mirror of
https://github.com/openjdk/jdk.git
synced 2026-02-27 10:40:29 +00:00
5a08374494
8377389: C2: Missed Ideal optimization opportunity in PhaseIterGVN for URShiftI and LShiftI Reviewed-by: qamai, chagedorn
2166 lines
81 KiB
C++
2166 lines
81 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/connode.hpp"
|
|
#include "opto/convertnode.hpp"
|
|
#include "opto/memnode.hpp"
|
|
#include "opto/mulnode.hpp"
|
|
#include "opto/phaseX.hpp"
|
|
#include "opto/rangeinference.hpp"
|
|
#include "opto/subnode.hpp"
|
|
#include "utilities/powerOfTwo.hpp"
|
|
|
|
// Portions of code courtesy of Clifford Click
|
|
|
|
|
|
//=============================================================================
|
|
//------------------------------hash-------------------------------------------
|
|
// Hash function over MulNodes. Needs to be commutative; i.e., I swap
|
|
// (commute) inputs to MulNodes willy-nilly so the hash function must return
|
|
// the same value in the presence of edge swapping.
|
|
uint MulNode::hash() const {
|
|
return (uintptr_t)in(1) + (uintptr_t)in(2) + Opcode();
|
|
}
|
|
|
|
//------------------------------Identity---------------------------------------
|
|
// Multiplying a one preserves the other argument
|
|
Node* MulNode::Identity(PhaseGVN* phase) {
|
|
const Type *one = mul_id(); // The multiplicative identity
|
|
if( phase->type( in(1) )->higher_equal( one ) ) return in(2);
|
|
if( phase->type( in(2) )->higher_equal( one ) ) return in(1);
|
|
|
|
return this;
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
// We also canonicalize the Node, moving constants to the right input,
|
|
// and flatten expressions (so that 1+x+2 becomes x+3).
|
|
Node *MulNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
Node* in1 = in(1);
|
|
Node* in2 = in(2);
|
|
Node* progress = nullptr; // Progress flag
|
|
|
|
// This code is used by And nodes too, but some conversions are
|
|
// only valid for the actual Mul nodes.
|
|
uint op = Opcode();
|
|
bool real_mul = (op == Op_MulI) || (op == Op_MulL) ||
|
|
(op == Op_MulF) || (op == Op_MulD) ||
|
|
(op == Op_MulHF);
|
|
|
|
// Convert "(-a)*(-b)" into "a*b".
|
|
if (real_mul && in1->is_Sub() && in2->is_Sub()) {
|
|
if (phase->type(in1->in(1))->is_zero_type() &&
|
|
phase->type(in2->in(1))->is_zero_type()) {
|
|
set_req_X(1, in1->in(2), phase);
|
|
set_req_X(2, in2->in(2), phase);
|
|
in1 = in(1);
|
|
in2 = in(2);
|
|
progress = this;
|
|
}
|
|
}
|
|
|
|
// convert "max(a,b) * min(a,b)" into "a*b".
|
|
if ((in(1)->Opcode() == max_opcode() && in(2)->Opcode() == min_opcode())
|
|
|| (in(1)->Opcode() == min_opcode() && in(2)->Opcode() == max_opcode())) {
|
|
Node *in11 = in(1)->in(1);
|
|
Node *in12 = in(1)->in(2);
|
|
|
|
Node *in21 = in(2)->in(1);
|
|
Node *in22 = in(2)->in(2);
|
|
|
|
if ((in11 == in21 && in12 == in22) ||
|
|
(in11 == in22 && in12 == in21)) {
|
|
set_req_X(1, in11, phase);
|
|
set_req_X(2, in12, phase);
|
|
in1 = in(1);
|
|
in2 = in(2);
|
|
progress = this;
|
|
}
|
|
}
|
|
|
|
const Type* t1 = phase->type(in1);
|
|
const Type* t2 = phase->type(in2);
|
|
|
|
// We are OK if right is a constant, or right is a load and
|
|
// left is a non-constant.
|
|
if( !(t2->singleton() ||
|
|
(in(2)->is_Load() && !(t1->singleton() || in(1)->is_Load())) ) ) {
|
|
if( t1->singleton() || // Left input is a constant?
|
|
// Otherwise, sort inputs (commutativity) to help value numbering.
|
|
(in(1)->_idx > in(2)->_idx) ) {
|
|
swap_edges(1, 2);
|
|
const Type *t = t1;
|
|
t1 = t2;
|
|
t2 = t;
|
|
progress = this; // Made progress
|
|
}
|
|
}
|
|
|
|
// If the right input is a constant, and the left input is a product of a
|
|
// constant, flatten the expression tree.
|
|
if( t2->singleton() && // Right input is a constant?
|
|
op != Op_MulF && // Float & double cannot reassociate
|
|
op != Op_MulD &&
|
|
op != Op_MulHF) {
|
|
if( t2 == Type::TOP ) return nullptr;
|
|
Node *mul1 = in(1);
|
|
#ifdef ASSERT
|
|
// Check for dead loop
|
|
int op1 = mul1->Opcode();
|
|
if ((mul1 == this) || (in(2) == this) ||
|
|
((op1 == mul_opcode() || op1 == add_opcode()) &&
|
|
((mul1->in(1) == this) || (mul1->in(2) == this) ||
|
|
(mul1->in(1) == mul1) || (mul1->in(2) == mul1)))) {
|
|
assert(false, "dead loop in MulNode::Ideal");
|
|
}
|
|
#endif
|
|
|
|
if( mul1->Opcode() == mul_opcode() ) { // Left input is a multiply?
|
|
// Mul of a constant?
|
|
const Type *t12 = phase->type( mul1->in(2) );
|
|
if( t12->singleton() && t12 != Type::TOP) { // Left input is an add of a constant?
|
|
// Compute new constant; check for overflow
|
|
const Type *tcon01 = ((MulNode*)mul1)->mul_ring(t2,t12);
|
|
if( tcon01->singleton() ) {
|
|
// The Mul of the flattened expression
|
|
set_req_X(1, mul1->in(1), phase);
|
|
set_req_X(2, phase->makecon(tcon01), phase);
|
|
t2 = tcon01;
|
|
progress = this; // Made progress
|
|
}
|
|
}
|
|
}
|
|
// If the right input is a constant, and the left input is an add of a
|
|
// constant, flatten the tree: (X+con1)*con0 ==> X*con0 + con1*con0
|
|
const Node *add1 = in(1);
|
|
if( add1->Opcode() == add_opcode() ) { // Left input is an add?
|
|
// Add of a constant?
|
|
const Type *t12 = phase->type( add1->in(2) );
|
|
if( t12->singleton() && t12 != Type::TOP ) { // Left input is an add of a constant?
|
|
assert( add1->in(1) != add1, "dead loop in MulNode::Ideal" );
|
|
// Compute new constant; check for overflow
|
|
const Type *tcon01 = mul_ring(t2,t12);
|
|
if( tcon01->singleton() ) {
|
|
|
|
// Convert (X+con1)*con0 into X*con0
|
|
Node *mul = clone(); // mul = ()*con0
|
|
mul->set_req(1,add1->in(1)); // mul = X*con0
|
|
mul = phase->transform(mul);
|
|
|
|
Node *add2 = add1->clone();
|
|
add2->set_req(1, mul); // X*con0 + con0*con1
|
|
add2->set_req(2, phase->makecon(tcon01) );
|
|
progress = add2;
|
|
}
|
|
}
|
|
} // End of is left input an add
|
|
} // End of is right input a Mul
|
|
|
|
return progress;
|
|
}
|
|
|
|
//------------------------------Value-----------------------------------------
|
|
const Type* MulNode::Value(PhaseGVN* phase) const {
|
|
const Type *t1 = phase->type( in(1) );
|
|
const Type *t2 = phase->type( in(2) );
|
|
// Either input is TOP ==> the result is TOP
|
|
if( t1 == Type::TOP ) return Type::TOP;
|
|
if( t2 == Type::TOP ) return Type::TOP;
|
|
|
|
// Either input is ZERO ==> the result is ZERO.
|
|
// Not valid for floats or doubles since +0.0 * -0.0 --> +0.0
|
|
int op = Opcode();
|
|
if( op == Op_MulI || op == Op_AndI || op == Op_MulL || op == Op_AndL ) {
|
|
const Type *zero = add_id(); // The multiplicative zero
|
|
if( t1->higher_equal( zero ) ) return zero;
|
|
if( t2->higher_equal( zero ) ) return zero;
|
|
}
|
|
|
|
// Either input is BOTTOM ==> the result is the local BOTTOM
|
|
if( t1 == Type::BOTTOM || t2 == Type::BOTTOM )
|
|
return bottom_type();
|
|
|
|
return mul_ring(t1,t2); // Local flavor of type multiplication
|
|
}
|
|
|
|
MulNode* MulNode::make(Node* in1, Node* in2, BasicType bt) {
|
|
switch (bt) {
|
|
case T_INT:
|
|
return new MulINode(in1, in2);
|
|
case T_LONG:
|
|
return new MulLNode(in1, in2);
|
|
default:
|
|
fatal("Not implemented for %s", type2name(bt));
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
MulNode* MulNode::make_and(Node* in1, Node* in2, BasicType bt) {
|
|
switch (bt) {
|
|
case T_INT:
|
|
return new AndINode(in1, in2);
|
|
case T_LONG:
|
|
return new AndLNode(in1, in2);
|
|
default:
|
|
fatal("Not implemented for %s", type2name(bt));
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
|
|
//=============================================================================
|
|
//------------------------------Ideal------------------------------------------
|
|
// Check for power-of-2 multiply, then try the regular MulNode::Ideal
|
|
Node *MulINode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
const jint con = in(2)->find_int_con(0);
|
|
if (con == 0) {
|
|
// If in(2) is not a constant, call Ideal() of the parent class to
|
|
// try to move constant to the right side.
