(pi4,abs));
Node *bol = _gvn.transform(new (C) BoolNode( cmp, BoolTest::lt ));
IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
set_control(opt_iff(r,iff));
phi->init_req(2, n);
Node* call = NULL;
switch (id) {
case vmIntrinsics::_dsin:
call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
CAST_FROM_FN_PTR(address, SharedRuntime::dsin),
"Sin", NULL, arg, top());
break;
case vmIntrinsics::_dcos:
call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
CAST_FROM_FN_PTR(address, SharedRuntime::dcos),
"Cos", NULL, arg, top());
break;
case vmIntrinsics::_dtan:
call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
"Tan", NULL, arg, top());
break;
}
assert(control()->in(0) == call, "");
Node* slow_result = _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
r->init_req(1, control());
phi->init_req(1, slow_result);
set_control(_gvn.transform(r));
record_for_igvn(r);
n = _gvn.transform(phi);
C->set_has_split_ifs(true); // Has chance for split-if optimization
}
set_result(n);
return true;
}
Node* LibraryCallKit::finish_pow_exp(Node* result, Node* x, Node* y, const TypeFunc* call_type, address funcAddr, const char* funcName) {
Node* cmpisnan = _gvn.transform(new (C) CmpDNode(result, result));
Node* bolisnum = _gvn.transform(new (C) BoolNode(cmpisnan, BoolTest::eq));
if (!too_many_traps(Deoptimization::Reason_intrinsic)) {
{ BuildCutout unless(this, bolisnum, PROB_STATIC_FREQUENT);
uncommon_trap(Deoptimization::Reason_intrinsic,
Deoptimization::Action_make_not_entrant);
}
return result;
} else {
IfNode* iff = create_and_xform_if(control(), bolisnum, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
Node* if_slow = _gvn.transform(new (C) IfFalseNode(iff));
Node* if_fast = _gvn.transform(new (C) IfTrueNode(iff));
if (!if_slow->is_top()) {
RegionNode* result_region = new (C) RegionNode(3);
PhiNode* result_val = new (C) PhiNode(result_region, Type::DOUBLE);
result_region->init_req(1, if_fast);
result_val->init_req(1, result);
set_control(if_slow);
const TypePtr* no_memory_effects = NULL;
Node* rt = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
no_memory_effects,
x, top(), y, y ? top() : NULL);
Node* value = _gvn.transform(new (C) ProjNode(rt, TypeFunc::Parms+0));
#ifdef ASSERT
Node* value_top = _gvn.transform(new (C) ProjNode(rt, TypeFunc::Parms+1));
assert(value_top == top(), "second value must be top");
#endif
result_region->init_req(2, control());
result_val->init_req(2, value);
set_control(_gvn.transform(result_region));
return _gvn.transform(result_val);
} else {
return result;
}
}
}
bool LibraryCallKit::inline_exp() {
Node* arg = round_double_node(argument(0));
Node* n = _gvn.transform(new (C) ExpDNode(C, control(), arg));
n = finish_pow_exp(n, arg, NULL, OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
set_result(n);
C->set_has_split_ifs(true); // Has chance for split-if optimization
return true;
}
bool LibraryCallKit::inline_pow() {
Node* x = round_double_node(argument(0));
Node* y = round_double_node(argument(2));
Node* result = NULL;
Node* const_two_node = makecon(TypeD::make(2.0));
Node* cmp_node = _gvn.transform(new (C) CmpDNode(y, const_two_node));
Node* bool_node = _gvn.transform(new (C) BoolNode(cmp_node, BoolTest::eq));
IfNode* if_node = create_and_xform_if(control(), bool_node, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
Node* if_true = _gvn.transform(new (C) IfTrueNode(if_node));
Node* if_false = _gvn.transform(new (C) IfFalseNode(if_node));
RegionNode* region_node = new (C) RegionNode(3);
region_node->init_req(1, if_true);
Node* phi_node = new (C) PhiNode(region_node, Type::DOUBLE);
phi_node->init_req(1, _gvn.transform(new (C) MulDNode(x, x)));
set_control(if_false);
if (!too_many_traps(Deoptimization::Reason_intrinsic)) {
result = _gvn.transform(new (C) PowDNode(C, control(), x, y));
} else {
RegionNode *r = new (C) RegionNode(4);
Node *phi = new (C) PhiNode(r, Type::DOUBLE);
Node *zeronode = makecon(TypeD::ZERO);
Node *cmp = _gvn.transform(new (C) CmpDNode(x, zeronode));
Node *bol1 = _gvn.transform(new (C) BoolNode( cmp, BoolTest::le ));
IfNode *if1 = create_and_xform_if(control(),bol1, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
Node *fast_taken = _gvn.transform(new (C) IfFalseNode(if1));
r->init_req(3,fast_taken); // Capture fast-control
Node *complex_path = _gvn.transform(new (C) IfTrueNode(if1));
Node *fast_result = _gvn.transform(new (C) PowDNode(C, control(), x, y));
phi->init_req(3, fast_result);
Node *longy = _gvn.transform(new (C) ConvD2LNode(y));
Node *doublelongy= _gvn.transform(new (C) ConvL2DNode(longy));
Node *cmplongy= _gvn.transform(new (C) CmpDNode(doublelongy, y));
Node *bol2 = _gvn.transform(new (C) BoolNode( cmplongy, BoolTest::ne ));
IfNode *if2 = create_and_xform_if(complex_path,bol2, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
Node* ylong_path = _gvn.transform(new (C) IfFalseNode(if2));
Node *slow_path = _gvn.transform(new (C) IfTrueNode(if2));
Node *conone = longcon(1);
Node *signnode= _gvn.transform(new (C) AndLNode(conone, longy));
Node* yplus1 = _gvn.transform(new (C) AddDNode(y, makecon(TypeD::make(1))));
Node *cmpyplus1= _gvn.transform(new (C) CmpDNode(yplus1, y));
Node *bolyplus1 = _gvn.transform(new (C) BoolNode( cmpyplus1, BoolTest::eq ));
Node* correctedsign = NULL;
if (ConditionalMoveLimit != 0) {
correctedsign = _gvn.transform( CMoveNode::make(C, NULL, bolyplus1, signnode, longcon(0), TypeLong::LONG));
} else {
IfNode *ifyplus1 = create_and_xform_if(ylong_path,bolyplus1, PROB_FAIR, COUNT_UNKNOWN);
RegionNode *r = new (C) RegionNode(3);
Node *phi = new (C) PhiNode(r, TypeLong::LONG);
r->init_req(1, _gvn.transform(new (C) IfFalseNode(ifyplus1)));
r->init_req(2, _gvn.transform(new (C) IfTrueNode(ifyplus1)));
phi->init_req(1, signnode);
phi->init_req(2, longcon(0));
correctedsign = _gvn.transform(phi);
ylong_path = _gvn.transform(r);
record_for_igvn(r);
}
Node *conzero = longcon(0);
Node *cmpeq1 = _gvn.transform(new (C) CmpLNode(correctedsign, conzero));
Node *bol3 = _gvn.transform(new (C) BoolNode( cmpeq1, BoolTest::ne ));
Node *absx=_gvn.transform(new (C) AbsDNode(x));
Node *absxpowy = _gvn.transform(new (C) PowDNode(C, control(), absx, y));
Node *negabsxpowy = _gvn.transform(new (C) NegDNode (absxpowy));
Node *signresult = NULL;
if (ConditionalMoveLimit != 0) {
signresult = _gvn.transform( CMoveNode::make(C, NULL, bol3, absxpowy, negabsxpowy, Type::DOUBLE));
} else {
IfNode *ifyeven = create_and_xform_if(ylong_path,bol3, PROB_FAIR, COUNT_UNKNOWN);
RegionNode *r = new (C) RegionNode(3);
Node *phi = new (C) PhiNode(r, Type::DOUBLE);
r->init_req(1, _gvn.transform(new (C) IfFalseNode(ifyeven)));
r->init_req(2, _gvn.transform(new (C) IfTrueNode(ifyeven)));
phi->init_req(1, absxpowy);
phi->init_req(2, negabsxpowy);
signresult = _gvn.transform(phi);
ylong_path = _gvn.transform(r);
record_for_igvn(r);
}
r->init_req(2, ylong_path);
phi->init_req(2, signresult);
static const jlong nan_bits = CONST64(0x7ff8000000000000);
Node *slow_result = makecon(TypeD::make(*(double*)&nan_bits)); // return NaN
r->init_req(1,slow_path);
phi->init_req(1,slow_result);
set_control(_gvn.transform(r));
record_for_igvn(r);
result = _gvn.transform(phi);
}
result = finish_pow_exp(result, x, y, OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
region_node->set_req(2, control());
phi_node->init_req(2, result);
set_control(_gvn.transform(region_node));
record_for_igvn(region_node);
set_result(_gvn.transform(phi_node));
C->set_has_split_ifs(true); // Has chance for split-if optimization
return true;
}
bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
"must be (DD)D or (D)D type");
Node* a = round_double_node(argument(0));
Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
const TypePtr* no_memory_effects = NULL;
Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
no_memory_effects,
a, top(), b, b ? top() : NULL);
Node* value = _gvn.transform(new (C) ProjNode(trig, TypeFunc::Parms+0));
#ifdef ASSERT
Node* value_top = _gvn.transform(new (C) ProjNode(trig, TypeFunc::Parms+1));
assert(value_top == top(), "second value must be top");
#endif
set_result(value);
return true;
}
bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
#define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
switch (id) {
case vmIntrinsics::_dcos: return Matcher::has_match_rule(Op_CosD) ? inline_trig(id) :
runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
case vmIntrinsics::_dsin: return Matcher::has_match_rule(Op_SinD) ? inline_trig(id) :
runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
case vmIntrinsics::_dtan: return Matcher::has_match_rule(Op_TanD) ? inline_trig(id) :
runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
case vmIntrinsics::_dlog: return Matcher::has_match_rule(Op_LogD) ? inline_math(id) :
runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
case vmIntrinsics::_dlog10: return Matcher::has_match_rule(Op_Log10D) ? inline_math(id) :
runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
case vmIntrinsics::_dexp: return Matcher::has_match_rule(Op_ExpD) ? inline_exp() :
runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
case vmIntrinsics::_dpow: return Matcher::has_match_rule(Op_PowD) ? inline_pow() :
runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
#undef FN_PTR
case vmIntrinsics::_datan2:
return false;
default:
fatal_unexpected_iid(id);
return false;
}
}
static bool is_simple_name(Node* n) {
return (n->req() == 1 // constant
|| (n->is_Type() && n->as_Type()->type()->singleton())
|| n->is_Proj() // parameter or return value
|| n->is_Phi() // local of some sort
);
}
bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
set_result(generate_min_max(id, argument(0), argument(1)));
return true;
}
void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
Node* bol = _gvn.transform( new (C) BoolNode(test, BoolTest::overflow) );
IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
Node* fast_path = _gvn.transform( new (C) IfFalseNode(check));
Node* slow_path = _gvn.transform( new (C) IfTrueNode(check) );
{
PreserveJVMState pjvms(this);
PreserveReexecuteState preexecs(this);
jvms()->set_should_reexecute(true);
set_control(slow_path);
set_i_o(i_o());
uncommon_trap(Deoptimization::Reason_intrinsic,
Deoptimization::Action_none);
}
set_control(fast_path);
set_result(math);
}
template <typename OverflowOp>
bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
typedef typename OverflowOp::MathOp MathOp;
MathOp* mathOp = new(C) MathOp(arg1, arg2);
Node* operation = _gvn.transform( mathOp );
Node* ofcheck = _gvn.transform( new(C) OverflowOp(arg1, arg2) );
inline_math_mathExact(operation, ofcheck);
return true;
}
bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
}
bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
}
bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
}
bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
}
bool LibraryCallKit::inline_math_negateExactI() {
return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
}
bool LibraryCallKit::inline_math_negateExactL() {
return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
}
bool LibraryCallKit::inline_math_multiplyExactI() {
return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
}
bool LibraryCallKit::inline_math_multiplyExactL() {
return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
}
Node*
LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
Node* xvalue = x0;
Node* yvalue = y0;
if (xvalue == yvalue) {
return xvalue;
}
bool want_max = (id == vmIntrinsics::_max);
const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
if (txvalue == NULL || tyvalue == NULL) return top();
int widen = MAX2(txvalue->_widen, tyvalue->_widen);
int cmp_op = Op_CmpI;
Node* xkey = xvalue;
Node* ykey = yvalue;
Node* ideal_cmpxy = _gvn.transform(new(C) CmpINode(xkey, ykey));
if (ideal_cmpxy->is_Cmp()) {
cmp_op = ideal_cmpxy->Opcode();
xkey = ideal_cmpxy->in(1);
ykey = ideal_cmpxy->in(2);
}
Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
Node* cmpxy = NULL;
Node* cmpyx = NULL;
for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
Node* cmp = start_from->fast_out(k);
if (cmp->outcnt() > 0 && // must have prior uses
cmp->in(0) == NULL && // must be context-independent
cmp->Opcode() == cmp_op) { // right kind of compare
if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
}
}
const int NCMPS = 2;
Node* cmps[NCMPS] = { cmpxy, cmpyx };
int cmpn;
for (cmpn = 0; cmpn < NCMPS; cmpn++) {
if (cmps[cmpn] != NULL) break; // find a result
}
if (cmpn < NCMPS) {
int depth = 0; // Limit search depth for speed
Node* dom = control();
for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
if (++depth >= 100) break;
Node* ifproj = dom;
if (!ifproj->is_Proj()) continue;
Node* iff = ifproj->in(0);
if (!iff->is_If()) continue;
Node* bol = iff->in(1);
if (!bol->is_Bool()) continue;
Node* cmp = bol->in(1);
if (cmp == NULL) continue;
for (cmpn = 0; cmpn < NCMPS; cmpn++)
if (cmps[cmpn] == cmp) break;
if (cmpn == NCMPS) continue;
BoolTest::mask btest = bol->as_Bool()->_test._test;
if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
switch (btest) {
case BoolTest::eq:
if (is_simple_name(yvalue) && !is_simple_name(xvalue))
return yvalue;
return xvalue;
case BoolTest::lt: // x < y
case BoolTest::le: // x <= y
return (want_max ? yvalue : xvalue);
case BoolTest::gt: // x > y
case BoolTest::ge: // x >= y
return (want_max ? xvalue : yvalue);
}
}
}
Node* best_bol = NULL;
BoolTest::mask best_btest = BoolTest::illegal;
for (cmpn = 0; cmpn < NCMPS; cmpn++) {
Node* cmp = cmps[cmpn];
if (cmp == NULL) continue;
for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
Node* bol = cmp->fast_out(j);
if (!bol->is_Bool()) continue;
BoolTest::mask btest = bol->as_Bool()->_test._test;
if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
best_bol = bol->as_Bool();
best_btest = btest;
}
}
}
Node* answer_if_true = NULL;
Node* answer_if_false = NULL;
switch (best_btest) {
default:
if (cmpxy == NULL)
cmpxy = ideal_cmpxy;
best_bol = _gvn.transform(new(C) BoolNode(cmpxy, BoolTest::lt));
case BoolTest::lt: // x < y
case BoolTest::le: // x <= y
answer_if_true = (want_max ? yvalue : xvalue);
answer_if_false = (want_max ? xvalue : yvalue);
break;
case BoolTest::gt: // x > y
case BoolTest::ge: // x >= y
answer_if_true = (want_max ? xvalue : yvalue);
answer_if_false = (want_max ? yvalue : xvalue);
break;
}
jint hi, lo;
if (want_max) {
hi = MAX2(txvalue->_hi, tyvalue->_hi);
lo = MAX2(txvalue->_lo, tyvalue->_lo);
} else {
hi = MIN2(txvalue->_hi, tyvalue->_hi);
lo = MIN2(txvalue->_lo, tyvalue->_lo);
}
Node* cmov = CMoveNode::make(C, NULL, best_bol,
answer_if_false, answer_if_true,
TypeInt::make(lo, hi, widen));
return _gvn.transform(cmov);
switch (id) {
case vmIntrinsics::_min:
result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
case vmIntrinsics::_max:
result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
default:
ShouldNotReachHere();
}
}
inline int
LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
const TypePtr* base_type = TypePtr::NULL_PTR;
if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
if (base_type == NULL) {
return Type::AnyPtr;
} else if (base_type == TypePtr::NULL_PTR) {
base = _gvn.transform(new (C) CastX2PNode(offset));
offset = MakeConX(0);
return Type::RawPtr;
} else if (base_type->base() == Type::RawPtr) {
return Type::RawPtr;
} else if (base_type->isa_oopptr()) {
if (base_type->ptr() == TypePtr::NotNull) {
return Type::OopPtr;
}
const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
if (offset_type != NULL &&
base_type->offset() == 0 && // (should always be?)
offset_type->_lo >= 0 &&
!MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
return Type::OopPtr;
}
return Type::AnyPtr;
} else {
return Type::AnyPtr;
}
}
inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
int kind = classify_unsafe_addr(base, offset);
if (kind == Type::RawPtr) {
return basic_plus_adr(top(), base, offset);
} else {
return basic_plus_adr(base, offset);
}
}
bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
Node* arg = argument(0);
Node* n = NULL;
switch (id) {
case vmIntrinsics::_numberOfLeadingZeros_i: n = new (C) CountLeadingZerosINode( arg); break;
case vmIntrinsics::_numberOfLeadingZeros_l: n = new (C) CountLeadingZerosLNode( arg); break;
case vmIntrinsics::_numberOfTrailingZeros_i: n = new (C) CountTrailingZerosINode(arg); break;
case vmIntrinsics::_numberOfTrailingZeros_l: n = new (C) CountTrailingZerosLNode(arg); break;
case vmIntrinsics::_bitCount_i: n = new (C) PopCountINode( arg); break;
case vmIntrinsics::_bitCount_l: n = new (C) PopCountLNode( arg); break;
case vmIntrinsics::_reverseBytes_c: n = new (C) ReverseBytesUSNode(0, arg); break;
case vmIntrinsics::_reverseBytes_s: n = new (C) ReverseBytesSNode( 0, arg); break;
case vmIntrinsics::_reverseBytes_i: n = new (C) ReverseBytesINode( 0, arg); break;
case vmIntrinsics::_reverseBytes_l: n = new (C) ReverseBytesLNode( 0, arg); break;
default: fatal_unexpected_iid(id); break;
}
set_result(_gvn.transform(n));
return true;
}
const static BasicType T_ADDRESS_HOLDER = T_LONG;
void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
Node* pre_val, bool need_mem_bar) {
if (!UseG1GC && !need_mem_bar)
return;
const TypeX* otype = offset->find_intptr_t_type();
if (otype != NULL && otype->is_con() &&
otype->get_con() != java_lang_ref_Reference::referent_offset) {
return;
}
const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
if (btype != NULL) {
if (btype->isa_aryptr()) {
return;
}
const TypeInstPtr* itype = btype->isa_instptr();
if (itype != NULL) {
ciKlass* klass = itype->klass();
if ( klass->is_loaded() &&
!klass->is_subtype_of(env()->Reference_klass()) &&
!env()->Object_klass()->is_subtype_of(klass)) {
return;
}
}
}
float likely = PROB_LIKELY( 0.999);
float unlikely = PROB_UNLIKELY(0.999);
IdealKit ideal(this);
#define __ ideal.
Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
__ if_then(offset, BoolTest::eq, referent_off, unlikely); {
sync_kit(ideal);
Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
__ sync_kit(this);
Node* one = __ ConI(1);
__ if_then(is_instof, BoolTest::eq, one, unlikely); {
sync_kit(ideal);
pre_barrier(false /* do_load */,
__ ctrl(),
NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
pre_val /* pre_val */,
T_OBJECT);
if (need_mem_bar) {
insert_mem_bar(Op_MemBarCPUOrder);
}
__ sync_kit(this);
} __ end_if(); // _ref_type != ref_none
} __ end_if(); // offset == referent_offset
final_sync(ideal);
#undef __
}
extern jlong Unsafe_field_offset_to_byte_offset(jlong field_offset);
const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) {
ciKlass* sharpened_klass = NULL;
if (alias_type->field() != NULL) {
assert(!is_native_ptr, "native pointer op cannot use a java address");
if (alias_type->field()->type()->is_klass()) {
sharpened_klass = alias_type->field()->type()->as_klass();
}
}
if (adr_type->isa_aryptr()) {
if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
if (elem_type != NULL) {
sharpened_klass = elem_type->klass();
}
}
}
if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
#ifndef PRODUCT
if (C->print_intrinsics() || C->print_inlining()) {
tty->print(" from base type: "); adr_type->dump(); tty->cr();
tty->print(" sharpened value: "); tjp->dump(); tty->cr();
}
#endif
return tjp;
}
return NULL;
}
bool LibraryCallKit::inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, bool is_volatile, bool unaligned) {
if (callee()->is_static()) return false; // caller must have the capability!
assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
#ifndef PRODUCT
{
ResourceMark rm;
ciSignature* sig = callee()->signature();
#ifdef ASSERT
if (!is_store) {
BasicType rtype = sig->return_type()->basic_type();
if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name())
rtype = T_ADDRESS; // it is really a C void*
assert(rtype == type, "getter must return the expected value");
if (!is_native_ptr) {
assert(sig->count() == 2, "oop getter has 2 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
} else {
assert(sig->count() == 1, "native getter has 1 argument");
assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long");
}
} else {
assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
if (!is_native_ptr) {
assert(sig->count() == 3, "oop putter has 3 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
} else {
assert(sig->count() == 2, "native putter has 2 arguments");
assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long");
}
BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name())
vtype = T_ADDRESS; // it is really a C void*
assert(vtype == type, "putter must accept the expected value");
}
#endif // ASSERT
}
#endif //PRODUCT
C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
Node* receiver = argument(0); // type: oop
Node* adr;
Node* heap_base_oop = top();
Node* offset = top();
Node* val;
Node* base = argument(1); // type: oop
if (!is_native_ptr) {
offset = argument(2); // type: long
assert(Unsafe_field_offset_to_byte_offset(11) == 11,
"fieldOffset must be byte-scaled");
offset = ConvL2X(offset);
adr = make_unsafe_address(base, offset);
heap_base_oop = base;
val = is_store ? argument(4) : NULL;
} else {
Node* ptr = argument(1); // type: long
ptr = ConvL2X(ptr); // adjust Java long to machine word
adr = make_unsafe_address(NULL, ptr);
val = is_store ? argument(3) : NULL;
}
if ((_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) && type == T_OBJECT) {
return false; // off-heap oop accesses are not supported
}
const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
Compile::AliasType* alias_type = C->alias_type(adr_type);
assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
if (alias_type->adr_type() == TypeInstPtr::KLASS ||
alias_type->adr_type() == TypeAryPtr::RANGE) {
return false; // not supported
}
bool mismatched = false;
BasicType bt = alias_type->basic_type();
if (bt != T_ILLEGAL) {
assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
if (bt == T_BYTE && adr_type->isa_aryptr()) {
bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
}
if (bt == T_ARRAY || bt == T_NARROWOOP) {
bt = T_OBJECT;
}
if ((bt == T_OBJECT) != (type == T_OBJECT)) {
return false;
}
mismatched = (bt != type);
}
assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
const Type *value_type = Type::get_const_basic_type(type);
bool need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);
bool need_read_barrier = !is_native_ptr && !is_store &&
offset != top() && heap_base_oop != top();
if (!is_store && type == T_OBJECT) {
const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr);
if (tjp != NULL) {
value_type = tjp;
}
}
receiver = null_check(receiver);
if (stopped()) {
return true;
}
Node* load = NULL;
Node* store = NULL;
Node* leading_membar = NULL;
if (is_volatile) {
need_mem_bar = true;
if (is_store) {
leading_membar = insert_mem_bar(Op_MemBarRelease);
} else {
if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
leading_membar = insert_mem_bar(Op_MemBarVolatile);
}
}
}
if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
if (!is_store) {
MemNode::MemOrd mo = is_volatile ? MemNode::acquire : MemNode::unordered;
load = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, is_volatile, unaligned, mismatched);
switch (type) {
case T_BOOLEAN:
case T_CHAR:
case T_BYTE:
case T_SHORT:
case T_INT:
case T_LONG:
case T_FLOAT:
case T_DOUBLE:
break;
case T_OBJECT:
if (need_read_barrier) {
insert_pre_barrier(heap_base_oop, offset, load, !(is_volatile || need_mem_bar));
}
break;
case T_ADDRESS:
load = _gvn.transform(new (C) CastP2XNode(NULL, load));
load = ConvX2UL(load);
break;
default:
fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
break;
}
set_result(load);
} else {
switch (type) {
case T_DOUBLE:
val = dstore_rounding(val);
break;
case T_ADDRESS:
val = ConvL2X(val);
val = _gvn.transform(new (C) CastX2PNode(val));
break;
}
MemNode::MemOrd mo = is_volatile ? MemNode::release : MemNode::unordered;
if (type == T_OBJECT ) {
store = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
} else {
store = store_to_memory(control(), adr, val, type, adr_type, mo, is_volatile, unaligned, mismatched);
}
}
if (is_volatile) {
if (!is_store) {
Node* mb = insert_mem_bar(Op_MemBarAcquire, load);
mb->as_MemBar()->set_trailing_load();
} else {
if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
Node* mb = insert_mem_bar(Op_MemBarVolatile, store);
MemBarNode::set_store_pair(leading_membar->as_MemBar(), mb->as_MemBar());
}
}
}
if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
return true;
}
bool LibraryCallKit::inline_unsafe_prefetch(bool is_native_ptr, bool is_store, bool is_static) {
#ifndef PRODUCT
{
ResourceMark rm;
ciSignature* sig = callee()->signature();
#ifdef ASSERT
BasicType rtype = sig->return_type()->basic_type();
if (!is_native_ptr) {
assert(sig->count() == 2, "oop prefetch has 2 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "prefetch base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "prefetcha offset is correct");
} else {
assert(sig->count() == 1, "native prefetch has 1 argument");
assert(sig->type_at(0)->basic_type() == T_LONG, "prefetch base is long");
}
#endif // ASSERT
}
#endif // !PRODUCT
C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
const int idx = is_static ? 0 : 1;
if (!is_static) {
null_check_receiver();
if (stopped()) {
return true;
}
}
Node *adr;
if (!is_native_ptr) {
Node* base = argument(idx + 0); // type: oop
Node* offset = argument(idx + 1); // type: long
assert(Unsafe_field_offset_to_byte_offset(11) == 11,
"fieldOffset must be byte-scaled");
offset = ConvL2X(offset);
adr = make_unsafe_address(base, offset);
} else {
Node* ptr = argument(idx + 0); // type: long
ptr = ConvL2X(ptr); // adjust Java long to machine word
adr = make_unsafe_address(NULL, ptr);
}
Node *prefetch;
if (is_store) {
prefetch = new (C) PrefetchWriteNode(i_o(), adr);
} else {
prefetch = new (C) PrefetchReadNode(i_o(), adr);
}
prefetch->init_req(0, control());
set_i_o(_gvn.transform(prefetch));
return true;
}
bool LibraryCallKit::inline_unsafe_load_store(BasicType type, LoadStoreKind kind) {
if (callee()->is_static()) return false; // caller must have the capability!
