mirror of
https://github.com/openjdk/jdk.git
synced 2026-01-31 05:28:33 +00:00
1889 lines
67 KiB
C++
1889 lines
67 KiB
C++
/*
|
|
* Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
|
|
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
|
|
*
|
|
* This code is free software; you can redistribute it and/or modify it
|
|
* under the terms of the GNU General Public License version 2 only, as
|
|
* published by the Free Software Foundation.
|
|
*
|
|
* This code is distributed in the hope that it will be useful, but WITHOUT
|
|
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
|
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
* version 2 for more details (a copy is included in the LICENSE file that
|
|
* accompanied this code).
|
|
*
|
|
* You should have received a copy of the GNU General Public License version
|
|
* 2 along with this work; if not, write to the Free Software Foundation,
|
|
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
|
|
*
|
|
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
|
|
* or visit www.oracle.com if you need additional information or have any
|
|
* questions.
|
|
*
|
|
*/
|
|
|
|
#include "memory/allocation.inline.hpp"
|
|
#include "opto/addnode.hpp"
|
|
#include "opto/castnode.hpp"
|
|
#include "opto/cfgnode.hpp"
|
|
#include "opto/connode.hpp"
|
|
#include "opto/machnode.hpp"
|
|
#include "opto/movenode.hpp"
|
|
#include "opto/mulnode.hpp"
|
|
#include "opto/phaseX.hpp"
|
|
#include "opto/subnode.hpp"
|
|
#include "opto/utilities/xor.hpp"
|
|
#include "runtime/stubRoutines.hpp"
|
|
|
|
// Portions of code courtesy of Clifford Click
|
|
|
|
// Classic Add functionality. This covers all the usual 'add' behaviors for
|
|
// an algebraic ring. Add-integer, add-float, add-double, and binary-or are
|
|
// all inherited from this class. The various identity values are supplied
|
|
// by virtual functions.
|
|
|
|
|
|
//=============================================================================
|
|
//------------------------------hash-------------------------------------------
|
|
// Hash function over AddNodes. Needs to be commutative; i.e., I swap
|
|
// (commute) inputs to AddNodes willy-nilly so the hash function must return
|
|
// the same value in the presence of edge swapping.
|
|
uint AddNode::hash() const {
|
|
return (uintptr_t)in(1) + (uintptr_t)in(2) + Opcode();
|
|
}
|
|
|
|
//------------------------------Identity---------------------------------------
|
|
// If either input is a constant 0, return the other input.
|
|
Node* AddNode::Identity(PhaseGVN* phase) {
|
|
const Type *zero = add_id(); // The additive identity
|
|
if( phase->type( in(1) )->higher_equal( zero ) ) return in(2);
|
|
if( phase->type( in(2) )->higher_equal( zero ) ) return in(1);
|
|
return this;
|
|
}
|
|
|
|
//------------------------------commute----------------------------------------
|
|
// Commute operands to move loads and constants to the right.
|
|
static bool commute(PhaseGVN* phase, Node* add) {
|
|
Node *in1 = add->in(1);
|
|
Node *in2 = add->in(2);
|
|
|
|
// convert "max(a,b) + min(a,b)" into "a+b".
|
|
if ((in1->Opcode() == add->as_Add()->max_opcode() && in2->Opcode() == add->as_Add()->min_opcode())
|
|
|| (in1->Opcode() == add->as_Add()->min_opcode() && in2->Opcode() == add->as_Add()->max_opcode())) {
|
|
Node *in11 = in1->in(1);
|
|
Node *in12 = in1->in(2);
|
|
|
|
Node *in21 = in2->in(1);
|
|
Node *in22 = in2->in(2);
|
|
|
|
if ((in11 == in21 && in12 == in22) ||
|
|
(in11 == in22 && in12 == in21)) {
|
|
add->set_req_X(1, in11, phase);
|
|
add->set_req_X(2, in12, phase);
|
|
return true;
|
|
}
|
|
}
|
|
|
|
bool con_left = phase->type(in1)->singleton();
|
|
bool con_right = phase->type(in2)->singleton();
|
|
|
|
// Convert "1+x" into "x+1".
|
|
// Right is a constant; leave it
|
|
if( con_right ) return false;
|
|
// Left is a constant; move it right.
|
|
if( con_left ) {
|
|
add->swap_edges(1, 2);
|
|
return true;
|
|
}
|
|
|
|
// Convert "Load+x" into "x+Load".
|
|
// Now check for loads
|
|
if (in2->is_Load()) {
|
|
if (!in1->is_Load()) {
|
|
// already x+Load to return
|
|
return false;
|
|
}
|
|
// both are loads, so fall through to sort inputs by idx
|
|
} else if( in1->is_Load() ) {
|
|
// Left is a Load and Right is not; move it right.
|
|
add->swap_edges(1, 2);
|
|
return true;
|
|
}
|
|
|
|
PhiNode *phi;
|
|
// Check for tight loop increments: Loop-phi of Add of loop-phi
|
|
if (in1->is_Phi() && (phi = in1->as_Phi()) && phi->region()->is_Loop() && phi->in(2) == add)
|
|
return false;
|
|
if (in2->is_Phi() && (phi = in2->as_Phi()) && phi->region()->is_Loop() && phi->in(2) == add) {
|
|
add->swap_edges(1, 2);
|
|
return true;
|
|
}
|
|
|
|
// Otherwise, sort inputs (commutativity) to help value numbering.
|
|
if( in1->_idx > in2->_idx ) {
|
|
add->swap_edges(1, 2);
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
//------------------------------Idealize---------------------------------------
|
|
// If we get here, we assume we are associative!
|
|
Node *AddNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
const Type *t1 = phase->type(in(1));
|
|
const Type *t2 = phase->type(in(2));
|
|
bool con_left = t1->singleton();
|
|
bool con_right = t2->singleton();
|
|
|
|
// Check for commutative operation desired
|
|
if (commute(phase, this)) return this;
|
|
|
|
AddNode *progress = nullptr; // Progress flag
|
|
|
|
// Convert "(x+1)+2" into "x+(1+2)". If the right input is a
|
|
// constant, and the left input is an add of a constant, flatten the
|
|
// expression tree.
|
|
Node *add1 = in(1);
|
|
Node *add2 = in(2);
|
|
int add1_op = add1->Opcode();
|
|
int this_op = Opcode();
|
|
if (con_right && t2 != Type::TOP && // Right input is a constant?
|
|
add1_op == this_op) { // Left input is an Add?
|
|
|
|
// Type of left _in right input
|
|
const Type *t12 = phase->type(add1->in(2));
|
|
if (t12->singleton() && t12 != Type::TOP) { // Left input is an add of a constant?
|
|
// Check for rare case of closed data cycle which can happen inside
|
|
// unreachable loops. In these cases the computation is undefined.
|
|
#ifdef ASSERT
|
|
Node *add11 = add1->in(1);
|
|
int add11_op = add11->Opcode();
|
|
if ((add1 == add1->in(1))
|
|
|| (add11_op == this_op && add11->in(1) == add1)) {
|
|
assert(false, "dead loop in AddNode::Ideal");
|
|
}
|
|
#endif
|
|
// The Add of the flattened expression
|
|
Node *x1 = add1->in(1);
|
|
Node *x2 = phase->makecon(add1->as_Add()->add_ring(t2, t12));
|
|
set_req_X(2, x2, phase);
|
|
set_req_X(1, x1, phase);
|
|
progress = this; // Made progress
|
|
add1 = in(1);
|
|
add1_op = add1->Opcode();
|
|
}
|
|
}
|
|
|
|
// Convert "(x+1)+y" into "(x+y)+1". Push constants down the expression tree.
|
|
if (add1_op == this_op && !con_right) {
|
|
Node *a12 = add1->in(2);
|
|
const Type *t12 = phase->type( a12 );
|
|
if (t12->singleton() && t12 != Type::TOP && (add1 != add1->in(1)) &&
|
|
!(add1->in(1)->is_Phi() && (add1->in(1)->as_Phi()->is_tripcount(T_INT) || add1->in(1)->as_Phi()->is_tripcount(T_LONG)))) {
|
|
assert(add1->in(1) != this, "dead loop in AddNode::Ideal");
|
|
add2 = add1->clone();
|
|
add2->set_req(2, in(2));
|
|
add2 = phase->transform(add2);
|
|
set_req_X(1, add2, phase);
|
|
set_req_X(2, a12, phase);
|
|
progress = this;
|
|
add2 = a12;
|
|
}
|
|
}
|
|
|
|
// Convert "x+(y+1)" into "(x+y)+1". Push constants down the expression tree.
|
|
int add2_op = add2->Opcode();
|
|
if (add2_op == this_op && !con_left) {
|
|
Node *a22 = add2->in(2);
|
|
const Type *t22 = phase->type( a22 );
|
|
if (t22->singleton() && t22 != Type::TOP && (add2 != add2->in(1)) &&
|
|
!(add2->in(1)->is_Phi() && (add2->in(1)->as_Phi()->is_tripcount(T_INT) || add2->in(1)->as_Phi()->is_tripcount(T_LONG)))) {
|
|
assert(add2->in(1) != this, "dead loop in AddNode::Ideal");
|
|
Node *addx = add2->clone();
|
|
addx->set_req(1, in(1));
|
|
addx->set_req(2, add2->in(1));
|
|
addx = phase->transform(addx);
|
|
set_req_X(1, addx, phase);
|
|
set_req_X(2, a22, phase);
|
|
progress = this;
|
|
}
|
|
}
|
|
|
|
return progress;
|
|
}
|
|
|
|
//------------------------------Value-----------------------------------------
|
|
// An add node sums it's two _in. If one input is an RSD, we must mixin
|
|
// the other input's symbols.