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
// Now we have a constant Node on the right and the constant in con.
|
|
if (con == 1) {
|
|
// By one is handled by Identity call
|
|
return nullptr;
|
|
}
|
|
|
|
// Check for negative constant; if so negate the final result
|
|
bool sign_flip = false;
|
|
|
|
unsigned int abs_con = g_uabs(con);
|
|
if (abs_con != (unsigned int)con) {
|
|
sign_flip = true;
|
|
}
|
|
|
|
// Get low bit; check for being the only bit
|
|
Node *res = nullptr;
|
|
unsigned int bit1 = submultiple_power_of_2(abs_con);
|
|
if (bit1 == abs_con) { // Found a power of 2?
|
|
res = new LShiftINode(in(1), phase->intcon(log2i_exact(bit1)));
|
|
} else {
|
|
// Check for constant with 2 bits set
|
|
unsigned int bit2 = abs_con - bit1;
|
|
bit2 = bit2 & (0 - bit2); // Extract 2nd bit
|
|
if (bit2 + bit1 == abs_con) { // Found all bits in con?
|
|
Node *n1 = phase->transform(new LShiftINode(in(1), phase->intcon(log2i_exact(bit1))));
|
|
Node *n2 = phase->transform(new LShiftINode(in(1), phase->intcon(log2i_exact(bit2))));
|
|
res = new AddINode(n2, n1);
|
|
} else if (is_power_of_2(abs_con + 1)) {
|
|
// Sleezy: power-of-2 - 1. Next time be generic.
|
|
unsigned int temp = abs_con + 1;
|
|
Node *n1 = phase->transform(new LShiftINode(in(1), phase->intcon(log2i_exact(temp))));
|
|
res = new SubINode(n1, in(1));
|
|
} else {
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
}
|
|
|
|
if (sign_flip) { // Need to negate result?
|
|
res = phase->transform(res);// Transform, before making the zero con
|
|
res = new SubINode(phase->intcon(0),res);
|
|
}
|
|
|
|
return res; // Return final result
|
|
}
|
|
|
|
// This template class performs type multiplication for MulI/MulLNode. NativeType is either jint or jlong.
|
|
// In this class, the inputs of the MulNodes are named left and right with types [left_lo,left_hi] and [right_lo,right_hi].
|
|
//
|
|
// In general, the multiplication of two x-bit values could produce a result that consumes up to 2x bits if there is
|
|
// enough space to hold them all. We can therefore distinguish the following two cases for the product:
|
|
// - no overflow (i.e. product fits into x bits)
|
|
// - overflow (i.e. product does not fit into x bits)
|
|
//
|
|
// When multiplying the two x-bit inputs 'left' and 'right' with their x-bit types [left_lo,left_hi] and [right_lo,right_hi]
|
|
// we need to find the minimum and maximum of all possible products to define a new type. To do that, we compute the
|
|
// cross product of [left_lo,left_hi] and [right_lo,right_hi] in 2x-bit space where no over- or underflow can happen.
|
|
// The cross product consists of the following four multiplications with 2x-bit results:
|
|
// (1) left_lo * right_lo
|
|
// (2) left_lo * right_hi
|
|
// (3) left_hi * right_lo
|
|
// (4) left_hi * right_hi
|
|
//
|
|
// Let's define the following two functions:
|
|
// - Lx(i): Returns the lower x bits of the 2x-bit number i.
|
|
// - Ux(i): Returns the upper x bits of the 2x-bit number i.
|
|
//
|
|
// Let's first assume all products are positive where only overflows are possible but no underflows. If there is no
|
|
// overflow for a product p, then the upper x bits of the 2x-bit result p are all zero:
|
|
// Ux(p) = 0
|
|
// Lx(p) = p
|
|
//
|
|
// If none of the multiplications (1)-(4) overflow, we can truncate the upper x bits and use the following result type
|
|
// with x bits:
|
|
// [result_lo,result_hi] = [MIN(Lx(1),Lx(2),Lx(3),Lx(4)),MAX(Lx(1),Lx(2),Lx(3),Lx(4))]
|
|
//
|
|
// If any of these multiplications overflows, we could pessimistically take the bottom type for the x bit result
|
|
// (i.e. all values in the x-bit space could be possible):
|
|
// [result_lo,result_hi] = [NativeType_min,NativeType_max]
|
|
//
|
|
// However, in case of any overflow, we can do better by analyzing the upper x bits of all multiplications (1)-(4) with
|
|
// 2x-bit results. The upper x bits tell us something about how many times a multiplication has overflown the lower
|
|
// x bits. If the upper x bits of (1)-(4) are all equal, then we know that all of these multiplications overflowed
|
|
// the lower x bits the same number of times:
|
|
// Ux((1)) = Ux((2)) = Ux((3)) = Ux((4))
|
|
//
|
|
// If all upper x bits are equal, we can conclude:
|
|
// Lx(MIN((1),(2),(3),(4))) = MIN(Lx(1),Lx(2),Lx(3),Lx(4)))
|
|
// Lx(MAX((1),(2),(3),(4))) = MAX(Lx(1),Lx(2),Lx(3),Lx(4)))
|
|
//
|
|
// Therefore, we can use the same precise x-bit result type as for the no-overflow case:
|
|
// [result_lo,result_hi] = [(MIN(Lx(1),Lx(2),Lx(3),Lx(4))),MAX(Lx(1),Lx(2),Lx(3),Lx(4)))]
|
|
//
|
|
//
|
|
// Now let's assume that (1)-(4) are signed multiplications where over- and underflow could occur:
|
|
// Negative numbers are all sign extend with ones. Therefore, if a negative product does not underflow, then the
|
|
// upper x bits of the 2x-bit result are all set to ones which is minus one in two's complement. If there is an underflow,
|
|
// the upper x bits are decremented by the number of times an underflow occurred. The smallest possible negative product
|
|
// is NativeType_min*NativeType_max, where the upper x bits are set to NativeType_min / 2 (b11...0). It is therefore
|
|
// impossible to underflow the upper x bits. Thus, when having all ones (i.e. minus one) in the upper x bits, we know
|
|
// that there is no underflow.
|
|
//
|
|
// To be able to compare the number of over-/underflows of positive and negative products, respectively, we normalize
|
|
// the upper x bits of negative 2x-bit products by adding one. This way a product has no over- or underflow if the
|
|
// normalized upper x bits are zero. Now we can use the same improved type as for strictly positive products because we
|
|
// can compare the upper x bits in a unified way with N() being the normalization function:
|
|
// N(Ux((1))) = N(Ux((2))) = N(Ux((3)) = N(Ux((4)))
|
|
template<typename NativeType>
|
|
class IntegerTypeMultiplication {
|
|
|
|
NativeType _lo_left;
|
|
NativeType _lo_right;
|
|
NativeType _hi_left;
|
|
NativeType _hi_right;
|
|
short _widen_left;
|
|
short _widen_right;
|
|
|
|
static const Type* overflow_type();
|
|
static NativeType multiply_high(NativeType x, NativeType y);
|
|
const Type* create_type(NativeType lo, NativeType hi) const;
|
|
|
|
static NativeType multiply_high_signed_overflow_value(NativeType x, NativeType y) {
|
|
return normalize_overflow_value(x, y, multiply_high(x, y));
|
|
}
|
|
|
|
bool cross_product_not_same_overflow_value() const {
|
|
const NativeType lo_lo_high_product = multiply_high_signed_overflow_value(_lo_left, _lo_right);
|
|
const NativeType lo_hi_high_product = multiply_high_signed_overflow_value(_lo_left, _hi_right);
|
|
const NativeType hi_lo_high_product = multiply_high_signed_overflow_value(_hi_left, _lo_right);
|
|
const NativeType hi_hi_high_product = multiply_high_signed_overflow_value(_hi_left, _hi_right);
|
|
return lo_lo_high_product != lo_hi_high_product ||
|
|
lo_hi_high_product != hi_lo_high_product ||
|
|
hi_lo_high_product != hi_hi_high_product;
|
|
}
|
|
|
|
bool does_product_overflow(NativeType x, NativeType y) const {
|
|
return multiply_high_signed_overflow_value(x, y) != 0;
|
|
}
|
|
|
|
static NativeType normalize_overflow_value(const NativeType x, const NativeType y, NativeType result) {
|
|
return java_multiply(x, y) < 0 ? result + 1 : result;
|
|
}
|
|
|
|
public:
|
|
template<class IntegerType>
|
|
IntegerTypeMultiplication(const IntegerType* left, const IntegerType* right)
|
|
: _lo_left(left->_lo), _lo_right(right->_lo),
|
|
_hi_left(left->_hi), _hi_right(right->_hi),
|
|
_widen_left(left->_widen), _widen_right(right->_widen) {}
|
|
|
|
// Compute the product type by multiplying the two input type ranges. We take the minimum and maximum of all possible
|
|
// values (requires 4 multiplications of all possible combinations of the two range boundary values). If any of these
|
|
// multiplications overflows/underflows, we need to make sure that they all have the same number of overflows/underflows
|
|
// If that is not the case, we return the bottom type to cover all values due to the inconsistent overflows/underflows).