#ifndef PRODUCT
BasicType rtype;
{
ResourceMark rm;
ciSignature* sig = callee()->signature();
rtype = sig->return_type()->basic_type();
if (kind == LS_xadd || kind == LS_xchg) {
#ifdef ASSERT
assert(rtype == type, "get and set must return the expected type");
assert(sig->count() == 3, "get and set has 3 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
#endif // ASSERT
} else if (kind == LS_cmpxchg) {
#ifdef ASSERT
assert(rtype == T_BOOLEAN, "CAS must return boolean");
assert(sig->count() == 4, "CAS has 4 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
#endif // ASSERT
} else {
ShouldNotReachHere();
}
}
#endif //PRODUCT
C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
Node* receiver = NULL;
Node* base = NULL;
Node* offset = NULL;
Node* oldval = NULL;
Node* newval = NULL;
if (kind == LS_cmpxchg) {
const bool two_slot_type = type2size[type] == 2;
receiver = argument(0); // type: oop
base = argument(1); // type: oop
offset = argument(2); // type: long
oldval = argument(4); // type: oop, int, or long
newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
} else if (kind == LS_xadd || kind == LS_xchg){
receiver = argument(0); // type: oop
base = argument(1); // type: oop
offset = argument(2); // type: long
oldval = NULL;
newval = argument(4); // type: oop, int, or long
}
assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
offset = ConvL2X(offset);
Node* adr = make_unsafe_address(base, offset);
const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
Compile::AliasType* alias_type = C->alias_type(adr_type);
BasicType bt = alias_type->basic_type();
if (bt != T_ILLEGAL &&
((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
return false;
}
assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
const Type *value_type = Type::get_const_basic_type(type);
if (kind == LS_xchg && type == T_OBJECT) {
const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
if (tjp != NULL) {
value_type = tjp;
}
}
receiver = null_check(receiver);
if (stopped()) {
return true;
}
int alias_idx = C->get_alias_index(adr_type);
Node* leading_membar = insert_mem_bar(Op_MemBarRelease);
insert_mem_bar(Op_MemBarCPUOrder);
Node *mem = memory(alias_idx);
Node* load_store = NULL;
switch(type) {
case T_INT:
if (kind == LS_xadd) {
load_store = _gvn.transform(new (C) GetAndAddINode(control(), mem, adr, newval, adr_type));
} else if (kind == LS_xchg) {
load_store = _gvn.transform(new (C) GetAndSetINode(control(), mem, adr, newval, adr_type));
} else if (kind == LS_cmpxchg) {
load_store = _gvn.transform(new (C) CompareAndSwapINode(control(), mem, adr, newval, oldval));
} else {
ShouldNotReachHere();
}
break;
case T_LONG:
if (kind == LS_xadd) {
load_store = _gvn.transform(new (C) GetAndAddLNode(control(), mem, adr, newval, adr_type));
} else if (kind == LS_xchg) {
load_store = _gvn.transform(new (C) GetAndSetLNode(control(), mem, adr, newval, adr_type));
} else if (kind == LS_cmpxchg) {
load_store = _gvn.transform(new (C) CompareAndSwapLNode(control(), mem, adr, newval, oldval));
} else {
ShouldNotReachHere();
}
break;
case T_OBJECT:
if (_gvn.type(newval) == TypePtr::NULL_PTR)
newval = _gvn.makecon(TypePtr::NULL_PTR);
if (kind == LS_xchg) {
if (!can_move_pre_barrier()) {
pre_barrier(true /* do_load*/,
control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
NULL /* pre_val*/,
T_OBJECT);
} // Else move pre_barrier to use load_store value, see below.
} else if (kind == LS_cmpxchg) {
if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
oldval = _gvn.makecon(TypePtr::NULL_PTR);
}
pre_barrier(false /* do_load */,
control(), NULL, NULL, max_juint, NULL, NULL,
oldval /* pre_val */,
T_OBJECT);
} else {
ShouldNotReachHere();
}
#ifdef _LP64
if (adr->bottom_type()->is_ptr_to_narrowoop()) {
Node *newval_enc = _gvn.transform(new (C) EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
if (kind == LS_xchg) {
load_store = _gvn.transform(new (C) GetAndSetNNode(control(), mem, adr,
newval_enc, adr_type, value_type->make_narrowoop()));
} else {
assert(kind == LS_cmpxchg, "wrong LoadStore operation");
Node *oldval_enc = _gvn.transform(new (C) EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
load_store = _gvn.transform(new (C) CompareAndSwapNNode(control(), mem, adr,
newval_enc, oldval_enc));
}
} else
#endif
{
if (kind == LS_xchg) {
load_store = _gvn.transform(new (C) GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
} else {
assert(kind == LS_cmpxchg, "wrong LoadStore operation");
load_store = _gvn.transform(new (C) CompareAndSwapPNode(control(), mem, adr, newval, oldval));
}
}
post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
break;
default:
fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
break;
}
Node* proj = _gvn.transform(new (C) SCMemProjNode(load_store));
set_memory(proj, alias_idx);
Node* access = load_store;
if (type == T_OBJECT && kind == LS_xchg) {
#ifdef _LP64
if (adr->bottom_type()->is_ptr_to_narrowoop()) {
load_store = _gvn.transform(new (C) DecodeNNode(load_store, load_store->get_ptr_type()));
}
#endif
if (can_move_pre_barrier()) {
pre_barrier(false /* do_load */,
control(), NULL, NULL, max_juint, NULL, NULL,
load_store /* pre_val */,
T_OBJECT);
}
}
insert_mem_bar(Op_MemBarCPUOrder);
Node* mb = insert_mem_bar(Op_MemBarAcquire, access);
MemBarNode::set_load_store_pair(leading_membar->as_MemBar(), mb->as_MemBar());
assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
set_result(load_store);
return true;
}
bool LibraryCallKit::inline_unsafe_ordered_store(BasicType type) {
if (callee()->is_static()) return false; // caller must have the capability!
#ifndef PRODUCT
{
ResourceMark rm;
ciSignature* sig = callee()->signature();
#ifdef ASSERT
BasicType rtype = sig->return_type()->basic_type();
assert(rtype == T_VOID, "must return void");
assert(sig->count() == 3, "has 3 arguments");
assert(sig->type_at(0)->basic_type() == T_OBJECT, "base is object");
assert(sig->type_at(1)->basic_type() == T_LONG, "offset is long");
#endif // ASSERT
}
#endif //PRODUCT
C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
Node* receiver = argument(0); // type: oop
Node* base = argument(1); // type: oop
Node* offset = argument(2); // type: long
Node* val = argument(4); // type: oop, int, or long
receiver = null_check(receiver);
if (stopped()) {
return true;
}
assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
offset = ConvL2X(offset);
Node* adr = make_unsafe_address(base, offset);
const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
const Type *value_type = Type::get_const_basic_type(type);
Compile::AliasType* alias_type = C->alias_type(adr_type);
insert_mem_bar(Op_MemBarRelease);
insert_mem_bar(Op_MemBarCPUOrder);
const bool require_atomic_access = true;
Node* store;
if (type == T_OBJECT) // reference stores need a store barrier.
store = store_oop_to_unknown(control(), base, adr, adr_type, val, type, MemNode::release);
else {
store = store_to_memory(control(), adr, val, type, adr_type, MemNode::release, require_atomic_access);
}
insert_mem_bar(Op_MemBarCPUOrder);
return true;
}
bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
insert_mem_bar(Op_MemBarCPUOrder);
switch(id) {
case vmIntrinsics::_loadFence:
insert_mem_bar(Op_LoadFence);
return true;
case vmIntrinsics::_storeFence:
insert_mem_bar(Op_StoreFence);
return true;
case vmIntrinsics::_fullFence:
insert_mem_bar(Op_MemBarVolatile);
return true;
default:
fatal_unexpected_iid(id);
return false;
}
}
bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
if (!kls->is_Con()) {
return true;
}
const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
if (klsptr == NULL) {
return true;
}
ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
return !ik->is_initialized();
}
bool LibraryCallKit::inline_unsafe_allocate() {
if (callee()->is_static()) return false; // caller must have the capability!
null_check_receiver(); // null-check, then ignore
Node* cls = null_check(argument(1));
if (stopped()) return true;
Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
kls = null_check(kls);
if (stopped()) return true; // argument was like int.class
Node* test = NULL;
if (LibraryCallKit::klass_needs_init_guard(kls)) {
Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
Node* bits = intcon(InstanceKlass::fully_initialized);
test = _gvn.transform(new (C) SubINode(inst, bits));
}
Node* obj = new_instance(kls, test);
set_result(obj);
return true;
}
#ifdef JFR_HAVE_INTRINSICS
bool LibraryCallKit::inline_native_classID() {
Node* cls = null_check(argument(0), T_OBJECT);
Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
kls = null_check(kls, T_OBJECT);
ByteSize offset = KLASS_TRACE_ID_OFFSET;
Node* insp = basic_plus_adr(kls, in_bytes(offset));
Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
Node* clsused = longcon(0x01l); // set the class bit
Node* orl = _gvn.transform(new (C) OrLNode(tvalue, clsused));
const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
#ifdef TRACE_ID_META_BITS
Node* mbits = longcon(~TRACE_ID_META_BITS);
tvalue = _gvn.transform(new (C) AndLNode(tvalue, mbits));
#endif
#ifdef TRACE_ID_SHIFT
Node* cbits = intcon(TRACE_ID_SHIFT);
tvalue = _gvn.transform(new (C) URShiftLNode(tvalue, cbits));
#endif
set_result(tvalue);
return true;
}
bool LibraryCallKit::inline_native_getEventWriter() {
Node* tls_ptr = _gvn.transform(new (C) ThreadLocalNode());
Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR)
);
Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
Node* jobj_cmp_null = _gvn.transform( new (C) CmpPNode(jobj, null()) );
Node* test_jobj_eq_null = _gvn.transform( new (C) BoolNode(jobj_cmp_null, BoolTest::eq) );
IfNode* iff_jobj_null =
create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
enum { _normal_path = 1,
_null_path = 2,
PATH_LIMIT };
RegionNode* result_rgn = new (C) RegionNode(PATH_LIMIT);
PhiNode* result_val = new (C) PhiNode(result_rgn, TypePtr::BOTTOM);
Node* jobj_is_null = _gvn.transform(new (C) IfTrueNode(iff_jobj_null));
result_rgn->init_req(_null_path, jobj_is_null);
result_val->init_req(_null_path, null());
Node* jobj_is_not_null = _gvn.transform(new (C) IfFalseNode(iff_jobj_null));
result_rgn->init_req(_normal_path, jobj_is_not_null);
Node* res = make_load(jobj_is_not_null, jobj, TypeInstPtr::NOTNULL, T_OBJECT, MemNode::unordered);
result_val->init_req(_normal_path, res);
set_result(result_rgn, result_val);
return true;
}
#endif // JFR_HAVE_INTRINSICS
bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
const TypeFunc* tf = OptoRuntime::void_long_Type();
const TypePtr* no_memory_effects = NULL;
Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
Node* value = _gvn.transform(new (C) ProjNode(time, TypeFunc::Parms+0));
#ifdef ASSERT
Node* value_top = _gvn.transform(new (C) ProjNode(time, TypeFunc::Parms+1));
assert(value_top == top(), "second value must be top");
#endif
set_result(value);
return true;
}
bool LibraryCallKit::inline_native_currentThread() {
Node* junk = NULL;
set_result(generate_current_thread(junk));
return true;
}
bool LibraryCallKit::inline_native_isInterrupted() {
enum {
no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
PATH_LIMIT
};
insert_mem_bar(Op_MemBarCPUOrder);
RegionNode* result_rgn = new (C) RegionNode(PATH_LIMIT);
PhiNode* result_val = new (C) PhiNode(result_rgn, TypeInt::BOOL);
RegionNode* slow_region = new (C) RegionNode(1);
record_for_igvn(slow_region);
Node* rec_thr = argument(0);
Node* tls_ptr = NULL;
Node* cur_thr = generate_current_thread(tls_ptr);
Node* cmp_thr = _gvn.transform(new (C) CmpPNode(cur_thr, rec_thr));
Node* bol_thr = _gvn.transform(new (C) BoolNode(cmp_thr, BoolTest::ne));
generate_slow_guard(bol_thr, slow_region);
Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
Node* cmp_bit = _gvn.transform(new (C) CmpINode(int_bit, intcon(0)));
Node* bol_bit = _gvn.transform(new (C) BoolNode(cmp_bit, BoolTest::ne));
IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
Node* false_bit = _gvn.transform(new (C) IfFalseNode(iff_bit));
result_rgn->init_req(no_int_result_path, false_bit);
result_val->init_req(no_int_result_path, intcon(0));
set_control( _gvn.transform(new (C) IfTrueNode(iff_bit)));
#ifndef TARGET_OS_FAMILY_windows
Node* clr_arg = argument(1);
Node* cmp_arg = _gvn.transform(new (C) CmpINode(clr_arg, intcon(0)));
Node* bol_arg = _gvn.transform(new (C) BoolNode(cmp_arg, BoolTest::ne));
IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
Node* false_arg = _gvn.transform(new (C) IfFalseNode(iff_arg));
result_rgn->init_req(no_clear_result_path, false_arg);
result_val->init_req(no_clear_result_path, intcon(1));
set_control( _gvn.transform(new (C) IfTrueNode(iff_arg)));
#else
#endif // TARGET_OS_FAMILY_windows
slow_region->add_req(control());
set_control( _gvn.transform(slow_region));
if (stopped()) {
result_rgn->init_req(slow_result_path, top());
result_val->init_req(slow_result_path, top());
} else {
CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
Node* slow_val = set_results_for_java_call(slow_call);
Node* fast_io = slow_call->in(TypeFunc::I_O);
Node* fast_mem = slow_call->in(TypeFunc::Memory);
PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
result_rgn->init_req(slow_result_path, control());
result_io ->init_req(slow_result_path, i_o());
result_mem->init_req(slow_result_path, reset_memory());
result_val->init_req(slow_result_path, slow_val);
set_all_memory(_gvn.transform(result_mem));
set_i_o( _gvn.transform(result_io));
}
C->set_has_split_ifs(true); // Has chance for split-if optimization
set_result(result_rgn, result_val);
return true;
}
Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
}
Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
bool never_see_null,
RegionNode* region,
int null_path,
int offset) {
if (region == NULL) never_see_null = true;
Node* p = basic_plus_adr(mirror, offset);
const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
Node* null_ctl = top();
kls = null_check_oop(kls, &null_ctl, never_see_null);
if (region != NULL) {
region->init_req(null_path, null_ctl);
} else {
assert(null_ctl == top(), "no loose ends");
}
return kls;
}
Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
Node* mask = intcon(modifier_mask);
Node* bits = intcon(modifier_bits);
Node* mbit = _gvn.transform(new (C) AndINode(mods, mask));
Node* cmp = _gvn.transform(new (C) CmpINode(mbit, bits));
Node* bol = _gvn.transform(new (C) BoolNode(cmp, BoolTest::ne));
return generate_fair_guard(bol, region);
}
Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
}
bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
const Type* return_type = TypeInt::BOOL;
Node* prim_return_value = top(); // what happens if it's a primitive class?
bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
bool expect_prim = false; // most of these guys expect to work on refs
enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
Node* mirror = argument(0);
Node* obj = top();
switch (id) {
case vmIntrinsics::_isInstance:
prim_return_value = intcon(0);
obj = argument(1);
break;
case vmIntrinsics::_getModifiers:
prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
break;
case vmIntrinsics::_isInterface:
prim_return_value = intcon(0);
break;
case vmIntrinsics::_isArray:
prim_return_value = intcon(0);
expect_prim = true; // cf. ObjectStreamClass.getClassSignature
break;
case vmIntrinsics::_isPrimitive:
prim_return_value = intcon(1);
expect_prim = true; // obviously
break;
case vmIntrinsics::_getSuperclass:
prim_return_value = null();
return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
break;
case vmIntrinsics::_getComponentType:
prim_return_value = null();
return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
break;
case vmIntrinsics::_getClassAccessFlags:
prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
return_type = TypeInt::INT; // not bool! 6297094
break;
default:
fatal_unexpected_iid(id);
break;
}
const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
if (mirror_con == NULL) return false; // cannot happen?
#ifndef PRODUCT
if (C->print_intrinsics() || C->print_inlining()) {
ciType* k = mirror_con->java_mirror_type();
if (k) {
tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
k->print_name();
tty->cr();
}
}
#endif
RegionNode* region = new (C) RegionNode(PATH_LIMIT);
record_for_igvn(region);
PhiNode* phi = new (C) PhiNode(region, return_type);
mirror = null_check(mirror);
if (stopped()) return true;
if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
phi->init_req(_prim_path, prim_return_value);
if (stopped()) { set_result(region, phi); return true; }
bool safe_for_replace = (region->in(_prim_path) == top());
Node* p; // handy temp
Node* null_ctl;
Node* query_value = top();
switch (id) {
case vmIntrinsics::_isInstance:
query_value = gen_instanceof(obj, kls, safe_for_replace);
break;
case vmIntrinsics::_getModifiers:
p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
break;
case vmIntrinsics::_isInterface:
if (generate_interface_guard(kls, region) != NULL)
phi->add_req(intcon(1));
query_value = intcon(0);
break;
case vmIntrinsics::_isArray:
if (generate_array_guard(kls, region) != NULL)
phi->add_req(intcon(1));
query_value = intcon(0);
break;
case vmIntrinsics::_isPrimitive:
query_value = intcon(0); // "normal" path produces false
break;
case vmIntrinsics::_getSuperclass:
if (generate_interface_guard(kls, region) != NULL)
phi->add_req(null());
if (generate_array_guard(kls, region) != NULL)
phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
null_ctl = top();
kls = null_check_oop(kls, &null_ctl);
if (null_ctl != top()) {
region->add_req(null_ctl);
phi ->add_req(null());
}
if (!stopped()) {
query_value = load_mirror_from_klass(kls);
}
break;
case vmIntrinsics::_getComponentType:
if (generate_array_guard(kls, region) != NULL) {
Node* is_array_ctrl = region->in(region->req()-1);
Node* cma = basic_plus_adr(kls, in_bytes(ArrayKlass::component_mirror_offset()));
Node* cmo = make_load(is_array_ctrl, cma, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
phi->add_req(cmo);
}
query_value = null(); // non-array case is null
break;
case vmIntrinsics::_getClassAccessFlags:
p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
break;
default:
fatal_unexpected_iid(id);
break;
}
phi->init_req(1, query_value);
region->init_req(1, control());
C->set_has_split_ifs(true); // Has chance for split-if optimization
set_result(region, phi);
return true;
}
bool LibraryCallKit::inline_native_subtype_check() {
Node* args[2]; // two java.lang.Class mirrors: superc, subc
args[0] = argument(0);
args[1] = argument(1);
Node* klasses[2]; // corresponding Klasses: superk, subk
klasses[0] = klasses[1] = top();
enum {
_prim_0_path = 1, // {P,N} => false
_prim_same_path, // {P,P} & superc==subc => true
_prim_1_path, // {N,P} => false
_ref_subtype_path, // {N,N} & subtype check wins => true
_both_ref_path, // {N,N} & subtype check loses => false
PATH_LIMIT
};
RegionNode* region = new (C) RegionNode(PATH_LIMIT);
Node* phi = new (C) PhiNode(region, TypeInt::BOOL);
record_for_igvn(region);
const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
int which_arg;
for (which_arg = 0; which_arg <= 1; which_arg++) {
Node* arg = args[which_arg];
arg = null_check(arg);
if (stopped()) break;
args[which_arg] = arg;
Node* p = basic_plus_adr(arg, class_klass_offset);
Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
klasses[which_arg] = _gvn.transform(kls);
}
bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
for (which_arg = 0; which_arg <= 1; which_arg++) {
Node* kls = klasses[which_arg];
Node* null_ctl = top();
kls = null_check_oop(kls, &null_ctl, never_see_null);
int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
region->init_req(prim_path, null_ctl);
if (stopped()) break;
klasses[which_arg] = kls;
}
if (!stopped()) {
Node* subk = klasses[1]; // the argument to isAssignableFrom
Node* superk = klasses[0]; // the receiver
region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
region->set_req(_ref_subtype_path, control());
}
set_control(region->in(_prim_0_path)); // go back to first null check
if (!stopped()) {
Node* cmp_eq = _gvn.transform(new (C) CmpPNode(args[0], args[1]));
Node* bol_eq = _gvn.transform(new (C) BoolNode(cmp_eq, BoolTest::eq));
generate_guard(bol_eq, region, PROB_FAIR);
if (region->req() == PATH_LIMIT+1) {
region->swap_edges(PATH_LIMIT, _prim_same_path);
region->del_req(PATH_LIMIT);
}
region->set_req(_prim_0_path, control()); // Not equal after all.