|
|
const Type* AddNode::Value(PhaseGVN* phase) const {
|
|
// Either input is TOP ==> the result is TOP
|
|
const Type* t1 = phase->type(in(1));
|
|
const Type* t2 = phase->type(in(2));
|
|
if (t1 == Type::TOP || t2 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
|
|
// Check for an addition involving the additive identity
|
|
const Type* tadd = add_of_identity(t1, t2);
|
|
if (tadd != nullptr) {
|
|
return tadd;
|
|
}
|
|
|
|
return add_ring(t1, t2); // Local flavor of type addition
|
|
}
|
|
|
|
//------------------------------add_identity-----------------------------------
|
|
// Check for addition of the identity
|
|
const Type *AddNode::add_of_identity( const Type *t1, const Type *t2 ) const {
|
|
const Type *zero = add_id(); // The additive identity
|
|
if( t1->higher_equal( zero ) ) return t2;
|
|
if( t2->higher_equal( zero ) ) return t1;
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
AddNode* AddNode::make(Node* in1, Node* in2, BasicType bt) {
|
|
switch (bt) {
|
|
case T_INT:
|
|
return new AddINode(in1, in2);
|
|
case T_LONG:
|
|
return new AddLNode(in1, in2);
|
|
default:
|
|
fatal("Not implemented for %s", type2name(bt));
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
bool AddNode::is_not(PhaseGVN* phase, Node* n, BasicType bt) {
|
|
return n->Opcode() == Op_Xor(bt) && phase->type(n->in(2)) == TypeInteger::minus_1(bt);
|
|
}
|
|
|
|
AddNode* AddNode::make_not(PhaseGVN* phase, Node* n, BasicType bt) {
|
|
switch (bt) {
|
|
case T_INT:
|
|
return new XorINode(n, phase->intcon(-1));
|
|
case T_LONG:
|
|
return new XorLNode(n, phase->longcon(-1L));
|
|
default:
|
|
fatal("Not implemented for %s", type2name(bt));
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Idealize---------------------------------------
|
|
Node* AddNode::IdealIL(PhaseGVN* phase, bool can_reshape, BasicType bt) {
|
|
Node* in1 = in(1);
|
|
Node* in2 = in(2);
|
|
int op1 = in1->Opcode();
|
|
int op2 = in2->Opcode();
|
|
// Fold (con1-x)+con2 into (con1+con2)-x
|
|
if (op1 == Op_Add(bt) && op2 == Op_Sub(bt)) {
|
|
// Swap edges to try optimizations below
|
|
in1 = in2;
|
|
in2 = in(1);
|
|
op1 = op2;
|
|
op2 = in2->Opcode();
|
|
}
|
|
if (op1 == Op_Sub(bt)) {
|
|
const Type* t_sub1 = phase->type(in1->in(1));
|
|
const Type* t_2 = phase->type(in2 );
|
|
if (t_sub1->singleton() && t_2->singleton() && t_sub1 != Type::TOP && t_2 != Type::TOP) {
|
|
return SubNode::make(phase->makecon(add_ring(t_sub1, t_2)), in1->in(2), bt);
|
|
}
|
|
// Convert "(a-b)+(c-d)" into "(a+c)-(b+d)"
|
|
if (op2 == Op_Sub(bt)) {
|
|
// Check for dead cycle: d = (a-b)+(c-d)
|
|
assert( in1->in(2) != this && in2->in(2) != this,
|
|
"dead loop in AddINode::Ideal" );
|
|
Node* sub = SubNode::make(nullptr, nullptr, bt);
|
|
Node* sub_in1;
|
|
PhaseIterGVN* igvn = phase->is_IterGVN();
|
|
// During IGVN, if both inputs of the new AddNode are a tree of SubNodes, this same transformation will be applied
|
|
// to every node of the tree. Calling transform() causes the transformation to be applied recursively, once per
|
|
// tree node whether some subtrees are identical or not. Pushing to the IGVN worklist instead, causes the transform
|
|
// to be applied once per unique subtrees (because all uses of a subtree are updated with the result of the
|
|
// transformation). In case of a large tree, this can make a difference in compilation time.
|
|
if (igvn != nullptr) {
|
|
sub_in1 = igvn->register_new_node_with_optimizer(AddNode::make(in1->in(1), in2->in(1), bt));
|
|
} else {
|
|
sub_in1 = phase->transform(AddNode::make(in1->in(1), in2->in(1), bt));
|
|
}
|
|
Node* sub_in2;
|
|
if (igvn != nullptr) {
|
|
sub_in2 = igvn->register_new_node_with_optimizer(AddNode::make(in1->in(2), in2->in(2), bt));
|
|
} else {
|
|
sub_in2 = phase->transform(AddNode::make(in1->in(2), in2->in(2), bt));
|
|
}
|
|
sub->init_req(1, sub_in1);
|
|
sub->init_req(2, sub_in2);
|
|
return sub;
|
|
}
|
|
// Convert "(a-b)+(b+c)" into "(a+c)"
|
|
if (op2 == Op_Add(bt) && in1->in(2) == in2->in(1)) {
|
|
assert(in1->in(1) != this && in2->in(2) != this,"dead loop in AddINode::Ideal/AddLNode::Ideal");
|
|
return AddNode::make(in1->in(1), in2->in(2), bt);
|
|
}
|
|
// Convert "(a-b)+(c+b)" into "(a+c)"
|
|
if (op2 == Op_Add(bt) && in1->in(2) == in2->in(2)) {
|
|
assert(in1->in(1) != this && in2->in(1) != this,"dead loop in AddINode::Ideal/AddLNode::Ideal");
|
|
return AddNode::make(in1->in(1), in2->in(1), bt);
|
|
}
|
|
}
|
|
|
|
// Convert (con - y) + x into "(x - y) + con"
|
|
if (op1 == Op_Sub(bt) && in1->in(1)->Opcode() == Op_ConIL(bt)
|
|
&& in1 != in1->in(2) && !(in1->in(2)->is_Phi() && in1->in(2)->as_Phi()->is_tripcount(bt))) {
|
|
return AddNode::make(phase->transform(SubNode::make(in2, in1->in(2), bt)), in1->in(1), bt);
|
|
}
|
|
|
|
// Convert x + (con - y) into "(x - y) + con"
|
|
if (op2 == Op_Sub(bt) && in2->in(1)->Opcode() == Op_ConIL(bt)
|
|
&& in2 != in2->in(2) && !(in2->in(2)->is_Phi() && in2->in(2)->as_Phi()->is_tripcount(bt))) {
|
|
return AddNode::make(phase->transform(SubNode::make(in1, in2->in(2), bt)), in2->in(1), bt);
|
|
}
|
|
|
|
// Associative
|
|
if (op1 == Op_Mul(bt) && op2 == Op_Mul(bt)) {
|
|
Node* add_in1 = nullptr;
|
|
Node* add_in2 = nullptr;
|
|
Node* mul_in = nullptr;
|
|
|
|
if (in1->in(1) == in2->in(1)) {
|
|
// Convert "a*b+a*c into a*(b+c)
|
|
add_in1 = in1->in(2);
|
|
add_in2 = in2->in(2);
|
|
mul_in = in1->in(1);
|
|
} else if (in1->in(2) == in2->in(1)) {
|
|
// Convert a*b+b*c into b*(a+c)
|
|
add_in1 = in1->in(1);
|
|
add_in2 = in2->in(2);
|
|
mul_in = in1->in(2);
|
|
} else if (in1->in(2) == in2->in(2)) {
|
|
// Convert a*c+b*c into (a+b)*c
|
|
add_in1 = in1->in(1);
|
|
add_in2 = in2->in(1);
|
|
mul_in = in1->in(2);
|
|
} else if (in1->in(1) == in2->in(2)) {
|
|
// Convert a*b+c*a into a*(b+c)
|
|
add_in1 = in1->in(2);
|
|
add_in2 = in2->in(1);
|
|
mul_in = in1->in(1);
|
|
}
|
|
|
|
if (mul_in != nullptr) {
|
|
Node* add = phase->transform(AddNode::make(add_in1, add_in2, bt));
|
|
return MulNode::make(mul_in, add, bt);
|
|
}
|
|
}
|
|
|
|
// Convert (x >>> rshift) + (x << lshift) into RotateRight(x, rshift)
|
|
if (Matcher::match_rule_supported(Op_RotateRight) &&
|
|
((op1 == Op_URShift(bt) && op2 == Op_LShift(bt)) || (op1 == Op_LShift(bt) && op2 == Op_URShift(bt))) &&
|
|
in1->in(1) != nullptr && in1->in(1) == in2->in(1)) {
|
|
Node* rshift = op1 == Op_URShift(bt) ? in1->in(2) : in2->in(2);
|
|
Node* lshift = op1 == Op_URShift(bt) ? in2->in(2) : in1->in(2);
|
|
if (rshift != nullptr && lshift != nullptr) {
|
|
const TypeInt* rshift_t = phase->type(rshift)->isa_int();
|
|
const TypeInt* lshift_t = phase->type(lshift)->isa_int();
|
|
int bits = bt == T_INT ? 32 : 64;
|
|
int mask = bt == T_INT ? 0x1F : 0x3F;
|
|
if (lshift_t != nullptr && lshift_t->is_con() &&
|
|
rshift_t != nullptr && rshift_t->is_con() &&
|
|
((lshift_t->get_con() & mask) == (bits - (rshift_t->get_con() & mask)))) {
|
|
return new RotateRightNode(in1->in(1), phase->intcon(rshift_t->get_con() & mask), TypeInteger::bottom(bt));
|
|
}
|
|
}
|
|
}
|
|
|
|
// Collapse addition of the same terms into multiplications.
|
|
Node* collapsed = Ideal_collapse_variable_times_con(phase, bt);
|
|
if (collapsed != nullptr) {
|
|
return collapsed; // Skip AddNode::Ideal() since it may now be a multiplication node.
|
|
}
|
|
|
|
return AddNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
// Try to collapse addition of the same terms into a single multiplication. On success, a new MulNode is returned.
|
|
// Examples of this conversion includes:
|
|
// - a + a + ... + a => CON*a
|
|
// - (a * CON) + a => (CON + 1) * a
|
|
// - a + (a * CON) => (CON + 1) * a
|
|
//
|
|
// We perform such conversions incrementally during IGVN by transforming left most nodes first and work up to the root
|
|
// of the expression. In other words, we convert, at each iteration:
|
|
// a + a + a + ... + a
|
|
// => 2*a + a + ... + a
|
|
// => 3*a + ... + a
|
|
// => n*a
|
|
//
|
|
// Due to the iterative nature of IGVN, MulNode transformed from first few AddNode terms may be further transformed into
|
|
// power-of-2 pattern. (e.g., 2 * a => a << 1, 3 * a => (a << 2) + a). We can't guarantee we'll always pick up
|
|
// transformed power-of-2 patterns when term `a` is complex.
|
|
//
|
|
// Note this also converts, for example, original expression `(a*3) + a` into `4*a` and `(a<<2) + a` into `5*a`. A more
|
|
// generalized pattern `(a*b) + (a*c)` into `a*(b + c)` is handled by AddNode::IdealIL().
|
|
Node* AddNode::Ideal_collapse_variable_times_con(PhaseGVN* phase, BasicType bt) {
|
|
// We need to make sure that the current AddNode is not part of a MulNode that has already been optimized to a
|
|
// power-of-2 addition (e.g., 3 * a => (a << 2) + a). Without this check, GVN would keep trying to optimize the same
|
|
// node and can't progress. For example, 3 * a => (a << 2) + a => 3 * a => (a << 2) + a => ...