|
|
const Type* compute() const {
|
|
if (cross_product_not_same_overflow_value()) {
|
|
return overflow_type();
|
|
}
|
|
|
|
NativeType lo_lo_product = java_multiply(_lo_left, _lo_right);
|
|
NativeType lo_hi_product = java_multiply(_lo_left, _hi_right);
|
|
NativeType hi_lo_product = java_multiply(_hi_left, _lo_right);
|
|
NativeType hi_hi_product = java_multiply(_hi_left, _hi_right);
|
|
const NativeType min = MIN4(lo_lo_product, lo_hi_product, hi_lo_product, hi_hi_product);
|
|
const NativeType max = MAX4(lo_lo_product, lo_hi_product, hi_lo_product, hi_hi_product);
|
|
return create_type(min, max);
|
|
}
|
|
|
|
bool does_overflow() const {
|
|
return does_product_overflow(_lo_left, _lo_right) ||
|
|
does_product_overflow(_lo_left, _hi_right) ||
|
|
does_product_overflow(_hi_left, _lo_right) ||
|
|
does_product_overflow(_hi_left, _hi_right);
|
|
}
|
|
};
|
|
|
|
template <>
|
|
const Type* IntegerTypeMultiplication<jint>::overflow_type() {
|
|
return TypeInt::INT;
|
|
}
|
|
|
|
template <>
|
|
jint IntegerTypeMultiplication<jint>::multiply_high(const jint x, const jint y) {
|
|
const jlong x_64 = x;
|
|
const jlong y_64 = y;
|
|
const jlong product = x_64 * y_64;
|
|
return (jint)((uint64_t)product >> 32u);
|
|
}
|
|
|
|
template <>
|
|
const Type* IntegerTypeMultiplication<jint>::create_type(jint lo, jint hi) const {
|
|
return TypeInt::make(lo, hi, MAX2(_widen_left, _widen_right));
|
|
}
|
|
|
|
template <>
|
|
const Type* IntegerTypeMultiplication<jlong>::overflow_type() {
|
|
return TypeLong::LONG;
|
|
}
|
|
|
|
template <>
|
|
jlong IntegerTypeMultiplication<jlong>::multiply_high(const jlong x, const jlong y) {
|
|
return multiply_high_signed(x, y);
|
|
}
|
|
|
|
template <>
|
|
const Type* IntegerTypeMultiplication<jlong>::create_type(jlong lo, jlong hi) const {
|
|
return TypeLong::make(lo, hi, MAX2(_widen_left, _widen_right));
|
|
}
|
|
|
|
// Compute the product type of two integer ranges into this node.
|
|
const Type* MulINode::mul_ring(const Type* type_left, const Type* type_right) const {
|
|
const IntegerTypeMultiplication<jint> integer_multiplication(type_left->is_int(), type_right->is_int());
|
|
return integer_multiplication.compute();
|
|
}
|
|
|
|
bool MulINode::does_overflow(const TypeInt* type_left, const TypeInt* type_right) {
|
|
const IntegerTypeMultiplication<jint> integer_multiplication(type_left, type_right);
|
|
return integer_multiplication.does_overflow();
|
|
}
|
|
|
|
// Compute the product type of two long ranges into this node.
|
|
const Type* MulLNode::mul_ring(const Type* type_left, const Type* type_right) const {
|
|
const IntegerTypeMultiplication<jlong> integer_multiplication(type_left->is_long(), type_right->is_long());
|
|
return integer_multiplication.compute();
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Ideal------------------------------------------
|
|
// Check for power-of-2 multiply, then try the regular MulNode::Ideal
|
|
Node *MulLNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
const jlong con = in(2)->find_long_con(0);
|
|
if (con == 0) {
|
|
// If in(2) is not a constant, call Ideal() of the parent class to
|
|
// try to move constant to the right side.
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
// Now we have a constant Node on the right and the constant in con.
|
|
if (con == 1) {
|
|
// By one is handled by Identity call
|
|
return nullptr;
|
|
}
|
|
|
|
// Check for negative constant; if so negate the final result
|
|
bool sign_flip = false;
|
|
julong abs_con = g_uabs(con);
|
|
if (abs_con != (julong)con) {
|
|
sign_flip = true;
|
|
}
|
|
|
|
// Get low bit; check for being the only bit
|
|
Node *res = nullptr;
|
|
julong bit1 = submultiple_power_of_2(abs_con);
|
|
if (bit1 == abs_con) { // Found a power of 2?
|
|
res = new LShiftLNode(in(1), phase->intcon(log2i_exact(bit1)));
|
|
} else {
|
|
|
|
// Check for constant with 2 bits set
|
|
julong bit2 = abs_con-bit1;
|
|
bit2 = bit2 & (0-bit2); // Extract 2nd bit
|
|
if (bit2 + bit1 == abs_con) { // Found all bits in con?
|
|
Node *n1 = phase->transform(new LShiftLNode(in(1), phase->intcon(log2i_exact(bit1))));
|
|
Node *n2 = phase->transform(new LShiftLNode(in(1), phase->intcon(log2i_exact(bit2))));
|
|
res = new AddLNode(n2, n1);
|
|
|
|
} else if (is_power_of_2(abs_con+1)) {
|
|
// Sleezy: power-of-2 -1. Next time be generic.
|
|
julong temp = abs_con + 1;
|
|
Node *n1 = phase->transform( new LShiftLNode(in(1), phase->intcon(log2i_exact(temp))));
|
|
res = new SubLNode(n1, in(1));
|
|
} else {
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
}
|
|
|
|
if (sign_flip) { // Need to negate result?
|
|
res = phase->transform(res);// Transform, before making the zero con
|
|
res = new SubLNode(phase->longcon(0),res);
|
|
}
|
|
|
|
return res; // Return final result
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------mul_ring---------------------------------------
|
|
// Compute the product type of two double ranges into this node.
|
|
const Type *MulFNode::mul_ring(const Type *t0, const Type *t1) const {
|
|
if( t0 == Type::FLOAT || t1 == Type::FLOAT ) return Type::FLOAT;
|
|
return TypeF::make( t0->getf() * t1->getf() );
|
|
}
|
|
|
|
//------------------------------Ideal---------------------------------------
|
|
// Check to see if we are multiplying by a constant 2 and convert to add, then try the regular MulNode::Ideal
|
|
Node* MulFNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
const TypeF *t2 = phase->type(in(2))->isa_float_constant();
|
|
|
|
// x * 2 -> x + x
|
|
if (t2 != nullptr && t2->getf() == 2) {
|
|
Node* base = in(1);
|
|
return new AddFNode(base, base);
|
|
}
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Ideal------------------------------------------
|
|
// Check to see if we are multiplying by a constant 2 and convert to add, then try the regular MulNode::Ideal
|
|
Node* MulHFNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
const TypeH* t2 = phase->type(in(2))->isa_half_float_constant();
|
|
|
|
// x * 2 -> x + x
|
|
if (t2 != nullptr && t2->getf() == 2) {
|
|
Node* base = in(1);
|
|
return new AddHFNode(base, base);
|
|
}
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
// Compute the product type of two half float ranges into this node.
|
|
const Type* MulHFNode::mul_ring(const Type* t0, const Type* t1) const {
|
|
if (t0 == Type::HALF_FLOAT || t1 == Type::HALF_FLOAT) {
|
|
return Type::HALF_FLOAT;
|
|
}
|
|
return TypeH::make(t0->getf() * t1->getf());
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------mul_ring---------------------------------------
|
|
// Compute the product type of two double ranges into this node.
|
|
const Type *MulDNode::mul_ring(const Type *t0, const Type *t1) const {
|
|
if( t0 == Type::DOUBLE || t1 == Type::DOUBLE ) return Type::DOUBLE;
|
|
// We must be multiplying 2 double constants.