}
phi->set_req(_prim_same_path, intcon(1));
phi->set_req(_ref_subtype_path, intcon(1));
assert(region->req() == PATH_LIMIT, "sane region");
for (uint i = 1; i < region->req(); i++) {
Node* ctl = region->in(i);
if (ctl == NULL || ctl == top()) {
region->set_req(i, top());
phi ->set_req(i, top());
} else if (phi->in(i) == NULL) {
phi->set_req(i, intcon(0)); // all other paths produce 'false'
}
}
set_control(_gvn.transform(region));
set_result(_gvn.transform(phi));
return true;
}
Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
bool obj_array, bool not_array) {
jint layout_con = 0;
Node* layout_val = get_layout_helper(kls, layout_con);
if (layout_val == NULL) {
bool query = (obj_array
? Klass::layout_helper_is_objArray(layout_con)
: Klass::layout_helper_is_array(layout_con));
if (query == not_array) {
return NULL; // never a branch
} else { // always a branch
Node* always_branch = control();
if (region != NULL)
region->add_req(always_branch);
set_control(top());
return always_branch;
}
}
jint nval = (obj_array
? (jint)(Klass::_lh_array_tag_type_value
<< Klass::_lh_array_tag_shift)
: Klass::_lh_neutral_value);
Node* cmp = _gvn.transform(new(C) CmpINode(layout_val, intcon(nval)));
BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
if (not_array) btest = BoolTest(btest).negate();
Node* bol = _gvn.transform(new(C) BoolNode(cmp, btest));
return generate_fair_guard(bol, region);
}
bool LibraryCallKit::inline_native_newArray() {
Node* mirror = argument(0);
Node* count_val = argument(1);
mirror = null_check(mirror);
if (stopped()) return true;
enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
RegionNode* result_reg = new(C) RegionNode(PATH_LIMIT);
PhiNode* result_val = new(C) PhiNode(result_reg,
TypeInstPtr::NOTNULL);
PhiNode* result_io = new(C) PhiNode(result_reg, Type::ABIO);
PhiNode* result_mem = new(C) PhiNode(result_reg, Type::MEMORY,
TypePtr::BOTTOM);
bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
result_reg, _slow_path);
Node* normal_ctl = control();
Node* no_array_ctl = result_reg->in(_slow_path);
set_control(no_array_ctl);
if (!stopped()) {
PreserveJVMState pjvms(this);
CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
Node* slow_result = set_results_for_java_call(slow_call);
result_reg->set_req(_slow_path, control());
result_val->set_req(_slow_path, slow_result);
result_io ->set_req(_slow_path, i_o());
result_mem->set_req(_slow_path, reset_memory());
}
set_control(normal_ctl);
if (!stopped()) {
Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
result_reg->init_req(_normal_path, control());
result_val->init_req(_normal_path, obj);
result_io ->init_req(_normal_path, i_o());
result_mem->init_req(_normal_path, reset_memory());
}
set_i_o( _gvn.transform(result_io) );
set_all_memory( _gvn.transform(result_mem));
C->set_has_split_ifs(true); // Has chance for split-if optimization
set_result(result_reg, result_val);
return true;
}
bool LibraryCallKit::inline_native_getLength() {
if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
Node* array = null_check(argument(0));
if (stopped()) return true;
Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
if (non_array != NULL) {
PreserveJVMState pjvms(this);
set_control(non_array);
uncommon_trap(Deoptimization::Reason_intrinsic,
Deoptimization::Action_maybe_recompile);
}
if (stopped()) return true;
Node* result = load_array_length(array);
C->set_has_split_ifs(true); // Has chance for split-if optimization
set_result(result);
return true;
}
bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
Node* original = argument(0);
Node* start = is_copyOfRange? argument(1): intcon(0);
Node* end = is_copyOfRange? argument(2): argument(1);
Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
Node* newcopy = NULL;
{ PreserveReexecuteState preexecs(this);
jvms()->set_should_reexecute(true);
array_type_mirror = null_check(array_type_mirror);
original = null_check(original);
if (stopped()) return true;
Node* orig_length = load_array_length(original);
Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
klass_node = null_check(klass_node);
RegionNode* bailout = new (C) RegionNode(1);
record_for_igvn(bailout);
Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
if (not_objArray != NULL) {
ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
Node* cast = new (C) CastPPNode(klass_node, akls);
cast->init_req(0, control());
klass_node = _gvn.transform(cast);
}
generate_negative_guard(start, bailout, &start);
generate_negative_guard(end, bailout, &end);
Node* length = end;
if (_gvn.type(start) != TypeInt::ZERO) {
length = _gvn.transform(new (C) SubINode(end, start));
}
generate_negative_guard(length, bailout, &length);
if (bailout->req() > 1) {
PreserveJVMState pjvms(this);
set_control(_gvn.transform(bailout));
uncommon_trap(Deoptimization::Reason_intrinsic,
Deoptimization::Action_maybe_recompile);
}
if (!stopped()) {
Node* orig_tail = _gvn.transform(new (C) SubINode(orig_length, start));
Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
newcopy = new_array(klass_node, length, 0); // no argments to push
bool disjoint_bases = true;
bool length_never_negative = !is_copyOfRange;
generate_arraycopy(TypeAryPtr::OOPS, T_OBJECT,
original, start, newcopy, intcon(0), moved,
disjoint_bases, length_never_negative);
}
} // original reexecute is set back here
C->set_has_split_ifs(true); // Has chance for split-if optimization
if (!stopped()) {
set_result(newcopy);
}
return true;
}
Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
RegionNode* slow_region) {
ciMethod* method = callee();
int vtable_index = method->vtable_index();
assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
err_msg_res("bad index %d", vtable_index));
int entry_offset = (InstanceKlass::vtable_start_offset() +
vtable_index*vtableEntry::size()) * wordSize +
vtableEntry::method_offset_in_bytes();
Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
Node* native_call = makecon(native_call_addr);
Node* chk_native = _gvn.transform(new(C) CmpPNode(target_call, native_call));
Node* test_native = _gvn.transform(new(C) BoolNode(chk_native, BoolTest::ne));
return generate_slow_guard(test_native, slow_region);
}
CallJavaNode*
LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
guarantee(callee() != C->method(), "cannot make slow-call to self");
ciMethod* method = callee();
guarantee(method_id == method->intrinsic_id(), "must match");
const TypeFunc* tf = TypeFunc::make(method);
CallJavaNode* slow_call;
if (is_static) {
assert(!is_virtual, "");
slow_call = new(C) CallStaticJavaNode(C, tf,
SharedRuntime::get_resolve_static_call_stub(),
method, bci());
} else if (is_virtual) {
null_check_receiver();
int vtable_index = Method::invalid_vtable_index;
if (UseInlineCaches) {
} else {
vtable_index = method->vtable_index();
assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
err_msg_res("bad index %d", vtable_index));
}
slow_call = new(C) CallDynamicJavaNode(tf,
SharedRuntime::get_resolve_virtual_call_stub(),
method, vtable_index, bci());
} else { // neither virtual nor static: opt_virtual
null_check_receiver();
slow_call = new(C) CallStaticJavaNode(C, tf,
SharedRuntime::get_resolve_opt_virtual_call_stub(),
method, bci());
slow_call->set_optimized_virtual(true);
}
set_arguments_for_java_call(slow_call);
set_edges_for_java_call(slow_call);
return slow_call;
}
bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
assert(is_static == callee()->is_static(), "correct intrinsic selection");
assert(!(is_virtual && is_static), "either virtual, special, or static");
enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
RegionNode* result_reg = new(C) RegionNode(PATH_LIMIT);
PhiNode* result_val = new(C) PhiNode(result_reg, TypeInt::INT);
PhiNode* result_io = new(C) PhiNode(result_reg, Type::ABIO);
PhiNode* result_mem = new(C) PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
Node* obj = NULL;
if (!is_static) {
obj = null_check_receiver();
if (stopped()) return true; // unconditionally null
result_reg->init_req(_null_path, top());
result_val->init_req(_null_path, top());
} else {
obj = argument(0);
Node* null_ctl = top();
obj = null_check_oop(obj, &null_ctl);
result_reg->init_req(_null_path, null_ctl);
result_val->init_req(_null_path, _gvn.intcon(0));
}
if (stopped()) {
set_control( result_reg->in(_null_path));
if (!stopped())
set_result(result_val->in(_null_path));
return true;
}
RegionNode* slow_region = new (C) RegionNode(1);
record_for_igvn(slow_region);
if (is_virtual) {
Node* obj_klass = load_object_klass(obj);
generate_virtual_guard(obj_klass, slow_region);
}
Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
Node* no_ctrl = NULL;
Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
Node* lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
Node* lmasked_header = _gvn.transform(new (C) AndXNode(header, lock_mask));
Node* unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
Node* chk_unlocked = _gvn.transform(new (C) CmpXNode( lmasked_header, unlocked_val));
Node* test_unlocked = _gvn.transform(new (C) BoolNode( chk_unlocked, BoolTest::ne));
generate_slow_guard(test_unlocked, slow_region);
Node* hash_mask = _gvn.intcon(markOopDesc::hash_mask);
Node* hash_shift = _gvn.intcon(markOopDesc::hash_shift);
Node* hshifted_header= _gvn.transform(new (C) URShiftXNode(header, hash_shift));
hshifted_header = ConvX2I(hshifted_header);
Node* hash_val = _gvn.transform(new (C) AndINode(hshifted_header, hash_mask));
Node* no_hash_val = _gvn.intcon(markOopDesc::no_hash);
Node* chk_assigned = _gvn.transform(new (C) CmpINode( hash_val, no_hash_val));
Node* test_assigned = _gvn.transform(new (C) BoolNode( chk_assigned, BoolTest::eq));
generate_slow_guard(test_assigned, slow_region);
Node* init_mem = reset_memory();
result_io ->init_req(_null_path, i_o());
result_mem->init_req(_null_path, init_mem);
result_val->init_req(_fast_path, hash_val);
result_reg->init_req(_fast_path, control());
result_io ->init_req(_fast_path, i_o());
result_mem->init_req(_fast_path, init_mem);
set_control(_gvn.transform(slow_region));
if (!stopped()) {
set_all_memory(init_mem);
vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
Node* slow_result = set_results_for_java_call(slow_call);
result_reg->init_req(_slow_path, control());
result_val->init_req(_slow_path, slow_result);
result_io ->set_req(_slow_path, i_o());
result_mem ->set_req(_slow_path, reset_memory());
}
set_i_o( _gvn.transform(result_io) );
set_all_memory( _gvn.transform(result_mem));
set_result(result_reg, result_val);
return true;
}
bool LibraryCallKit::inline_native_getClass() {
Node* obj = null_check_receiver();
if (stopped()) return true;
set_result(load_mirror_from_klass(load_object_klass(obj)));
return true;
}
bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
#ifndef PRODUCT
if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
}
#endif
if (!jvms()->has_method()) {
#ifndef PRODUCT
if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
tty->print_cr(" Bailing out because intrinsic was inlined at top level");
}
#endif
return false;
}
JVMState* caller_jvms = jvms();
for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
ciMethod* m = caller_jvms->method();
switch (n) {
case 0:
fatal("current JVM state does not include the Reflection.getCallerClass frame");
break;
case 1:
if (!m->caller_sensitive()) {
#ifndef PRODUCT
if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
}
#endif
return false; // bail-out; let JVM_GetCallerClass do the work
}
break;
default:
if (!m->is_ignored_by_security_stack_walk()) {
ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
ciInstance* caller_mirror = caller_klass->java_mirror();
set_result(makecon(TypeInstPtr::make(caller_mirror)));
#ifndef PRODUCT
if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
tty->print_cr(" Succeeded: caller = %d) %s.%s, JVMS depth = %d", n, caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), jvms()->depth());
tty->print_cr(" JVM state at this point:");
for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
ciMethod* m = jvms()->of_depth(i)->method();
tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
}
}
#endif
return true;
}
break;
}
}
#ifndef PRODUCT
if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
tty->print_cr(" JVM state at this point:");
for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
ciMethod* m = jvms()->of_depth(i)->method();
tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
}
}
#endif
return false; // bail-out; let JVM_GetCallerClass do the work
}
bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
Node* arg = argument(0);
Node* result = NULL;
switch (id) {
case vmIntrinsics::_floatToRawIntBits: result = new (C) MoveF2INode(arg); break;
case vmIntrinsics::_intBitsToFloat: result = new (C) MoveI2FNode(arg); break;
case vmIntrinsics::_doubleToRawLongBits: result = new (C) MoveD2LNode(arg); break;
case vmIntrinsics::_longBitsToDouble: result = new (C) MoveL2DNode(arg); break;
case vmIntrinsics::_doubleToLongBits: {
RegionNode *r = new (C) RegionNode(3);
Node *phi = new (C) PhiNode(r, TypeLong::LONG);
Node *cmpisnan = _gvn.transform(new (C) CmpDNode(arg, arg));
Node *bolisnan = _gvn.transform(new (C) BoolNode(cmpisnan, BoolTest::ne));
IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
Node *opt_isnan = _gvn.transform(ifisnan);
assert( opt_isnan->is_If(), "Expect an IfNode");
IfNode *opt_ifisnan = (IfNode*)opt_isnan;
Node *iftrue = _gvn.transform(new (C) IfTrueNode(opt_ifisnan));
set_control(iftrue);
static const jlong nan_bits = CONST64(0x7ff8000000000000);
Node *slow_result = longcon(nan_bits); // return NaN
phi->init_req(1, _gvn.transform( slow_result ));
r->init_req(1, iftrue);
Node *iffalse = _gvn.transform(new (C) IfFalseNode(opt_ifisnan));
set_control(iffalse);
phi->init_req(2, _gvn.transform(new (C) MoveD2LNode(arg)));
r->init_req(2, iffalse);
set_control(_gvn.transform(r));
record_for_igvn(r);
C->set_has_split_ifs(true); // Has chance for split-if optimization
result = phi;
assert(result->bottom_type()->isa_long(), "must be");
break;
}
case vmIntrinsics::_floatToIntBits: {
RegionNode *r = new (C) RegionNode(3);
Node *phi = new (C) PhiNode(r, TypeInt::INT);
Node *cmpisnan = _gvn.transform(new (C) CmpFNode(arg, arg));
Node *bolisnan = _gvn.transform(new (C) BoolNode(cmpisnan, BoolTest::ne));
IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
Node *opt_isnan = _gvn.transform(ifisnan);
assert( opt_isnan->is_If(), "Expect an IfNode");
IfNode *opt_ifisnan = (IfNode*)opt_isnan;
Node *iftrue = _gvn.transform(new (C) IfTrueNode(opt_ifisnan));
set_control(iftrue);
static const jint nan_bits = 0x7fc00000;
Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
phi->init_req(1, _gvn.transform( slow_result ));
r->init_req(1, iftrue);
Node *iffalse = _gvn.transform(new (C) IfFalseNode(opt_ifisnan));
set_control(iffalse);
phi->init_req(2, _gvn.transform(new (C) MoveF2INode(arg)));
r->init_req(2, iffalse);
set_control(_gvn.transform(r));
record_for_igvn(r);
C->set_has_split_ifs(true); // Has chance for split-if optimization
result = phi;
assert(result->bottom_type()->isa_int(), "must be");
break;
}
default:
fatal_unexpected_iid(id);
break;
}
set_result(_gvn.transform(result));
return true;
}
#ifdef _LP64
#define XTOP ,top() /*additional argument*/
#else //_LP64
#define XTOP /*no additional argument*/
#endif //_LP64
bool LibraryCallKit::inline_unsafe_copyMemory() {
if (callee()->is_static()) return false; // caller must have the capability!
null_check_receiver(); // null-check receiver
if (stopped()) return true;
C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
Node* src_ptr = argument(1); // type: oop
Node* src_off = ConvL2X(argument(2)); // type: long
Node* dst_ptr = argument(4); // type: oop
Node* dst_off = ConvL2X(argument(5)); // type: long
Node* size = ConvL2X(argument(7)); // type: long
assert(Unsafe_field_offset_to_byte_offset(11) == 11,
"fieldOffset must be byte-scaled");
Node* src = make_unsafe_address(src_ptr, src_off);
Node* dst = make_unsafe_address(dst_ptr, dst_off);
insert_mem_bar(Op_MemBarCPUOrder);
make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::fast_arraycopy_Type(),
StubRoutines::unsafe_arraycopy(),
"unsafe_arraycopy",
TypeRawPtr::BOTTOM,
src, dst, size XTOP);
insert_mem_bar(Op_MemBarCPUOrder);
return true;
}
void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
assert(obj_size != NULL, "");
Node* raw_obj = alloc_obj->in(1);
assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
AllocateNode* alloc = NULL;
if (ReduceBulkZeroing) {
alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
alloc->initialization()->set_complete_with_arraycopy();
}
Node* src = obj;
Node* dest = alloc_obj;
Node* size = _gvn.transform(obj_size);
int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
instanceOopDesc::base_offset_in_bytes();
if (base_off % BytesPerLong != 0) {
assert(UseCompressedClassPointers, "");
if (is_array) {
base_off += sizeof(int);
} else {
base_off = instanceOopDesc::klass_offset_in_bytes();
}
assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
}
src = basic_plus_adr(src, base_off);
dest = basic_plus_adr(dest, base_off);
Node* countx = size;
countx = _gvn.transform(new (C) SubXNode(countx, MakeConX(base_off)));
countx = _gvn.transform(new (C) URShiftXNode(countx, intcon(LogBytesPerLong) ));
const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
bool disjoint_bases = true;
generate_unchecked_arraycopy(raw_adr_type, T_LONG, disjoint_bases,
src, NULL, dest, NULL, countx,
if (card_mark) {
assert(!is_array, "");
Node* no_particular_value = NULL;
Node* no_particular_field = NULL;
int raw_adr_idx = Compile::AliasIdxRaw;
post_barrier(control(),
memory(raw_adr_type),
alloc_obj,
no_particular_field,
raw_adr_idx,
no_particular_value,
T_OBJECT,
false);
}
if (alloc != NULL) {
insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
} else {
insert_mem_bar(Op_MemBarCPUOrder);
}
}
bool LibraryCallKit::inline_native_clone(bool is_virtual) {
PhiNode* result_val;
{ PreserveReexecuteState preexecs(this);
jvms()->set_should_reexecute(true);
Node* obj = null_check_receiver();
if (stopped()) return true;
Node* obj_klass = load_object_klass(obj);
const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
const TypeOopPtr* toop = ((tklass != NULL)
? tklass->as_instance_type()
: TypeInstPtr::NOTNULL);
insert_mem_bar(Op_MemBarCPUOrder);
enum {
_slow_path = 1, // out-of-line call to clone method (virtual or not)
_objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
_array_path, // plain array allocation, plus arrayof_long_arraycopy
_instance_path, // plain instance allocation, plus arrayof_long_arraycopy
PATH_LIMIT
};
RegionNode* result_reg = new(C) RegionNode(PATH_LIMIT);
result_val = new(C) PhiNode(result_reg,
TypeInstPtr::NOTNULL);
PhiNode* result_i_o = new(C) PhiNode(result_reg, Type::ABIO);
PhiNode* result_mem = new(C) PhiNode(result_reg, Type::MEMORY,
TypePtr::BOTTOM);
record_for_igvn(result_reg);
const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
int raw_adr_idx = Compile::AliasIdxRaw;
Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
if (array_ctl != NULL) {
PreserveJVMState pjvms(this);
set_control(array_ctl);
Node* obj_length = load_array_length(obj);
Node* obj_size = NULL;
Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
if (!use_ReduceInitialCardMarks()) {
Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
if (is_obja != NULL) {
PreserveJVMState pjvms2(this);
set_control(is_obja);
bool disjoint_bases = true;
bool length_never_negative = true;
generate_arraycopy(TypeAryPtr::OOPS, T_OBJECT,
obj, intcon(0), alloc_obj, intcon(0),
obj_length,
disjoint_bases, length_never_negative);
result_reg->init_req(_objArray_path, control());
result_val->init_req(_objArray_path, alloc_obj);
result_i_o ->set_req(_objArray_path, i_o());
result_mem ->set_req(_objArray_path, reset_memory());
}
}
if (!stopped()) {
copy_to_clone(obj, alloc_obj, obj_size, true, false);
result_reg->init_req(_array_path, control());
result_val->init_req(_array_path, alloc_obj);
result_i_o ->set_req(_array_path, i_o());
result_mem ->set_req(_array_path, reset_memory());
}
}
RegionNode* slow_region = new (C) RegionNode(1);
record_for_igvn(slow_region);
if (!stopped()) {
if (is_virtual) {
generate_virtual_guard(obj_klass, slow_region);
}
generate_access_flags_guard(obj_klass,
JVM_ACC_IS_CLONEABLE | JVM_ACC_HAS_FINALIZER,
JVM_ACC_IS_CLONEABLE,
slow_region);
}
if (!stopped()) {
PreserveJVMState pjvms(this);
Node* obj_size = NULL;
Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
result_reg->init_req(_instance_path, control());
result_val->init_req(_instance_path, alloc_obj);
result_i_o ->set_req(_instance_path, i_o());
result_mem ->set_req(_instance_path, reset_memory());
}
set_control(_gvn.transform(slow_region));
if (!stopped()) {
PreserveJVMState pjvms(this);
CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
Node* slow_result = set_results_for_java_call(slow_call);
result_reg->init_req(_slow_path, control());
result_val->init_req(_slow_path, slow_result);
result_i_o ->set_req(_slow_path, i_o());
result_mem ->set_req(_slow_path, reset_memory());
}
set_control( _gvn.transform(result_reg));
set_i_o( _gvn.transform(result_i_o));
set_all_memory( _gvn.transform(result_mem));
} // original reexecute is set back here
set_result(_gvn.transform(result_val));
return true;
}
address LibraryCallKit::basictype2arraycopy(BasicType t,
Node* src_offset,
Node* dest_offset,
bool disjoint_bases,
const char* &name,
bool dest_uninitialized) {
const TypeInt* src_offset_inttype = gvn().find_int_type(src_offset);;
const TypeInt* dest_offset_inttype = gvn().find_int_type(dest_offset);;
bool aligned = false;
bool disjoint = disjoint_bases;
if (src_offset_inttype != NULL && src_offset_inttype->is_con() &&
dest_offset_inttype != NULL && dest_offset_inttype->is_con()) {
int s_offs = src_offset_inttype->get_con();
int d_offs = dest_offset_inttype->get_con();
int element_size = type2aelembytes(t);
aligned = ((arrayOopDesc::base_offset_in_bytes(t) + s_offs * element_size) % HeapWordSize == 0) &&
((arrayOopDesc::base_offset_in_bytes(t) + d_offs * element_size) % HeapWordSize == 0);
if (s_offs >= d_offs) disjoint = true;
} else if (src_offset == dest_offset && src_offset != NULL) {
disjoint = true;
}
return StubRoutines::select_arraycopy_function(t, aligned, disjoint, name, dest_uninitialized);
}
bool LibraryCallKit::inline_arraycopy() {
Node* src = argument(0); // type: oop
Node* src_offset = argument(1); // type: int
Node* dest = argument(2); // type: oop
Node* dest_offset = argument(3); // type: int
Node* length = argument(4); // type: int
const Type* src_type = src->Value(&_gvn);
const Type* dest_type = dest->Value(&_gvn);
const TypeAryPtr* top_src = src_type->isa_aryptr();
const TypeAryPtr* top_dest = dest_type->isa_aryptr();
bool has_src = (top_src != NULL && top_src->klass() != NULL);
bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
bool src_spec = false;
bool dest_spec = false;
if (!has_src || !has_dest) {
bool could_have_src = has_src;
bool could_have_dest = has_dest;
ciKlass* src_k = NULL;
if (!has_src) {
src_k = src_type->speculative_type();
if (src_k != NULL && src_k->is_array_klass()) {
could_have_src = true;
}
}
ciKlass* dest_k = NULL;
if (!has_dest) {
dest_k = dest_type->speculative_type();
if (dest_k != NULL && dest_k->is_array_klass()) {
could_have_dest = true;
}
}
if (could_have_src && could_have_dest) {
if (!has_src) {
src = maybe_cast_profiled_obj(src, src_k);
src_type = _gvn.type(src);
top_src = src_type->isa_aryptr();
has_src = (top_src != NULL && top_src->klass() != NULL);
src_spec = true;
}
if (!has_dest) {
dest = maybe_cast_profiled_obj(dest, dest_k);
dest_type = _gvn.type(dest);
top_dest = dest_type->isa_aryptr();
has_dest = (top_dest != NULL && top_dest->klass() != NULL);
dest_spec = true;
}
}
}
if (!has_src || !has_dest) {
insert_mem_bar(Op_MemBarCPUOrder);
generate_arraycopy(TypeRawPtr::BOTTOM, T_CONFLICT,
src, src_offset, dest, dest_offset, length);
if (!InsertMemBarAfterArraycopy)
insert_mem_bar(Op_MemBarCPUOrder);
return true;
}
BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
if (src_elem == T_ARRAY) src_elem = T_OBJECT;
if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
if (src_elem != dest_elem || dest_elem == T_VOID) {
generate_slow_arraycopy(TypePtr::BOTTOM,
src, src_offset, dest, dest_offset, length,
return true;
}
if (src_elem == T_OBJECT) {
bool could_have_src = src_spec;
bool could_have_dest = dest_spec;
ciKlass* src_k = top_src->klass();
ciKlass* dest_k = top_dest->klass();
if (!src_spec) {
src_k = src_type->speculative_type();
if (src_k != NULL && src_k->is_array_klass()) {
could_have_src = true;
}
}
if (!dest_spec) {
dest_k = dest_type->speculative_type();
if (dest_k != NULL && dest_k->is_array_klass()) {
could_have_dest = true;
}
}
if (could_have_src && could_have_dest) {
if (could_have_src && !src_spec) {
src = maybe_cast_profiled_obj(src, src_k);
}
if (could_have_dest && !dest_spec) {
dest = maybe_cast_profiled_obj(dest, dest_k);
}
}
}
RegionNode* slow_region = new (C) RegionNode(1);
record_for_igvn(slow_region);
src = null_check(src, T_ARRAY);
dest = null_check(dest, T_ARRAY);
generate_negative_guard(src_offset, slow_region);
generate_negative_guard(dest_offset, slow_region);
generate_limit_guard(src_offset, length,
load_array_length(src),
slow_region);
generate_limit_guard(dest_offset, length,
load_array_length(dest),
slow_region);
const TypePtr* adr_type = TypeAryPtr::get_array_body_type(dest_elem);
generate_arraycopy(adr_type, dest_elem,
src, src_offset, dest, dest_offset, length,
false, false, slow_region);
return true;
}
void
LibraryCallKit::generate_arraycopy(const TypePtr* adr_type,
BasicType basic_elem_type,
Node* src, Node* src_offset,
Node* dest, Node* dest_offset,
Node* copy_length,
bool disjoint_bases,
bool length_never_negative,
RegionNode* slow_region) {
if (slow_region == NULL) {
slow_region = new(C) RegionNode(1);
record_for_igvn(slow_region);
}
Node* original_dest = dest;
AllocateArrayNode* alloc = NULL; // used for zeroing, if needed
bool dest_uninitialized = false;
if (ReduceBulkZeroing
&& !ZeroTLAB // pointless if already zeroed
&& basic_elem_type != T_CONFLICT // avoid corner case
&& !src->eqv_uncast(dest)
&& ((alloc = tightly_coupled_allocation(dest, slow_region))
!= NULL)
&& _gvn.find_int_con(alloc->in(AllocateNode::ALength), 1) > 0
&& alloc->maybe_set_complete(&_gvn)) {
InitializeNode* init = alloc->initialization();
assert(init->is_complete(), "we just did this");
init->set_complete_with_arraycopy();
assert(dest->is_CheckCastPP(), "sanity");
assert(dest->in(0)->in(0) == init, "dest pinned");
adr_type = TypeRawPtr::BOTTOM; // all initializations are into raw memory
dest_uninitialized = true;
} else {
alloc = NULL;
}
enum { fast_path = 1, // normal void-returning assembly stub
checked_path = 2, // special assembly stub with cleanup
slow_call_path = 3, // something went wrong; call the VM
zero_path = 4, // bypass when length of copy is zero
bcopy_path = 5, // copy primitive array by 64-bit blocks
PATH_LIMIT = 6
};
RegionNode* result_region = new(C) RegionNode(PATH_LIMIT);
PhiNode* result_i_o = new(C) PhiNode(result_region, Type::ABIO);
PhiNode* result_memory = new(C) PhiNode(result_region, Type::MEMORY, adr_type);
record_for_igvn(result_region);
_gvn.set_type_bottom(result_i_o);
_gvn.set_type_bottom(result_memory);
assert(adr_type != TypePtr::BOTTOM, "must be RawMem or a T[] slice");
Node* slow_control;
Node* slow_i_o = i_o();
Node* slow_mem = memory(adr_type);
debug_only(slow_control = (Node*) badAddress);
Node* checked_control = top();
Node* checked_mem = NULL;
Node* checked_i_o = NULL;
Node* checked_value = NULL;
if (basic_elem_type == T_CONFLICT) {
assert(!dest_uninitialized, "");
Node* cv = generate_generic_arraycopy(adr_type,
src, src_offset, dest, dest_offset,
copy_length, dest_uninitialized);
if (cv == NULL) cv = intcon(-1); // failure (no stub available)
checked_control = control();
checked_i_o = i_o();
checked_mem = memory(adr_type);
checked_value = cv;
set_control(top()); // no fast path
}
Node* not_pos = generate_nonpositive_guard(copy_length, length_never_negative);
if (not_pos != NULL) {
PreserveJVMState pjvms(this);
set_control(not_pos);
if (!length_never_negative) {
generate_negative_guard(copy_length, slow_region);
}
if (!stopped() && dest_uninitialized) {
Node* dest_length = alloc->in(AllocateNode::ALength);
if (copy_length->eqv_uncast(dest_length)
|| _gvn.find_int_con(dest_length, 1) <= 0) {
} else {
generate_clear_array(adr_type, dest, basic_elem_type,
intcon(0), NULL,
alloc->in(AllocateNode::AllocSize));
InitializeNode* init = insert_mem_bar_volatile(Op_Initialize,
Compile::AliasIdxRaw,
top())->as_Initialize();
init->set_complete(&_gvn); // (there is no corresponding AllocateNode)
}
}
result_region->init_req(zero_path, control());
result_i_o ->init_req(zero_path, i_o());
result_memory->init_req(zero_path, memory(adr_type));
}
if (!stopped() && dest_uninitialized) {
Node* dest_size = alloc->in(AllocateNode::AllocSize);
Node* dest_length = alloc->in(AllocateNode::ALength);
Node* dest_tail = _gvn.transform(new(C) AddINode(dest_offset,
copy_length));
if (find_int_con(dest_offset, -1) != 0) {
generate_clear_array(adr_type, dest, basic_elem_type,
intcon(0), dest_offset,
NULL);
}
Node* tail_ctl = NULL;
if (!stopped() && !dest_tail->eqv_uncast(dest_length)) {
Node* cmp_lt = _gvn.transform(new(C) CmpINode(dest_tail, dest_length));
Node* bol_lt = _gvn.transform(new(C) BoolNode(cmp_lt, BoolTest::lt));
tail_ctl = generate_slow_guard(bol_lt, NULL);
assert(tail_ctl != NULL || !stopped(), "must be an outcome");
}
if (!stopped() && alloc != NULL && basic_elem_type != T_OBJECT) {
bool didit = false;
{ PreserveJVMState pjvms(this);
didit = generate_block_arraycopy(adr_type, basic_elem_type, alloc,
src, src_offset, dest, dest_offset,
dest_size, dest_uninitialized);
if (didit) {
result_region->init_req(bcopy_path, control());
result_i_o ->init_req(bcopy_path, i_o());
result_memory->init_req(bcopy_path, memory(adr_type));
}
}
if (didit)
set_control(top()); // no regular fast path
}
if (tail_ctl != NULL) {
Node* notail_ctl = stopped() ? NULL : control();
set_control(tail_ctl);
if (notail_ctl == NULL) {
generate_clear_array(adr_type, dest, basic_elem_type,
dest_tail, NULL,
dest_size);
} else {
Node* done_ctl = new(C) RegionNode(3);
Node* done_mem = new(C) PhiNode(done_ctl, Type::MEMORY, adr_type);
done_ctl->init_req(1, notail_ctl);
done_mem->init_req(1, memory(adr_type));
generate_clear_array(adr_type, dest, basic_elem_type,
dest_tail, NULL,
dest_size);
done_ctl->init_req(2, control());
done_mem->init_req(2, memory(adr_type));
set_control( _gvn.transform(done_ctl));
set_memory( _gvn.transform(done_mem), adr_type );
}
}
}
BasicType copy_type = basic_elem_type;
assert(basic_elem_type != T_ARRAY, "caller must fix this");
if (!stopped() && copy_type == T_OBJECT) {
Node* src_klass = load_object_klass(src);
Node* dest_klass = load_object_klass(dest);
Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
if (not_subtype_ctrl != top()) {
PreserveJVMState pjvms(this);
set_control(not_subtype_ctrl);
int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
Node* p1 = basic_plus_adr(dest_klass, ek_offset);
Node* n1 = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p1, TypeRawPtr::BOTTOM);
Node* dest_elem_klass = _gvn.transform(n1);
Node* cv = generate_checkcast_arraycopy(adr_type,
dest_elem_klass,
src, src_offset, dest, dest_offset,
ConvI2X(copy_length), dest_uninitialized);
if (cv == NULL) cv = intcon(-1); // failure (no stub available)
checked_control = control();
checked_i_o = i_o();
checked_mem = memory(adr_type);
checked_value = cv;
}
if (alloc != NULL && use_ReduceInitialCardMarks()) {
copy_type = LP64_ONLY(UseCompressedOops ? T_INT : T_LONG) NOT_LP64(T_INT);
assert(type2aelembytes(basic_elem_type) == type2aelembytes(copy_type),
"sizes agree");
}
}
if (!stopped()) {
PreserveJVMState pjvms(this);
generate_unchecked_arraycopy(adr_type, copy_type, disjoint_bases,
src, src_offset, dest, dest_offset,
ConvI2X(copy_length), dest_uninitialized);
result_region->init_req(fast_path, control());
result_i_o ->init_req(fast_path, i_o());
result_memory->init_req(fast_path, memory(adr_type));
}
slow_control = top();
if (slow_region != NULL)
slow_control = _gvn.transform(slow_region);
DEBUG_ONLY(slow_region = (RegionNode*)badAddress);
set_control(checked_control);
if (!stopped()) {
Node* cmp = _gvn.transform(new(C) CmpINode(checked_value, intcon(0)));
Node* bol = _gvn.transform(new(C) BoolNode(cmp, BoolTest::eq));
IfNode* iff = create_and_map_if(control(), bol, PROB_MAX, COUNT_UNKNOWN);
Node* checks_done = _gvn.transform(new(C) IfTrueNode(iff));
result_region->init_req(checked_path, checks_done);
result_i_o ->init_req(checked_path, checked_i_o);
result_memory->init_req(checked_path, checked_mem);
set_control( _gvn.transform(new(C) IfFalseNode(iff) ));
RegionNode* slow_reg2 = new(C) RegionNode(3);
PhiNode* slow_i_o2 = new(C) PhiNode(slow_reg2, Type::ABIO);
PhiNode* slow_mem2 = new(C) PhiNode(slow_reg2, Type::MEMORY, adr_type);
record_for_igvn(slow_reg2);
slow_reg2 ->init_req(1, slow_control);
slow_i_o2 ->init_req(1, slow_i_o);
slow_mem2 ->init_req(1, slow_mem);
slow_reg2 ->init_req(2, control());
slow_i_o2 ->init_req(2, checked_i_o);
slow_mem2 ->init_req(2, checked_mem);
slow_control = _gvn.transform(slow_reg2);
slow_i_o = _gvn.transform(slow_i_o2);
slow_mem = _gvn.transform(slow_mem2);
if (alloc != NULL) {
} else {
Node* checked_offset = _gvn.transform(new(C) XorINode(checked_value, intcon(-1)));
Node* slow_offset = new(C) PhiNode(slow_reg2, TypeInt::INT);
slow_offset->init_req(1, intcon(0));
slow_offset->init_req(2, checked_offset);
slow_offset = _gvn.transform(slow_offset);
Node* src_off_plus = _gvn.transform(new(C) AddINode(src_offset, slow_offset));
Node* dest_off_plus = _gvn.transform(new(C) AddINode(dest_offset, slow_offset));
Node* length_minus = _gvn.transform(new(C) SubINode(copy_length, slow_offset));
src_offset = src_off_plus;
dest_offset = dest_off_plus;
copy_length = length_minus;
}
}
set_control(slow_control);
if (!stopped()) {
PreserveJVMState pjvms(this); // replace_in_map may trash the map
set_memory(slow_mem, adr_type);
set_i_o(slow_i_o);
if (dest_uninitialized) {
generate_clear_array(adr_type, dest, basic_elem_type,
intcon(0), NULL,
alloc->in(AllocateNode::AllocSize));
}
generate_slow_arraycopy(adr_type,
src, src_offset, dest, dest_offset,
copy_length, /*dest_uninitialized*/false);
result_region->init_req(slow_call_path, control());
result_i_o ->init_req(slow_call_path, i_o());
result_memory->init_req(slow_call_path, memory(adr_type));
}
for (uint i = 1; i < result_region->req(); i++) {
if (result_region->in(i) == NULL)
result_region->init_req(i, top());
}
set_control( _gvn.transform(result_region));
set_i_o( _gvn.transform(result_i_o) );
set_memory( _gvn.transform(result_memory), adr_type );
if (alloc != NULL) {
insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
} else if (InsertMemBarAfterArraycopy)
insert_mem_bar(Op_MemBarCPUOrder);
}
AllocateArrayNode*
LibraryCallKit::tightly_coupled_allocation(Node* ptr,
RegionNode* slow_region) {
if (stopped()) return NULL; // no fast path
if (C->AliasLevel() == 0) return NULL; // no MergeMems around
AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
if (alloc == NULL) return NULL;
Node* rawmem = memory(Compile::AliasIdxRaw);
if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
return NULL;
}
rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
return NULL;
}
for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
Node* obs = ptr->fast_out(i);
if (obs != this->map()) {
return NULL;
}
}
Node* alloc_ctl = ptr->in(0);
assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
Node* ctl = control();
while (ctl != alloc_ctl) {
if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
IfNode* iff = ctl->in(0)->as_If();
Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
ctl = iff->in(0); // This test feeds the known slow_region.