|
|
if (Multiplication::find_power_of_two_addition_pattern(this, bt).is_valid()) {
|
|
return nullptr;
|
|
}
|
|
|
|
Node* lhs = in(1);
|
|
Node* rhs = in(2);
|
|
|
|
Multiplication mul = Multiplication::find_collapsible_addition_patterns(lhs, rhs, bt);
|
|
if (!mul.is_valid_with(rhs)) {
|
|
// Swap lhs and rhs then try again
|
|
mul = Multiplication::find_collapsible_addition_patterns(rhs, lhs, bt);
|
|
if (!mul.is_valid_with(lhs)) {
|
|
return nullptr;
|
|
}
|
|
}
|
|
|
|
Node* con;
|
|
if (bt == T_INT) {
|
|
con = phase->intcon(java_add(static_cast<jint>(mul.multiplier()), 1));
|
|
} else {
|
|
con = phase->longcon(java_add(mul.multiplier(), CONST64(1)));
|
|
}
|
|
|
|
return MulNode::make(con, mul.variable(), bt);
|
|
}
|
|
|
|
// Find a pattern of collapsable additions that can be converted to a multiplication.
|
|
// When matching the LHS `a * CON`, we match with best efforts by looking for the following patterns:
|
|
// - (1) Simple addition: LHS = a + a
|
|
// - (2) Simple lshift: LHS = a << CON
|
|
// - (3) Simple multiplication: LHS = CON * a
|
|
// - (4) Power-of-two addition: LHS = (a << CON1) + (a << CON2)
|
|
AddNode::Multiplication AddNode::Multiplication::find_collapsible_addition_patterns(const Node* a, const Node* pattern, BasicType bt) {
|
|
// (1) Simple addition pattern (e.g., lhs = a + a)
|
|
Multiplication mul = find_simple_addition_pattern(a, bt);
|
|
if (mul.is_valid_with(pattern)) {
|
|
return mul;
|
|
}
|
|
|
|
// (2) Simple lshift pattern (e.g., lhs = a << CON)
|
|
mul = find_simple_lshift_pattern(a, bt);
|
|
if (mul.is_valid_with(pattern)) {
|
|
return mul;
|
|
}
|
|
|
|
// (3) Simple multiplication pattern (e.g., lhs = CON * a)
|
|
mul = find_simple_multiplication_pattern(a, bt);
|
|
if (mul.is_valid_with(pattern)) {
|
|
return mul;
|
|
}
|
|
|
|
// (4) Power-of-two addition pattern (e.g., lhs = (a << CON1) + (a << CON2))
|
|
// While multiplications can be potentially optimized to power-of-2 subtractions (e.g., a * 7 => (a << 3) - a),
|
|
// (x - y) + y => x is already handled by the Identity() methods. So, we don't need to check for that pattern here.
|
|
mul = find_power_of_two_addition_pattern(a, bt);
|
|
if (mul.is_valid_with(pattern)) {
|
|
return mul;
|
|
}
|
|
|
|
// We've tried everything.
|
|
return make_invalid();
|
|
}
|
|
|
|
// Try to match `n = a + a`. On success, return a struct with `.valid = true`, `variable = a`, and `multiplier = 2`.
|
|
// The method matches `n` for pattern: a + a.
|
|
AddNode::Multiplication AddNode::Multiplication::find_simple_addition_pattern(const Node* n, BasicType bt) {
|
|
if (n->Opcode() == Op_Add(bt) && n->in(1) == n->in(2)) {
|
|
return Multiplication(n->in(1), 2);
|
|
}
|
|
|
|
return make_invalid();
|
|
}
|
|
|
|
// Try to match `n = a << CON`. On success, return a struct with `.valid = true`, `variable = a`, and
|
|
// `multiplier = 1 << CON`.
|
|
// Match `n` for pattern: a << CON.
|
|
// Note that the power-of-2 multiplication optimization could potentially convert a MulNode to this pattern.
|
|
AddNode::Multiplication AddNode::Multiplication::find_simple_lshift_pattern(const Node* n, BasicType bt) {
|
|
// Note that power-of-2 multiplication optimization could potentially convert a MulNode to this pattern
|
|
if (n->Opcode() == Op_LShift(bt) && n->in(2)->is_Con()) {
|
|
Node* con = n->in(2);
|
|
if (!con->is_top()) {
|
|
return Multiplication(n->in(1), java_shift_left(1, con->get_int(), bt));
|
|
}
|
|
}
|
|
|
|
return make_invalid();
|
|
}
|
|
|
|
// Try to match `n = CON * a`. On success, return a struct with `.valid = true`, `variable = a`, and `multiplier = CON`.
|
|
// Match `n` for patterns: CON * a
|
|
// Note that `CON` will always be the second input node of a Mul node canonicalized by Ideal(). If this is not the case,
|
|
// `n` has not been processed by iGVN. So we skip the optimization for the current add node and wait for to be added to
|
|
// the queue again.
|
|
AddNode::Multiplication AddNode::Multiplication::find_simple_multiplication_pattern(const Node* n, BasicType bt) {
|
|
if (n->Opcode() == Op_Mul(bt) && n->in(2)->is_Con()) {
|
|
Node* con = n->in(2);
|
|
Node* base = n->in(1);
|
|
|
|
if (!con->is_top()) {
|
|
return Multiplication(base, con->get_integer_as_long(bt));
|
|
}
|
|
}
|
|
|
|
return make_invalid();
|
|
}
|
|
|
|
// Try to match `n = (a << CON1) + (a << CON2)`. On success, return a struct with `.valid = true`, `variable = a`, and
|
|
// `multiplier = (1 << CON1) + (1 << CON2)`.
|
|
// Match `n` for patterns:
|
|
// - (1) (a << CON) + (a << CON)
|
|
// - (2) (a << CON) + a
|
|
// - (3) a + (a << CON)
|
|
// - (4) a + a
|
|
// Note that one or both of the term of the addition could simply be `a` (i.e., a << 0) as in pattern (4).
|
|
AddNode::Multiplication AddNode::Multiplication::find_power_of_two_addition_pattern(const Node* n, BasicType bt) {
|
|
if (n->Opcode() == Op_Add(bt) && n->in(1) != n->in(2)) {
|
|
const Multiplication lhs = find_simple_lshift_pattern(n->in(1), bt);
|
|
const Multiplication rhs = find_simple_lshift_pattern(n->in(2), bt);
|
|
|
|
// Pattern (1)
|
|
{
|
|
const Multiplication res = lhs.add(rhs);
|
|
if (res.is_valid()) {
|
|
return res;
|
|
}
|
|
}
|
|
|
|
// Pattern (2)
|
|
if (lhs.is_valid_with(n->in(2))) {
|
|
return Multiplication(lhs.variable(), java_add(lhs.multiplier(), CONST64(1)));
|
|
}
|
|
|
|
// Pattern (3)
|
|
if (rhs.is_valid_with(n->in(1))) {
|
|
return Multiplication(rhs.variable(), java_add(rhs.multiplier(), CONST64(1)));
|
|
}
|
|
|
|
// Pattern (4), which is equivalent to a simple addition pattern
|
|
return find_simple_addition_pattern(n, bt);
|
|
}
|
|
|
|
return make_invalid();
|
|
}
|
|
|
|
Node* AddINode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
Node* in1 = in(1);
|
|
Node* in2 = in(2);
|
|
int op1 = in1->Opcode();
|
|
int op2 = in2->Opcode();
|
|
|
|
// Convert (x>>>z)+y into (x+(y<<z))>>>z for small constant z and y.
|
|
// Helps with array allocation math constant folding
|
|
// See 4790063:
|
|
// Unrestricted transformation is unsafe for some runtime values of 'x'
|
|
// ( x == 0, z == 1, y == -1 ) fails
|
|
// ( x == -5, z == 1, y == 1 ) fails
|
|
// Transform works for small z and small negative y when the addition
|
|
// (x + (y << z)) does not cross zero.
|
|
// Implement support for negative y and (x >= -(y << z))
|
|
// Have not observed cases where type information exists to support
|
|
// positive y and (x <= -(y << z))
|
|
if (op1 == Op_URShiftI && op2 == Op_ConI &&
|
|
in1->in(2)->Opcode() == Op_ConI) {
|
|
jint z = phase->type(in1->in(2))->is_int()->get_con() & 0x1f; // only least significant 5 bits matter
|
|
jint y = phase->type(in2)->is_int()->get_con();
|
|
|
|
if (z < 5 && -5 < y && y < 0) {
|
|
const Type* t_in11 = phase->type(in1->in(1));
|
|
if( t_in11 != Type::TOP && (t_in11->is_int()->_lo >= -(y << z))) {
|
|
Node* a = phase->transform(new AddINode( in1->in(1), phase->intcon(y<<z)));
|
|
return new URShiftINode(a, in1->in(2));
|
|
}
|
|
}
|
|
}
|
|
|
|
return AddNode::IdealIL(phase, can_reshape, T_INT);
|
|
}
|
|
|
|
|
|
//------------------------------Identity---------------------------------------
|
|
// Fold (x-y)+y OR y+(x-y) into x
|
|
Node* AddINode::Identity(PhaseGVN* phase) {
|
|
if (in(1)->Opcode() == Op_SubI && in(1)->in(2) == in(2)) {
|
|
return in(1)->in(1);
|
|
} else if (in(2)->Opcode() == Op_SubI && in(2)->in(2) == in(1)) {
|
|
return in(2)->in(1);
|
|
}
|
|
return AddNode::Identity(phase);
|
|
}
|
|
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
// Supplied function returns the sum of the inputs. Guaranteed never
|
|
// to be passed a TOP or BOTTOM type, these are filtered out by
|
|
// pre-check.
|
|
const Type *AddINode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
const TypeInt *r0 = t0->is_int(); // Handy access
|
|
const TypeInt *r1 = t1->is_int();
|
|
int lo = java_add(r0->_lo, r1->_lo);
|
|
int hi = java_add(r0->_hi, r1->_hi);
|
|
if( !(r0->is_con() && r1->is_con()) ) {
|
|
// Not both constants, compute approximate result
|
|
if( (r0->_lo & r1->_lo) < 0 && lo >= 0 ) {
|
|
lo = min_jint; hi = max_jint; // Underflow on the low side
|
|
}
|
|
if( (~(r0->_hi | r1->_hi)) < 0 && hi < 0 ) {
|
|
lo = min_jint; hi = max_jint; // Overflow on the high side
|
|
}
|
|
if( lo > hi ) { // Handle overflow
|
|
lo = min_jint; hi = max_jint;
|
|
}
|
|
} else {
|
|
// both constants, compute precise result using 'lo' and 'hi'
|
|
// Semantics define overflow and underflow for integer addition
|
|
// as expected. In particular: 0x80000000 + 0x80000000 --> 0x0
|
|
}
|
|
return TypeInt::make( lo, hi, MAX2(r0->_widen,r1->_widen) );
|
|
}
|
|
|
|
|
|
//=============================================================================
|
|
//------------------------------Idealize---------------------------------------
|
|
Node* AddLNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
return AddNode::IdealIL(phase, can_reshape, T_LONG);
|
|
}
|
|
|
|
|
|
//------------------------------Identity---------------------------------------
|
|
// Fold (x-y)+y OR y+(x-y) into x
|
|
Node* AddLNode::Identity(PhaseGVN* phase) {
|
|
if (in(1)->Opcode() == Op_SubL && in(1)->in(2) == in(2)) {
|
|
return in(1)->in(1);
|
|
} else if (in(2)->Opcode() == Op_SubL && in(2)->in(2) == in(1)) {
|
|
return in(2)->in(1);
|
|
}
|
|
return AddNode::Identity(phase);
|
|
}
|
|
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
// Supplied function returns the sum of the inputs. Guaranteed never
|
|
// to be passed a TOP or BOTTOM type, these are filtered out by
|
|
// pre-check.