|
|
return TypeD::make( t0->getd() * t1->getd() );
|
|
}
|
|
|
|
//------------------------------Ideal---------------------------------------
|
|
// Check to see if we are multiplying by a constant 2 and convert to add, then try the regular MulNode::Ideal
|
|
Node* MulDNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
const TypeD *t2 = phase->type(in(2))->isa_double_constant();
|
|
|
|
// x * 2 -> x + x
|
|
if (t2 != nullptr && t2->getd() == 2) {
|
|
Node* base = in(1);
|
|
return new AddDNode(base, base);
|
|
}
|
|
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Value------------------------------------------
|
|
const Type* MulHiLNode::Value(PhaseGVN* phase) const {
|
|
const Type *t1 = phase->type( in(1) );
|
|
const Type *t2 = phase->type( in(2) );
|
|
const Type *bot = bottom_type();
|
|
return MulHiValue(t1, t2, bot);
|
|
}
|
|
|
|
const Type* UMulHiLNode::Value(PhaseGVN* phase) const {
|
|
const Type *t1 = phase->type( in(1) );
|
|
const Type *t2 = phase->type( in(2) );
|
|
const Type *bot = bottom_type();
|
|
return MulHiValue(t1, t2, bot);
|
|
}
|
|
|
|
// A common routine used by UMulHiLNode and MulHiLNode
|
|
const Type* MulHiValue(const Type *t1, const Type *t2, const Type *bot) {
|
|
// Either input is TOP ==> the result is TOP
|
|
if( t1 == Type::TOP ) return Type::TOP;
|
|
if( t2 == Type::TOP ) return Type::TOP;
|
|
|
|
// Either input is BOTTOM ==> the result is the local BOTTOM
|
|
if( (t1 == bot) || (t2 == bot) ||
|
|
(t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) )
|
|
return bot;
|
|
|
|
// It is not worth trying to constant fold this stuff!
|
|
return TypeLong::LONG;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------mul_ring---------------------------------------
|
|
// Supplied function returns the product of the inputs IN THE CURRENT RING.
|
|
// For the logical operations the ring's MUL is really a logical AND 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* AndINode::mul_ring(const Type* t1, const Type* t2) const {
|
|
return RangeInference::infer_and(t1->is_int(), t2->is_int());
|
|
}
|
|
|
|
static bool AndIL_is_zero_element_under_mask(const PhaseGVN* phase, const Node* expr, const Node* mask, BasicType bt);
|
|
|
|
const Type* AndINode::Value(PhaseGVN* phase) const {
|
|
if (AndIL_is_zero_element_under_mask(phase, in(1), in(2), T_INT) ||
|
|
AndIL_is_zero_element_under_mask(phase, in(2), in(1), T_INT)) {
|
|
return TypeInt::ZERO;
|
|
}
|
|
|
|
return MulNode::Value(phase);
|
|
}
|
|
|
|
//------------------------------Identity---------------------------------------
|
|
// Masking off the high bits of an unsigned load is not required
|
|
Node* AndINode::Identity(PhaseGVN* phase) {
|
|
|
|
// x & x => x
|
|
if (in(1) == in(2)) {
|
|
return in(1);
|
|
}
|
|
|
|
Node* in1 = in(1);
|
|
uint op = in1->Opcode();
|
|
const TypeInt* t2 = phase->type(in(2))->isa_int();
|
|
if (t2 && t2->is_con()) {
|
|
int con = t2->get_con();
|
|
// Masking off high bits which are always zero is useless.
|
|
const TypeInt* t1 = phase->type(in(1))->isa_int();
|
|
if (t1 != nullptr && t1->_lo >= 0) {
|
|
jint t1_support = right_n_bits(1 + log2i_graceful(t1->_hi));
|
|
if ((t1_support & con) == t1_support)
|
|
return in1;
|
|
}
|
|
// Masking off the high bits of a unsigned-shift-right is not
|
|
// needed either.
|
|
if (op == Op_URShiftI) {
|
|
const TypeInt* t12 = phase->type(in1->in(2))->isa_int();
|
|
if (t12 && t12->is_con()) { // Shift is by a constant
|
|
int shift = t12->get_con();
|
|
shift &= BitsPerJavaInteger - 1; // semantics of Java shifts
|
|
int mask = max_juint >> shift;
|
|
if ((mask & con) == mask) // If AND is useless, skip it
|
|
return in1;
|
|
}
|
|
}
|
|
}
|
|
return MulNode::Identity(phase);
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
Node *AndINode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
// Simplify (v1 + v2) & mask to v1 & mask or v2 & mask when possible.
|
|
Node* progress = AndIL_sum_and_mask(phase, T_INT);
|
|
if (progress != nullptr) {
|
|
return progress;
|
|
}
|
|
|
|
// Convert "(~a) & (~b)" into "~(a | b)"
|
|
if (AddNode::is_not(phase, in(1), T_INT) && AddNode::is_not(phase, in(2), T_INT)) {
|
|
Node* or_a_b = new OrINode(in(1)->in(1), in(2)->in(1));
|
|
Node* tn = phase->transform(or_a_b);
|
|
return AddNode::make_not(phase, tn, T_INT);
|
|
}
|
|
|
|
// Special case constant AND mask
|
|
const TypeInt *t2 = phase->type( in(2) )->isa_int();
|
|
if( !t2 || !t2->is_con() ) return MulNode::Ideal(phase, can_reshape);
|
|
const int mask = t2->get_con();
|
|
Node *load = in(1);
|
|
uint lop = load->Opcode();
|
|
|
|
// Masking bits off of a Character? Hi bits are already zero.
|
|
if( lop == Op_LoadUS &&
|
|
(mask & 0xFFFF0000) ) // Can we make a smaller mask?
|
|
return new AndINode(load,phase->intcon(mask&0xFFFF));
|
|
|
|
// Masking bits off of a Short? Loading a Character does some masking
|
|
if (can_reshape &&
|
|
load->outcnt() == 1 && load->unique_out() == this) {
|
|
if (lop == Op_LoadS && (mask & 0xFFFF0000) == 0 ) {
|
|
Node* ldus = load->as_Load()->convert_to_unsigned_load(*phase);
|
|
ldus = phase->transform(ldus);
|
|
return new AndINode(ldus, phase->intcon(mask & 0xFFFF));
|
|
}
|
|
|
|
// Masking sign bits off of a Byte? Do an unsigned byte load plus
|
|
// an and.
|
|
if (lop == Op_LoadB && (mask & 0xFFFFFF00) == 0) {
|
|
Node* ldub = load->as_Load()->convert_to_unsigned_load(*phase);
|
|
ldub = phase->transform(ldub);
|
|
return new AndINode(ldub, phase->intcon(mask));
|
|
}
|
|
}
|
|
|
|
// Masking off sign bits? Dont make them!
|
|
if( lop == Op_RShiftI ) {
|
|
const TypeInt *t12 = phase->type(load->in(2))->isa_int();
|
|
if( t12 && t12->is_con() ) { // Shift is by a constant
|
|
int shift = t12->get_con();
|
|
shift &= BitsPerJavaInteger-1; // semantics of Java shifts
|
|
const int sign_bits_mask = ~right_n_bits(BitsPerJavaInteger - shift);
|
|
// If the AND'ing of the 2 masks has no bits, then only original shifted
|
|
// bits survive. NO sign-extension bits survive the maskings.
|
|
if( (sign_bits_mask & mask) == 0 ) {
|
|
// Use zero-fill shift instead
|
|
Node *zshift = phase->transform(new URShiftINode(load->in(1),load->in(2)));
|
|
return new AndINode( zshift, in(2) );
|
|
}
|
|
}
|
|
}
|
|
|
|
// Check for 'negate/and-1', a pattern emitted when someone asks for
|
|
// 'mod 2'. Negate leaves the low order bit unchanged (think: complement
|
|
// plus 1) and the mask is of the low order bit. Skip the negate.
|
|
if( lop == Op_SubI && mask == 1 && load->in(1) &&
|
|
phase->type(load->in(1)) == TypeInt::ZERO )
|
|
return new AndINode( load->in(2), in(2) );
|
|
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------mul_ring---------------------------------------
|
|
// Supplied function returns the product of the inputs IN THE CURRENT RING.
|
|
// For the logical operations the ring's MUL is really a logical AND 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* AndLNode::mul_ring(const Type* t1, const Type* t2) const {
|
|
return RangeInference::infer_and(t1->is_long(), t2->is_long());
|
|
}
|
|
|
|
const Type* AndLNode::Value(PhaseGVN* phase) const {
|
|
if (AndIL_is_zero_element_under_mask(phase, in(1), in(2), T_LONG) ||
|
|
AndIL_is_zero_element_under_mask(phase, in(2), in(1), T_LONG)) {
|
|
return TypeLong::ZERO;
|
|
}
|
|
|
|
return MulNode::Value(phase);
|
|
}
|
|
|
|
//------------------------------Identity---------------------------------------
|
|
// Masking off the high bits of an unsigned load is not required
|
|
Node* AndLNode::Identity(PhaseGVN* phase) {
|
|
|
|
// x & x => x
|
|
if (in(1) == in(2)) {
|
|
return in(1);
|
|
}
|
|
|
|
Node *usr = in(1);
|
|
const TypeLong *t2 = phase->type( in(2) )->isa_long();
|
|
if( t2 && t2->is_con() ) {
|
|
jlong con = t2->get_con();
|
|
// Masking off high bits which are always zero is useless.