continue;
}
bool found_trap = false;
for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
Node* obs = not_ctl->fast_out(j);
if (obs->in(0) == not_ctl && obs->is_Call() &&
(obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
found_trap = true; break;
}
}
if (found_trap) {
ctl = iff->in(0); // This test feeds a harmless uncommon trap.
continue;
}
}
return NULL;
}
return alloc;
}
void
LibraryCallKit::generate_clear_array(const TypePtr* adr_type,
Node* dest,
BasicType basic_elem_type,
Node* slice_idx,
Node* slice_len,
Node* dest_size) {
assert((slice_len != NULL? 1: 0) + (dest_size != NULL? 1: 0) == 1, "");
if (slice_len == NULL) slice_len = top();
if (dest_size == NULL) dest_size = top();
Node* mem = memory(adr_type); // memory slice to operate on
int scale = exact_log2(type2aelembytes(basic_elem_type));
int abase = arrayOopDesc::base_offset_in_bytes(basic_elem_type);
int clear_low = (-1 << scale) & (BytesPerInt - 1);
int bump_bit = (-1 << scale) & BytesPerInt;
const intptr_t BIG_NEG = -128;
assert(BIG_NEG + 2*abase < 0, "neg enough");
intptr_t slice_idx_con = (intptr_t) find_int_con(slice_idx, BIG_NEG);
intptr_t slice_len_con = (intptr_t) find_int_con(slice_len, BIG_NEG);
if (slice_len_con == 0) {
return; // nothing to do here
}
intptr_t start_con = (abase + (slice_idx_con << scale)) & ~clear_low;
intptr_t end_con = find_intptr_t_con(dest_size, -1);
if (slice_idx_con >= 0 && slice_len_con >= 0) {
assert(end_con < 0, "not two cons");
end_con = round_to(abase + ((slice_idx_con + slice_len_con) << scale),
BytesPerLong);
}
if (start_con >= 0 && end_con >= 0) {
mem = ClearArrayNode::clear_memory(control(), mem, dest,
start_con, end_con, &_gvn);
} else if (start_con >= 0 && dest_size != top()) {
Node* end = dest_size;
mem = ClearArrayNode::clear_memory(control(), mem, dest,
start_con, end, &_gvn);
} else if (start_con >= 0 && slice_len != top()) {
intptr_t end_base = abase + (slice_idx_con << scale);
int end_round = (-1 << scale) & (BytesPerLong - 1);
Node* end = ConvI2X(slice_len);
if (scale != 0)
end = _gvn.transform(new(C) LShiftXNode(end, intcon(scale) ));
end_base += end_round;
end = _gvn.transform(new(C) AddXNode(end, MakeConX(end_base)));
end = _gvn.transform(new(C) AndXNode(end, MakeConX(~end_round)));
mem = ClearArrayNode::clear_memory(control(), mem, dest,
start_con, end, &_gvn);
} else if (start_con < 0 && dest_size != top()) {
Node* start = slice_idx;
start = ConvI2X(start);
if (scale != 0)
start = _gvn.transform(new(C) LShiftXNode( start, intcon(scale) ));
start = _gvn.transform(new(C) AddXNode(start, MakeConX(abase)));
if ((bump_bit | clear_low) != 0) {
int to_clear = (bump_bit | clear_low);
if (((abase + bump_bit) & ~to_clear) - bump_bit
< arrayOopDesc::length_offset_in_bytes() + BytesPerInt) {
bump_bit = 0;
assert((abase & to_clear) == 0, "array base must be long-aligned");
} else {
start = _gvn.transform(new(C) AddXNode(start, MakeConX(bump_bit)));
assert((abase & clear_low) == 0, "array base must be int-aligned");
}
start = _gvn.transform(new(C) AndXNode(start, MakeConX(~to_clear)));
if (bump_bit != 0) {
Node* x1 = _gvn.transform(new(C) AddXNode(start, MakeConX(-bump_bit)));
Node* p1 = basic_plus_adr(dest, x1);
mem = StoreNode::make(_gvn, control(), mem, p1, adr_type, intcon(0), T_INT, MemNode::unordered);
mem = _gvn.transform(mem);
}
}
Node* end = dest_size; // pre-rounded
mem = ClearArrayNode::clear_memory(control(), mem, dest,
start, end, &_gvn);
} else {
ShouldNotReachHere(); // fix caller
}
set_memory(mem, adr_type);
}
bool
LibraryCallKit::generate_block_arraycopy(const TypePtr* adr_type,
BasicType basic_elem_type,
AllocateNode* alloc,
Node* src, Node* src_offset,
Node* dest, Node* dest_offset,
Node* dest_size, bool dest_uninitialized) {
int scale = exact_log2(type2aelembytes(basic_elem_type));
if (scale >= LogBytesPerLong)
return false; // it is already a block transfer
int abase = arrayOopDesc::base_offset_in_bytes(basic_elem_type);
intptr_t src_off_con = (intptr_t) find_int_con(src_offset, -1);
intptr_t dest_off_con = (intptr_t) find_int_con(dest_offset, -1);
if (src_off_con < 0 || dest_off_con < 0)
return false;
intptr_t src_off = abase + (src_off_con << scale);
intptr_t dest_off = abase + (dest_off_con << scale);
if (((src_off | dest_off) & (BytesPerLong-1)) != 0) {
if (((src_off | dest_off) & (BytesPerLong-1)) == BytesPerInt &&
((src_off ^ dest_off) & (BytesPerLong-1)) == 0) {
Node* sptr = basic_plus_adr(src, src_off);
Node* dptr = basic_plus_adr(dest, dest_off);
const TypePtr* s_adr_type = _gvn.type(sptr)->is_ptr();
assert(s_adr_type->isa_aryptr(), "impossible slice");
Node* sval = make_load(control(), sptr, TypeInt::INT, T_INT, s_adr_type, MemNode::unordered);
store_to_memory(control(), dptr, sval, T_INT, adr_type, MemNode::unordered);
src_off += BytesPerInt;
dest_off += BytesPerInt;
} else {
return false;
}
}
assert(src_off % BytesPerLong == 0, "");
assert(dest_off % BytesPerLong == 0, "");
Node* sptr = basic_plus_adr(src, src_off);
Node* dptr = basic_plus_adr(dest, dest_off);
Node* countx = dest_size;
countx = _gvn.transform(new (C) SubXNode(countx, MakeConX(dest_off)));
countx = _gvn.transform(new (C) URShiftXNode(countx, intcon(LogBytesPerLong)));
bool disjoint_bases = true; // since alloc != NULL
generate_unchecked_arraycopy(adr_type, T_LONG, disjoint_bases,
sptr, NULL, dptr, NULL, countx, dest_uninitialized);
return true;
}
void
LibraryCallKit::generate_slow_arraycopy(const TypePtr* adr_type,
Node* src, Node* src_offset,
Node* dest, Node* dest_offset,
Node* copy_length, bool dest_uninitialized) {
assert(!dest_uninitialized, "Invariant");
Node* call = make_runtime_call(RC_NO_LEAF | RC_UNCOMMON,
OptoRuntime::slow_arraycopy_Type(),
OptoRuntime::slow_arraycopy_Java(),
"slow_arraycopy", adr_type,
src, src_offset, dest, dest_offset,
copy_length);
make_slow_call_ex(call, env()->Throwable_klass(), false);
}
Node*
LibraryCallKit::generate_checkcast_arraycopy(const TypePtr* adr_type,
Node* dest_elem_klass,
Node* src, Node* src_offset,
Node* dest, Node* dest_offset,
Node* copy_length, bool dest_uninitialized) {
if (stopped()) return NULL;
address copyfunc_addr = StubRoutines::checkcast_arraycopy(dest_uninitialized);
if (copyfunc_addr == NULL) { // Stub was not generated, go slow path.
return NULL;
}
int sco_offset = in_bytes(Klass::super_check_offset_offset());
Node* p3 = basic_plus_adr(dest_elem_klass, sco_offset);
Node* n3 = new(C) LoadINode(NULL, memory(p3), p3, _gvn.type(p3)->is_ptr(), TypeInt::INT, MemNode::unordered);
Node* check_offset = ConvI2X(_gvn.transform(n3));
Node* check_value = dest_elem_klass;
Node* src_start = array_element_address(src, src_offset, T_OBJECT);
Node* dest_start = array_element_address(dest, dest_offset, T_OBJECT);
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::checkcast_arraycopy_Type(),
copyfunc_addr, "checkcast_arraycopy", adr_type,
src_start, dest_start,
copy_length XTOP,
check_offset XTOP,
check_value);
return _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
}
Node*
LibraryCallKit::generate_generic_arraycopy(const TypePtr* adr_type,
Node* src, Node* src_offset,
Node* dest, Node* dest_offset,
Node* copy_length, bool dest_uninitialized) {
assert(!dest_uninitialized, "Invariant");
if (stopped()) return NULL;
address copyfunc_addr = StubRoutines::generic_arraycopy();
if (copyfunc_addr == NULL) { // Stub was not generated, go slow path.
return NULL;
}
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::generic_arraycopy_Type(),
copyfunc_addr, "generic_arraycopy", adr_type,
src, src_offset, dest, dest_offset, copy_length);
return _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
}
void
LibraryCallKit::generate_unchecked_arraycopy(const TypePtr* adr_type,
BasicType basic_elem_type,
bool disjoint_bases,
Node* src, Node* src_offset,
Node* dest, Node* dest_offset,
Node* copy_length, bool dest_uninitialized) {
if (stopped()) return; // nothing to do
Node* src_start = src;
Node* dest_start = dest;
if (src_offset != NULL || dest_offset != NULL) {
assert(src_offset != NULL && dest_offset != NULL, "");
src_start = array_element_address(src, src_offset, basic_elem_type);
dest_start = array_element_address(dest, dest_offset, basic_elem_type);
}
const char* copyfunc_name = "arraycopy";
address copyfunc_addr =
basictype2arraycopy(basic_elem_type, src_offset, dest_offset,
disjoint_bases, copyfunc_name, dest_uninitialized);
make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::fast_arraycopy_Type(),
copyfunc_addr, copyfunc_name, adr_type,
src_start, dest_start, copy_length XTOP);
}
bool LibraryCallKit::inline_encodeISOArray() {
assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
Node *src = argument(0);
Node *src_offset = argument(1);
Node *dst = argument(2);
Node *dst_offset = argument(3);
Node *length = argument(4);
const Type* src_type = src->Value(&_gvn);
const Type* dst_type = dst->Value(&_gvn);
const TypeAryPtr* top_src = src_type->isa_aryptr();
const TypeAryPtr* top_dest = dst_type->isa_aryptr();
if (top_src == NULL || top_src->klass() == NULL ||
top_dest == NULL || top_dest->klass() == NULL) {
return false;
}
BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (src_elem != T_CHAR || dst_elem != T_BYTE) {
return false;
}
Node* src_start = array_element_address(src, src_offset, src_elem);
Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
const TypeAryPtr* mtype = TypeAryPtr::BYTES;
Node* enc = new (C) EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
enc = _gvn.transform(enc);
Node* res_mem = _gvn.transform(new (C) SCMemProjNode(enc));
set_memory(res_mem, mtype);
set_result(enc);
return true;
}
bool LibraryCallKit::inline_multiplyToLen() {
assert(UseMultiplyToLenIntrinsic, "not implementated on this platform");
address stubAddr = StubRoutines::multiplyToLen();
if (stubAddr == NULL) {
return false; // Intrinsic's stub is not implemented on this platform
}
const char* stubName = "multiplyToLen";
assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
Node* x = argument(0);
Node* xlen = argument(1);
Node* y = argument(2);
Node* ylen = argument(3);
Node* z = argument(4);
const Type* x_type = x->Value(&_gvn);
const Type* y_type = y->Value(&_gvn);
const TypeAryPtr* top_x = x_type->isa_aryptr();
const TypeAryPtr* top_y = y_type->isa_aryptr();
if (top_x == NULL || top_x->klass() == NULL ||
top_y == NULL || top_y->klass() == NULL) {
return false;
}
BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (x_elem != T_INT || y_elem != T_INT) {
return false;
}
{ PreserveReexecuteState preexecs(this);
jvms()->set_should_reexecute(true);
Node* x_start = array_element_address(x, intcon(0), x_elem);
Node* y_start = array_element_address(y, intcon(0), y_elem);
Node* zlen = _gvn.transform(new(C) AddINode(xlen, ylen));
ciKlass* klass = ciTypeArrayKlass::make(T_INT);
Node* klass_node = makecon(TypeKlassPtr::make(klass));
IdealKit ideal(this);
#define __ ideal.
Node* one = __ ConI(1);
Node* zero = __ ConI(0);
IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
__ set(need_alloc, zero);
__ set(z_alloc, z);
__ if_then(z, BoolTest::eq, null()); {
__ increment (need_alloc, one);
} __ else_(); {
sync_kit(ideal);
Node* zlen_arg = load_array_length(z);
__ sync_kit(this);
__ if_then(zlen_arg, BoolTest::lt, zlen); {
__ increment (need_alloc, one);
} __ end_if();
} __ end_if();
__ if_then(__ value(need_alloc), BoolTest::ne, zero); {
sync_kit(ideal);
Node * narr = new_array(klass_node, zlen, 1);
__ sync_kit(this);
__ set(z_alloc, narr);
} __ end_if();
sync_kit(ideal);
z = __ value(z_alloc);
_gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
final_sync(ideal);
#undef __
Node* z_start = array_element_address(z, intcon(0), T_INT);
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::multiplyToLen_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
x_start, xlen, y_start, ylen, z_start, zlen);
} // original reexecute is set back here
C->set_has_split_ifs(true); // Has chance for split-if optimization
set_result(z);
return true;
}
bool LibraryCallKit::inline_squareToLen() {
assert(UseSquareToLenIntrinsic, "not implementated on this platform");
address stubAddr = StubRoutines::squareToLen();
if (stubAddr == NULL) {
return false; // Intrinsic's stub is not implemented on this platform
}
const char* stubName = "squareToLen";
assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
Node* x = argument(0);
Node* len = argument(1);
Node* z = argument(2);
Node* zlen = argument(3);
const Type* x_type = x->Value(&_gvn);
const Type* z_type = z->Value(&_gvn);
const TypeAryPtr* top_x = x_type->isa_aryptr();
const TypeAryPtr* top_z = z_type->isa_aryptr();
if (top_x == NULL || top_x->klass() == NULL ||
top_z == NULL || top_z->klass() == NULL) {
return false;
}
BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (x_elem != T_INT || z_elem != T_INT) {
return false;
}
Node* x_start = array_element_address(x, intcon(0), x_elem);
Node* z_start = array_element_address(z, intcon(0), z_elem);
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::squareToLen_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
x_start, len, z_start, zlen);
set_result(z);
return true;
}
bool LibraryCallKit::inline_mulAdd() {
assert(UseMulAddIntrinsic, "not implementated on this platform");
address stubAddr = StubRoutines::mulAdd();
if (stubAddr == NULL) {
return false; // Intrinsic's stub is not implemented on this platform
}
const char* stubName = "mulAdd";
assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
Node* out = argument(0);
Node* in = argument(1);
Node* offset = argument(2);
Node* len = argument(3);
Node* k = argument(4);
const Type* out_type = out->Value(&_gvn);
const Type* in_type = in->Value(&_gvn);
const TypeAryPtr* top_out = out_type->isa_aryptr();
const TypeAryPtr* top_in = in_type->isa_aryptr();
if (top_out == NULL || top_out->klass() == NULL ||
top_in == NULL || top_in->klass() == NULL) {
return false;
}
BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (out_elem != T_INT || in_elem != T_INT) {
return false;
}
Node* outlen = load_array_length(out);
Node* new_offset = _gvn.transform(new (C) SubINode(outlen, offset));
Node* out_start = array_element_address(out, intcon(0), out_elem);
Node* in_start = array_element_address(in, intcon(0), in_elem);
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::mulAdd_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
out_start,in_start, new_offset, len, k);
Node* result = _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
set_result(result);
return true;
}
bool LibraryCallKit::inline_montgomeryMultiply() {
address stubAddr = StubRoutines::montgomeryMultiply();
if (stubAddr == NULL) {
return false; // Intrinsic's stub is not implemented on this platform
}
assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
const char* stubName = "montgomery_multiply";
assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
Node* a = argument(0);
Node* b = argument(1);
Node* n = argument(2);
Node* len = argument(3);
Node* inv = argument(4);
Node* m = argument(6);
const Type* a_type = a->Value(&_gvn);
const TypeAryPtr* top_a = a_type->isa_aryptr();
const Type* b_type = b->Value(&_gvn);
const TypeAryPtr* top_b = b_type->isa_aryptr();
const Type* n_type = a->Value(&_gvn);
const TypeAryPtr* top_n = n_type->isa_aryptr();
const Type* m_type = a->Value(&_gvn);
const TypeAryPtr* top_m = m_type->isa_aryptr();
if (top_a == NULL || top_a->klass() == NULL ||
top_b == NULL || top_b->klass() == NULL ||
top_n == NULL || top_n->klass() == NULL ||
top_m == NULL || top_m->klass() == NULL) {
return false;
}
BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
return false;
}
{
Node* a_start = array_element_address(a, intcon(0), a_elem);
Node* b_start = array_element_address(b, intcon(0), b_elem);
Node* n_start = array_element_address(n, intcon(0), n_elem);
Node* m_start = array_element_address(m, intcon(0), m_elem);
Node* call = NULL;
if (CCallingConventionRequiresIntsAsLongs) {
Node* len_I2L = ConvI2L(len);
call = make_runtime_call(RC_LEAF,
OptoRuntime::montgomeryMultiply_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
a_start, b_start, n_start, len_I2L XTOP, inv,
top(), m_start);
} else {
call = make_runtime_call(RC_LEAF,
OptoRuntime::montgomeryMultiply_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
a_start, b_start, n_start, len, inv, top(),
m_start);
}
set_result(m);
}
return true;
}
bool LibraryCallKit::inline_montgomerySquare() {
address stubAddr = StubRoutines::montgomerySquare();
if (stubAddr == NULL) {
return false; // Intrinsic's stub is not implemented on this platform
}
assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
const char* stubName = "montgomery_square";
assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
Node* a = argument(0);
Node* n = argument(1);
Node* len = argument(2);
Node* inv = argument(3);
Node* m = argument(5);
const Type* a_type = a->Value(&_gvn);
const TypeAryPtr* top_a = a_type->isa_aryptr();
const Type* n_type = a->Value(&_gvn);
const TypeAryPtr* top_n = n_type->isa_aryptr();
const Type* m_type = a->Value(&_gvn);
const TypeAryPtr* top_m = m_type->isa_aryptr();
if (top_a == NULL || top_a->klass() == NULL ||
top_n == NULL || top_n->klass() == NULL ||
top_m == NULL || top_m->klass() == NULL) {
return false;
}
BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
return false;
}
{
Node* a_start = array_element_address(a, intcon(0), a_elem);
Node* n_start = array_element_address(n, intcon(0), n_elem);
Node* m_start = array_element_address(m, intcon(0), m_elem);
Node* call = NULL;
if (CCallingConventionRequiresIntsAsLongs) {
Node* len_I2L = ConvI2L(len);
call = make_runtime_call(RC_LEAF,
OptoRuntime::montgomerySquare_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
a_start, n_start, len_I2L XTOP, inv, top(),
m_start);
} else {
call = make_runtime_call(RC_LEAF,
OptoRuntime::montgomerySquare_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
a_start, n_start, len, inv, top(),
m_start);
}
set_result(m);
}
return true;
}
bool LibraryCallKit::inline_updateCRC32() {
assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
assert(callee()->signature()->size() == 2, "update has 2 parameters");
Node* crc = argument(0); // type: int
Node* b = argument(1); // type: int
Node* M1 = intcon(-1);
crc = _gvn.transform(new (C) XorINode(crc, M1));
Node* result = _gvn.transform(new (C) XorINode(crc, b));
result = _gvn.transform(new (C) AndINode(result, intcon(0xFF)));
Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
Node* offset = _gvn.transform(new (C) LShiftINode(result, intcon(0x2)));
Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
crc = _gvn.transform(new (C) URShiftINode(crc, intcon(8)));
result = _gvn.transform(new (C) XorINode(crc, result));
result = _gvn.transform(new (C) XorINode(result, M1));
set_result(result);
return true;
}
bool LibraryCallKit::inline_updateBytesCRC32() {
assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
Node* crc = argument(0); // type: int
Node* src = argument(1); // type: oop
Node* offset = argument(2); // type: int
Node* length = argument(3); // type: int
const Type* src_type = src->Value(&_gvn);
const TypeAryPtr* top_src = src_type->isa_aryptr();
if (top_src == NULL || top_src->klass() == NULL) {
return false;
}
BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (src_elem != T_BYTE) {
return false;
}
Node* src_start = array_element_address(src, offset, src_elem);
address stubAddr = StubRoutines::updateBytesCRC32();
const char *stubName = "updateBytesCRC32";
Node* call;
if (CCallingConventionRequiresIntsAsLongs) {
call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
crc XTOP, src_start, length XTOP);
} else {
call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
crc, src_start, length);
}
Node* result = _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
set_result(result);
return true;
}
bool LibraryCallKit::inline_updateByteBufferCRC32() {
assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
Node* crc = argument(0); // type: int
Node* src = argument(1); // type: long
Node* offset = argument(3); // type: int
Node* length = argument(4); // type: int
src = ConvL2X(src); // adjust Java long to machine word
Node* base = _gvn.transform(new (C) CastX2PNode(src));
offset = ConvI2X(offset);
Node* src_start = basic_plus_adr(top(), base, offset);
address stubAddr = StubRoutines::updateBytesCRC32();
const char *stubName = "updateBytesCRC32";
Node* call;
if (CCallingConventionRequiresIntsAsLongs) {
call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
crc XTOP, src_start, length XTOP);
} else {
call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
crc, src_start, length);
}
Node* result = _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
set_result(result);
return true;
}
bool LibraryCallKit::inline_reference_get() {
const int referent_offset = java_lang_ref_Reference::referent_offset;