|
|
const Type *AddLNode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
const TypeLong *r0 = t0->is_long(); // Handy access
|
|
const TypeLong *r1 = t1->is_long();
|
|
jlong lo = java_add(r0->_lo, r1->_lo);
|
|
jlong hi = java_add(r0->_hi, r1->_hi);
|
|
if( !(r0->is_con() && r1->is_con()) ) {
|
|
// Not both constants, compute approximate result
|
|
if( (r0->_lo & r1->_lo) < 0 && lo >= 0 ) {
|
|
lo =min_jlong; hi = max_jlong; // Underflow on the low side
|
|
}
|
|
if( (~(r0->_hi | r1->_hi)) < 0 && hi < 0 ) {
|
|
lo = min_jlong; hi = max_jlong; // Overflow on the high side
|
|
}
|
|
if( lo > hi ) { // Handle overflow
|
|
lo = min_jlong; hi = max_jlong;
|
|
}
|
|
} else {
|
|
// both constants, compute precise result using 'lo' and 'hi'
|
|
// Semantics define overflow and underflow for integer addition
|
|
// as expected. In particular: 0x80000000 + 0x80000000 --> 0x0
|
|
}
|
|
return TypeLong::make( lo, hi, MAX2(r0->_widen,r1->_widen) );
|
|
}
|
|
|
|
|
|
//=============================================================================
|
|
//------------------------------add_of_identity--------------------------------
|
|
// Check for addition of the identity
|
|
const Type *AddFNode::add_of_identity( const Type *t1, const Type *t2 ) const {
|
|
// x ADD 0 should return x unless 'x' is a -zero
|
|
//
|
|
// const Type *zero = add_id(); // The additive identity
|
|
// jfloat f1 = t1->getf();
|
|
// jfloat f2 = t2->getf();
|
|
//
|
|
// if( t1->higher_equal( zero ) ) return t2;
|
|
// if( t2->higher_equal( zero ) ) return t1;
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
// Supplied function returns the sum of the inputs.
|
|
// This also type-checks the inputs for sanity. Guaranteed never to
|
|
// be passed a TOP or BOTTOM type, these are filtered out by pre-check.
|
|
const Type *AddFNode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
if (!t0->isa_float_constant() || !t1->isa_float_constant()) {
|
|
return bottom_type();
|
|
}
|
|
return TypeF::make( t0->getf() + t1->getf() );
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
Node *AddFNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
// Floating point additions are not associative because of boundary conditions (infinity)
|
|
return commute(phase, this) ? this : nullptr;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------add_of_identity--------------------------------
|
|
// Check for addition of the identity
|
|
const Type* AddHFNode::add_of_identity(const Type* t1, const Type* t2) const {
|
|
return nullptr;
|
|
}
|
|
|
|
// Supplied function returns the sum of the inputs.
|
|
// This also type-checks the inputs for sanity. Guaranteed never to
|
|
// be passed a TOP or BOTTOM type, these are filtered out by pre-check.
|
|
const Type* AddHFNode::add_ring(const Type* t0, const Type* t1) const {
|
|
if (!t0->isa_half_float_constant() || !t1->isa_half_float_constant()) {
|
|
return bottom_type();
|
|
}
|
|
return TypeH::make(t0->getf() + t1->getf());
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------add_of_identity--------------------------------
|
|
// Check for addition of the identity
|
|
const Type *AddDNode::add_of_identity( const Type *t1, const Type *t2 ) const {
|
|
// x ADD 0 should return x unless 'x' is a -zero
|
|
//
|
|
// const Type *zero = add_id(); // The additive identity
|
|
// jfloat f1 = t1->getf();
|
|
// jfloat f2 = t2->getf();
|
|
//
|
|
// if( t1->higher_equal( zero ) ) return t2;
|
|
// if( t2->higher_equal( zero ) ) return t1;
|
|
|
|
return nullptr;
|
|
}
|
|
//------------------------------add_ring---------------------------------------
|
|
// Supplied function returns the sum of the inputs.
|
|
// This also type-checks the inputs for sanity. Guaranteed never to
|
|
// be passed a TOP or BOTTOM type, these are filtered out by pre-check.
|
|
const Type *AddDNode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
if (!t0->isa_double_constant() || !t1->isa_double_constant()) {
|
|
return bottom_type();
|
|
}
|
|
return TypeD::make( t0->getd() + t1->getd() );
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
Node *AddDNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
// Floating point additions are not associative because of boundary conditions (infinity)
|
|
return commute(phase, this) ? this : nullptr;
|
|
}
|
|
|
|
|
|
//=============================================================================
|
|
//------------------------------Identity---------------------------------------
|
|
// If one input is a constant 0, return the other input.
|
|
Node* AddPNode::Identity(PhaseGVN* phase) {
|
|
return ( phase->type( in(Offset) )->higher_equal( TypeX_ZERO ) ) ? in(Address) : this;
|
|
}
|
|
|
|
//------------------------------Idealize---------------------------------------
|
|
Node *AddPNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
// Bail out if dead inputs
|
|
if( phase->type( in(Address) ) == Type::TOP ) return nullptr;
|
|
|
|
// If the left input is an add of a constant, flatten the expression tree.
|
|
const Node *n = in(Address);
|
|
if (n->is_AddP() && n->in(Base) == in(Base)) {
|
|
const AddPNode *addp = n->as_AddP(); // Left input is an AddP
|
|
assert( !addp->in(Address)->is_AddP() ||
|
|
addp->in(Address)->as_AddP() != addp,
|
|
"dead loop in AddPNode::Ideal" );
|
|
// Type of left input's right input
|
|
const Type *t = phase->type( addp->in(Offset) );
|
|
if( t == Type::TOP ) return nullptr;
|
|
const TypeX *t12 = t->is_intptr_t();
|
|
if( t12->is_con() ) { // Left input is an add of a constant?
|
|
// If the right input is a constant, combine constants
|
|
const Type *temp_t2 = phase->type( in(Offset) );
|
|
if( temp_t2 == Type::TOP ) return nullptr;
|
|
const TypeX *t2 = temp_t2->is_intptr_t();
|
|
Node* address;
|
|
Node* offset;
|
|
if( t2->is_con() ) {
|
|
// The Add of the flattened expression
|
|
address = addp->in(Address);
|
|
offset = phase->MakeConX(t2->get_con() + t12->get_con());
|
|
} else {
|
|
// Else move the constant to the right. ((A+con)+B) into ((A+B)+con)
|
|
address = phase->transform(new AddPNode(in(Base),addp->in(Address),in(Offset)));
|
|
offset = addp->in(Offset);
|
|
}
|
|
set_req_X(Address, address, phase);
|
|
set_req_X(Offset, offset, phase);
|
|
return this;
|
|
}
|
|
}
|
|
|
|
// Raw pointers?
|
|
if( in(Base)->bottom_type() == Type::TOP ) {
|
|
// If this is a null+long form (from unsafe accesses), switch to a rawptr.
|
|
if (phase->type(in(Address)) == TypePtr::NULL_PTR) {
|
|
Node* offset = in(Offset);
|
|
return new CastX2PNode(offset);
|
|
}
|
|
}
|
|
|
|
// If the right is an add of a constant, push the offset down.
|
|
// Convert: (ptr + (offset+con)) into (ptr+offset)+con.
|
|
// The idea is to merge array_base+scaled_index groups together,
|
|
// and only have different constant offsets from the same base.
|
|
const Node *add = in(Offset);
|
|
if( add->Opcode() == Op_AddX && add->in(1) != add ) {
|
|
const Type *t22 = phase->type( add->in(2) );
|
|
if( t22->singleton() && (t22 != Type::TOP) ) { // Right input is an add of a constant?
|
|
set_req(Address, phase->transform(new AddPNode(in(Base),in(Address),add->in(1))));
|
|
set_req_X(Offset, add->in(2), phase); // puts add on igvn worklist if needed
|
|
return this; // Made progress
|
|
}
|
|
}
|
|
|
|
return nullptr; // No progress
|
|
}
|
|
|
|
//------------------------------bottom_type------------------------------------
|
|
// Bottom-type is the pointer-type with unknown offset.
|
|
const Type *AddPNode::bottom_type() const {
|
|
if (in(Address) == nullptr) return TypePtr::BOTTOM;
|
|
const TypePtr *tp = in(Address)->bottom_type()->isa_ptr();
|
|
if( !tp ) return Type::TOP; // TOP input means TOP output
|
|
assert( in(Offset)->Opcode() != Op_ConP, "" );
|
|
const Type *t = in(Offset)->bottom_type();
|
|
if( t == Type::TOP )
|
|
return tp->add_offset(Type::OffsetTop);
|
|
const TypeX *tx = t->is_intptr_t();
|
|
intptr_t txoffset = Type::OffsetBot;
|
|
if (tx->is_con()) { // Left input is an add of a constant?
|
|
txoffset = tx->get_con();
|
|
}
|
|
return tp->add_offset(txoffset);
|
|
}
|
|
|
|
//------------------------------Value------------------------------------------
|
|
const Type* AddPNode::Value(PhaseGVN* phase) const {
|
|
// Either input is TOP ==> the result is TOP
|
|
const Type *t1 = phase->type( in(Address) );
|
|
const Type *t2 = phase->type( in(Offset) );
|
|
if( t1 == Type::TOP ) return Type::TOP;
|
|
if( t2 == Type::TOP ) return Type::TOP;
|
|
|
|
// Left input is a pointer
|
|
const TypePtr *p1 = t1->isa_ptr();
|
|
// Right input is an int
|
|
const TypeX *p2 = t2->is_intptr_t();
|
|
// Add 'em
|
|
intptr_t p2offset = Type::OffsetBot;
|
|
if (p2->is_con()) { // Left input is an add of a constant?
|
|
p2offset = p2->get_con();
|
|
}
|
|
return p1->add_offset(p2offset);
|
|
}
|
|
|
|
//------------------------Ideal_base_and_offset--------------------------------
|
|
// Split an oop pointer into a base and offset.
|
|
// (The offset might be Type::OffsetBot in the case of an array.)
|
|
// Return the base, or null if failure.