|
|
const TypeLong* t1 = phase->type( in(1) )->isa_long();
|
|
if (t1 != nullptr && t1->_lo >= 0) {
|
|
int bit_count = log2i_graceful(t1->_hi) + 1;
|
|
jlong t1_support = jlong(max_julong >> (BitsPerJavaLong - bit_count));
|
|
if ((t1_support & con) == t1_support)
|
|
return usr;
|
|
}
|
|
uint lop = usr->Opcode();
|
|
// Masking off the high bits of a unsigned-shift-right is not
|
|
// needed either.
|
|
if( lop == Op_URShiftL ) {
|
|
const TypeInt *t12 = phase->type( usr->in(2) )->isa_int();
|
|
if( t12 && t12->is_con() ) { // Shift is by a constant
|
|
int shift = t12->get_con();
|
|
shift &= BitsPerJavaLong - 1; // semantics of Java shifts
|
|
jlong mask = max_julong >> shift;
|
|
if( (mask&con) == mask ) // If AND is useless, skip it
|
|
return usr;
|
|
}
|
|
}
|
|
}
|
|
return MulNode::Identity(phase);
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
Node *AndLNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
// Simplify (v1 + v2) & mask to v1 & mask or v2 & mask when possible.
|
|
Node* progress = AndIL_sum_and_mask(phase, T_LONG);
|
|
if (progress != nullptr) {
|
|
return progress;
|
|
}
|
|
|
|
// Convert "(~a) & (~b)" into "~(a | b)"
|
|
if (AddNode::is_not(phase, in(1), T_LONG) && AddNode::is_not(phase, in(2), T_LONG)) {
|
|
Node* or_a_b = new OrLNode(in(1)->in(1), in(2)->in(1));
|
|
Node* tn = phase->transform(or_a_b);
|
|
return AddNode::make_not(phase, tn, T_LONG);
|
|
}
|
|
|
|
// Special case constant AND mask
|
|
const TypeLong *t2 = phase->type( in(2) )->isa_long();
|
|
if( !t2 || !t2->is_con() ) return MulNode::Ideal(phase, can_reshape);
|
|
const jlong mask = t2->get_con();
|
|
|
|
Node* in1 = in(1);
|
|
int op = in1->Opcode();
|
|
|
|
// Are we masking a long that was converted from an int with a mask
|
|
// that fits in 32-bits? Commute them and use an AndINode. Don't
|
|
// convert masks which would cause a sign extension of the integer
|
|
// value. This check includes UI2L masks (0x00000000FFFFFFFF) which
|
|
// would be optimized away later in Identity.
|
|
if (op == Op_ConvI2L && (mask & UCONST64(0xFFFFFFFF80000000)) == 0) {
|
|
Node* andi = new AndINode(in1->in(1), phase->intcon(mask));
|
|
andi = phase->transform(andi);
|
|
return new ConvI2LNode(andi);
|
|
}
|
|
|
|
// Masking off sign bits? Dont make them!
|
|
if (op == Op_RShiftL) {
|
|
const TypeInt* t12 = phase->type(in1->in(2))->isa_int();
|
|
if( t12 && t12->is_con() ) { // Shift is by a constant
|
|
int shift = t12->get_con();
|
|
shift &= BitsPerJavaLong - 1; // semantics of Java shifts
|
|
if (shift != 0) {
|
|
const julong sign_bits_mask = ~(((julong)CONST64(1) << (julong)(BitsPerJavaLong - shift)) -1);
|
|
// If the AND'ing of the 2 masks has no bits, then only original shifted
|
|
// bits survive. NO sign-extension bits survive the maskings.
|
|
if( (sign_bits_mask & mask) == 0 ) {
|
|
// Use zero-fill shift instead
|
|
Node *zshift = phase->transform(new URShiftLNode(in1->in(1), in1->in(2)));
|
|
return new AndLNode(zshift, in(2));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return MulNode::Ideal(phase, can_reshape);
|
|
}
|
|
|
|
LShiftNode* LShiftNode::make(Node* in1, Node* in2, BasicType bt) {
|
|
switch (bt) {
|
|
case T_INT:
|
|
return new LShiftINode(in1, in2);
|
|
case T_LONG:
|
|
return new LShiftLNode(in1, in2);
|
|
default:
|
|
fatal("Not implemented for %s", type2name(bt));
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
// Returns whether the shift amount is constant. If so, sets count.
|
|
static bool const_shift_count(PhaseGVN* phase, const Node* shift_node, int* count) {
|
|
const TypeInt* tcount = phase->type(shift_node->in(2))->isa_int();
|
|
if (tcount != nullptr && tcount->is_con()) {
|
|
*count = tcount->get_con();
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
// Returns whether the shift amount is constant. If so, sets real_shift and masked_shift.
|
|
static bool mask_shift_amount(PhaseGVN* phase, const Node* shift_node, uint nBits, int& real_shift, uint& masked_shift) {
|
|
if (const_shift_count(phase, shift_node, &real_shift)) {
|
|
masked_shift = real_shift & (nBits - 1);
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
// Convenience for when we don't care about the real amount
|
|
static bool mask_shift_amount(PhaseGVN* phase, const Node* shift_node, uint nBits, uint& masked_shift) {
|
|
int real_shift;
|
|
return mask_shift_amount(phase, shift_node, nBits, real_shift, masked_shift);
|
|
}
|
|
|
|
// Use this in ::Ideal only with shiftNode == this!
|
|
// Returns the masked shift amount if constant or 0 if not constant.
|
|
static uint mask_and_replace_shift_amount(PhaseGVN* phase, Node* shift_node, uint nBits) {
|
|
int real_shift;
|
|
uint masked_shift;
|
|
if (mask_shift_amount(phase, shift_node, nBits, real_shift, masked_shift)) {
|
|
if (masked_shift == 0) {
|
|
// Let Identity() handle 0 shift count.
|
|
return 0;
|
|
}
|
|
|
|
if (real_shift != (int)masked_shift) {
|
|
PhaseIterGVN* igvn = phase->is_IterGVN();
|
|
if (igvn != nullptr) {
|
|
igvn->_worklist.push(shift_node);
|
|
}
|
|
shift_node->set_req(2, phase->intcon(masked_shift)); // Replace shift count with masked value.
|
|
}
|
|
return masked_shift;
|
|
}
|
|
// Not a shift by a constant.
|
|
return 0;
|
|
}
|
|
|
|
// Called with
|
|
// outer_shift = (_ << rhs_outer)
|
|
// We are looking for the pattern:
|
|
// outer_shift = ((X << rhs_inner) << rhs_outer)
|
|
// where rhs_outer and rhs_inner are constant
|
|
// we denote inner_shift the nested expression (X << rhs_inner)
|
|
// con_inner = rhs_inner % nbits and con_outer = rhs_outer % nbits
|
|
// where nbits is the number of bits of the shifts
|
|
//
|
|
// There are 2 cases:
|
|
// if con_outer + con_inner >= nbits => 0
|
|
// if con_outer + con_inner < nbits => X << (con_outer + con_inner)
|
|
static Node* collapse_nested_shift_left(PhaseGVN* phase, const Node* outer_shift, uint con_outer, BasicType bt) {
|
|
assert(bt == T_LONG || bt == T_INT, "Unexpected type");
|
|
const Node* inner_shift = outer_shift->in(1);
|
|
if (inner_shift->Opcode() != Op_LShift(bt)) {
|
|
return nullptr;
|
|
}
|
|
|
|
uint nbits = bits_per_java_integer(bt);
|
|
uint con_inner;
|
|
if (!mask_shift_amount(phase, inner_shift, nbits, con_inner)) {
|
|
return nullptr;
|
|
}
|
|
|
|
if (con_inner == 0) {
|
|
// We let the Identity() of the inner shift do its job.
|
|
return nullptr;
|
|
}
|
|
|
|
if (con_outer + con_inner >= nbits) {
|
|
// While it might be tempting to use
|
|
// phase->zerocon(bt);
|
|
// it would be incorrect: zerocon caches nodes, while Ideal is only allowed
|
|
// to return a new node, this or nullptr, but not an old (cached) node.