guarantee(referent_offset > 0, "should have already been set");
Node* reference_obj = null_check_receiver();
if (stopped()) return true;
Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
ciInstanceKlass* klass = env()->Object_klass();
const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
Node* no_ctrl = NULL;
Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
pre_barrier(false /* do_load */,
control(),
NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
result /* pre_val */,
T_OBJECT);
insert_mem_bar(Op_MemBarCPUOrder);
set_result(result);
return true;
}
Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
bool is_exact=true, bool is_static=false) {
const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
assert(tinst != NULL, "obj is null");
assert(tinst->klass()->is_loaded(), "obj is not loaded");
assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
ciField* field = tinst->klass()->as_instance_klass()->get_field_by_name(ciSymbol::make(fieldName),
ciSymbol::make(fieldTypeString),
is_static);
if (field == NULL) return (Node *) NULL;
assert (field != NULL, "undefined field");
int offset = field->offset_in_bytes();
bool is_vol = field->is_volatile();
ciType* field_klass = field->type();
assert(field_klass->is_loaded(), "should be loaded");
const TypePtr* adr_type = C->alias_type(field)->adr_type();
Node *adr = basic_plus_adr(fromObj, fromObj, offset);
BasicType bt = field->layout_type();
const Type *type;
if (bt == T_OBJECT) {
type = TypeOopPtr::make_from_klass(field_klass->as_klass());
} else {
type = Type::get_const_basic_type(bt);
}
Node* leading_membar = NULL;
if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
leading_membar = insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
}
MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
if (is_vol) {
Node* mb = insert_mem_bar(Op_MemBarAcquire, loadedField);
mb->as_MemBar()->set_trailing_load();
}
return loadedField;
}
bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
address stubAddr = NULL;
const char *stubName;
assert(UseAES, "need AES instruction support");
switch(id) {
case vmIntrinsics::_aescrypt_encryptBlock:
stubAddr = StubRoutines::aescrypt_encryptBlock();
stubName = "aescrypt_encryptBlock";
break;
case vmIntrinsics::_aescrypt_decryptBlock:
stubAddr = StubRoutines::aescrypt_decryptBlock();
stubName = "aescrypt_decryptBlock";
break;
}
if (stubAddr == NULL) return false;
Node* aescrypt_object = argument(0);
Node* src = argument(1);
Node* src_offset = argument(2);
Node* dest = argument(3);
Node* dest_offset = argument(4);
const Type* src_type = src->Value(&_gvn);
const Type* dest_type = dest->Value(&_gvn);
const TypeAryPtr* top_src = src_type->isa_aryptr();
const TypeAryPtr* top_dest = dest_type->isa_aryptr();
assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
Node* src_start = src;
Node* dest_start = dest;
if (src_offset != NULL || dest_offset != NULL) {
assert(src_offset != NULL && dest_offset != NULL, "");
src_start = array_element_address(src, src_offset, T_BYTE);
dest_start = array_element_address(dest, dest_offset, T_BYTE);
}
Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
if (k_start == NULL) return false;
if (Matcher::pass_original_key_for_aes()) {
Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
if (original_k_start == NULL) return false;
make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
src_start, dest_start, k_start, original_k_start);
} else {
make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
src_start, dest_start, k_start);
}
return true;
}
bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
address stubAddr = NULL;
const char *stubName = NULL;
assert(UseAES, "need AES instruction support");
switch(id) {
case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
stubName = "cipherBlockChaining_encryptAESCrypt";
break;
case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
stubName = "cipherBlockChaining_decryptAESCrypt";
break;
}
if (stubAddr == NULL) return false;
Node* cipherBlockChaining_object = argument(0);
Node* src = argument(1);
Node* src_offset = argument(2);
Node* len = argument(3);
Node* dest = argument(4);
Node* dest_offset = argument(5);
const Type* src_type = src->Value(&_gvn);
const Type* dest_type = dest->Value(&_gvn);
const TypeAryPtr* top_src = src_type->isa_aryptr();
const TypeAryPtr* top_dest = dest_type->isa_aryptr();
assert (top_src != NULL && top_src->klass() != NULL
&& top_dest != NULL && top_dest->klass() != NULL, "args are strange");
Node* src_start = src;
Node* dest_start = dest;
if (src_offset != NULL || dest_offset != NULL) {
assert(src_offset != NULL && dest_offset != NULL, "");
src_start = array_element_address(src, src_offset, T_BYTE);
dest_start = array_element_address(dest, dest_offset, T_BYTE);
}
Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
if (embeddedCipherObj == NULL) return false;
const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
assert(tinst != NULL, "CBC obj is null");
assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
const TypeOopPtr* xtype = aklass->as_instance_type();
Node* aescrypt_object = new(C) CheckCastPPNode(control(), embeddedCipherObj, xtype);
aescrypt_object = _gvn.transform(aescrypt_object);
Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
if (k_start == NULL) return false;
Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
if (objRvec == NULL) return false;
Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
Node* cbcCrypt;
if (Matcher::pass_original_key_for_aes()) {
Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
if (original_k_start == NULL) return false;
cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::cipherBlockChaining_aescrypt_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
src_start, dest_start, k_start, r_start, len, original_k_start);
} else {
cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::cipherBlockChaining_aescrypt_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
src_start, dest_start, k_start, r_start, len);
}
Node* retvalue = _gvn.transform(new (C) ProjNode(cbcCrypt, TypeFunc::Parms));
set_result(retvalue);
return true;
}
Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
#ifdef PPC64
Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
if (objSessionK == NULL) {
return (Node *) NULL;
}
Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
#else
Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
#endif // PPC64
assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
if (objAESCryptKey == NULL) return (Node *) NULL;
Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
return k_start;
}
Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
if (objAESCryptKey == NULL) return (Node *) NULL;
Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
return original_k_start;
}
Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
Node* objCBC = argument(0);
Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
assert(tinst != NULL, "CBCobj is null");
assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
if (!klass_AESCrypt->is_loaded()) {
Node* ctrl = control();
set_control(top()); // no regular fast path
return ctrl;
}
ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
Node* cmp_instof = _gvn.transform(new (C) CmpINode(instof, intcon(1)));
Node* bool_instof = _gvn.transform(new (C) BoolNode(cmp_instof, BoolTest::ne));
Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
if (!decrypting)
return instof_false; // even if it is NULL
RegionNode* region = new(C) RegionNode(3);
region->init_req(1, instof_false);
Node* src = argument(1);
Node* dest = argument(4);
Node* cmp_src_dest = _gvn.transform(new (C) CmpPNode(src, dest));
Node* bool_src_dest = _gvn.transform(new (C) BoolNode(cmp_src_dest, BoolTest::eq));
Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
region->init_req(2, src_dest_conjoint);
record_for_igvn(region);
return _gvn.transform(region);
}
bool LibraryCallKit::inline_ghash_processBlocks() {
address stubAddr;
const char *stubName;
assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
stubAddr = StubRoutines::ghash_processBlocks();
stubName = "ghash_processBlocks";
Node* data = argument(0);
Node* offset = argument(1);
Node* len = argument(2);
Node* state = argument(3);
Node* subkeyH = argument(4);
Node* state_start = array_element_address(state, intcon(0), T_LONG);
assert(state_start, "state is NULL");
Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
assert(subkeyH_start, "subkeyH is NULL");
Node* data_start = array_element_address(data, offset, T_BYTE);
assert(data_start, "data is NULL");
Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::ghash_processBlocks_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
state_start, subkeyH_start, data_start, len);
return true;
}
bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
Node* sha_obj = argument(0);
Node* src = argument(1); // type oop
Node* ofs = argument(2); // type int
const Type* src_type = src->Value(&_gvn);
const TypeAryPtr* top_src = src_type->isa_aryptr();
if (top_src == NULL || top_src->klass() == NULL) {
return false;
}
BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (src_elem != T_BYTE) {
return false;
}
Node* src_start = array_element_address(src, ofs, src_elem);
Node* state = NULL;
address stubAddr;
const char *stubName;
switch(id) {
case vmIntrinsics::_sha_implCompress:
assert(UseSHA1Intrinsics, "need SHA1 instruction support");
state = get_state_from_sha_object(sha_obj);
stubAddr = StubRoutines::sha1_implCompress();
stubName = "sha1_implCompress";
break;
case vmIntrinsics::_sha2_implCompress:
assert(UseSHA256Intrinsics, "need SHA256 instruction support");
state = get_state_from_sha_object(sha_obj);
stubAddr = StubRoutines::sha256_implCompress();
stubName = "sha256_implCompress";
break;
case vmIntrinsics::_sha5_implCompress:
assert(UseSHA512Intrinsics, "need SHA512 instruction support");
state = get_state_from_sha5_object(sha_obj);
stubAddr = StubRoutines::sha512_implCompress();
stubName = "sha512_implCompress";
break;
default:
fatal_unexpected_iid(id);
return false;
}
if (state == NULL) return false;
Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
src_start, state);
return true;
}
bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
"need SHA1/SHA256/SHA512 instruction support");
assert((uint)predicate < 3, "sanity");
assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
Node* src = argument(1); // byte[] array
Node* ofs = argument(2); // type int
Node* limit = argument(3); // type int
const Type* src_type = src->Value(&_gvn);
const TypeAryPtr* top_src = src_type->isa_aryptr();
if (top_src == NULL || top_src->klass() == NULL) {
return false;
}
BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
if (src_elem != T_BYTE) {
return false;
}
Node* src_start = array_element_address(src, ofs, src_elem);
const char* klass_SHA_name = NULL;
const char* stub_name = NULL;
address stub_addr = NULL;
bool long_state = false;
switch (predicate) {
case 0:
if (UseSHA1Intrinsics) {
klass_SHA_name = "sun/security/provider/SHA";
stub_name = "sha1_implCompressMB";
stub_addr = StubRoutines::sha1_implCompressMB();
}
break;
case 1:
if (UseSHA256Intrinsics) {
klass_SHA_name = "sun/security/provider/SHA2";
stub_name = "sha256_implCompressMB";
stub_addr = StubRoutines::sha256_implCompressMB();
}
break;
case 2:
if (UseSHA512Intrinsics) {
klass_SHA_name = "sun/security/provider/SHA5";
stub_name = "sha512_implCompressMB";
stub_addr = StubRoutines::sha512_implCompressMB();
long_state = true;
}
break;
default:
fatal(err_msg_res("unknown SHA intrinsic predicate: %d", predicate));
}
if (klass_SHA_name != NULL) {
const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
assert(tinst != NULL, "digestBase_obj is not instance???");
assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
}
return false;
}
bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
bool long_state, address stubAddr, const char *stubName,
Node* src_start, Node* ofs, Node* limit) {
const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
const TypeOopPtr* xtype = aklass->as_instance_type();
Node* sha_obj = new (C) CheckCastPPNode(control(), digestBase_obj, xtype);
sha_obj = _gvn.transform(sha_obj);
Node* state;
if (long_state) {
state = get_state_from_sha5_object(sha_obj);
} else {
state = get_state_from_sha_object(sha_obj);
}
if (state == NULL) return false;
Node *call;
if (CCallingConventionRequiresIntsAsLongs) {
call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::digestBase_implCompressMB_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
src_start, state, ofs XTOP, limit XTOP);
} else {
call = make_runtime_call(RC_LEAF|RC_NO_FP,
OptoRuntime::digestBase_implCompressMB_Type(),
stubAddr, stubName, TypePtr::BOTTOM,
src_start, state, ofs, limit);
}
Node* result = _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
set_result(result);
return true;
}
Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
if (sha_state == NULL) return (Node *) NULL;
Node* state = array_element_address(sha_state, intcon(0), T_INT);
return state;
}
Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
if (sha_state == NULL) return (Node *) NULL;
Node* state = array_element_address(sha_state, intcon(0), T_LONG);
return state;
}
Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
"need SHA1/SHA256/SHA512 instruction support");
assert((uint)predicate < 3, "sanity");
Node* digestBaseObj = argument(0);
const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
assert(tinst != NULL, "digestBaseObj is null");
assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
const char* klass_SHA_name = NULL;
switch (predicate) {
case 0:
if (UseSHA1Intrinsics) {
klass_SHA_name = "sun/security/provider/SHA";
}
break;
case 1:
if (UseSHA256Intrinsics) {
klass_SHA_name = "sun/security/provider/SHA2";
}
break;
case 2:
if (UseSHA512Intrinsics) {
klass_SHA_name = "sun/security/provider/SHA5";
}
break;
default:
fatal(err_msg_res("unknown SHA intrinsic predicate: %d", predicate));
}
ciKlass* klass_SHA = NULL;
if (klass_SHA_name != NULL) {
klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
}
if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
Node* ctrl = control();
set_control(top()); // no intrinsic path
return ctrl;
}
ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
Node* cmp_instof = _gvn.transform(new (C) CmpINode(instofSHA, intcon(1)));
Node* bool_instof = _gvn.transform(new (C) BoolNode(cmp_instof, BoolTest::ne));
Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
return instof_false; // even if it is NULL
}
bool LibraryCallKit::inline_profileBoolean() {
Node* counts = argument(1);
const TypeAryPtr* ary = NULL;
ciArray* aobj = NULL;
if (counts->is_Con()
&& (ary = counts->bottom_type()->isa_aryptr()) != NULL
&& (aobj = ary->const_oop()->as_array()) != NULL
&& (aobj->length() == 2)) {
jint false_cnt = aobj->element_value(0).as_int();
jint true_cnt = aobj->element_value(1).as_int();
if (C->log() != NULL) {
C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
false_cnt, true_cnt);
}
if (false_cnt + true_cnt == 0) {
uncommon_trap_exact(Deoptimization::Reason_intrinsic,
Deoptimization::Action_reinterpret);
return true;
}
Node* result = argument(0);
if (false_cnt == 0 || true_cnt == 0) {
int expected_val = (false_cnt == 0) ? 1 : 0;
Node* cmp = _gvn.transform(new (C) CmpINode(result, intcon(expected_val)));
Node* test = _gvn.transform(new (C) BoolNode(cmp, BoolTest::eq));
IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
Node* fast_path = _gvn.transform(new (C) IfTrueNode(check));
Node* slow_path = _gvn.transform(new (C) IfFalseNode(check));
{ // Slow path: uncommon trap for never seen value and then reexecute
PreserveJVMState pjvms(this);
PreserveReexecuteState preexecs(this);
jvms()->set_should_reexecute(true);
set_control(slow_path);
set_i_o(i_o());
uncommon_trap_exact(Deoptimization::Reason_intrinsic,
Deoptimization::Action_reinterpret);
}
set_control(fast_path);
result = intcon(expected_val);
}
Node* profile = _gvn.transform(new (C) ProfileBooleanNode(result, false_cnt, true_cnt));
C->record_for_igvn(profile);
set_result(profile);
return true;
} else {
return false;
}
}
C:\hotspot-69087d08d473\src\share\vm/opto/live.cpp
#include "precompiled.hpp"
#include "memory/allocation.inline.hpp"
#include "opto/callnode.hpp"
#include "opto/chaitin.hpp"
#include "opto/live.hpp"
#include "opto/machnode.hpp"
PhaseLive::PhaseLive( const PhaseCFG &cfg, const LRG_List &names, Arena *arena ) : Phase(LIVE), _cfg(cfg), _names(names), _arena(arena), _live(0) {
}
void PhaseLive::compute(uint maxlrg) {
_maxlrg = maxlrg;
_worklist = new (_arena) Block_List();
_live = (IndexSet*)_arena->Amalloc(sizeof(IndexSet) * _cfg.number_of_blocks());
uint i;
for (i = 0; i < _cfg.number_of_blocks(); i++) {
_live[i].initialize(_maxlrg);
}
ResourceMark rm; // Nuke temp storage on exit
_defs = NEW_RESOURCE_ARRAY(IndexSet,_cfg.number_of_blocks());
for (i = 0; i < _cfg.number_of_blocks(); i++) {
_defs[i].initialize(_maxlrg);
}
_deltas = NEW_RESOURCE_ARRAY(IndexSet*,_cfg.number_of_blocks());
memset( _deltas, 0, sizeof(IndexSet*)* _cfg.number_of_blocks());
_free_IndexSet = NULL;
VectorSet first_pass(Thread::current()->resource_area());
for (uint j = _cfg.number_of_blocks(); j > 0; j--) {
Block* block = _cfg.get_block(j - 1);
IndexSet* use = getset(block);
IndexSet* def = &_defs[block->_pre_order-1];
DEBUG_ONLY(IndexSet *def_outside = getfreeset();)
uint i;
for (i = block->number_of_nodes(); i > 1; i--) {
Node* n = block->get_node(i-1);
if (n->is_Phi()) {
break;
}
uint r = _names.at(n->_idx);
assert(!def_outside->member(r), "Use of external LRG overlaps the same LRG defined in this block");
def->insert( r );
use->remove( r );
uint cnt = n->req();
for (uint k = 1; k < cnt; k++) {
Node *nk = n->in(k);
uint nkidx = nk->_idx;
if (_cfg.get_block_for_node(nk) != block) {
uint u = _names.at(nkidx);
use->insert(u);
DEBUG_ONLY(def_outside->insert(u);)
}
}
}
#ifdef ASSERT
def_outside->set_next(_free_IndexSet);
_free_IndexSet = def_outside; // Drop onto free list
#endif
for (uint k = i; k > 0; k--) {
uint r = _names.at(block->get_node(k - 1)->_idx);
def->insert(r);
use->remove(r);
}
for (uint l = 1; l < block->num_preds(); l++) {
Block* p = _cfg.get_block_for_node(block->pred(l));
add_liveout(p, use, first_pass);
for (uint k = i; k > 0; k--) {
add_liveout(p, _names.at(block->get_node(k-1)->in(l)->_idx), first_pass);
}
}
freeset(block);
first_pass.set(block->_pre_order);
while (_worklist->size()) {
Block* block = _worklist->pop();
IndexSet *delta = getset(block);
assert( delta->count(), "missing delta set" );
for (uint l = 1; l < block->num_preds(); l++) {
Block* predecessor = _cfg.get_block_for_node(block->pred(l));
add_liveout(predecessor, delta, first_pass);
}
freeset(block);
} // End of while-worklist-not-empty
} // End of for-all-blocks-outer-loop
for (i = 0; i < _cfg.number_of_blocks(); i++) {
_defs[i].clear();
if (_deltas[i]) {
_deltas[i]->clear();
}
}
IndexSet *free = _free_IndexSet;
while (free != NULL) {
IndexSet *temp = free;
free = free->next();
temp->clear();
}
}
#ifndef PRODUCT
void PhaseLive::stats(uint iters) const {
}
#endif
IndexSet *PhaseLive::getset( Block *p ) {
IndexSet *delta = _deltas[p->_pre_order-1];
if( !delta ) // Not on worklist?
delta = _deltas[p->_pre_order-1] = getfreeset();
return delta; // Return set of new live-out items
}
IndexSet *PhaseLive::getfreeset( ) {
IndexSet *f = _free_IndexSet;
if( !f ) {
f = new IndexSet;
f->initialize(_maxlrg, Thread::current()->resource_area());
} else {
_free_IndexSet = f->next();
f->initialize(_maxlrg, Thread::current()->resource_area());
}
return f;
}
void PhaseLive::freeset( const Block *p ) {
IndexSet *f = _deltas[p->_pre_order-1];
f->set_next(_free_IndexSet);
_free_IndexSet = f; // Drop onto free list
_deltas[p->_pre_order-1] = NULL;
}
void PhaseLive::add_liveout( Block *p, uint r, VectorSet &first_pass ) {
IndexSet *live = &_live[p->_pre_order-1];
if( live->insert(r) ) { // If actually inserted...
if( !_defs[p->_pre_order-1].member( r ) ) {
if( !_deltas[p->_pre_order-1] && // Not on worklist?
first_pass.test(p->_pre_order) )
_worklist->push(p); // Actually go on worklist if already 1st pass
getset(p)->insert(r);
}
}
}
void PhaseLive::add_liveout( Block *p, IndexSet *lo, VectorSet &first_pass ) {
IndexSet *live = &_live[p->_pre_order-1];
IndexSet *defs = &_defs[p->_pre_order-1];
IndexSet *on_worklist = _deltas[p->_pre_order-1];
IndexSet *delta = on_worklist ? on_worklist : getfreeset();
IndexSetIterator elements(lo);
uint r;
while ((r = elements.next()) != 0) {
if( live->insert(r) && // If actually inserted...
!defs->member( r ) ) // and not defined locally
delta->insert(r); // Then add to live-in set
}
if( delta->count() ) { // If actually added things
_deltas[p->_pre_order-1] = delta; // Flag as on worklist now
if( !on_worklist && // Not on worklist?