|
|
Node* AddPNode::Ideal_base_and_offset(Node* ptr, PhaseValues* phase,
|
|
// second return value:
|
|
intptr_t& offset) {
|
|
if (ptr->is_AddP()) {
|
|
Node* base = ptr->in(AddPNode::Base);
|
|
Node* addr = ptr->in(AddPNode::Address);
|
|
Node* offs = ptr->in(AddPNode::Offset);
|
|
if (base == addr || base->is_top()) {
|
|
offset = phase->find_intptr_t_con(offs, Type::OffsetBot);
|
|
if (offset != Type::OffsetBot) {
|
|
return addr;
|
|
}
|
|
}
|
|
}
|
|
offset = Type::OffsetBot;
|
|
return nullptr;
|
|
}
|
|
|
|
//------------------------------unpack_offsets----------------------------------
|
|
// Collect the AddP offset values into the elements array, giving up
|
|
// if there are more than length.
|
|
int AddPNode::unpack_offsets(Node* elements[], int length) const {
|
|
int count = 0;
|
|
Node const* addr = this;
|
|
Node* base = addr->in(AddPNode::Base);
|
|
while (addr->is_AddP()) {
|
|
if (addr->in(AddPNode::Base) != base) {
|
|
// give up
|
|
return -1;
|
|
}
|
|
elements[count++] = addr->in(AddPNode::Offset);
|
|
if (count == length) {
|
|
// give up
|
|
return -1;
|
|
}
|
|
addr = addr->in(AddPNode::Address);
|
|
}
|
|
if (addr != base) {
|
|
return -1;
|
|
}
|
|
return count;
|
|
}
|
|
|
|
//------------------------------match_edge-------------------------------------
|
|
// Do we Match on this edge index or not? Do not match base pointer edge
|
|
uint AddPNode::match_edge(uint idx) const {
|
|
return idx > Base;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Identity---------------------------------------
|
|
Node* OrINode::Identity(PhaseGVN* phase) {
|
|
// x | x => x
|
|
if (in(1) == in(2)) {
|
|
return in(1);
|
|
}
|
|
|
|
return AddNode::Identity(phase);
|
|
}
|
|
|
|
// Find shift value for Integer or Long OR.
|
|
static Node* rotate_shift(PhaseGVN* phase, Node* lshift, Node* rshift, int mask) {
|
|
// val << norm_con_shift | val >> ({32|64} - norm_con_shift) => rotate_left val, norm_con_shift
|
|
const TypeInt* lshift_t = phase->type(lshift)->isa_int();
|
|
const TypeInt* rshift_t = phase->type(rshift)->isa_int();
|
|
if (lshift_t != nullptr && lshift_t->is_con() &&
|
|
rshift_t != nullptr && rshift_t->is_con() &&
|
|
((lshift_t->get_con() & mask) == ((mask + 1) - (rshift_t->get_con() & mask)))) {
|
|
return phase->intcon(lshift_t->get_con() & mask);
|
|
}
|
|
// val << var_shift | val >> ({0|32|64} - var_shift) => rotate_left val, var_shift
|
|
if (rshift->Opcode() == Op_SubI && rshift->in(2) == lshift && rshift->in(1)->is_Con()){
|
|
const TypeInt* shift_t = phase->type(rshift->in(1))->isa_int();
|
|
if (shift_t != nullptr && shift_t->is_con() &&
|
|
(shift_t->get_con() == 0 || shift_t->get_con() == (mask + 1))) {
|
|
return lshift;
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
Node* OrINode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
int lopcode = in(1)->Opcode();
|
|
int ropcode = in(2)->Opcode();
|
|
if (Matcher::match_rule_supported(Op_RotateLeft) &&
|
|
lopcode == Op_LShiftI && ropcode == Op_URShiftI && in(1)->in(1) == in(2)->in(1)) {
|
|
Node* lshift = in(1)->in(2);
|
|
Node* rshift = in(2)->in(2);
|
|
Node* shift = rotate_shift(phase, lshift, rshift, 0x1F);
|
|
if (shift != nullptr) {
|
|
return new RotateLeftNode(in(1)->in(1), shift, TypeInt::INT);
|
|
}
|
|
return nullptr;
|
|
}
|
|
if (Matcher::match_rule_supported(Op_RotateRight) &&
|
|
lopcode == Op_URShiftI && ropcode == Op_LShiftI && in(1)->in(1) == in(2)->in(1)) {
|
|
Node* rshift = in(1)->in(2);
|
|
Node* lshift = in(2)->in(2);
|
|
Node* shift = rotate_shift(phase, rshift, lshift, 0x1F);
|
|
if (shift != nullptr) {
|
|
return new RotateRightNode(in(1)->in(1), shift, TypeInt::INT);
|
|
}
|
|
}
|
|
|
|
// Convert "~a | ~b" into "~(a & b)"
|
|
if (AddNode::is_not(phase, in(1), T_INT) && AddNode::is_not(phase, in(2), T_INT)) {
|
|
Node* and_a_b = new AndINode(in(1)->in(1), in(2)->in(1));
|
|
Node* tn = phase->transform(and_a_b);
|
|
return AddNode::make_not(phase, tn, T_INT);
|
|
}
|
|
return AddNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
// Supplied function returns the sum of the inputs IN THE CURRENT RING. For
|
|
// the logical operations the ring's ADD is really a logical OR function.
|
|
// This also type-checks the inputs for sanity. Guaranteed never to
|
|
// be passed a TOP or BOTTOM type, these are filtered out by pre-check.
|
|
const Type *OrINode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
const TypeInt *r0 = t0->is_int(); // Handy access
|
|
const TypeInt *r1 = t1->is_int();
|
|
|
|
// If both args are bool, can figure out better types
|
|
if ( r0 == TypeInt::BOOL ) {
|
|
if ( r1 == TypeInt::ONE) {
|
|
return TypeInt::ONE;
|
|
} else if ( r1 == TypeInt::BOOL ) {
|
|
return TypeInt::BOOL;
|
|
}
|
|
} else if ( r0 == TypeInt::ONE ) {
|
|
if ( r1 == TypeInt::BOOL ) {
|
|
return TypeInt::ONE;
|
|
}
|
|
}
|
|
|
|
// If either input is all ones, the output is all ones.
|
|
// x | ~0 == ~0 <==> x | -1 == -1
|
|
if (r0 == TypeInt::MINUS_1 || r1 == TypeInt::MINUS_1) {
|
|
return TypeInt::MINUS_1;
|
|
}
|
|
|
|
// If either input is not a constant, just return all integers.
|
|
if( !r0->is_con() || !r1->is_con() )
|
|
return TypeInt::INT; // Any integer, but still no symbols.
|
|
|
|
// Otherwise just OR them bits.
|
|
return TypeInt::make( r0->get_con() | r1->get_con() );
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Identity---------------------------------------
|
|
Node* OrLNode::Identity(PhaseGVN* phase) {
|
|
// x | x => x
|
|
if (in(1) == in(2)) {
|
|
return in(1);
|
|
}
|
|
|
|
return AddNode::Identity(phase);
|
|
}
|
|
|
|
Node* OrLNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
int lopcode = in(1)->Opcode();
|
|
int ropcode = in(2)->Opcode();
|
|
if (Matcher::match_rule_supported(Op_RotateLeft) &&
|
|
lopcode == Op_LShiftL && ropcode == Op_URShiftL && in(1)->in(1) == in(2)->in(1)) {
|
|
Node* lshift = in(1)->in(2);
|
|
Node* rshift = in(2)->in(2);
|
|
Node* shift = rotate_shift(phase, lshift, rshift, 0x3F);
|
|
if (shift != nullptr) {
|
|
return new RotateLeftNode(in(1)->in(1), shift, TypeLong::LONG);
|
|
}
|
|
return nullptr;
|
|
}
|
|
if (Matcher::match_rule_supported(Op_RotateRight) &&
|
|
lopcode == Op_URShiftL && ropcode == Op_LShiftL && in(1)->in(1) == in(2)->in(1)) {
|
|
Node* rshift = in(1)->in(2);
|
|
Node* lshift = in(2)->in(2);
|
|
Node* shift = rotate_shift(phase, rshift, lshift, 0x3F);
|
|
if (shift != nullptr) {
|
|
return new RotateRightNode(in(1)->in(1), shift, TypeLong::LONG);
|
|
}
|
|
}
|
|
|
|
// Convert "~a | ~b" into "~(a & b)"
|
|
if (AddNode::is_not(phase, in(1), T_LONG) && AddNode::is_not(phase, in(2), T_LONG)) {
|
|
Node* and_a_b = new AndLNode(in(1)->in(1), in(2)->in(1));
|
|
Node* tn = phase->transform(and_a_b);
|
|
return AddNode::make_not(phase, tn, T_LONG);
|
|
}
|
|
|
|
return AddNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
const Type *OrLNode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
const TypeLong *r0 = t0->is_long(); // Handy access
|
|
const TypeLong *r1 = t1->is_long();
|
|
|
|
// If either input is all ones, the output is all ones.
|
|
// x | ~0 == ~0 <==> x | -1 == -1
|
|
if (r0 == TypeLong::MINUS_1 || r1 == TypeLong::MINUS_1) {
|
|
return TypeLong::MINUS_1;
|
|
}
|
|
|
|
// If either input is not a constant, just return all integers.
|
|
if( !r0->is_con() || !r1->is_con() )
|
|
return TypeLong::LONG; // Any integer, but still no symbols.
|
|
|
|
// Otherwise just OR them bits.
|
|
return TypeLong::make( r0->get_con() | r1->get_con() );
|
|
}
|
|
|
|
//---------------------------Helper -------------------------------------------
|
|
/* Decide if the given node is used only in arithmetic expressions(add or sub).
|
|
*/
|
|
static bool is_used_in_only_arithmetic(Node* n, BasicType bt) {
|
|
for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
|
|
Node* u = n->fast_out(i);
|
|
if (u->Opcode() != Op_Add(bt) && u->Opcode() != Op_Sub(bt)) {
|
|
return false;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Idealize---------------------------------------
|
|
Node* XorINode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
Node* in1 = in(1);
|
|
Node* in2 = in(2);
|
|
|
|
// Convert ~x into -1-x when ~x is used in an arithmetic expression
|
|
// or x itself is an expression.
|
|
if (phase->type(in2) == TypeInt::MINUS_1) { // follows LHS^(-1), i.e., ~LHS
|
|
if (phase->is_IterGVN()) {
|
|
if (is_used_in_only_arithmetic(this, T_INT)
|
|
// LHS is arithmetic
|
|
|| (in1->Opcode() == Op_AddI || in1->Opcode() == Op_SubI)) {
|
|
return new SubINode(in2, in1);
|
|
}
|
|
} else {
|
|
// graph could be incomplete in GVN so we postpone to IGVN
|
|
phase->record_for_igvn(this);
|
|
}
|
|
}
|
|
|
|
// Propagate xor through constant cmoves. This pattern can occur after expansion of Conv2B nodes.