|
|
return ConNode::make(TypeInteger::zero(bt));
|
|
}
|
|
|
|
// con0 + con1 < nbits ==> actual shift happens now
|
|
Node* con0_plus_con1 = phase->intcon(con_outer + con_inner);
|
|
return LShiftNode::make(inner_shift->in(1), con0_plus_con1, bt);
|
|
}
|
|
|
|
//------------------------------Identity---------------------------------------
|
|
Node* LShiftINode::Identity(PhaseGVN* phase) {
|
|
return IdentityIL(phase, T_INT);
|
|
}
|
|
|
|
Node* LShiftNode::IdealIL(PhaseGVN* phase, bool can_reshape, BasicType bt) {
|
|
uint con = mask_and_replace_shift_amount(phase, this, bits_per_java_integer(bt));
|
|
if (con == 0) {
|
|
return nullptr;
|
|
}
|
|
|
|
// If the right input is a constant, and the left input is an add of a
|
|
// constant, flatten the tree: (X+con1)<<con0 ==> X<<con0 + con1<<con0
|
|
Node* add1 = in(1);
|
|
int add1_op = add1->Opcode();
|
|
if (add1_op == Op_Add(bt)) { // Left input is an add?
|
|
assert(add1 != add1->in(1), "dead loop in LShiftINode::Ideal");
|
|
|
|
// Transform is legal, but check for profit. Avoid breaking 'i2s'
|
|
// and 'i2b' patterns which typically fold into 'StoreC/StoreB'.
|
|
if (bt != T_INT || con < 16) {
|
|
// Left input is an add of the same number?
|
|
if (con != (bits_per_java_integer(bt) - 1) && add1->in(1) == add1->in(2)) {
|
|
// Convert "(x + x) << c0" into "x << (c0 + 1)"
|
|
// In general, this optimization cannot be applied for c0 == 31 (for LShiftI) since
|
|
// 2x << 31 != x << 32 = x << 0 = x (e.g. x = 1: 2 << 31 = 0 != 1)
|
|
// or c0 != 63 (for LShiftL) because:
|
|
// (x + x) << 63 = 2x << 63, while
|
|
// (x + x) << 63 --transform--> x << 64 = x << 0 = x (!= 2x << 63, for example for x = 1)
|
|
// According to the Java spec, chapter 15.19, we only consider the six lowest-order bits of the right-hand operand
|
|
// (i.e. "right-hand operand" & 0b111111). Therefore, x << 64 is the same as x << 0 (64 = 0b10000000 & 0b0111111 = 0).
|
|
return LShiftNode::make(add1->in(1), phase->intcon(con + 1), bt);
|
|
}
|
|
|
|
// Left input is an add of a constant?
|
|
const TypeInteger* t12 = phase->type(add1->in(2))->isa_integer(bt);
|
|
if (t12 != nullptr && t12->is_con()) { // Left input is an add of a con?
|
|
// Compute X << con0
|
|
Node* lsh = phase->transform(LShiftNode::make(add1->in(1), in(2), bt));
|
|
// Compute X<<con0 + (con1<<con0)
|
|
return AddNode::make(lsh, phase->integercon(java_shift_left(t12->get_con_as_long(bt), con, bt), bt), bt);
|
|
}
|
|
}
|
|
}
|
|
// Check for "(con0 - X) << con1"
|
|
// Transform is legal, but check for profit. Avoid breaking 'i2s'
|
|
// and 'i2b' patterns which typically fold into 'StoreC/StoreB'.
|
|
if (add1_op == Op_Sub(bt) && (bt != T_INT || con < 16)) { // Left input is a sub?
|
|
// Left input is a sub from a constant?
|
|
const TypeInteger* t11 = phase->type(add1->in(1))->isa_integer(bt);
|
|
if (t11 != nullptr && t11->is_con()) {
|
|
// Compute X << con0
|
|
Node* lsh = phase->transform(LShiftNode::make(add1->in(2), in(2), bt));
|
|
// Compute (con1<<con0) - (X<<con0)
|
|
return SubNode::make(phase->integercon(java_shift_left(t11->get_con_as_long(bt), con, bt), bt), lsh, bt);
|
|
}
|
|
}
|
|
|
|
// Check for "(x >> C1) << C2"
|
|
if (add1_op == Op_RShift(bt) || add1_op == Op_URShift(bt)) {
|
|
int add1Con = 0;
|
|
const_shift_count(phase, add1, &add1Con);
|
|
|
|
// Special case C1 == C2, which just masks off low bits
|
|
if (add1Con > 0 && con == (uint)add1Con) {
|
|
// Convert to "(x & -(1 << C2))"
|
|
return MulNode::make_and(add1->in(1), phase->integercon(java_negate(java_shift_left(1, con, bt), bt), bt), bt);
|
|
} else {
|
|
// Wait until the right shift has been sharpened to the correct count
|
|
if (add1Con > 0 && (uint)add1Con < bits_per_java_integer(bt)) {
|
|
// As loop parsing can produce LShiftI nodes, we should wait until the graph is fully formed
|
|
// to apply optimizations, otherwise we can inadvertently stop vectorization opportunities.
|
|
if (phase->is_IterGVN()) {
|
|
if (con > (uint)add1Con) {
|
|
// Creates "(x << (C2 - C1)) & -(1 << C2)"
|
|
Node* lshift = phase->transform(LShiftNode::make(add1->in(1), phase->intcon(con - add1Con), bt));
|
|
return MulNode::make_and(lshift, phase->integercon(java_negate(java_shift_left(1, con, bt), bt), bt), bt);
|
|
} else {
|
|
assert(con < (uint)add1Con, "must be (%d < %d)", con, add1Con);
|
|
// Creates "(x >> (C1 - C2)) & -(1 << C2)"
|
|
|
|
// Handle logical and arithmetic shifts
|
|
Node* rshift;
|
|
if (add1_op == Op_RShift(bt)) {
|
|
rshift = phase->transform(RShiftNode::make(add1->in(1), phase->intcon(add1Con - con), bt));
|
|
} else {
|
|
rshift = phase->transform(URShiftNode::make(add1->in(1), phase->intcon(add1Con - con), bt));
|
|
}
|
|
|
|
return MulNode::make_and(rshift, phase->integercon(java_negate(java_shift_left(1, con, bt)), bt), bt);
|
|
}
|
|
} else {
|
|
phase->record_for_igvn(this);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Check for "((x >> C1) & Y) << C2"
|
|
if (add1_op == Op_And(bt)) {
|
|
Node* add2 = add1->in(1);
|
|
int add2_op = add2->Opcode();
|
|
if (add2_op == Op_RShift(bt) || add2_op == Op_URShift(bt)) {
|
|
// Special case C1 == C2, which just masks off low bits
|
|
if (add2->in(2) == in(2)) {
|
|
// Convert to "(x & (Y << C2))"
|
|
Node* y_sh = phase->transform(LShiftNode::make(add1->in(2), phase->intcon(con), bt));
|
|
return MulNode::make_and(add2->in(1), y_sh, bt);
|
|
}
|
|
|
|
int add2Con = 0;
|
|
const_shift_count(phase, add2, &add2Con);
|
|
if (add2Con > 0 && (uint)add2Con < bits_per_java_integer(bt)) {
|
|
if (phase->is_IterGVN()) {
|
|
// Convert to "((x >> C1) << C2) & (Y << C2)"
|
|
|
|
// Make "(x >> C1) << C2", which will get folded away by the rule above
|
|
Node* x_sh = phase->transform(LShiftNode::make(add2, phase->intcon(con), bt));
|
|
// Make "Y << C2", which will simplify when Y is a constant
|
|
Node* y_sh = phase->transform(LShiftNode::make(add1->in(2), phase->intcon(con), bt));
|
|
|
|
return MulNode::make_and(x_sh, y_sh, bt);
|
|
} else {
|
|
phase->record_for_igvn(this);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// Check for ((x & ((1<<(32-c0))-1)) << c0) which ANDs off high bits
|
|
// before shifting them away.
|
|
const jlong bits_mask = max_unsigned_integer(bt) >> con;
|
|
assert(bt != T_INT || bits_mask == right_n_bits(bits_per_java_integer(bt)-con), "inconsistent");
|
|
if (add1_op == Op_And(bt) &&
|
|
phase->type(add1->in(2)) == TypeInteger::make(bits_mask, bt)) {
|
|
return LShiftNode::make(add1->in(1), in(2), bt);
|
|
}
|
|
|
|
// Collapse nested left-shifts with constant rhs:
|
|
// (X << con1) << con2 ==> X << (con1 + con2)
|
|
Node* doubleShift = collapse_nested_shift_left(phase, this, con, bt);
|
|
if (doubleShift != nullptr) {
|
|
return doubleShift;
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
Node* LShiftINode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
return IdealIL(phase, can_reshape, T_INT);
|
|
}
|
|
|
|
const Type* LShiftNode::ValueIL(PhaseGVN* phase, BasicType bt) const {
|
|
const Type* t1 = phase->type(in(1));
|
|
const Type* t2 = phase->type(in(2));
|
|
// Either input is TOP ==> the result is TOP
|
|
if (t1 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
if (t2 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
|
|
// Left input is ZERO ==> the result is ZERO.