first_pass.test(p->_pre_order) )
_worklist->push(p); // Actually go on worklist if already 1st pass
} else { // Nothing there; just free it
delta->set_next(_free_IndexSet);
_free_IndexSet = delta; // Drop onto free list
}
}
#ifndef PRODUCT
void PhaseLive::dump( const Block *b ) const {
tty->print("Block %d: ",b->_pre_order);
tty->print("LiveOut: "); _live[b->_pre_order-1].dump();
uint cnt = b->number_of_nodes();
for( uint i=0; i<cnt; i++ ) {
tty->print("L%d/", _names.at(b->get_node(i)->_idx));
b->get_node(i)->dump();
}
tty->print("\n");
}
void PhaseChaitin::verify_base_ptrs( ResourceArea *a ) const {
#ifdef ASSERT
Unique_Node_List worklist(a);
for (uint i = 0; i < _cfg.number_of_blocks(); i++) {
Block* block = _cfg.get_block(i);
for (uint j = block->end_idx() + 1; j > 1; j--) {
Node* n = block->get_node(j-1);
if (n->is_Phi()) {
break;
}
if (n->is_MachSafePoint()) {
MachSafePointNode *sfpt = n->as_MachSafePoint();
JVMState* jvms = sfpt->jvms();
if (jvms != NULL) {
if (jvms->oopoff() < sfpt->req()) {
for (uint idx = jvms->oopoff(); idx < sfpt->req(); idx++) {
Node *check = sfpt->in(idx);
bool is_derived = ((idx - jvms->oopoff()) & 1) == 0;
worklist.clear();
worklist.push(check);
uint k = 0;
while( k < worklist.size() ) {
check = worklist.at(k);
assert(check,"Bad base or derived pointer");
int isc = check->is_Copy();
if( isc ) {
worklist.push(check->in(isc));
} else if( check->is_Phi() ) {
for (uint m = 1; m < check->req(); m++)
worklist.push(check->in(m));
} else if( check->is_Con() ) {
if (is_derived) {
assert(!is_derived || check->bottom_type()->is_ptr()->ptr() == TypePtr::Null,"Bad derived pointer");
} else {
assert(check->bottom_type()->is_ptr()->_offset == 0,"Bad base pointer");
if (check->is_Mach()) {
assert(check->as_Mach()->ideal_Opcode() == Op_ConP,"Bad base pointer");
} else {
assert(check->Opcode() == Op_ConP &&
check->bottom_type()->is_ptr()->ptr() == TypePtr::Null,"Bad base pointer");
}
}
} else if( check->bottom_type()->is_ptr()->_offset == 0 ) {
if(check->is_Proj() || check->is_Mach() &&
(check->as_Mach()->ideal_Opcode() == Op_CreateEx ||
check->as_Mach()->ideal_Opcode() == Op_ThreadLocal ||
check->as_Mach()->ideal_Opcode() == Op_CMoveP ||
check->as_Mach()->ideal_Opcode() == Op_CheckCastPP ||
#ifdef _LP64
UseCompressedOops && check->as_Mach()->ideal_Opcode() == Op_CastPP ||
UseCompressedOops && check->as_Mach()->ideal_Opcode() == Op_DecodeN ||
UseCompressedClassPointers && check->as_Mach()->ideal_Opcode() == Op_DecodeNKlass ||
#endif
check->as_Mach()->ideal_Opcode() == Op_LoadP ||
check->as_Mach()->ideal_Opcode() == Op_LoadKlass)) {
} else {
check->dump();
assert(false,"Bad base or derived pointer");
}
} else {
assert(is_derived,"Bad base pointer");
assert(check->is_Mach() && check->as_Mach()->ideal_Opcode() == Op_AddP,"Bad derived pointer");
}
k++;
assert(k < 100000,"Derived pointer checking in infinite loop");
} // End while
}
} // End of check for derived pointers
} // End of Kcheck for debug info
} // End of if found a safepoint
} // End of forall instructions in block
} // End of forall blocks
#endif
}
void PhaseChaitin::verify( ResourceArea *a, bool verify_ifg ) const {
#ifdef ASSERT
if( VerifyOpto || VerifyRegisterAllocator ) {
_cfg.verify();
verify_base_ptrs(a);
if(verify_ifg)
_ifg->verify(this);
}
#endif
}
#endif
C:\hotspot-69087d08d473\src\share\vm/opto/live.hpp
#ifndef SHARE_VM_OPTO_LIVE_HPP
#define SHARE_VM_OPTO_LIVE_HPP
#include "libadt/port.hpp"
#include "libadt/vectset.hpp"
#include "opto/block.hpp"
#include "opto/indexSet.hpp"
#include "opto/phase.hpp"
#include "opto/regmask.hpp"
class Block;
class PhaseCFG;
class VectorSet;
class IndexSet;
typedef GrowableArray<uint> LRG_List;
class PhaseLive : public Phase {
IndexSet *_live;
IndexSet *_defs;
IndexSet **_deltas;
IndexSet *_free_IndexSet; // Free list of same
Block_List *_worklist; // Worklist for iterative solution
const PhaseCFG &_cfg; // Basic blocks
const LRG_List &_names; // Mapping from Nodes to live ranges
uint _maxlrg; // Largest live-range number
Arena *_arena;
IndexSet *getset( Block *p );
IndexSet *getfreeset( );
void freeset( const Block *p );
void add_liveout( Block *p, uint r, VectorSet &first_pass );
void add_liveout( Block *p, IndexSet *lo, VectorSet &first_pass );
public:
PhaseLive(const PhaseCFG &cfg, const LRG_List &names, Arena *arena);
~PhaseLive() {}
void compute(uint maxlrg);
void reset() { _live = NULL; }
IndexSet *live( const Block * b ) { return &_live[b->_pre_order-1]; }
#ifndef PRODUCT
void dump( const Block *b ) const;
void stats(uint iters) const;
#endif
};
#endif // SHARE_VM_OPTO_LIVE_HPP
C:\hotspot-69087d08d473\src\share\vm/opto/locknode.cpp
#include "precompiled.hpp"
#include "opto/locknode.hpp"
#include "opto/parse.hpp"
#include "opto/rootnode.hpp"
#include "opto/runtime.hpp"
const RegMask &BoxLockNode::in_RegMask(uint i) const {
return _inmask;
}
const RegMask &BoxLockNode::out_RegMask() const {
return *Matcher::idealreg2regmask[Op_RegP];
}
uint BoxLockNode::size_of() const { return sizeof(*this); }
BoxLockNode::BoxLockNode( int slot ) : Node( Compile::current()->root() ),
_slot(slot), _is_eliminated(false) {
init_class_id(Class_BoxLock);
init_flags(Flag_rematerialize);
OptoReg::Name reg = OptoReg::stack2reg(_slot);
_inmask.Insert(reg);
}
uint BoxLockNode::hash() const {
if (EliminateNestedLocks)
return NO_HASH; // Each locked region has own BoxLock node
return Node::hash() + _slot + (_is_eliminated ? Compile::current()->fixed_slots() : 0);
}
uint BoxLockNode::cmp( const Node &n ) const {
if (EliminateNestedLocks)
return (&n == this); // Always fail except on self
const BoxLockNode &bn = (const BoxLockNode &)n;
return bn._slot == _slot && bn._is_eliminated == _is_eliminated;
}
BoxLockNode* BoxLockNode::box_node(Node* box) {
while (!box->is_BoxLock()) {
assert(box->is_SpillCopy() || box->is_Phi(), "Bad spill of Lock.");
box = box->in(1);
}
return box->as_BoxLock();
}
OptoReg::Name BoxLockNode::reg(Node* box) {
return box_node(box)->in_RegMask(0).find_first_elem();
}
bool BoxLockNode::is_simple_lock_region(LockNode** unique_lock, Node* obj) {
LockNode* lock = NULL;
bool has_one_lock = false;
for (uint i = 0; i < this->outcnt(); i++) {
Node* n = this->raw_out(i);
assert(!n->is_Phi(), "should not merge BoxLock nodes");
if (n->is_AbstractLock()) {
AbstractLockNode* alock = n->as_AbstractLock();
if (alock->box_node() == this) {
if (alock->obj_node()->eqv_uncast(obj)) {
if ((unique_lock != NULL) && alock->is_Lock()) {
if (lock == NULL) {
lock = alock->as_Lock();
has_one_lock = true;
} else if (lock != alock->as_Lock()) {
has_one_lock = false;
}
}
} else {
return false; // Different objects
}
}
}
}
#ifdef ASSERT
for (uint i = 0; i < this->outcnt(); i++) {
Node* n = this->raw_out(i);
if (n->is_FastLock()) {
FastLockNode* flock = n->as_FastLock();
assert((flock->box_node() == this) && flock->obj_node()->eqv_uncast(obj),"");
}
}
#endif
if (unique_lock != NULL && has_one_lock) {
}
return true;
}
uint FastLockNode::hash() const { return NO_HASH; }
uint FastLockNode::size_of() const { return sizeof(*this); }
uint FastLockNode::cmp( const Node &n ) const {
return (&n == this); // Always fail except on self
}
uint FastUnlockNode::hash() const { return NO_HASH; }
uint FastUnlockNode::cmp( const Node &n ) const {
return (&n == this); // Always fail except on self
}
void FastLockNode::create_lock_counter(JVMState* state) {
BiasedLockingNamedCounter* blnc = (BiasedLockingNamedCounter*)
OptoRuntime::new_named_counter(state, NamedCounter::BiasedLockingCounter);
_counters = blnc->counters();
}
void FastLockNode::create_rtm_lock_counter(JVMState* state) {
#if INCLUDE_RTM_OPT
Compile* C = Compile::current();
if (C->profile_rtm() || (PrintPreciseRTMLockingStatistics && C->use_rtm())) {
RTMLockingNamedCounter* rlnc = (RTMLockingNamedCounter*)
OptoRuntime::new_named_counter(state, NamedCounter::RTMLockingCounter);
_rtm_counters = rlnc->counters();
if (UseRTMForStackLocks) {
rlnc = (RTMLockingNamedCounter*)
OptoRuntime::new_named_counter(state, NamedCounter::RTMLockingCounter);
_stack_rtm_counters = rlnc->counters();
}
}
#endif
}
void Parse::do_monitor_enter() {
kill_dead_locals();
Node* obj = null_check(peek());
if (stopped()) return;
pop();
shared_lock(obj);
}
void Parse::do_monitor_exit() {
kill_dead_locals();
pop(); // Pop oop to unlock
shared_unlock(map()->peek_monitor_box(), map()->peek_monitor_obj());
}
C:\hotspot-69087d08d473\src\share\vm/opto/locknode.hpp
#ifndef SHARE_VM_OPTO_LOCKNODE_HPP
#define SHARE_VM_OPTO_LOCKNODE_HPP
#include "opto/node.hpp"
#include "opto/opcodes.hpp"
#include "opto/subnode.hpp"
#if defined AD_MD_HPP
# include AD_MD_HPP
#elif defined TARGET_ARCH_MODEL_x86_32
# include "adfiles/ad_x86_32.hpp"
#elif defined TARGET_ARCH_MODEL_x86_64
# include "adfiles/ad_x86_64.hpp"
#elif defined TARGET_ARCH_MODEL_aarch64
# include "adfiles/ad_aarch64.hpp"
#elif defined TARGET_ARCH_MODEL_sparc
# include "adfiles/ad_sparc.hpp"
#elif defined TARGET_ARCH_MODEL_zero
# include "adfiles/ad_zero.hpp"
#elif defined TARGET_ARCH_MODEL_ppc_64
# include "adfiles/ad_ppc_64.hpp"
#endif
class BoxLockNode : public Node {
const int _slot; // stack slot
RegMask _inmask; // OptoReg corresponding to stack slot
bool _is_eliminated; // Associated locks were safely eliminated
public:
BoxLockNode( int lock );
virtual int Opcode() const;
virtual void emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const;
virtual uint size(PhaseRegAlloc *ra_) const;
virtual const RegMask &in_RegMask(uint) const;
virtual const RegMask &out_RegMask() const;
virtual uint size_of() const;
virtual uint hash() const;
virtual uint cmp( const Node &n ) const;
virtual const class Type *bottom_type() const { return TypeRawPtr::BOTTOM; }
virtual uint ideal_reg() const { return Op_RegP; }
static OptoReg::Name reg(Node* box_node);
static BoxLockNode* box_node(Node* box_node);
static bool same_slot(Node* box1, Node* box2) {
return box1->as_BoxLock()->_slot == box2->as_BoxLock()->_slot;
}
int stack_slot() const { return _slot; }
bool is_eliminated() const { return _is_eliminated; }
void set_eliminated() { _is_eliminated = true; }
bool is_simple_lock_region(LockNode** unique_lock, Node* obj);
#ifndef PRODUCT
virtual void format( PhaseRegAlloc *, outputStream *st ) const;
virtual void dump_spec(outputStream *st) const { st->print(" Lock %d",_slot); }
#endif
};
class FastLockNode: public CmpNode {
private:
BiasedLockingCounters* _counters;
RTMLockingCounters* _rtm_counters; // RTM lock counters for inflated locks
RTMLockingCounters* _stack_rtm_counters; // RTM lock counters for stack locks
public:
FastLockNode(Node *ctrl, Node *oop, Node *box) : CmpNode(oop,box) {
init_req(0,ctrl);
init_class_id(Class_FastLock);
_counters = NULL;
_rtm_counters = NULL;
_stack_rtm_counters = NULL;
}
Node* obj_node() const { return in(1); }
Node* box_node() const { return in(2); }
void set_box_node(Node* box) { set_req(2, box); }
virtual uint hash() const ; // { return NO_HASH; }
virtual uint size_of() const;
virtual uint cmp( const Node &n ) const ; // Always fail, except on self
virtual int Opcode() const;
virtual const Type *Value( PhaseTransform *phase ) const { return TypeInt::CC; }
const Type *sub(const Type *t1, const Type *t2) const { return TypeInt::CC;}
void create_lock_counter(JVMState* s);
void create_rtm_lock_counter(JVMState* state);
BiasedLockingCounters* counters() const { return _counters; }
RTMLockingCounters* rtm_counters() const { return _rtm_counters; }
RTMLockingCounters* stack_rtm_counters() const { return _stack_rtm_counters; }
};
class FastUnlockNode: public CmpNode {
public:
FastUnlockNode(Node *ctrl, Node *oop, Node *box) : CmpNode(oop,box) {
init_req(0,ctrl);
init_class_id(Class_FastUnlock);
}
Node* obj_node() const { return in(1); }
Node* box_node() const { return in(2); }
virtual uint hash() const ; // { return NO_HASH; }
virtual uint cmp( const Node &n ) const ; // Always fail, except on self
virtual int Opcode() const;
virtual const Type *Value( PhaseTransform *phase ) const { return TypeInt::CC; }
const Type *sub(const Type *t1, const Type *t2) const { return TypeInt::CC;}
};
#endif // SHARE_VM_OPTO_LOCKNODE_HPP
C:\hotspot-69087d08d473\src\share\vm/opto/loopnode.cpp
#include "precompiled.hpp"
#include "ci/ciMethodData.hpp"
#include "compiler/compileLog.hpp"
#include "libadt/vectset.hpp"
#include "memory/allocation.inline.hpp"
#include "opto/addnode.hpp"
#include "opto/callnode.hpp"
#include "opto/connode.hpp"
#include "opto/divnode.hpp"
#include "opto/idealGraphPrinter.hpp"
#include "opto/loopnode.hpp"
#include "opto/mulnode.hpp"
#include "opto/rootnode.hpp"
#include "opto/superword.hpp"
const Node* Node::is_loop_iv() const {
if (this->is_Phi() && !this->as_Phi()->is_copy() &&
this->as_Phi()->region()->is_CountedLoop() &&
this->as_Phi()->region()->as_CountedLoop()->phi() == this) {
return this;
} else {
return NULL;
}
}
#ifndef PRODUCT
void LoopNode::dump_spec(outputStream *st) const {
if (is_inner_loop()) st->print( "inner " );
if (is_partial_peel_loop()) st->print( "partial_peel " );
if (partial_peel_has_failed()) st->print( "partial_peel_failed " );
}
#endif
bool LoopNode::is_valid_counted_loop() const {
if (is_CountedLoop()) {
CountedLoopNode* l = as_CountedLoop();
CountedLoopEndNode* le = l->loopexit();
if (le != NULL &&
le->proj_out(1 /* true */) == l->in(LoopNode::LoopBackControl)) {
Node* phi = l->phi();
Node* exit = le->proj_out(0 /* false */);
if (exit != NULL && exit->Opcode() == Op_IfFalse &&
phi != NULL && phi->is_Phi() &&
phi->in(LoopNode::LoopBackControl) == l->incr() &&
le->loopnode() == l && le->stride_is_con()) {
return true;
}
}
}
return false;
}
Node *PhaseIdealLoop::get_early_ctrl( Node *n ) {
assert( !n->is_Phi() && !n->is_CFG(), "this code only handles data nodes" );
uint i;
Node *early;
if (n->in(0) && !n->is_expensive()) {
early = n->in(0);
if (!early->is_CFG()) // Might be a non-CFG multi-def
early = get_ctrl(early); // So treat input as a straight data input
i = 1;
} else {
early = get_ctrl(n->in(1));
i = 2;
}
uint e_d = dom_depth(early);
assert( early, "" );
for (; i < n->req(); i++) {
Node *cin = get_ctrl(n->in(i));
assert( cin, "" );
uint c_d = dom_depth(cin);
if (c_d > e_d) { // Deeper guy?
early = cin; // Keep deepest found so far
e_d = c_d;
} else if (c_d == e_d && // Same depth?
early != cin) { // If not equal, must use slower algorithm
Node *n1 = early;
Node *n2 = cin;
while (1) {
n1 = idom(n1); // Walk up until break cycle
n2 = idom(n2);
if (n1 == cin || // Walked early up to cin
dom_depth(n2) < c_d)
break; // early is deeper; keep him
if (n2 == early || // Walked cin up to early
dom_depth(n1) < c_d) {
early = cin; // cin is deeper; keep him
break;
}
}
e_d = dom_depth(early); // Reset depth register cache
}
}
assert(early == find_non_split_ctrl(early), "unexpected early control");
if (n->is_expensive()) {
assert(n->in(0), "should have control input");
early = get_early_ctrl_for_expensive(n, early);
}
return early;
}
Node *PhaseIdealLoop::get_early_ctrl_for_expensive(Node *n, Node* earliest) {
assert(n->in(0) && n->is_expensive(), "expensive node with control input here");
assert(OptimizeExpensiveOps, "optimization off?");
Node* ctl = n->in(0);
assert(ctl->is_CFG(), "expensive input 0 must be cfg");
uint min_dom_depth = dom_depth(earliest);
#ifdef ASSERT
if (!is_dominator(ctl, earliest) && !is_dominator(earliest, ctl)) {
dump_bad_graph("Bad graph detected in get_early_ctrl_for_expensive", n, earliest, ctl);
assert(false, "Bad graph detected in get_early_ctrl_for_expensive");
}
#endif
if (dom_depth(ctl) < min_dom_depth) {
return earliest;
}
while (1) {
Node *next = ctl;
if (ctl->is_CountedLoop() && ctl->in(1) != NULL && ctl->in(1)->in(0) != NULL && ctl->in(1)->in(0)->is_If()) {
if (!ctl->in(1)->as_Proj()->is_uncommon_trap_if_pattern(Deoptimization::Reason_none)) {
break;
}
next = idom(ctl->in(1)->in(0));
} else if (ctl->is_Proj()) {
Node* parent_ctl = ctl->in(0);
if (parent_ctl == NULL) {
break;
} else if (parent_ctl->is_CountedLoopEnd() && parent_ctl->as_CountedLoopEnd()->loopnode() != NULL) {
next = parent_ctl->as_CountedLoopEnd()->loopnode()->init_control();
} else if (parent_ctl->is_If()) {
if (!ctl->as_Proj()->is_uncommon_trap_if_pattern(Deoptimization::Reason_none)) {
break;
}
assert(idom(ctl) == parent_ctl, "strange");
next = idom(parent_ctl);
} else if (ctl->is_CatchProj()) {
if (ctl->as_Proj()->_con != CatchProjNode::fall_through_index) {
break;
}
assert(parent_ctl->in(0)->in(0)->is_Call(), "strange graph");
next = parent_ctl->in(0)->in(0)->in(0);
} else {
int nb_ctl_proj = 0;
for (DUIterator_Fast imax, i = parent_ctl->fast_outs(imax); i < imax; i++) {
Node *p = parent_ctl->fast_out(i);
if (p->is_Proj() && p->is_CFG()) {
nb_ctl_proj++;
if (nb_ctl_proj > 1) {
break;
}
}
}
if (nb_ctl_proj > 1) {
break;
}
assert(parent_ctl->is_Start() || parent_ctl->is_MemBar() || parent_ctl->is_Call(), "unexpected node");
assert(idom(ctl) == parent_ctl, "strange");
next = idom(parent_ctl);
}
} else {
next = idom(ctl);
}
if (next->is_Root() || next->is_Start() || dom_depth(next) < min_dom_depth) {
break;
}
ctl = next;
}
if (ctl != n->in(0)) {
_igvn.hash_delete(n);
n->set_req(0, ctl);
_igvn.hash_insert(n);
}
return ctl;
}
void PhaseIdealLoop::set_early_ctrl( Node *n ) {
Node *early = get_early_ctrl(n);
set_ctrl(n, early);
}
void PhaseIdealLoop::set_subtree_ctrl( Node *n ) {
if( _nodes[n->_idx] ) return;
uint i;
for( i = 0; i < n->req(); ++i ) {
Node *m = n->in(i);
if( m && m != C->root() )
set_subtree_ctrl( m );
}
set_early_ctrl( n );
}
bool PhaseIdealLoop::is_counted_loop( Node *x, IdealLoopTree *loop ) {
PhaseGVN *gvn = &_igvn;
if (x->in(LoopNode::Self) == NULL || x->req() != 3 || loop->_irreducible) {
return false;
}
Node *init_control = x->in(LoopNode::EntryControl);
Node *back_control = x->in(LoopNode::LoopBackControl);
if (init_control == NULL || back_control == NULL) // Partially dead
return false;
if (init_control->is_top() || back_control->is_top())
return false;
if (back_control->Opcode() == Op_SafePoint) {
if (UseCountedLoopSafepoints) {
return false;
}
back_control = back_control->in(TypeFunc::Control);
}
Node *iftrue = back_control;
uint iftrue_op = iftrue->Opcode();
if (iftrue_op != Op_IfTrue &&
iftrue_op != Op_IfFalse)
return false; // Obscure back-control
Node *iff = iftrue->in(0);
if (get_loop(iff) != loop || !iff->in(1)->is_Bool())
return false;
BoolNode *test = iff->in(1)->as_Bool();
BoolTest::mask bt = test->_test._test;
float cl_prob = iff->as_If()->_prob;
if (iftrue_op == Op_IfFalse) {
bt = BoolTest(bt).negate();
cl_prob = 1.0 - cl_prob;
}
Node *cmp = test->in(1);
int cmp_op = cmp->Opcode();
if (cmp_op != Op_CmpI)
return false; // Avoid pointer & float compares
Node *incr = cmp->in(1);
Node *limit = cmp->in(2);
if (!is_member(loop, get_ctrl(incr))) { // Swapped trip counter and limit?
Node *tmp = incr; // Then reverse order into the CmpI
incr = limit;
limit = tmp;
bt = BoolTest(bt).commute(); // And commute the exit test
}
if (is_member(loop, get_ctrl(limit))) // Limit must be loop-invariant
return false;
if (!is_member(loop, get_ctrl(incr))) // Trip counter must be loop-variant
return false;
Node* phi_incr = NULL;
if (incr->is_Phi()) {
if (incr->as_Phi()->region() != x || incr->req() != 3)
return false; // Not simple trip counter expression
phi_incr = incr;
incr = phi_incr->in(LoopNode::LoopBackControl); // Assume incr is on backedge of Phi
if (!is_member(loop, get_ctrl(incr))) // Trip counter must be loop-variant
return false;
}
Node* trunc1 = NULL;
Node* trunc2 = NULL;
const TypeInt* iv_trunc_t = NULL;
if (!(incr = CountedLoopNode::match_incr_with_optional_truncation(incr, &trunc1, &trunc2, &iv_trunc_t))) {
return false; // Funny increment opcode
}
assert(incr->Opcode() == Op_AddI, "wrong increment code");
Node *xphi = incr->in(1);
Node *stride = incr->in(2);
if (!stride->is_Con()) { // Oops, swap these
if (!xphi->is_Con()) // Is the other guy a constant?
return false; // Nope, unknown stride, bail out
Node *tmp = xphi; // 'incr' is commutative, so ok to swap
xphi = stride;
stride = tmp;
}
int stride_con = stride->get_int();
if (stride_con == 0)
return false; // missed some peephole opt
if (!xphi->is_Phi())
return false; // Too much math on the trip counter
if (phi_incr != NULL && phi_incr != xphi)
return false;
PhiNode *phi = xphi->as_Phi();
if (phi->region() != x)
return false;
if (trunc1 == NULL && phi->in(LoopNode::LoopBackControl) != incr ||
trunc1 != NULL && phi->in(LoopNode::LoopBackControl) != trunc1) {
return false;
}
Node *init_trip = phi->in(LoopNode::EntryControl);
if (!TypeInt::INT->higher_equal(iv_trunc_t)) {
assert(trunc1 != NULL, "must have found some truncation");
const TypeInt* phi_ft = filtered_type(phi);
if (stride_con > 0) {
if (iv_trunc_t->_hi - phi_ft->_hi < stride_con ||
iv_trunc_t->_lo > phi_ft->_lo) {
return false; // truncation may occur
}
} else if (stride_con < 0) {
if (iv_trunc_t->_lo - phi_ft->_lo > stride_con ||
iv_trunc_t->_hi < phi_ft->_hi) {
return false; // truncation may occur
}
}
} else {
assert(trunc1 == NULL && trunc2 == NULL, "no truncation for int");
}
if (bt == BoolTest::eq || // Bail out, but this loop trips at most twice!