|
|
const TypeInt* in2_type = phase->type(in2)->isa_int();
|
|
if (in1->Opcode() == Op_CMoveI && in2_type != nullptr && in2_type->is_con()) {
|
|
int in2_val = in2_type->get_con();
|
|
|
|
// Get types of both sides of the CMove
|
|
const TypeInt* left = phase->type(in1->in(CMoveNode::IfFalse))->isa_int();
|
|
const TypeInt* right = phase->type(in1->in(CMoveNode::IfTrue))->isa_int();
|
|
|
|
// Ensure that both sides are int constants
|
|
if (left != nullptr && right != nullptr && left->is_con() && right->is_con()) {
|
|
Node* cond = in1->in(CMoveNode::Condition);
|
|
|
|
// Check that the comparison is a bool and that the cmp node type is correct
|
|
if (cond->is_Bool()) {
|
|
int cmp_op = cond->in(1)->Opcode();
|
|
|
|
if (cmp_op == Op_CmpI || cmp_op == Op_CmpP) {
|
|
int l_val = left->get_con();
|
|
int r_val = right->get_con();
|
|
|
|
return new CMoveINode(cond, phase->intcon(l_val ^ in2_val), phase->intcon(r_val ^ in2_val), TypeInt::INT);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return AddNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
const Type* XorINode::Value(PhaseGVN* phase) const {
|
|
Node* in1 = in(1);
|
|
Node* in2 = in(2);
|
|
const Type* t1 = phase->type(in1);
|
|
const Type* t2 = phase->type(in2);
|
|
if (t1 == Type::TOP || t2 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
// x ^ x ==> 0
|
|
if (in1->eqv_uncast(in2)) {
|
|
return add_id();
|
|
}
|
|
return AddNode::Value(phase);
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
// Supplied function returns the sum of the inputs IN THE CURRENT RING. For
|
|
// the logical operations the ring's ADD is really a logical OR function.
|
|
// This also type-checks the inputs for sanity. Guaranteed never to
|
|
// be passed a TOP or BOTTOM type, these are filtered out by pre-check.
|
|
const Type *XorINode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
const TypeInt *r0 = t0->is_int(); // Handy access
|
|
const TypeInt *r1 = t1->is_int();
|
|
|
|
if (r0->is_con() && r1->is_con()) {
|
|
// compute constant result
|
|
return TypeInt::make(r0->get_con() ^ r1->get_con());
|
|
}
|
|
|
|
// At least one of the arguments is not constant
|
|
|
|
if (r0->_lo >= 0 && r1->_lo >= 0) {
|
|
// Combine [r0->_lo, r0->_hi] ^ [r0->_lo, r1->_hi] -> [0, upper_bound]
|
|
jint upper_bound = xor_upper_bound_for_ranges<jint, juint>(r0->_hi, r1->_hi);
|
|
return TypeInt::make(0, upper_bound, MAX2(r0->_widen, r1->_widen));
|
|
}
|
|
|
|
return TypeInt::INT;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------add_ring---------------------------------------
|
|
const Type *XorLNode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
const TypeLong *r0 = t0->is_long(); // Handy access
|
|
const TypeLong *r1 = t1->is_long();
|
|
|
|
if (r0->is_con() && r1->is_con()) {
|
|
// compute constant result
|
|
return TypeLong::make(r0->get_con() ^ r1->get_con());
|
|
}
|
|
|
|
// At least one of the arguments is not constant
|
|
|
|
if (r0->_lo >= 0 && r1->_lo >= 0) {
|
|
// Combine [r0->_lo, r0->_hi] ^ [r0->_lo, r1->_hi] -> [0, upper_bound]
|
|
julong upper_bound = xor_upper_bound_for_ranges<jlong, julong>(r0->_hi, r1->_hi);
|
|
return TypeLong::make(0, upper_bound, MAX2(r0->_widen, r1->_widen));
|
|
}
|
|
|
|
return TypeLong::LONG;
|
|
}
|
|
|
|
Node* XorLNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
Node* in1 = in(1);
|
|
Node* in2 = in(2);
|
|
|
|
// Convert ~x into -1-x when ~x is used in an arithmetic expression
|
|
// or x itself is an arithmetic expression.
|
|
if (phase->type(in2) == TypeLong::MINUS_1) { // follows LHS^(-1), i.e., ~LHS
|
|
if (phase->is_IterGVN()) {
|
|
if (is_used_in_only_arithmetic(this, T_LONG)
|
|
// LHS is arithmetic
|
|
|| (in1->Opcode() == Op_AddL || in1->Opcode() == Op_SubL)) {
|
|
return new SubLNode(in2, in1);
|
|
}
|
|
} else {
|
|
// graph could be incomplete in GVN so we postpone to IGVN
|
|
phase->record_for_igvn(this);
|
|
}
|
|
}
|
|
return AddNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
const Type* XorLNode::Value(PhaseGVN* phase) const {
|
|
Node* in1 = in(1);
|
|
Node* in2 = in(2);
|
|
const Type* t1 = phase->type(in1);
|
|
const Type* t2 = phase->type(in2);
|
|
if (t1 == Type::TOP || t2 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
// x ^ x ==> 0
|
|
if (in1->eqv_uncast(in2)) {
|
|
return add_id();
|
|
}
|
|
|
|
return AddNode::Value(phase);
|
|
}
|
|
|
|
Node* MaxNode::build_min_max_int(Node* a, Node* b, bool is_max) {
|
|
if (is_max) {
|
|
return new MaxINode(a, b);
|
|
} else {
|
|
return new MinINode(a, b);
|
|
}
|
|
}
|
|
|
|
Node* MaxNode::build_min_max_long(PhaseGVN* phase, Node* a, Node* b, bool is_max) {
|
|
if (is_max) {
|
|
return new MaxLNode(phase->C, a, b);
|
|
} else {
|
|
return new MinLNode(phase->C, a, b);
|
|
}
|
|
}
|
|
|
|
Node* MaxNode::build_min_max(Node* a, Node* b, bool is_max, bool is_unsigned, const Type* t, PhaseGVN& gvn) {
|
|
bool is_int = gvn.type(a)->isa_int();
|
|
assert(is_int || gvn.type(a)->isa_long(), "int or long inputs");
|
|
assert(is_int == (gvn.type(b)->isa_int() != nullptr), "inconsistent inputs");
|
|
BasicType bt = is_int ? T_INT: T_LONG;
|
|
Node* hook = nullptr;
|
|
if (gvn.is_IterGVN()) {
|
|
// Make sure a and b are not destroyed
|
|
hook = new Node(2);
|
|
hook->init_req(0, a);
|
|
hook->init_req(1, b);
|
|
}
|
|
Node* res = nullptr;
|
|
if (is_int && !is_unsigned) {
|
|
res = gvn.transform(build_min_max_int(a, b, is_max));
|
|
assert(gvn.type(res)->is_int()->_lo >= t->is_int()->_lo && gvn.type(res)->is_int()->_hi <= t->is_int()->_hi, "type doesn't match");
|
|
} else {
|
|
Node* cmp = nullptr;
|
|
if (is_max) {
|
|
cmp = gvn.transform(CmpNode::make(a, b, bt, is_unsigned));
|
|
} else {
|
|
cmp = gvn.transform(CmpNode::make(b, a, bt, is_unsigned));
|
|
}
|
|
Node* bol = gvn.transform(new BoolNode(cmp, BoolTest::lt));
|
|
res = gvn.transform(CMoveNode::make(bol, a, b, t));
|
|
}
|
|
if (hook != nullptr) {
|
|
hook->destruct(&gvn);
|
|
}
|
|
return res;
|
|
}
|
|
|
|
Node* MaxNode::build_min_max_diff_with_zero(Node* a, Node* b, bool is_max, const Type* t, PhaseGVN& gvn) {
|
|
bool is_int = gvn.type(a)->isa_int();
|
|
assert(is_int || gvn.type(a)->isa_long(), "int or long inputs");
|
|
assert(is_int == (gvn.type(b)->isa_int() != nullptr), "inconsistent inputs");
|
|
BasicType bt = is_int ? T_INT: T_LONG;
|
|
Node* zero = gvn.integercon(0, bt);
|
|
Node* hook = nullptr;
|
|
if (gvn.is_IterGVN()) {
|
|
// Make sure a and b are not destroyed
|
|
hook = new Node(2);
|
|
hook->init_req(0, a);
|
|
hook->init_req(1, b);
|
|
}
|
|
Node* cmp = nullptr;
|
|
if (is_max) {
|
|
cmp = gvn.transform(CmpNode::make(a, b, bt, false));
|
|
} else {
|
|
cmp = gvn.transform(CmpNode::make(b, a, bt, false));
|
|
}
|
|
Node* sub = gvn.transform(SubNode::make(a, b, bt));
|
|
Node* bol = gvn.transform(new BoolNode(cmp, BoolTest::lt));
|
|
Node* res = gvn.transform(CMoveNode::make(bol, sub, zero, t));
|
|
if (hook != nullptr) {
|
|
hook->destruct(&gvn);
|
|
}
|
|
return res;
|
|
}
|
|
|
|
// Check if addition of an integer with type 't' and a constant 'c' can overflow.
|
|
static bool can_overflow(const TypeInt* t, jint c) {
|
|
jint t_lo = t->_lo;
|
|
jint t_hi = t->_hi;
|
|
return ((c < 0 && (java_add(t_lo, c) > t_lo)) ||
|
|
(c > 0 && (java_add(t_hi, c) < t_hi)));
|
|
}
|
|
|
|
// Check if addition of a long with type 't' and a constant 'c' can overflow.
|
|
static bool can_overflow(const TypeLong* t, jlong c) {
|
|
jlong t_lo = t->_lo;
|
|
jlong t_hi = t->_hi;
|
|
return ((c < 0 && (java_add(t_lo, c) > t_lo)) ||
|
|
(c > 0 && (java_add(t_hi, c) < t_hi)));
|
|
}
|
|
|
|
// Let <x, x_off> = x_operands and <y, y_off> = y_operands.
|
|
// If x == y and neither add(x, x_off) nor add(y, y_off) overflow, return
|
|
// add(x, op(x_off, y_off)). Otherwise, return nullptr.
|
|
Node* MaxNode::extract_add(PhaseGVN* phase, ConstAddOperands x_operands, ConstAddOperands y_operands) {
|
|
Node* x = x_operands.first;
|
|
Node* y = y_operands.first;
|
|
int opcode = Opcode();
|
|
assert(opcode == Op_MaxI || opcode == Op_MinI, "Unexpected opcode");
|
|
const TypeInt* tx = phase->type(x)->isa_int();
|
|
jint x_off = x_operands.second;
|
|
jint y_off = y_operands.second;
|
|
if (x == y && tx != nullptr &&
|
|
!can_overflow(tx, x_off) &&
|
|
!can_overflow(tx, y_off)) {
|
|
jint c = opcode == Op_MinI ? MIN2(x_off, y_off) : MAX2(x_off, y_off);
|
|
return new AddINode(x, phase->intcon(c));
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
// Try to cast n as an integer addition with a constant. Return:
|
|
// <x, C>, if n == add(x, C), where 'C' is a non-TOP constant;
|
|
// <nullptr, 0>, if n == add(x, C), where 'C' is a TOP constant; or
|
|
// <n, 0>, otherwise.