|
|
if (t1 == TypeInteger::zero(bt)) {
|
|
return TypeInteger::zero(bt);
|
|
}
|
|
// Shift by zero does nothing
|
|
if (t2 == TypeInt::ZERO) {
|
|
return t1;
|
|
}
|
|
|
|
// Either input is BOTTOM ==> the result is BOTTOM
|
|
if ((t1 == TypeInteger::bottom(bt)) || (t2 == TypeInt::INT) ||
|
|
(t1 == Type::BOTTOM) || (t2 == Type::BOTTOM)) {
|
|
return TypeInteger::bottom(bt);
|
|
}
|
|
|
|
const TypeInteger* r1 = t1->is_integer(bt); // Handy access
|
|
const TypeInt* r2 = t2->is_int(); // Handy access
|
|
|
|
if (!r2->is_con()) {
|
|
return TypeInteger::bottom(bt);
|
|
}
|
|
|
|
uint shift = r2->get_con();
|
|
shift &= bits_per_java_integer(bt) - 1; // semantics of Java shifts
|
|
// Shift by a multiple of 32/64 does nothing:
|
|
if (shift == 0) {
|
|
return t1;
|
|
}
|
|
|
|
// If the shift is a constant, shift the bounds of the type,
|
|
// unless this could lead to an overflow.
|
|
if (!r1->is_con()) {
|
|
jlong lo = r1->lo_as_long(), hi = r1->hi_as_long();
|
|
#ifdef ASSERT
|
|
if (bt == T_INT) {
|
|
jint lo_int = r1->is_int()->_lo, hi_int = r1->is_int()->_hi;
|
|
assert((java_shift_right(java_shift_left(lo, shift, bt), shift, bt) == lo) == (((lo_int << shift) >> shift) == lo_int), "inconsistent");
|
|
assert((java_shift_right(java_shift_left(hi, shift, bt), shift, bt) == hi) == (((hi_int << shift) >> shift) == hi_int), "inconsistent");
|
|
}
|
|
#endif
|
|
if (java_shift_right(java_shift_left(lo, shift, bt), shift, bt) == lo &&
|
|
java_shift_right(java_shift_left(hi, shift, bt), shift, bt) == hi) {
|
|
// No overflow. The range shifts up cleanly.
|
|
return TypeInteger::make(java_shift_left(lo, shift, bt),
|
|
java_shift_left(hi, shift, bt),
|
|
MAX2(r1->_widen, r2->_widen), bt);
|
|
}
|
|
return TypeInteger::bottom(bt);
|
|
}
|
|
|
|
return TypeInteger::make(java_shift_left(r1->get_con_as_long(bt), shift, bt), bt);
|
|
}
|
|
|
|
//------------------------------Value------------------------------------------
|
|
const Type* LShiftINode::Value(PhaseGVN* phase) const {
|
|
return ValueIL(phase, T_INT);
|
|
}
|
|
|
|
Node* LShiftNode::IdentityIL(PhaseGVN* phase, BasicType bt) {
|
|
int count = 0;
|
|
if (const_shift_count(phase, this, &count) && (count & (bits_per_java_integer(bt) - 1)) == 0) {
|
|
// Shift by a multiple of 32/64 does nothing
|
|
return in(1);
|
|
}
|
|
return this;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Identity---------------------------------------
|
|
Node* LShiftLNode::Identity(PhaseGVN* phase) {
|
|
return IdentityIL(phase, T_LONG);
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
Node* LShiftLNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
return IdealIL(phase, can_reshape, T_LONG);
|
|
}
|
|
|
|
//------------------------------Value------------------------------------------
|
|
const Type* LShiftLNode::Value(PhaseGVN* phase) const {
|
|
return ValueIL(phase, T_LONG);
|
|
}
|
|
|
|
RShiftNode* RShiftNode::make(Node* in1, Node* in2, BasicType bt) {
|
|
switch (bt) {
|
|
case T_INT:
|
|
return new RShiftINode(in1, in2);
|
|
case T_LONG:
|
|
return new RShiftLNode(in1, in2);
|
|
default:
|
|
fatal("Not implemented for %s", type2name(bt));
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
|
|
//=============================================================================
|
|
//------------------------------Identity---------------------------------------
|
|
Node* RShiftNode::IdentityIL(PhaseGVN* phase, BasicType bt) {
|
|
int count = 0;
|
|
if (const_shift_count(phase, this, &count)) {
|
|
if ((count & (bits_per_java_integer(bt) - 1)) == 0) {
|
|
// Shift by a multiple of 32/64 does nothing
|
|
return in(1);
|
|
}
|
|
// Check for useless sign-masking
|
|
int lshift_count = 0;
|
|
if (in(1)->Opcode() == Op_LShift(bt) &&
|
|
in(1)->req() == 3 &&
|
|
// Compare shift counts by value, not by node pointer, to also match a not-yet-normalized
|
|
// negative constant (e.g. -1 vs 31)
|
|
const_shift_count(phase, in(1), &lshift_count)) {
|
|
count &= bits_per_java_integer(bt) - 1; // semantics of Java shifts
|
|
lshift_count &= bits_per_java_integer(bt) - 1;
|
|
if (count == lshift_count) {
|
|
// Compute masks for which this shifting doesn't change
|
|
jlong lo = (CONST64(-1) << (bits_per_java_integer(bt) - ((uint)count)-1)); // FFFF8000
|
|
jlong hi = ~lo; // 00007FFF
|
|
const TypeInteger* t11 = phase->type(in(1)->in(1))->isa_integer(bt);
|
|
if (t11 == nullptr) {
|
|
return this;
|
|
}
|
|
// Does actual value fit inside of mask?
|
|
if (lo <= t11->lo_as_long() && t11->hi_as_long() <= hi) {
|
|
return in(1)->in(1); // Then shifting is a nop
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return this;
|
|
}
|
|
|
|
Node* RShiftINode::Identity(PhaseGVN* phase) {
|
|
return IdentityIL(phase, T_INT);
|
|
}
|
|
|
|
Node* RShiftNode::IdealIL(PhaseGVN* phase, bool can_reshape, BasicType bt) {
|
|
// Inputs may be TOP if they are dead.
|
|
const TypeInteger* t1 = phase->type(in(1))->isa_integer(bt);
|
|
if (t1 == nullptr) {
|
|
return NodeSentinel; // Left input is an integer
|
|
}
|
|
int shift = mask_and_replace_shift_amount(phase, this, bits_per_java_integer(bt));
|
|
if (shift == 0) {
|
|
return NodeSentinel;
|
|
}
|
|
|
|
// Check for (x & 0xFF000000) >> 24, whose mask can be made smaller.
|
|
// and convert to (x >> 24) & (0xFF000000 >> 24) = x >> 24
|
|
// Such expressions arise normally from shift chains like (byte)(x >> 24).
|
|
const Node* and_node = in(1);
|
|
if (and_node->Opcode() != Op_And(bt)) {
|
|
return nullptr;
|
|
}
|
|
const TypeInteger* mask_t = phase->type(and_node->in(2))->isa_integer(bt);
|
|
if (mask_t != nullptr && mask_t->is_con()) {
|
|
jlong maskbits = mask_t->get_con_as_long(bt);
|
|
// Convert to "(x >> shift) & (mask >> shift)"
|
|
Node* shr_nomask = phase->transform(RShiftNode::make(and_node->in(1), in(2), bt));
|
|
return MulNode::make_and(shr_nomask, phase->integercon(maskbits >> shift, bt), bt);
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
Node* RShiftINode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
Node* progress = IdealIL(phase, can_reshape, T_INT);
|
|
if (progress == NodeSentinel) {
|
|
return nullptr;
|
|
}
|
|
if (progress != nullptr) {
|
|
return progress;
|
|
}
|
|
int shift = mask_and_replace_shift_amount(phase, this, BitsPerJavaInteger);
|
|
assert(shift != 0, "handled by IdealIL");
|
|
|
|
// Check for "(short[i] <<16)>>16" which simply sign-extends
|
|
const Node *shl = in(1);
|
|
if (shl->Opcode() != Op_LShiftI) {
|
|
return nullptr;
|
|
}
|
|
|
|
const TypeInt* left_shift_t = phase->type(shl->in(2))->isa_int();
|
|
if (left_shift_t == nullptr) {
|
|
return nullptr;
|
|
}
|
|
if (shift == 16 && left_shift_t->is_con(16)) {
|
|
Node *ld = shl->in(1);
|
|
if (ld->Opcode() == Op_LoadS) {
|
|
// Sign extension is just useless here. Return a RShiftI of zero instead
|
|
// returning 'ld' directly. We cannot return an old Node directly as
|
|
// that is the job of 'Identity' calls and Identity calls only work on
|
|
// direct inputs ('ld' is an extra Node removed from 'this'). The
|
|
// combined optimization requires Identity only return direct inputs.