bt == BoolTest::ne && stride_con != 1 && stride_con != -1 ||
(bt == BoolTest::le || bt == BoolTest::lt) && stride_con < 0 ||
(bt == BoolTest::ge || bt == BoolTest::gt) && stride_con > 0) {
return false; // Bail out
}
const TypeInt* init_t = gvn->type(init_trip)->is_int();
const TypeInt* limit_t = gvn->type(limit)->is_int();
if (stride_con > 0) {
jlong init_p = (jlong)init_t->_lo + stride_con;
if (init_p > (jlong)max_jint || init_p > (jlong)limit_t->_hi)
return false; // cyclic loop or this loop trips only once
} else {
jlong init_p = (jlong)init_t->_hi + stride_con;
if (init_p < (jlong)min_jint || init_p < (jlong)limit_t->_lo)
return false; // cyclic loop or this loop trips only once
}
if (phi_incr != NULL) {
if (stride_con > 0) {
if (init_t->_hi > max_jint - stride_con) {
return false;
}
} else {
if (init_t->_lo < min_jint - stride_con) {
return false;
}
}
}
assert(x->Opcode() == Op_Loop, "regular loops only");
C->print_method(PHASE_BEFORE_CLOOPS, 3);
Node *hook = new (C) Node(6);
if (LoopLimitCheck) {
bool incl_limit = (bt == BoolTest::le || bt == BoolTest::ge);
int stride_m = stride_con - (incl_limit ? 0 : (stride_con > 0 ? 1 : -1));
if (phi_incr != NULL) {
stride_m += stride_con;
}
if (limit->is_Con()) {
int limit_con = limit->get_int();
if ((stride_con > 0 && limit_con > (max_jint - stride_m)) ||
(stride_con < 0 && limit_con < (min_jint - stride_m))) {
return false;
}
} else if ((stride_con > 0 && limit_t->_hi <= (max_jint - stride_m)) ||
(stride_con < 0 && limit_t->_lo >= (min_jint - stride_m))) {
} else {
ProjNode *limit_check_proj = find_predicate_insertion_point(init_control, Deoptimization::Reason_loop_limit_check);
if (!limit_check_proj) {
#ifdef ASSERT
if (TraceLoopLimitCheck) {
tty->print("missing loop limit check:");
loop->dump_head();
x->dump(1);
}
#endif
return false;
}
IfNode* check_iff = limit_check_proj->in(0)->as_If();
Node* cmp_limit;
Node* bol;
if (stride_con > 0) {
cmp_limit = new (C) CmpINode(limit, _igvn.intcon(max_jint - stride_m));
bol = new (C) BoolNode(cmp_limit, BoolTest::le);
} else {
cmp_limit = new (C) CmpINode(limit, _igvn.intcon(min_jint - stride_m));
bol = new (C) BoolNode(cmp_limit, BoolTest::ge);
}
cmp_limit = _igvn.register_new_node_with_optimizer(cmp_limit);
bol = _igvn.register_new_node_with_optimizer(bol);
set_subtree_ctrl(bol);
assert(check_iff->in(1)->Opcode() == Op_Conv2B &&
check_iff->in(1)->in(1)->Opcode() == Op_Opaque1, "");
Node* opq = check_iff->in(1)->in(1);
_igvn.hash_delete(opq);
opq->set_req(1, bol);
set_ctrl(opq, check_iff->in(0));
set_ctrl(check_iff->in(1), check_iff->in(0));
#ifndef PRODUCT
if (TraceLoopLimitCheck) {
tty->print_cr("Counted Loop Limit Check generated:");
debug_only( bol->dump(2); )
}
#endif
}
if (phi_incr != NULL) {
limit = gvn->transform(new (C) AddINode(limit, stride));
}
if (bt == BoolTest::ne) {
assert(stride_con == 1 || stride_con == -1, "simple increment only");
if (stride_con > 0 && init_t->_hi < limit_t->_lo)
bt = BoolTest::lt;
if (stride_con < 0 && init_t->_lo > limit_t->_hi)
bt = BoolTest::gt;
}
if (incl_limit) {
Node* one = (stride_con > 0) ? gvn->intcon( 1) : gvn->intcon(-1);
limit = gvn->transform(new (C) AddINode(limit, one));
if (bt == BoolTest::le)
bt = BoolTest::lt;
else if (bt == BoolTest::ge)
bt = BoolTest::gt;
else
ShouldNotReachHere();
}
set_subtree_ctrl( limit );
} else { // LoopLimitCheck
if (cmp->in(1) == phi || cmp->in(2) == phi)
limit = gvn->transform(new (C) AddINode(limit,stride));
Node *one_p = gvn->intcon( 1);
Node *one_m = gvn->intcon(-1);
Node *trip_count = NULL;
switch( bt ) {
case BoolTest::eq:
ShouldNotReachHere();
case BoolTest::ne: // Ahh, the case we desire
if (stride_con == 1)
trip_count = gvn->transform(new (C) SubINode(limit,init_trip));
else if (stride_con == -1)
trip_count = gvn->transform(new (C) SubINode(init_trip,limit));
else
ShouldNotReachHere();
set_subtree_ctrl(trip_count);
break;
case BoolTest::le: // Maybe convert to '<' case
limit = gvn->transform(new (C) AddINode(limit,one_p));
set_subtree_ctrl( limit );
hook->init_req(4, limit);
bt = BoolTest::lt;
case BoolTest::lt: { // Maybe convert to '!=' case
if (stride_con < 0) // Count down loop rolls through MAXINT
ShouldNotReachHere();
Node *range = gvn->transform(new (C) SubINode(limit,init_trip));
set_subtree_ctrl( range );
hook->init_req(0, range);
Node *bias = gvn->transform(new (C) AddINode(range,stride));
set_subtree_ctrl( bias );
hook->init_req(1, bias);
Node *bias1 = gvn->transform(new (C) AddINode(bias,one_m));
set_subtree_ctrl( bias1 );
hook->init_req(2, bias1);
trip_count = gvn->transform(new (C) DivINode(0,bias1,stride));
set_subtree_ctrl( trip_count );
hook->init_req(3, trip_count);
break;
}
case BoolTest::ge: // Maybe convert to '>' case
limit = gvn->transform(new (C) AddINode(limit,one_m));
set_subtree_ctrl( limit );
hook->init_req(4 ,limit);
bt = BoolTest::gt;
case BoolTest::gt: { // Maybe convert to '!=' case
if (stride_con > 0) // count up loop rolls through MININT
ShouldNotReachHere();
Node *range = gvn->transform(new (C) SubINode(limit,init_trip));
set_subtree_ctrl( range );
hook->init_req(0, range);
Node *bias = gvn->transform(new (C) AddINode(range,stride));
set_subtree_ctrl( bias );
hook->init_req(1, bias);
Node *bias1 = gvn->transform(new (C) AddINode(bias,one_p));
set_subtree_ctrl( bias1 );
hook->init_req(2, bias1);
trip_count = gvn->transform(new (C) DivINode(0,bias1,stride));
set_subtree_ctrl( trip_count );
hook->init_req(3, trip_count);
break;
}
} // switch( bt )
Node *span = gvn->transform(new (C) MulINode(trip_count,stride));
set_subtree_ctrl( span );
hook->init_req(5, span);
limit = gvn->transform(new (C) AddINode(span,init_trip));
set_subtree_ctrl( limit );
} // LoopLimitCheck
if (!UseCountedLoopSafepoints) {
Node *sfpt = x->in(LoopNode::LoopBackControl);
if (sfpt->Opcode() == Op_SafePoint && is_deleteable_safept(sfpt)) {
lazy_replace( sfpt, iftrue );
if (loop->_safepts != NULL) {
loop->_safepts->yank(sfpt);
}
loop->_tail = iftrue;
}
}
incr = incr->clone();
incr->set_req(1,phi);
incr->set_req(2,stride);
incr = _igvn.register_new_node_with_optimizer(incr);
set_early_ctrl( incr );
_igvn.hash_delete(phi);
phi->set_req_X( LoopNode::LoopBackControl, incr, &_igvn );
if (!TypeInt::INT->higher_equal(phi->bottom_type())) {
Node* nphi = PhiNode::make(phi->in(0), phi->in(LoopNode::EntryControl), TypeInt::INT);
nphi->set_req(LoopNode::LoopBackControl, phi->in(LoopNode::LoopBackControl));
nphi = _igvn.register_new_node_with_optimizer(nphi);
set_ctrl(nphi, get_ctrl(phi));
_igvn.replace_node(phi, nphi);
phi = nphi->as_Phi();
}
cmp = cmp->clone();
cmp->set_req(1,incr);
cmp->set_req(2,limit);
cmp = _igvn.register_new_node_with_optimizer(cmp);
set_ctrl(cmp, iff->in(0));
test = test->clone()->as_Bool();
(*(BoolTest*)&test->_test)._test = bt;
test->set_req(1,cmp);
_igvn.register_new_node_with_optimizer(test);
set_ctrl(test, iff->in(0));
Node *lex = _igvn.register_new_node_with_optimizer(new (C) CountedLoopEndNode( iff->in(0), test, cl_prob, iff->as_If()->_fcnt ));
IfNode *le = lex->as_If();
uint dd = dom_depth(iff);
set_idom(le, le->in(0), dd); // Update dominance for loop exit
set_loop(le, loop);
Node *iffalse = iff->as_If()->proj_out(!(iftrue_op == Op_IfTrue));
if (iftrue_op == Op_IfFalse) {
Node *ift2=_igvn.register_new_node_with_optimizer(new (C) IfTrueNode (le));
Node *iff2=_igvn.register_new_node_with_optimizer(new (C) IfFalseNode(le));
loop->_tail = back_control = ift2;
set_loop(ift2, loop);
set_loop(iff2, get_loop(iffalse));
lazy_replace(iffalse, iff2);
lazy_replace(iftrue, ift2);
iffalse = iff2;
iftrue = ift2;
} else {
_igvn.hash_delete(iffalse);
_igvn.hash_delete(iftrue);
iffalse->set_req_X( 0, le, &_igvn );
iftrue ->set_req_X( 0, le, &_igvn );
}
set_idom(iftrue, le, dd+1);
set_idom(iffalse, le, dd+1);
assert(iff->outcnt() == 0, "should be dead now");
lazy_replace( iff, le ); // fix 'get_ctrl'
CountedLoopNode *l = new (C) CountedLoopNode(init_control, back_control);
l->set_unswitch_count(x->as_Loop()->unswitch_count()); // Preserve
_igvn.register_new_node_with_optimizer(l);
set_loop(l, loop);
loop->_head = l;
lazy_replace( x, l );
set_idom(l, init_control, dom_depth(x));
if (!UseCountedLoopSafepoints) {
Node *sfpt2 = le->in(0);
if (sfpt2->Opcode() == Op_SafePoint && is_deleteable_safept(sfpt2)) {
lazy_replace( sfpt2, sfpt2->in(TypeFunc::Control));
if (loop->_safepts != NULL) {
loop->_safepts->yank(sfpt2);
}
}
}
_igvn.remove_dead_node(hook);
#ifdef ASSERT
assert(l->is_valid_counted_loop(), "counted loop shape is messed up");
assert(l == loop->_head && l->phi() == phi && l->loopexit() == lex, "" );
#endif
#ifndef PRODUCT
if (TraceLoopOpts) {
tty->print("Counted ");
loop->dump_head();
}
#endif
C->print_method(PHASE_AFTER_CLOOPS, 3);
return true;
}
Node* PhaseIdealLoop::exact_limit( IdealLoopTree *loop ) {
assert(loop->_head->is_CountedLoop(), "");
CountedLoopNode *cl = loop->_head->as_CountedLoop();
assert(cl->is_valid_counted_loop(), "");
if (!LoopLimitCheck || ABS(cl->stride_con()) == 1 ||
cl->limit()->Opcode() == Op_LoopLimit) {
return cl->limit();
}
Node *limit = NULL;
#ifdef ASSERT
BoolTest::mask bt = cl->loopexit()->test_trip();
assert(bt == BoolTest::lt || bt == BoolTest::gt, "canonical test is expected");
#endif
if (cl->has_exact_trip_count()) {
int stride_con = cl->stride_con();
jlong init_con = cl->init_trip()->get_int();
jlong limit_con = cl->limit()->get_int();
julong trip_cnt = cl->trip_count();
jlong final_con = init_con + trip_cnt*stride_con;
int final_int = (int)final_con;
assert(final_con == (jlong)final_int, "final value should be integer");
limit = _igvn.intcon(final_int);
} else {
limit = new (C) LoopLimitNode(C, cl->init_trip(), cl->limit(), cl->stride());
register_new_node(limit, cl->in(LoopNode::EntryControl));
}
assert(limit != NULL, "sanity");
return limit;
}
Node *LoopNode::Ideal(PhaseGVN *phase, bool can_reshape) {
if (!can_be_counted_loop(phase)) {
phase->C->set_major_progress();
}
return RegionNode::Ideal(phase, can_reshape);
}
Node *CountedLoopNode::Ideal(PhaseGVN *phase, bool can_reshape) {
return RegionNode::Ideal(phase, can_reshape);
}
#ifndef PRODUCT
void CountedLoopNode::dump_spec(outputStream *st) const {
LoopNode::dump_spec(st);
if (stride_is_con()) {
st->print("stride: %d ",stride_con());
}
if (is_pre_loop ()) st->print("pre of N%d" , _main_idx);
if (is_main_loop()) st->print("main of N%d", _idx);
if (is_post_loop()) st->print("post of N%d", _main_idx);
}
#endif
int CountedLoopEndNode::stride_con() const {
return stride()->bottom_type()->is_int()->get_con();
}
const Type *LoopLimitNode::Value( PhaseTransform *phase ) const {
const Type* init_t = phase->type(in(Init));
const Type* limit_t = phase->type(in(Limit));
const Type* stride_t = phase->type(in(Stride));
if (init_t == Type::TOP) return Type::TOP;
if (limit_t == Type::TOP) return Type::TOP;
if (stride_t == Type::TOP) return Type::TOP;
int stride_con = stride_t->is_int()->get_con();
if (stride_con == 1)
return NULL; // Identity
if (init_t->is_int()->is_con() && limit_t->is_int()->is_con()) {
jlong init_con = init_t->is_int()->get_con();
jlong limit_con = limit_t->is_int()->get_con();
int stride_m = stride_con - (stride_con > 0 ? 1 : -1);
jlong trip_count = (limit_con - init_con + stride_m)/stride_con;
jlong final_con = init_con + stride_con*trip_count;
int final_int = (int)final_con;
assert(final_con == (jlong)final_int, "final value should be integer");
return TypeInt::make(final_int);
}
return bottom_type(); // TypeInt::INT
}
Node *LoopLimitNode::Ideal(PhaseGVN *phase, bool can_reshape) {
if (phase->type(in(Init)) == Type::TOP ||
phase->type(in(Limit)) == Type::TOP ||
phase->type(in(Stride)) == Type::TOP)
return NULL; // Dead
int stride_con = phase->type(in(Stride))->is_int()->get_con();
if (stride_con == 1)
return NULL; // Identity
if (in(Init)->is_Con() && in(Limit)->is_Con())
return NULL; // Value
if (!can_reshape || phase->C->major_progress())
return NULL;
const TypeInt* init_t = phase->type(in(Init) )->is_int();
const TypeInt* limit_t = phase->type(in(Limit))->is_int();
int stride_p;
jlong lim, ini;
julong max;
if (stride_con > 0) {
stride_p = stride_con;
lim = limit_t->_hi;
ini = init_t->_lo;
max = (julong)max_jint;
} else {
stride_p = -stride_con;
lim = init_t->_hi;
ini = limit_t->_lo;
max = (julong)min_jint;
}
julong range = lim - ini + stride_p;
if (range <= max) {
Node* stride_m = phase->intcon(stride_con - (stride_con > 0 ? 1 : -1));
Node *range = phase->transform(new (phase->C) SubINode(in(Limit), in(Init)));
Node *bias = phase->transform(new (phase->C) AddINode(range, stride_m));
Node *trip = phase->transform(new (phase->C) DivINode(0, bias, in(Stride)));
Node *span = phase->transform(new (phase->C) MulINode(trip, in(Stride)));
return new (phase->C) AddINode(span, in(Init)); // exact limit
}
if (is_power_of_2(stride_p) || // divisor is 2^n
!Matcher::has_match_rule(Op_LoopLimit)) { // or no specialized Mach node?
Node* init = phase->transform( new (phase->C) ConvI2LNode(in(Init)));
Node* limit = phase->transform( new (phase->C) ConvI2LNode(in(Limit)));
Node* stride = phase->longcon(stride_con);
Node* stride_m = phase->longcon(stride_con - (stride_con > 0 ? 1 : -1));
Node *range = phase->transform(new (phase->C) SubLNode(limit, init));
Node *bias = phase->transform(new (phase->C) AddLNode(range, stride_m));
Node *span;
if (stride_con > 0 && is_power_of_2(stride_p)) {
Node* neg_stride = phase->longcon(-stride_con);
span = phase->transform(new (phase->C) AndLNode(bias, neg_stride));
} else {
Node *trip = phase->transform(new (phase->C) DivLNode(0, bias, stride));
span = phase->transform(new (phase->C) MulLNode(trip, stride));
}
Node *span_int = phase->transform(new (phase->C) ConvL2INode(span));
return new (phase->C) AddINode(span_int, in(Init)); // exact limit
}
return NULL; // No progress
}
Node *LoopLimitNode::Identity( PhaseTransform *phase ) {
int stride_con = phase->type(in(Stride))->is_int()->get_con();
if (stride_con == 1 || stride_con == -1)
return in(Limit);
return this;
}
Node* CountedLoopNode::match_incr_with_optional_truncation(
Node* expr, Node** trunc1, Node** trunc2, const TypeInt** trunc_type) {
if (expr == NULL || expr->req() != 3) return NULL;
Node *t1 = NULL;
Node *t2 = NULL;
const TypeInt* trunc_t = TypeInt::INT;
Node* n1 = expr;
int n1op = n1->Opcode();
if (n1op == Op_AndI &&
n1->in(2)->is_Con() &&
n1->in(2)->bottom_type()->is_int()->get_con() == 0x7fff) {
t1 = n1;
n1 = t1->in(1);
n1op = n1->Opcode();
trunc_t = TypeInt::CHAR;
} else if (n1op == Op_RShiftI &&
n1->in(1) != NULL &&
n1->in(1)->Opcode() == Op_LShiftI &&
n1->in(2) == n1->in(1)->in(2) &&
n1->in(2)->is_Con()) {
jint shift = n1->in(2)->bottom_type()->is_int()->get_con();
if (shift == 16 || shift == 8) {
t1 = n1;
t2 = t1->in(1);
n1 = t2->in(1);
n1op = n1->Opcode();
if (shift == 16) {
trunc_t = TypeInt::SHORT;
} else if (shift == 8) {
trunc_t = TypeInt::BYTE;
}
}
}
if (n1op == Op_AddI) {
return n1;
}
return NULL;
}
const TypeInt* PhaseIdealLoop::filtered_type( Node *n, Node* n_ctrl) {
assert(n && n->bottom_type()->is_int(), "must be int");
const TypeInt* filtered_t = NULL;
if (!n->is_Phi()) {
assert(n_ctrl != NULL || n_ctrl == C->top(), "valid control");
filtered_t = filtered_type_from_dominators(n, n_ctrl);
} else {
Node* phi = n->as_Phi();
Node* region = phi->in(0);
assert(n_ctrl == NULL || n_ctrl == region, "ctrl parameter must be region");
if (region && region != C->top()) {
for (uint i = 1; i < phi->req(); i++) {
Node* val = phi->in(i);
Node* use_c = region->in(i);
const TypeInt* val_t = filtered_type_from_dominators(val, use_c);
if (val_t != NULL) {
if (filtered_t == NULL) {
filtered_t = val_t;
} else {
filtered_t = filtered_t->meet(val_t)->is_int();
}
}
}
}
}
const TypeInt* n_t = _igvn.type(n)->is_int();
if (filtered_t != NULL) {
n_t = n_t->join(filtered_t)->is_int();
}
return n_t;
}
const TypeInt* PhaseIdealLoop::filtered_type_from_dominators( Node* val, Node *use_ctrl) {
if (val->is_Con()) {
return val->bottom_type()->is_int();
}
uint if_limit = 10; // Max number of dominating if's visited
const TypeInt* rtn_t = NULL;
if (use_ctrl && use_ctrl != C->top()) {
Node* val_ctrl = get_ctrl(val);
uint val_dom_depth = dom_depth(val_ctrl);
Node* pred = use_ctrl;
uint if_cnt = 0;
while (if_cnt < if_limit) {
if ((pred->Opcode() == Op_IfTrue || pred->Opcode() == Op_IfFalse)) {
if_cnt++;
const TypeInt* if_t = IfNode::filtered_int_type(&_igvn, val, pred);
if (if_t != NULL) {
if (rtn_t == NULL) {
rtn_t = if_t;
} else {
rtn_t = rtn_t->join(if_t)->is_int();
}
}
}
pred = idom(pred);
if (pred == NULL || pred == C->top()) {
break;
}
if (dom_depth(pred) < val_dom_depth) {
break;
}
}
}
return rtn_t;
}
#ifndef PRODUCT
void CountedLoopEndNode::dump_spec(outputStream *st) const {
if( in(TestValue)->is_Bool() ) {
BoolTest bt( test_trip()); // Added this for g++.
st->print("[");
bt.dump_on(st);
st->print("]");
}
st->print(" ");
IfNode::dump_spec(st);
}
#endif
int IdealLoopTree::is_member( const IdealLoopTree *l ) const {
while( l->_nest > _nest ) l = l->_parent;
return l == this;
}
int IdealLoopTree::set_nest( uint depth ) {
_nest = depth;
int bits = _has_call;
if( _child ) bits |= _child->set_nest(depth+1);
if( bits ) _has_call = 1;
if( _next ) bits |= _next ->set_nest(depth );
return bits;
}
void IdealLoopTree::split_fall_in( PhaseIdealLoop *phase, int fall_in_cnt ) {
PhaseIterGVN &igvn = phase->_igvn;
uint i;
Node *landing_pad = new (phase->C) RegionNode( fall_in_cnt+1 );
phase->set_loop(landing_pad,_parent);
uint icnt = fall_in_cnt;
uint oreq = _head->req();
for( i = oreq-1; i>0; i-- )
if( !phase->is_member( this, _head->in(i) ) )
landing_pad->set_req(icnt--,_head->in(i));
for (DUIterator_Fast jmax, j = _head->fast_outs(jmax); j < jmax; j++) {
Node *oj = _head->fast_out(j);
if( oj->is_Phi() ) {
PhiNode* old_phi = oj->as_Phi();
assert( old_phi->region() == _head, "" );
igvn.hash_delete(old_phi); // Yank from hash before hacking edges
Node *p = PhiNode::make_blank(landing_pad, old_phi);
uint icnt = fall_in_cnt;
for( i = oreq-1; i>0; i-- ) {
if( !phase->is_member( this, _head->in(i) ) ) {
p->init_req(icnt--, old_phi->in(i));
old_phi->del_req(i);
}
}
Node *p2 = igvn.hash_find_insert(p); // Look for a CSE
if( p2 ) { // Found CSE
p->destruct(); // Recover useless new node
p = p2; // Use old node
} else {
igvn.register_new_node_with_optimizer(p, old_phi);
}
old_phi->add_req(p);
Node *id_old_phi = old_phi->Identity( &igvn );
if( id_old_phi != old_phi ) { // Found a simple identity?
for (DUIterator_Last imin, i = old_phi->last_outs(imin); i >= imin; ) {
Node* use = old_phi->last_out(i);
igvn.rehash_node_delayed(use);
uint uses_found = 0;
for (uint j = 0; j < use->len(); j++) {
if (use->in(j) == old_phi) {
if (j < use->req()) use->set_req (j, id_old_phi);
else use->set_prec(j, id_old_phi);
uses_found++;
}
}
i -= uses_found; // we deleted 1 or more copies of this edge
}
}
igvn._worklist.push(old_phi);
}
}
for( i = oreq-1; i>0; i-- ) {
if( !phase->is_member( this, _head->in(i) ) ) {
_head->del_req(i);
}
}
igvn.register_new_node_with_optimizer(landing_pad, _head);
_head->add_req(landing_pad);
}
void IdealLoopTree::split_outer_loop( PhaseIdealLoop *phase ) {
PhaseIterGVN &igvn = phase->_igvn;
uint outer_idx = 1;
while( _head->in(outer_idx) != _tail ) outer_idx++;
Node *ctl = _head->in(LoopNode::EntryControl);
Node *outer = new (phase->C) LoopNode( ctl, _head->in(outer_idx) );
outer = igvn.register_new_node_with_optimizer(outer, _head);
phase->set_created_loop_node();
_head->set_req(LoopNode::EntryControl, outer);
_head->del_req(outer_idx);
for (DUIterator_Fast jmax, j = _head->fast_outs(jmax); j < jmax; j++) {
Node *out = _head->fast_out(j);
if( out->is_Phi() ) {
PhiNode *old_phi = out->as_Phi();
assert( old_phi->region() == _head, "" );
Node *phi = PhiNode::make_blank(outer, old_phi);
phi->init_req(LoopNode::EntryControl, old_phi->in(LoopNode::EntryControl));
phi->init_req(LoopNode::LoopBackControl, old_phi->in(outer_idx));
phi = igvn.register_new_node_with_optimizer(phi, old_phi);
igvn.replace_input_of(old_phi, LoopNode::EntryControl, phi);
old_phi->del_req(outer_idx);
}
}
_head = outer;
phase->set_loop(_head, this);
}
static void fix_parent( IdealLoopTree *loop, IdealLoopTree *parent ) {
loop->_parent = parent;
if( loop->_child ) fix_parent( loop->_child, loop );
if( loop->_next ) fix_parent( loop->_next , parent );
}
static float estimate_path_freq( Node *n ) {
IfNode *iff;
for( int i = 0; i < 50; i++ ) { // Skip through a bunch of uncommon tests
uint nop = n->Opcode();
if( nop == Op_SafePoint ) { // Skip any safepoint
n = n->in(0);
continue;
}
if( nop == Op_CatchProj ) { // Get count from a prior call
assert( n->is_CatchProj(), "" );
if( ((CatchProjNode*)n)->_con != CatchProjNode::fall_through_index )
return 0.0f; // Assume call exception path is rare
Node *call = n->in(0)->in(0)->in(0);
assert( call->is_Call(), "expect a call here" );
const JVMState *jvms = ((CallNode*)call)->jvms();
ciMethodData* methodData = jvms->method()->method_data();
if (!methodData->is_mature()) return 0.0f; // No call-site data
ciProfileData* data = methodData->bci_to_data(jvms->bci());
if ((data == NULL) || !data->is_CounterData()) {
n = n->in(0);
continue;
}
return data->as_CounterData()->count()/FreqCountInvocations;
}
Node *n_c = n->in(0);
if( !n_c->is_If() ) break; // No estimate available
iff = n_c->as_If();
if( iff->_fcnt != COUNT_UNKNOWN ) // Have a valid count?
return ((nop == Op_IfTrue) ? iff->_prob : 1.0f - iff->_prob) * iff->_fcnt;
if( (nop == Op_IfTrue && iff->_prob < PROB_LIKELY_MAG(5)) ||
(nop == Op_IfFalse && iff->_prob > PROB_UNLIKELY_MAG(5)) )
break;
n = iff->in(0);
}
return 0.0f; // No estimate available
}
void IdealLoopTree::merge_many_backedges( PhaseIdealLoop *phase ) {
uint i;
float hotcnt = 0.0f;
float warmcnt = 0.0f;
uint hot_idx = 0;
for( i = 2; i < _head->req(); i++ ) {
float cnt = estimate_path_freq(_head->in(i));
if( cnt > hotcnt ) { // Grab hottest path
warmcnt = hotcnt;
hotcnt = cnt;
hot_idx = i;
} else if( cnt > warmcnt ) { // And 2nd hottest path
warmcnt = cnt;
}
}
if( hotcnt <= 0.0001 ||
hotcnt < 2.0*warmcnt ) hot_idx = 0;// No hot backedge
PhaseIterGVN &igvn = phase->_igvn;
Node *hot_tail = NULL;
Node *r = new (phase->C) RegionNode(1);
for( i = 2; i < _head->req(); i++ ) {
if( i != hot_idx )
r->add_req( _head->in(i) );
else hot_tail = _head->in(i);
}
igvn.register_new_node_with_optimizer(r, _head);
while( _head->req() > 3 ) _head->del_req( _head->req()-1 );
_head->set_req(2, r);
if( hot_idx ) _head->add_req(hot_tail);
for (DUIterator_Fast jmax, j = _head->fast_outs(jmax); j < jmax; j++) {
Node *out = _head->fast_out(j);
if( out->is_Phi() ) {
PhiNode* n = out->as_Phi();
igvn.hash_delete(n); // Delete from hash before hacking edges
Node *hot_phi = NULL;
Node *phi = new (phase->C) PhiNode(r, n->type(), n->adr_type());
uint j = 1;
for( uint i = 2; i < n->req(); i++ ) {
if( i != hot_idx )
phi->set_req( j++, n->in(i) );
else hot_phi = n->in(i);
}
igvn.register_new_node_with_optimizer(phi, n);
while( n->req() > 3 ) n->del_req( n->req()-1 );
n->set_req(2, phi);
if( hot_idx ) n->add_req(hot_phi);
}
}
IdealLoopTree *ilt = new IdealLoopTree( phase, _head, _tail );
phase->set_loop(_tail,ilt); // Adjust tail
_tail = r; // Self's tail is new merge point
phase->set_loop(r,this);
ilt->_child = _child; // New guy has my children
_child = ilt; // Self has new guy as only child
ilt->_parent = this; // new guy has self for parent
ilt->_nest = _nest; // Same nesting depth (for now)
IdealLoopTree **pilt = &_child;
while( ilt ) {
if( ilt->_head == _head ) {
uint i;
for( i = 2; i < _head->req(); i++ )
if( _head->in(i) == ilt->_tail )
break; // Still a loop
if( i == _head->req() ) { // No longer a loop
IdealLoopTree **cp = &ilt->_child;
while( *cp ) cp = &(*cp)->_next; // Find end of child list
ilt->_head = NULL; // Flag as a loop UNIONED into parent
ilt = ilt->_child; // Repeat using new ilt
continue; // do not advance over ilt->_child
}
assert( ilt->_tail == hot_tail, "expected to only find the hot inner loop here" );
phase->set_loop(_head,ilt);
}
pilt = &ilt->_child; // Advance to next
ilt = *pilt;
}
if( _child ) fix_parent( _child, this );
}
bool IdealLoopTree::beautify_loops( PhaseIdealLoop *phase ) {
bool result = false;
PhaseIterGVN &igvn = phase->_igvn;
igvn.hash_delete(_head); // Yank from hash before hacking edges
int fall_in_cnt = 0;
for( uint i = 1; i < _head->req(); i++ )
if( !phase->is_member( this, _head->in(i) ) )
fall_in_cnt++;
assert( fall_in_cnt, "at least 1 fall-in path" );
if( fall_in_cnt > 1 ) // Need a loop landing pad to merge fall-ins
split_fall_in( phase, fall_in_cnt );
fall_in_cnt = 1;
while( phase->is_member( this, _head->in(fall_in_cnt) ) )
fall_in_cnt++;
if( fall_in_cnt > 1 ) {
Node *tmp = _head->in(1);
_head->set_req( 1, _head->in(fall_in_cnt) );
_head->set_req( fall_in_cnt, tmp );
for (DUIterator_Fast imax, i = _head->fast_outs(imax); i < imax; i++) {
Node* phi = _head->fast_out(i);
if( phi->is_Phi() ) {
igvn.hash_delete(phi); // Yank from hash before hacking edges
tmp = phi->in(1);
phi->set_req( 1, phi->in(fall_in_cnt) );
phi->set_req( fall_in_cnt, tmp );
}
}
}
assert( !phase->is_member( this, _head->in(1) ), "left edge is fall-in" );
assert( phase->is_member( this, _head->in(2) ), "right edge is loop" );
if (_head->req() > 3) {
if (!_irreducible) {
merge_many_backedges( phase );
}
result = true;
}
if (_head->req() > 3 && !_irreducible) {
split_outer_loop( phase );
result = true;
} else if (!_head->is_Loop() && !_irreducible) {
Node *l = new (phase->C) LoopNode( _head->in(1), _head->in(2) );
l = igvn.register_new_node_with_optimizer(l, _head);
phase->set_created_loop_node();
phase->_igvn.replace_node( _head, l );
_head = l;
phase->set_loop(_head, this);
}
if( _child ) result |= _child->beautify_loops( phase );
if( _next ) result |= _next ->beautify_loops( phase );
return result;
}
void IdealLoopTree::allpaths_check_safepts(VectorSet &visited, Node_List &stack) {
assert(stack.size() == 0, "empty stack");
stack.push(_tail);
visited.Clear();
visited.set(_tail->_idx);
while (stack.size() > 0) {
Node* n = stack.pop();
if (n->is_Call() && n->as_Call()->guaranteed_safepoint()) {
} else if (n->Opcode() == Op_SafePoint) {
if (_phase->get_loop(n) != this) {
if (_required_safept == NULL) _required_safept = new Node_List();
_required_safept->push(n); // save the one closest to the tail
}
} else {
uint start = n->is_Region() ? 1 : 0;
uint end = n->is_Region() && !n->is_Loop() ? n->req() : start + 1;
for (uint i = start; i < end; i++) {
Node* in = n->in(i);
assert(in->is_CFG(), "must be");
if (!visited.test_set(in->_idx) && is_member(_phase->get_loop(in))) {
stack.push(in);
}
}
}
}
}
void IdealLoopTree::check_safepts(VectorSet &visited, Node_List &stack) {
IdealLoopTree* ch = _child;
if (_child) _child->check_safepts(visited, stack);
if (_next) _next ->check_safepts(visited, stack);
if (!_head->is_CountedLoop() && !_has_sfpt && _parent != NULL && !_irreducible) {
bool has_call = false; // call on dom-path
bool has_local_ncsfpt = false; // ncsfpt on dom-path at this loop depth
Node* nonlocal_ncsfpt = NULL; // ncsfpt on dom-path at a deeper depth
for (Node* n = tail(); n != _head; n = _phase->idom(n)) {
if (n->is_Call() && n->as_Call()->guaranteed_safepoint()) {
has_call = true;
_has_sfpt = 1; // Then no need for a safept!
break;
} else if (n->Opcode() == Op_SafePoint) {
if (_phase->get_loop(n) == this) {
has_local_ncsfpt = true;
break;
}
if (nonlocal_ncsfpt == NULL) {
nonlocal_ncsfpt = n; // save the one closest to the tail
}
} else {
IdealLoopTree* nlpt = _phase->get_loop(n);
if (this != nlpt) {
assert(is_member(nlpt), "nested loop");
Node* tail = nlpt->_tail;
if (tail->in(0)->is_If()) tail = tail->in(0);
if (n == tail) {
if (nlpt->_has_sfpt) {
has_call = true;
_has_sfpt = 1;
break;
}
assert(_phase->is_dominator(_head, nlpt->_head), "inner head dominated by outer head");
n = nlpt->_head;
}
}
}
}
if (_child != NULL && !has_call && !has_local_ncsfpt) {
if (nonlocal_ncsfpt != NULL) {
if (_required_safept == NULL) _required_safept = new Node_List();
_required_safept->push(nonlocal_ncsfpt);
} else {
allpaths_check_safepts(visited, stack);
}
}
}
}
bool PhaseIdealLoop::is_deleteable_safept(Node* sfpt) {
assert(sfpt->Opcode() == Op_SafePoint, "");
IdealLoopTree* lp = get_loop(sfpt)->_parent;
while (lp != NULL) {
Node_List* sfpts = lp->_required_safept;
if (sfpts != NULL) {
for (uint i = 0; i < sfpts->size(); i++) {
if (sfpt == sfpts->at(i))
return false;
}
}
lp = lp->_parent;
}
return true;
}
void PhaseIdealLoop::replace_parallel_iv(IdealLoopTree *loop) {
assert(loop->_head->is_CountedLoop(), "");
CountedLoopNode *cl = loop->_head->as_CountedLoop();
if (!cl->is_valid_counted_loop())
return; // skip malformed counted loop
Node *incr = cl->incr();
if (incr == NULL)
return; // Dead loop?