|
|
static ConstAddOperands as_add_with_constant(Node* n) {
|
|
if (n->Opcode() != Op_AddI) {
|
|
return ConstAddOperands(n, 0);
|
|
}
|
|
Node* x = n->in(1);
|
|
Node* c = n->in(2);
|
|
if (!c->is_Con()) {
|
|
return ConstAddOperands(n, 0);
|
|
}
|
|
const Type* c_type = c->bottom_type();
|
|
if (c_type == Type::TOP) {
|
|
return ConstAddOperands(nullptr, 0);
|
|
}
|
|
return ConstAddOperands(x, c_type->is_int()->get_con());
|
|
}
|
|
|
|
Node* MaxNode::IdealI(PhaseGVN* phase, bool can_reshape) {
|
|
int opcode = Opcode();
|
|
assert(opcode == Op_MinI || opcode == Op_MaxI, "Unexpected opcode");
|
|
// Try to transform the following pattern, in any of its four possible
|
|
// permutations induced by op's commutativity:
|
|
// op(op(add(inner, inner_off), inner_other), add(outer, outer_off))
|
|
// into
|
|
// op(add(inner, op(inner_off, outer_off)), inner_other),
|
|
// where:
|
|
// op is either MinI or MaxI, and
|
|
// inner == outer, and
|
|
// the additions cannot overflow.
|
|
for (uint inner_op_index = 1; inner_op_index <= 2; inner_op_index++) {
|
|
if (in(inner_op_index)->Opcode() != opcode) {
|
|
continue;
|
|
}
|
|
Node* outer_add = in(inner_op_index == 1 ? 2 : 1);
|
|
ConstAddOperands outer_add_operands = as_add_with_constant(outer_add);
|
|
if (outer_add_operands.first == nullptr) {
|
|
return nullptr; // outer_add has a TOP input, no need to continue.
|
|
}
|
|
// One operand is a MinI/MaxI and the other is an integer addition with
|
|
// constant. Test the operands of the inner MinI/MaxI.
|
|
for (uint inner_add_index = 1; inner_add_index <= 2; inner_add_index++) {
|
|
Node* inner_op = in(inner_op_index);
|
|
Node* inner_add = inner_op->in(inner_add_index);
|
|
ConstAddOperands inner_add_operands = as_add_with_constant(inner_add);
|
|
if (inner_add_operands.first == nullptr) {
|
|
return nullptr; // inner_add has a TOP input, no need to continue.
|
|
}
|
|
// Try to extract the inner add.
|
|
Node* add_extracted = extract_add(phase, inner_add_operands, outer_add_operands);
|
|
if (add_extracted == nullptr) {
|
|
continue;
|
|
}
|
|
Node* add_transformed = phase->transform(add_extracted);
|
|
Node* inner_other = inner_op->in(inner_add_index == 1 ? 2 : 1);
|
|
return build_min_max_int(add_transformed, inner_other, opcode == Op_MaxI);
|
|
}
|
|
}
|
|
// Try to transform
|
|
// op(add(x, x_off), add(y, y_off))
|
|
// into
|
|
// add(x, op(x_off, y_off)),
|
|
// where:
|
|
// op is either MinI or MaxI, and
|
|
// inner == outer, and
|
|
// the additions cannot overflow.
|
|
ConstAddOperands xC = as_add_with_constant(in(1));
|
|
ConstAddOperands yC = as_add_with_constant(in(2));
|
|
if (xC.first == nullptr || yC.first == nullptr) return nullptr;
|
|
return extract_add(phase, xC, yC);
|
|
}
|
|
|
|
// Ideal transformations for MaxINode
|
|
Node* MaxINode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
return IdealI(phase, can_reshape);
|
|
}
|
|
|
|
Node* MaxINode::Identity(PhaseGVN* phase) {
|
|
const TypeInt* t1 = phase->type(in(1))->is_int();
|
|
const TypeInt* t2 = phase->type(in(2))->is_int();
|
|
|
|
// Can we determine the maximum statically?
|
|
if (t1->_lo >= t2->_hi) {
|
|
return in(1);
|
|
} else if (t2->_lo >= t1->_hi) {
|
|
return in(2);
|
|
}
|
|
|
|
return MaxNode::Identity(phase);
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------add_ring---------------------------------------
|
|
// Supplied function returns the sum of the inputs.
|
|
const Type *MaxINode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
const TypeInt *r0 = t0->is_int(); // Handy access
|
|
const TypeInt *r1 = t1->is_int();
|
|
|
|
// Otherwise just MAX them bits.
|
|
return TypeInt::make( MAX2(r0->_lo,r1->_lo), MAX2(r0->_hi,r1->_hi), MAX2(r0->_widen,r1->_widen) );
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Idealize---------------------------------------
|
|
// MINs show up in range-check loop limit calculations. Look for
|
|
// "MIN2(x+c0,MIN2(y,x+c1))". Pick the smaller constant: "MIN2(x+c0,y)"
|
|
Node* MinINode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
return IdealI(phase, can_reshape);
|
|
}
|
|
|
|
Node* MinINode::Identity(PhaseGVN* phase) {
|
|
const TypeInt* t1 = phase->type(in(1))->is_int();
|
|
const TypeInt* t2 = phase->type(in(2))->is_int();
|
|
|
|
// Can we determine the minimum statically?
|
|
if (t1->_lo >= t2->_hi) {
|
|
return in(2);
|
|
} else if (t2->_lo >= t1->_hi) {
|
|
return in(1);
|
|
}
|
|
|
|
return MaxNode::Identity(phase);
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
// Supplied function returns the sum of the inputs.
|
|
const Type *MinINode::add_ring( const Type *t0, const Type *t1 ) const {
|
|
const TypeInt *r0 = t0->is_int(); // Handy access
|
|
const TypeInt *r1 = t1->is_int();
|
|
|
|
// Otherwise just MIN them bits.
|
|
return TypeInt::make( MIN2(r0->_lo,r1->_lo), MIN2(r0->_hi,r1->_hi), MAX2(r0->_widen,r1->_widen) );
|
|
}
|
|
|
|
// Collapse the "addition with overflow-protection" pattern, and the symmetrical
|
|
// "subtraction with underflow-protection" pattern. These are created during the
|
|
// unrolling, when we have to adjust the limit by subtracting the stride, but want
|
|
// to protect against underflow: MaxL(SubL(limit, stride), min_jint).
|
|
// If we have more than one of those in a sequence:
|
|
//
|
|
// x con2
|
|
// | |
|
|
// AddL clamp2
|
|
// | |
|
|
// Max/MinL con1
|
|
// | |
|
|
// AddL clamp1
|
|
// | |
|
|
// Max/MinL (n)
|
|
//
|
|
// We want to collapse it to:
|
|
//
|
|
// x con1 con2
|
|
// | | |
|
|
// | AddLNode (new_con)
|
|
// | |
|
|
// AddLNode clamp1
|
|
// | |
|
|
// Max/MinL (n)
|
|
//
|
|
// Note: we assume that SubL was already replaced by an AddL, and that the stride
|
|
// has its sign flipped: SubL(limit, stride) -> AddL(limit, -stride).
|
|
//
|
|
// Proof MaxL collapsed version equivalent to original (MinL version similar):
|
|
// is_sub_con ensures that con1, con2 ∈ [min_int, 0[
|
|
//
|
|
// Original:
|
|
// - AddL2 underflow => x + con2 ∈ ]max_long - min_int, max_long], ALWAYS BAILOUT as x + con1 + con2 surely fails can_overflow (*)
|
|
// - AddL2 no underflow => x + con2 ∈ [min_long, max_long]
|
|
// - MaxL2 clamp => min_int
|
|
// - AddL1 underflow: NOT POSSIBLE: cannot underflow since min_int + con1 ∈ [2 * min_int, min_int] always > min_long
|
|
// - AddL1 no underflow => min_int + con1 ∈ [2 * min_int, min_int]
|
|
// - MaxL1 clamp => min_int (RESULT 1)
|
|
// - MaxL1 no clamp: NOT POSSIBLE: min_int + con1 ∈ [2 * min_int, min_int] always <= min_int, so clamp always taken
|
|
// - MaxL2 no clamp => x + con2 ∈ [min_int, max_long]
|
|
// - AddL1 underflow: NOT POSSIBLE: cannot underflow since x + con2 + con1 ∈ [2 * min_int, max_long] always > min_long
|
|
// - AddL1 no underflow => x + con2 + con1 ∈ [2 * min_int, max_long]
|
|
// - MaxL1 clamp => min_int (RESULT 2)
|
|
// - MaxL1 no clamp => x + con2 + con1 ∈ ]min_int, max_long] (RESULT 3)
|
|
//
|
|
// Collapsed:
|
|
// - AddL2 (cannot underflow) => con2 + con1 ∈ [2 * min_int, 0]
|
|
// - AddL1 underflow: NOT POSSIBLE: would have bailed out at can_overflow (*)
|
|
// - AddL1 no underflow => x + con2 + con1 ∈ [min_long, max_long]
|
|
// - MaxL clamp => min_int (RESULT 1 and RESULT 2)
|
|
// - MaxL no clamp => x + con2 + con1 ∈ ]min_int, max_long] (RESULT 3)
|
|
//
|
|
static Node* fold_subI_no_underflow_pattern(Node* n, PhaseGVN* phase) {
|
|
assert(n->Opcode() == Op_MaxL || n->Opcode() == Op_MinL, "sanity");
|
|
// Check that the two clamps have the correct values.
|
|
jlong clamp = (n->Opcode() == Op_MaxL) ? min_jint : max_jint;
|
|
auto is_clamp = [&](Node* c) {
|
|
const TypeLong* t = phase->type(c)->isa_long();
|
|
return t != nullptr && t->is_con() &&
|
|
t->get_con() == clamp;
|
|
};
|
|
// Check that the constants are negative if MaxL, and positive if MinL.
|
|
auto is_sub_con = [&](Node* c) {
|
|
const TypeLong* t = phase->type(c)->isa_long();
|
|
return t != nullptr && t->is_con() &&
|
|
t->get_con() < max_jint && t->get_con() > min_jint &&
|
|
(t->get_con() < 0) == (n->Opcode() == Op_MaxL);
|
|
};
|
|
// Verify the graph level by level:
|
|
Node* add1 = n->in(1);
|
|
Node* clamp1 = n->in(2);
|
|
if (add1->Opcode() == Op_AddL && is_clamp(clamp1)) {
|
|
Node* max2 = add1->in(1);
|
|
Node* con1 = add1->in(2);
|
|
if (max2->Opcode() == n->Opcode() && is_sub_con(con1)) {
|
|
Node* add2 = max2->in(1);
|
|
Node* clamp2 = max2->in(2);
|
|
if (add2->Opcode() == Op_AddL && is_clamp(clamp2)) {
|
|
Node* x = add2->in(1);
|
|
Node* con2 = add2->in(2);
|
|
if (is_sub_con(con2)) {
|
|
// The graph could be dying (i.e. x is top) in which case type(x) is not a long.