Node *init = cl->init_trip();
Node *phi = cl->phi();
int stride_con = cl->stride_con();
for (DUIterator i = cl->outs(); cl->has_out(i); i++) {
Node *out = cl->out(i);
if (!out->is_Phi() || out == phi || !has_node(out))
continue;
PhiNode* phi2 = out->as_Phi();
Node *incr2 = phi2->in( LoopNode::LoopBackControl );
if (phi2->region() != loop->_head ||
incr2->req() != 3 ||
incr2->in(1) != phi2 ||
incr2 == incr ||
incr2->Opcode() != Op_AddI ||
!incr2->in(2)->is_Con())
continue;
Node *init2 = phi2->in( LoopNode::EntryControl );
int stride_con2 = incr2->in(2)->get_int();
if (stride_con2 == min_jint && stride_con == -1) {
continue;
}
int ratio_con = stride_con2/stride_con;
if ((ratio_con * stride_con) == stride_con2) { // Check for exact
#ifndef PRODUCT
if (TraceLoopOpts) {
tty->print("Parallel IV: %d ", phi2->_idx);
loop->dump_head();
}
#endif
Node* ratio = _igvn.intcon(ratio_con);
set_ctrl(ratio, C->root());
Node* ratio_init = new (C) MulINode(init, ratio);
_igvn.register_new_node_with_optimizer(ratio_init, init);
set_early_ctrl(ratio_init);
Node* diff = new (C) SubINode(init2, ratio_init);
_igvn.register_new_node_with_optimizer(diff, init2);
set_early_ctrl(diff);
Node* ratio_idx = new (C) MulINode(phi, ratio);
_igvn.register_new_node_with_optimizer(ratio_idx, phi);
set_ctrl(ratio_idx, cl);
Node* add = new (C) AddINode(ratio_idx, diff);
_igvn.register_new_node_with_optimizer(add);
set_ctrl(add, cl);
_igvn.replace_node( phi2, add );
if (add->outcnt() == 0) {
_igvn.remove_dead_node(add);
}
--i; // deleted this phi; rescan starting with next position
continue;
}
}
}
void IdealLoopTree::remove_safepoints(PhaseIdealLoop* phase, bool keep_one) {
Node* keep = NULL;
if (keep_one) {
for (Node* i = tail(); i != _head; i = phase->idom(i)) {
if (i->Opcode() == Op_SafePoint && phase->get_loop(i) == this) {
keep = i;
break; // Found one
}
}
}
bool prune = !keep_one || keep != NULL;
Node_List* sfpts = _safepts;
if (prune && sfpts != NULL) {
assert(keep == NULL || keep->Opcode() == Op_SafePoint, "not safepoint");
for (uint i = 0; i < sfpts->size(); i++) {
Node* n = sfpts->at(i);
assert(phase->get_loop(n) == this, "");
if (n != keep && phase->is_deleteable_safept(n)) {
phase->lazy_replace(n, n->in(TypeFunc::Control));
}
}
}
}
void IdealLoopTree::counted_loop( PhaseIdealLoop *phase ) {
if (!_child) {
if (_head->is_Loop()) _head->as_Loop()->set_inner_loop();
}
if (_head->is_CountedLoop() ||
phase->is_counted_loop(_head, this)) {
if (!UseCountedLoopSafepoints) {
_has_sfpt = 1;
}
bool keep_one_sfpt = !(_has_call || _has_sfpt);
remove_safepoints(phase, keep_one_sfpt);
phase->replace_parallel_iv(this);
} else if (_parent != NULL && !_irreducible) {
bool keep_one_sfpt = true;
remove_safepoints(phase, keep_one_sfpt);
}
if (_child) _child->counted_loop( phase );
if (_next) _next ->counted_loop( phase );
}
#ifndef PRODUCT
void IdealLoopTree::dump_head( ) const {
for (uint i=0; i<_nest; i++)
tty->print(" ");
tty->print("Loop: N%d/N%d ",_head->_idx,_tail->_idx);
if (_irreducible) tty->print(" IRREDUCIBLE");
Node* entry = _head->in(LoopNode::EntryControl);
if (LoopLimitCheck) {
Node* predicate = PhaseIdealLoop::find_predicate_insertion_point(entry, Deoptimization::Reason_loop_limit_check);
if (predicate != NULL ) {
tty->print(" limit_check");
entry = entry->in(0)->in(0);
}
}
if (UseLoopPredicate) {
entry = PhaseIdealLoop::find_predicate_insertion_point(entry, Deoptimization::Reason_predicate);
if (entry != NULL) {
tty->print(" predicated");
}
}
if (_head->is_CountedLoop()) {
CountedLoopNode *cl = _head->as_CountedLoop();
tty->print(" counted");
Node* init_n = cl->init_trip();
if (init_n != NULL && init_n->is_Con())
tty->print(" [%d,", cl->init_trip()->get_int());
else
tty->print(" [int,");
Node* limit_n = cl->limit();
if (limit_n != NULL && limit_n->is_Con())
tty->print("%d),", cl->limit()->get_int());
else
tty->print("int),");
int stride_con = cl->stride_con();
if (stride_con > 0) tty->print("+");
tty->print("%d", stride_con);
tty->print(" (%d iters) ", (int)cl->profile_trip_cnt());
if (cl->is_pre_loop ()) tty->print(" pre" );
if (cl->is_main_loop()) tty->print(" main");
if (cl->is_post_loop()) tty->print(" post");
}
if (_has_call) tty->print(" has_call");
if (_has_sfpt) tty->print(" has_sfpt");
if (_rce_candidate) tty->print(" rce");
if (_safepts != NULL && _safepts->size() > 0) {
tty->print(" sfpts={"); _safepts->dump_simple(); tty->print(" }");
}
if (_required_safept != NULL && _required_safept->size() > 0) {
tty->print(" req={"); _required_safept->dump_simple(); tty->print(" }");
}
tty->cr();
}
void IdealLoopTree::dump( ) const {
dump_head();
if (_child) _child->dump();
if (_next) _next ->dump();
}
#endif
static void log_loop_tree(IdealLoopTree* root, IdealLoopTree* loop, CompileLog* log) {
if (loop == root) {
if (loop->_child != NULL) {
log->begin_head("loop_tree");
log->end_head();
if( loop->_child ) log_loop_tree(root, loop->_child, log);
log->tail("loop_tree");
assert(loop->_next == NULL, "what?");
}
} else {
Node* head = loop->_head;
log->begin_head("loop");
log->print(" idx='%d' ", head->_idx);
if (loop->_irreducible) log->print("irreducible='1' ");
if (head->is_Loop()) {
if (head->as_Loop()->is_inner_loop()) log->print("inner_loop='1' ");
if (head->as_Loop()->is_partial_peel_loop()) log->print("partial_peel_loop='1' ");
}
if (head->is_CountedLoop()) {
CountedLoopNode* cl = head->as_CountedLoop();
if (cl->is_pre_loop()) log->print("pre_loop='%d' ", cl->main_idx());
if (cl->is_main_loop()) log->print("main_loop='%d' ", cl->_idx);
if (cl->is_post_loop()) log->print("post_loop='%d' ", cl->main_idx());
}
log->end_head();
if( loop->_child ) log_loop_tree(root, loop->_child, log);
log->tail("loop");
if( loop->_next ) log_loop_tree(root, loop->_next, log);
}
}
void PhaseIdealLoop::collect_potentially_useful_predicates(
IdealLoopTree * loop, Unique_Node_List &useful_predicates) {
if (loop->_child) { // child
collect_potentially_useful_predicates(loop->_child, useful_predicates);
}
if (loop->_head->is_Loop() &&
!loop->_irreducible &&
!loop->tail()->is_top()) {
LoopNode* lpn = loop->_head->as_Loop();
Node* entry = lpn->in(LoopNode::EntryControl);
Node* predicate_proj = find_predicate(entry); // loop_limit_check first
if (predicate_proj != NULL ) { // right pattern that can be used by loop predication
assert(entry->in(0)->in(1)->in(1)->Opcode() == Op_Opaque1, "must be");
useful_predicates.push(entry->in(0)->in(1)->in(1)); // good one
entry = entry->in(0)->in(0);
}
predicate_proj = find_predicate(entry); // Predicate
if (predicate_proj != NULL ) {
useful_predicates.push(entry->in(0)->in(1)->in(1)); // good one
}
}
if (loop->_next) { // sibling
collect_potentially_useful_predicates(loop->_next, useful_predicates);
}
}
void PhaseIdealLoop::eliminate_useless_predicates() {
if (C->predicate_count() == 0)
return; // no predicate left
Unique_Node_List useful_predicates; // to store useful predicates
if (C->has_loops()) {
collect_potentially_useful_predicates(_ltree_root->_child, useful_predicates);
}
for (int i = C->predicate_count(); i > 0; i--) {
Node * n = C->predicate_opaque1_node(i-1);
assert(n->Opcode() == Op_Opaque1, "must be");
if (!useful_predicates.member(n)) { // not in the useful list
_igvn.replace_node(n, n->in(1));
}
}
}
bool PhaseIdealLoop::process_expensive_nodes() {
assert(OptimizeExpensiveOps, "optimization off?");
C->sort_expensive_nodes();
bool progress = false;
for (int i = 0; i < C->expensive_count(); ) {
Node* n = C->expensive_node(i);
int start = i;
i++;
for (; i < C->expensive_count() && Compile::cmp_expensive_nodes(n, C->expensive_node(i)) == 0; i++);
int end = i;
for (int j = start; j < end; j++) {
Node* n1 = C->expensive_node(j);
if (is_node_unreachable(n1)) {
continue;
}
for (int k = j+1; k < end; k++) {
Node* n2 = C->expensive_node(k);
if (is_node_unreachable(n2)) {
continue;
}
assert(n1 != n2, "should be pair of nodes");
Node* c1 = n1->in(0);
Node* c2 = n2->in(0);
Node* parent_c1 = c1;
Node* parent_c2 = c2;
if (c1->is_Loop()) {
parent_c1 = c1->in(1);
}
if (c2->is_Loop()) {
parent_c2 = c2->in(1);
}
if (parent_c1 == parent_c2) {
_igvn._worklist.push(n1);
_igvn._worklist.push(n2);
continue;
}
if (is_dominator(c1, c2)) {
c2 = c1;
} else if (is_dominator(c2, c1)) {
c1 = c2;
} else if (parent_c1->is_Proj() && parent_c1->in(0)->is_If() &&
parent_c2->is_Proj() && parent_c1->in(0) == parent_c2->in(0)) {
c1 = c2 = idom(parent_c1->in(0));
}
if (n1->in(0) != c1) {
_igvn.hash_delete(n1);
n1->set_req(0, c1);
_igvn.hash_insert(n1);
_igvn._worklist.push(n1);
progress = true;
}
if (n2->in(0) != c2) {
_igvn.hash_delete(n2);
n2->set_req(0, c2);
_igvn.hash_insert(n2);
_igvn._worklist.push(n2);
progress = true;
}
}
}
}
return progress;
}
void PhaseIdealLoop::build_and_optimize(bool do_split_ifs, bool skip_loop_opts) {
ResourceMark rm;
int old_progress = C->major_progress();
uint orig_worklist_size = _igvn._worklist.size();
C->clear_major_progress();
#ifndef PRODUCT
uint unique = C->unique();
_loop_invokes++;
_loop_work += unique;
#endif
_has_irreducible_loops = false;
_created_loop_node = false;
Arena *a = Thread::current()->resource_area();
VectorSet visited(a);
_nodes.map(C->unique(), NULL);
memset(_nodes.adr(), 0, wordSize * C->unique());
_ltree_root = new IdealLoopTree( this, C->root(), C->root() );
_ltree_root->_has_sfpt = 1;
_idom_size = 0;
_idom = NULL;
_dom_depth = NULL;
_dom_stk = NULL;
allocate_preorders();
build_loop_tree();
if (C->failing()) {
return;
}
if( !_ltree_root->_child && !_verify_only ) C->set_has_loops(false);
if (!has_node(C->root())) {
if (!_verify_only) {
C->clear_major_progress();
C->record_method_not_compilable("empty program detected during loop optimization");
}
return;
}
bool stop_early = !C->has_loops() && !skip_loop_opts && !do_split_ifs && !_verify_me && !_verify_only;
bool do_expensive_nodes = C->should_optimize_expensive_nodes(_igvn);
if (stop_early && !do_expensive_nodes) {
_igvn.optimize(); // Cleanup NeverBranches
return;
}
_ltree_root->set_nest( 0 );
if( !_verify_me && !_verify_only && _ltree_root->_child ) {
C->print_method(PHASE_BEFORE_BEAUTIFY_LOOPS, 3);
if( _ltree_root->_child->beautify_loops( this ) ) {
_ltree_root->_child = NULL;
_nodes.clear();
reallocate_preorders();
build_loop_tree();
if (C->failing()) {
return;
}
_ltree_root->set_nest( 0 );
C->print_method(PHASE_AFTER_BEAUTIFY_LOOPS, 3);
}
}
_idom_size = C->unique();
_idom = NEW_RESOURCE_ARRAY( Node*, _idom_size );
_dom_depth = NEW_RESOURCE_ARRAY( uint, _idom_size );
_dom_stk = NULL; // Allocated on demand in recompute_dom_depth
memset( _dom_depth, 0, _idom_size * sizeof(uint) );
Dominators();
if (!_verify_only) {
for( uint i = 1; i < C->root()->req(); i++ ) {
if( !_nodes[C->root()->in(i)->_idx] ) { // Dead path into Root?
_igvn.delete_input_of(C->root(), i);
i--; // Rerun same iteration on compressed edges
}
}
Node_List cisstack(a);
_ltree_root->check_safepts(visited, cisstack);
}
int stack_size = (C->live_nodes() >> 1) + 16; // (live_nodes>>1)+16 from Java2D stats
Node_Stack nstack( a, stack_size );
visited.Clear();
Node_List worklist(a);
worklist.push( C->top() );
visited.set( C->top()->_idx ); // Set C->top() as visited now
build_loop_early( visited, worklist, nstack );
if( !_verify_me && !_verify_only )
_ltree_root->counted_loop( this );
visited.Clear();
init_dom_lca_tags();
worklist.push( C->root() );
NOT_PRODUCT( C->verify_graph_edges(); )
worklist.push( C->top() );
build_loop_late( visited, worklist, nstack );
if (_verify_only) {
for (int i = 0; i < old_progress; i++)
C->set_major_progress();
assert(C->unique() == unique, "verification mode made Nodes? ? ?");
assert(_igvn._worklist.size() == orig_worklist_size, "shouldn't push anything");
return;
}
while (_deadlist.size()) {
_igvn.remove_globally_dead_node(_deadlist.pop());
}
if (stop_early) {
assert(do_expensive_nodes, "why are we here?");
if (process_expensive_nodes()) {
C->set_major_progress();
}
_igvn.optimize();
return;
}
if (UseLoopPredicate || LoopLimitCheck) {
eliminate_useless_predicates();
}
#ifndef PRODUCT
C->verify_graph_edges();
if (_verify_me) { // Nested verify pass?
assert(C->unique() == unique, "non-optimize mode made Nodes? ? ?");
return;
}
if(VerifyLoopOptimizations) verify();
if(TraceLoopOpts && C->has_loops()) {
_ltree_root->dump();
}
#endif
if (skip_loop_opts) {
for (int i = 0; i < old_progress; i++) {
C->set_major_progress();
}
_igvn.optimize();
if (C->log() != NULL) {
log_loop_tree(_ltree_root, _ltree_root, C->log());
}
return;
}
if (ReassociateInvariants) {
for (LoopTreeIterator iter(_ltree_root); !iter.done(); iter.next()) {
IdealLoopTree* lpt = iter.current();
if (!lpt->is_counted() || !lpt->is_inner()) continue;
lpt->reassociate_invariants(this);
if (SplitIfBlocks && do_split_ifs) {
if (lpt->policy_range_check(this)) {
lpt->_rce_candidate = 1; // = true
}
}
}
}
if( SplitIfBlocks && do_split_ifs ) {
visited.Clear();
split_if_with_blocks( visited, nstack );
NOT_PRODUCT( if( VerifyLoopOptimizations ) verify(); );
}
if (!C->major_progress() && do_expensive_nodes && process_expensive_nodes()) {
C->set_major_progress();
}
if (C->has_loops() && !C->major_progress() && (C->predicate_count() > 0)) {
_ltree_root->_child->loop_predication(this);
}
if (OptimizeFill && UseLoopPredicate && C->has_loops() && !C->major_progress()) {
if (do_intrinsify_fill()) {
C->set_major_progress();
}
}
if (C->has_loops() && !C->major_progress()) {
memset( worklist.adr(), 0, worklist.Size()*sizeof(Node*) );
_ltree_root->_child->iteration_split( this, worklist );
}
NOT_PRODUCT( C->verify_graph_edges(); );
if (!do_split_ifs) {
C->set_major_progress();
}
if (created_loop_node()) {
C->set_major_progress();
}
if (!C->major_progress() && (C->predicate_count() > 0)) {
C->cleanup_loop_predicates(_igvn);
#ifndef PRODUCT
if (TraceLoopOpts) {
tty->print_cr("PredicatesOff");
}
#endif
C->set_major_progress();
}
if (UseSuperWord && C->has_loops() && !C->major_progress()) {
SuperWord sw(this);
for (LoopTreeIterator iter(_ltree_root); !iter.done(); iter.next()) {
IdealLoopTree* lpt = iter.current();
if (lpt->is_counted()) {
sw.transform_loop(lpt);
}
}
}
_igvn.optimize();
if (C->log() != NULL) {
log_loop_tree(_ltree_root, _ltree_root, C->log());
}
}
#ifndef PRODUCT
int PhaseIdealLoop::_loop_invokes=0;// Count of PhaseIdealLoop invokes
int PhaseIdealLoop::_loop_work=0; // Sum of PhaseIdealLoop x unique
void PhaseIdealLoop::print_statistics() {
tty->print_cr("PhaseIdealLoop=%d, sum _unique=%d", _loop_invokes, _loop_work);
}
static int fail; // debug only, so its multi-thread dont care
void PhaseIdealLoop::verify() const {
int old_progress = C->major_progress();
ResourceMark rm;
PhaseIdealLoop loop_verify( _igvn, this );
VectorSet visited(Thread::current()->resource_area());
fail = 0;
verify_compare( C->root(), &loop_verify, visited );
assert( fail == 0, "verify loops failed" );
_ltree_root->verify_tree(loop_verify._ltree_root, NULL);
for( int i=0; i<old_progress; i++ )
C->set_major_progress();
}
void PhaseIdealLoop::verify_compare( Node *n, const PhaseIdealLoop *loop_verify, VectorSet &visited ) const {
if( !n ) return;
if( visited.test_set( n->_idx ) ) return;
if( !_nodes[n->_idx] ) { // Unreachable
assert( !loop_verify->_nodes[n->_idx], "both should be unreachable" );
return;
}
uint i;
for( i = 0; i < n->req(); i++ )
verify_compare( n->in(i), loop_verify, visited );
i = n->_idx;
if( has_ctrl(n) ) { // We have control; verify has loop or ctrl
if( _nodes[i] != loop_verify->_nodes[i] &&
get_ctrl_no_update(n) != loop_verify->get_ctrl_no_update(n) ) {
tty->print("Mismatched control setting for: ");
n->dump();
if( fail++ > 10 ) return;
Node *c = get_ctrl_no_update(n);
tty->print("We have it as: ");
if( c->in(0) ) c->dump();
else tty->print_cr("N%d",c->_idx);
tty->print("Verify thinks: ");
if( loop_verify->has_ctrl(n) )
loop_verify->get_ctrl_no_update(n)->dump();
else
loop_verify->get_loop_idx(n)->dump();
tty->cr();
}
} else { // We have a loop
IdealLoopTree *us = get_loop_idx(n);
if( loop_verify->has_ctrl(n) ) {
tty->print("Mismatched loop setting for: ");
n->dump();
if( fail++ > 10 ) return;
tty->print("We have it as: ");
us->dump();
tty->print("Verify thinks: ");
loop_verify->get_ctrl_no_update(n)->dump();
tty->cr();
} else if (!C->major_progress()) {
IdealLoopTree *them = loop_verify->get_loop_idx(n);
if( us->_head != them->_head || us->_tail != them->_tail ) {
tty->print("Unequals loops for: ");
n->dump();
if( fail++ > 10 ) return;
tty->print("We have it as: ");
us->dump();
tty->print("Verify thinks: ");
them->dump();
tty->cr();
}
}
}
if( i >= _idom_size ) {
if( !n->is_CFG() ) return;
tty->print("CFG Node with no idom: ");
n->dump();
return;
}
if( !n->is_CFG() ) return;
if( n == C->root() ) return; // No IDOM here
assert(n->_idx == i, "sanity");
Node *id = idom_no_update(n);
if( id != loop_verify->idom_no_update(n) ) {
tty->print("Unequals idoms for: ");
n->dump();
if( fail++ > 10 ) return;
tty->print("We have it as: ");
id->dump();
tty->print("Verify thinks: ");
loop_verify->idom_no_update(n)->dump();
tty->cr();
}
}
void IdealLoopTree::verify_tree(IdealLoopTree *loop, const IdealLoopTree *parent) const {
assert( _parent == parent, "Badly formed loop tree" );
if( _head != loop->_head ) {
IdealLoopTree **pp = &loop->_parent->_child;
while( *pp != loop )
pp = &((*pp)->_next);
IdealLoopTree **nn = &loop->_next;
while( (*nn) && (*nn)->_head != _head )
nn = &((*nn)->_next);
if( !(*nn) ) {
if (_irreducible && Compile::current()->major_progress()) return;
assert( 0, "failed to match loop tree" );
}
IdealLoopTree *hit = *nn;
hit->_next = loop;
loop = hit;
}
assert( _head == loop->_head , "mismatched loop head" );
Node *tail = _tail; // Inline a non-updating version of
while( !tail->in(0) ) // the 'tail()' call.
tail = tail->in(1);
assert( tail == loop->_tail, "mismatched loop tail" );
if( _head->is_CountedLoop() && _head->as_CountedLoop()->is_main_loop() ) {
CountedLoopNode *cl = _head->as_CountedLoop();
Node *init = cl->init_trip();
Node *ctrl = cl->in(LoopNode::EntryControl);
assert( ctrl->Opcode() == Op_IfTrue || ctrl->Opcode() == Op_IfFalse, "" );
Node *iff = ctrl->in(0);
assert( iff->Opcode() == Op_If, "" );
Node *bol = iff->in(1);
assert( bol->Opcode() == Op_Bool, "" );
Node *cmp = bol->in(1);
assert( cmp->Opcode() == Op_CmpI, "" );
Node *add = cmp->in(1);
Node *opaq;
if( add->Opcode() == Op_Opaque1 ) {
opaq = add;
} else {
assert( add->Opcode() == Op_AddI || add->Opcode() == Op_ConI , "" );
assert( add == init, "" );
opaq = cmp->in(2);
}
assert( opaq->Opcode() == Op_Opaque1, "" );
}
ssssssssss54
最新推荐文章于 2024-11-14 19:49:43 发布