|
|
const TypeLong* x_long = phase->type(x)->isa_long();
|
|
// Collapsed graph not equivalent if potential over/underflow -> bailing out (*)
|
|
if (x_long == nullptr || can_overflow(x_long, con1->get_long() + con2->get_long())) {
|
|
return nullptr;
|
|
}
|
|
Node* new_con = phase->transform(new AddLNode(con1, con2));
|
|
Node* new_sub = phase->transform(new AddLNode(x, new_con));
|
|
n->set_req_X(1, new_sub, phase);
|
|
return n;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
const Type* MaxLNode::add_ring(const Type* t0, const Type* t1) const {
|
|
const TypeLong* r0 = t0->is_long();
|
|
const TypeLong* r1 = t1->is_long();
|
|
|
|
return TypeLong::make(MAX2(r0->_lo, r1->_lo), MAX2(r0->_hi, r1->_hi), MAX2(r0->_widen, r1->_widen));
|
|
}
|
|
|
|
Node* MaxLNode::Identity(PhaseGVN* phase) {
|
|
const TypeLong* t1 = phase->type(in(1))->is_long();
|
|
const TypeLong* t2 = phase->type(in(2))->is_long();
|
|
|
|
// Can we determine maximum statically?
|
|
if (t1->_lo >= t2->_hi) {
|
|
return in(1);
|
|
} else if (t2->_lo >= t1->_hi) {
|
|
return in(2);
|
|
}
|
|
|
|
return MaxNode::Identity(phase);
|
|
}
|
|
|
|
Node* MaxLNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
Node* n = AddNode::Ideal(phase, can_reshape);
|
|
if (n != nullptr) {
|
|
return n;
|
|
}
|
|
if (can_reshape) {
|
|
return fold_subI_no_underflow_pattern(this, phase);
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
const Type* MinLNode::add_ring(const Type* t0, const Type* t1) const {
|
|
const TypeLong* r0 = t0->is_long();
|
|
const TypeLong* r1 = t1->is_long();
|
|
|
|
return TypeLong::make(MIN2(r0->_lo, r1->_lo), MIN2(r0->_hi, r1->_hi), MAX2(r0->_widen, r1->_widen));
|
|
}
|
|
|
|
Node* MinLNode::Identity(PhaseGVN* phase) {
|
|
const TypeLong* t1 = phase->type(in(1))->is_long();
|
|
const TypeLong* t2 = phase->type(in(2))->is_long();
|
|
|
|
// Can we determine minimum statically?
|
|
if (t1->_lo >= t2->_hi) {
|
|
return in(2);
|
|
} else if (t2->_lo >= t1->_hi) {
|
|
return in(1);
|
|
}
|
|
|
|
return MaxNode::Identity(phase);
|
|
}
|
|
|
|
Node* MinLNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
Node* n = AddNode::Ideal(phase, can_reshape);
|
|
if (n != nullptr) {
|
|
return n;
|
|
}
|
|
if (can_reshape) {
|
|
return fold_subI_no_underflow_pattern(this, phase);
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
int MaxNode::opposite_opcode() const {
|
|
if (Opcode() == max_opcode()) {
|
|
return min_opcode();
|
|
} else {
|
|
assert(Opcode() == min_opcode(), "Caller should be either %s or %s, but is %s", NodeClassNames[max_opcode()], NodeClassNames[min_opcode()], NodeClassNames[Opcode()]);
|
|
return max_opcode();
|
|
}
|
|
}
|
|
|
|
// Given a redundant structure such as Max/Min(A, Max/Min(B, C)) where A == B or A == C, return the useful part of the structure.
|
|
// 'operation' is the node expected to be the inner 'Max/Min(B, C)', and 'operand' is the node expected to be the 'A' operand of the outer node.
|
|
Node* MaxNode::find_identity_operation(Node* operation, Node* operand) {
|
|
if (operation->Opcode() == Opcode() || operation->Opcode() == opposite_opcode()) {
|
|
Node* n1 = operation->in(1);
|
|
Node* n2 = operation->in(2);
|
|
|
|
// Given Op(A, Op(B, C)), see if either A == B or A == C is true.
|
|
if (n1 == operand || n2 == operand) {
|
|
// If the operations are the same return the inner operation, as Max(A, Max(A, B)) == Max(A, B).
|
|
if (operation->Opcode() == Opcode()) {
|
|
return operation;
|
|
}
|
|
|
|
// If the operations are different return the operand 'A', as Max(A, Min(A, B)) == A if the value isn't floating point.
|
|
// With floating point values, the identity doesn't hold if B == NaN.
|
|
const Type* type = bottom_type();
|
|
if (type->isa_int() || type->isa_long()) {
|
|
return operand;
|
|
}
|
|
}
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
Node* MaxNode::Identity(PhaseGVN* phase) {
|
|
if (in(1) == in(2)) {
|
|
return in(1);
|
|
}
|
|
|
|
Node* identity_1 = MaxNode::find_identity_operation(in(2), in(1));
|
|
if (identity_1 != nullptr) {
|
|
return identity_1;
|
|
}
|
|
|
|
Node* identity_2 = MaxNode::find_identity_operation(in(1), in(2));
|
|
if (identity_2 != nullptr) {
|
|
return identity_2;
|
|
}
|
|
|
|
return AddNode::Identity(phase);
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
const Type* MinHFNode::add_ring(const Type* t0, const Type* t1) const {
|
|
const TypeH* r0 = t0->isa_half_float_constant();
|
|
const TypeH* r1 = t1->isa_half_float_constant();
|
|
if (r0 == nullptr || r1 == nullptr) {
|
|
return bottom_type();
|
|
}
|
|
|
|
if (r0->is_nan()) {
|
|
return r0;
|
|
}
|
|
if (r1->is_nan()) {
|
|
return r1;
|
|
}
|
|
|
|
float f0 = r0->getf();
|
|
float f1 = r1->getf();
|
|
if (f0 != 0.0f || f1 != 0.0f) {
|
|
return f0 < f1 ? r0 : r1;
|
|
}
|
|
|
|
// As per IEEE 754 specification, floating point comparison consider +ve and -ve
|
|
// zeros as equals. Thus, performing signed integral comparison for min value
|
|
// detection.
|
|
return (jint_cast(f0) < jint_cast(f1)) ? r0 : r1;
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
const Type* MinFNode::add_ring(const Type* t0, const Type* t1 ) const {
|
|
const TypeF* r0 = t0->isa_float_constant();
|
|
const TypeF* r1 = t1->isa_float_constant();
|
|
if (r0 == nullptr || r1 == nullptr) {
|
|
return bottom_type();
|
|
}
|
|
|
|
if (r0->is_nan()) {
|
|
return r0;
|
|
}
|
|
if (r1->is_nan()) {
|
|
return r1;
|
|
}
|
|
|
|
float f0 = r0->getf();
|
|
float f1 = r1->getf();
|
|
if (f0 != 0.0f || f1 != 0.0f) {
|
|
return f0 < f1 ? r0 : r1;
|
|
}
|
|
|
|
// handle min of 0.0, -0.0 case.
|
|
return (jint_cast(f0) < jint_cast(f1)) ? r0 : r1;
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
const Type* MinDNode::add_ring(const Type* t0, const Type* t1) const {
|
|
const TypeD* r0 = t0->isa_double_constant();
|
|
const TypeD* r1 = t1->isa_double_constant();
|
|
if (r0 == nullptr || r1 == nullptr) {
|
|
return bottom_type();
|
|
}
|
|
|
|
if (r0->is_nan()) {
|
|
return r0;
|
|
}
|
|
if (r1->is_nan()) {
|
|
return r1;
|
|
}
|
|
|
|
double d0 = r0->getd();
|
|
double d1 = r1->getd();
|
|
if (d0 != 0.0 || d1 != 0.0) {
|
|
return d0 < d1 ? r0 : r1;
|
|
}
|
|
|
|
// handle min of 0.0, -0.0 case.
|
|
return (jlong_cast(d0) < jlong_cast(d1)) ? r0 : r1;
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
const Type* MaxHFNode::add_ring(const Type* t0, const Type* t1) const {
|
|
const TypeH* r0 = t0->isa_half_float_constant();
|
|
const TypeH* r1 = t1->isa_half_float_constant();
|
|
if (r0 == nullptr || r1 == nullptr) {
|
|
return bottom_type();
|
|
}
|
|
|
|
if (r0->is_nan()) {
|
|
return r0;
|
|
}
|
|
if (r1->is_nan()) {
|
|
return r1;
|
|
}
|
|
|
|
float f0 = r0->getf();
|
|
float f1 = r1->getf();
|
|
if (f0 != 0.0f || f1 != 0.0f) {
|
|
return f0 > f1 ? r0 : r1;
|
|
}
|
|
|
|
// As per IEEE 754 specification, floating point comparison consider +ve and -ve
|
|
// zeros as equals. Thus, performing signed integral comparison for max value
|
|
// detection.
|
|
return (jint_cast(f0) > jint_cast(f1)) ? r0 : r1;
|
|
}
|
|
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
const Type* MaxFNode::add_ring(const Type* t0, const Type* t1) const {
|
|
const TypeF* r0 = t0->isa_float_constant();
|
|
const TypeF* r1 = t1->isa_float_constant();
|
|
if (r0 == nullptr || r1 == nullptr) {
|
|
return bottom_type();
|
|
}
|
|
|
|
if (r0->is_nan()) {
|
|
return r0;
|
|
}
|
|
if (r1->is_nan()) {
|
|
return r1;
|
|
}
|
|
|
|
float f0 = r0->getf();
|
|
float f1 = r1->getf();
|
|
if (f0 != 0.0f || f1 != 0.0f) {
|
|
return f0 > f1 ? r0 : r1;
|
|
}
|
|
|
|
// handle max of 0.0,-0.0 case.
|
|
return (jint_cast(f0) > jint_cast(f1)) ? r0 : r1;
|
|
}
|
|
|
|
//------------------------------add_ring---------------------------------------
|
|
const Type* MaxDNode::add_ring(const Type* t0, const Type* t1) const {
|
|
const TypeD* r0 = t0->isa_double_constant();
|
|
const TypeD* r1 = t1->isa_double_constant();
|
|
if (r0 == nullptr || r1 == nullptr) {
|
|
return bottom_type();
|
|
}
|
|
|
|
if (r0->is_nan()) {
|
|
return r0;
|
|
}
|
|
if (r1->is_nan()) {
|
|
return r1;
|
|
}
|
|
|
|
double d0 = r0->getd();
|
|
double d1 = r1->getd();
|
|
if (d0 != 0.0 || d1 != 0.0) {
|
|
return d0 > d1 ? r0 : r1;
|
|
}
|
|
|
|
// handle max of 0.0, -0.0 case.
|
|
return (jlong_cast(d0) > jlong_cast(d1)) ? r0 : r1;
|
|
}
|