|
|
set_req_X(1, ld, phase);
|
|
set_req_X(2, phase->intcon(0), phase);
|
|
return this;
|
|
}
|
|
else if (can_reshape &&
|
|
ld->Opcode() == Op_LoadUS &&
|
|
ld->outcnt() == 1 && ld->unique_out() == shl)
|
|
// Replace zero-extension-load with sign-extension-load
|
|
return ld->as_Load()->convert_to_signed_load(*phase);
|
|
}
|
|
|
|
// Check for "(byte[i] <<24)>>24" which simply sign-extends
|
|
if (shift == 24 && left_shift_t->is_con(24)) {
|
|
Node *ld = shl->in(1);
|
|
if (ld->Opcode() == Op_LoadB) {
|
|
// Sign extension is just useless here
|
|
set_req_X(1, ld, phase);
|
|
set_req_X(2, phase->intcon(0), phase);
|
|
return this;
|
|
}
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
const Type* RShiftNode::ValueIL(PhaseGVN* phase, BasicType bt) const {
|
|
const Type* t1 = phase->type(in(1));
|
|
const Type* t2 = phase->type(in(2));
|
|
// Either input is TOP ==> the result is TOP
|
|
if (t1 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
if (t2 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
|
|
// Left input is ZERO ==> the result is ZERO.
|
|
if (t1 == TypeInteger::zero(bt)) {
|
|
return TypeInteger::zero(bt);
|
|
}
|
|
// Shift by zero does nothing
|
|
if (t2 == TypeInt::ZERO) {
|
|
return t1;
|
|
}
|
|
|
|
// Either input is BOTTOM ==> the result is BOTTOM
|
|
if (t1 == Type::BOTTOM || t2 == Type::BOTTOM) {
|
|
return TypeInteger::bottom(bt);
|
|
}
|
|
|
|
const TypeInteger* r1 = t1->isa_integer(bt);
|
|
const TypeInt* r2 = t2->isa_int();
|
|
|
|
// If the shift is a constant, just shift the bounds of the type.
|
|
// For example, if the shift is 31/63, we just propagate sign bits.
|
|
if (!r1->is_con() && r2->is_con()) {
|
|
uint shift = r2->get_con();
|
|
shift &= bits_per_java_integer(bt) - 1; // semantics of Java shifts
|
|
// Shift by a multiple of 32/64 does nothing:
|
|
if (shift == 0) {
|
|
return t1;
|
|
}
|
|
// Calculate reasonably aggressive bounds for the result.
|
|
// This is necessary if we are to correctly type things
|
|
// like (x<<24>>24) == ((byte)x).
|
|
jlong lo = r1->lo_as_long() >> (jint)shift;
|
|
jlong hi = r1->hi_as_long() >> (jint)shift;
|
|
assert(lo <= hi, "must have valid bounds");
|
|
#ifdef ASSERT
|
|
if (bt == T_INT) {
|
|
jint lo_verify = checked_cast<jint>(r1->lo_as_long()) >> (jint)shift;
|
|
jint hi_verify = checked_cast<jint>(r1->hi_as_long()) >> (jint)shift;
|
|
assert((checked_cast<jint>(lo) == lo_verify) && (checked_cast<jint>(hi) == hi_verify), "inconsistent");
|
|
}
|
|
#endif
|
|
const TypeInteger* ti = TypeInteger::make(lo, hi, MAX2(r1->_widen,r2->_widen), bt);
|
|
#ifdef ASSERT
|
|
// Make sure we get the sign-capture idiom correct.
|
|
if (shift == bits_per_java_integer(bt) - 1) {
|
|
if (r1->lo_as_long() >= 0) {
|
|
assert(ti == TypeInteger::zero(bt), ">>31/63 of + is 0");
|
|
}
|
|
if (r1->hi_as_long() < 0) {
|
|
assert(ti == TypeInteger::minus_1(bt), ">>31/63 of - is -1");
|
|
}
|
|
}
|
|
#endif
|
|
return ti;
|
|
}
|
|
|
|
if (!r1->is_con() || !r2->is_con()) {
|
|
// If the left input is non-negative the result must also be non-negative, regardless of what the right input is.
|
|
if (r1->lo_as_long() >= 0) {
|
|
return TypeInteger::make(0, r1->hi_as_long(), MAX2(r1->_widen, r2->_widen), bt);
|
|
}
|
|
|
|
// Conversely, if the left input is negative then the result must be negative.
|
|
if (r1->hi_as_long() <= -1) {
|
|
return TypeInteger::make(r1->lo_as_long(), -1, MAX2(r1->_widen, r2->_widen), bt);
|
|
}
|
|
|
|
return TypeInteger::bottom(bt);
|
|
}
|
|
|
|
// Signed shift right
|
|
return TypeInteger::make(r1->get_con_as_long(bt) >> (r2->get_con() & (bits_per_java_integer(bt) - 1)), bt);
|
|
}
|
|
|
|
const Type* RShiftINode::Value(PhaseGVN* phase) const {
|
|
return ValueIL(phase, T_INT);
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Identity---------------------------------------
|
|
Node* RShiftLNode::Identity(PhaseGVN* phase) {
|
|
return IdentityIL(phase, T_LONG);
|
|
}
|
|
|
|
Node* RShiftLNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
Node* progress = IdealIL(phase, can_reshape, T_LONG);
|
|
if (progress == NodeSentinel) {
|
|
return nullptr;
|
|
}
|
|
return progress;
|
|
}
|
|
|
|
const Type* RShiftLNode::Value(PhaseGVN* phase) const {
|
|
return ValueIL(phase, T_LONG);
|
|
}
|
|
|
|
URShiftNode* URShiftNode::make(Node* in1, Node* in2, BasicType bt) {
|
|
switch (bt) {
|
|
case T_INT:
|
|
return new URShiftINode(in1, in2);
|
|
case T_LONG:
|
|
return new URShiftLNode(in1, in2);
|
|
default:
|
|
fatal("Not implemented for %s", type2name(bt));
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Identity---------------------------------------
|
|
Node* URShiftINode::Identity(PhaseGVN* phase) {
|
|
int count = 0;
|
|
if (const_shift_count(phase, this, &count) && (count & (BitsPerJavaInteger - 1)) == 0) {
|
|
// Shift by a multiple of 32 does nothing
|
|
return in(1);
|
|
}
|
|
|
|
// Check for "((x << LogBytesPerWord) + (wordSize-1)) >> LogBytesPerWord" which is just "x".
|
|
// Happens during new-array length computation.
|
|
// Safe if 'x' is in the range [0..(max_int>>LogBytesPerWord)]
|
|
Node *add = in(1);
|
|
if (add->Opcode() == Op_AddI) {
|
|
const TypeInt *t2 = phase->type(add->in(2))->isa_int();
|
|
if (t2 && t2->is_con(wordSize - 1) &&
|
|
add->in(1)->Opcode() == Op_LShiftI) {
|
|
// Check that shift_counts are LogBytesPerWord.
|
|
Node *lshift_count = add->in(1)->in(2);
|
|
const TypeInt *t_lshift_count = phase->type(lshift_count)->isa_int();
|
|
if (t_lshift_count && t_lshift_count->is_con(LogBytesPerWord) &&
|
|
t_lshift_count == phase->type(in(2))) {
|
|
Node *x = add->in(1)->in(1);
|
|
const TypeInt *t_x = phase->type(x)->isa_int();
|
|
if (t_x != nullptr && 0 <= t_x->_lo && t_x->_hi <= (max_jint>>LogBytesPerWord)) {
|
|
return x;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
return (phase->type(in(2))->higher_equal(TypeInt::ZERO)) ? in(1) : this;
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
Node* URShiftINode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
int con = mask_and_replace_shift_amount(phase, this, BitsPerJavaInteger);
|
|
if (con == 0) {
|
|
return nullptr;
|
|
}
|
|
|
|
// We'll be wanting the right-shift amount as a mask of that many bits
|
|
const int mask = right_n_bits(BitsPerJavaInteger - con);
|
|
|
|
int in1_op = in(1)->Opcode();
|
|
|
|
// Check for ((x>>>a)>>>b) and replace with (x>>>(a+b)) when a+b < 32
|
|
if( in1_op == Op_URShiftI ) {
|
|
const TypeInt *t12 = phase->type( in(1)->in(2) )->isa_int();
|
|
if( t12 && t12->is_con() ) { // Right input is a constant
|
|
assert( in(1) != in(1)->in(1), "dead loop in URShiftINode::Ideal" );
|
|
const int con2 = t12->get_con() & 31; // Shift count is always masked
|
|
const int con3 = con+con2;
|
|
if( con3 < 32 ) // Only merge shifts if total is < 32
|
|
return new URShiftINode( in(1)->in(1), phase->intcon(con3) );
|
|
}
|
|
}
|
|
|
|
// Check for ((x << z) + Y) >>> z. Replace with x + con>>>z
|
|
// The idiom for rounding to a power of 2 is "(Q+(2^z-1)) >>> z".
|
|
// If Q is "X << z" the rounding is useless. Look for patterns like
|
|
// ((X<<Z) + Y) >>> Z and replace with (X + Y>>>Z) & Z-mask.
|
|
Node *add = in(1);
|
|
if (in1_op == Op_AddI) {
|
|
Node *lshl = add->in(1);
|
|
// Compare shift counts by value, not by node pointer, to also match a not-yet-normalized
|
|
// negative constant (e.g. -1 vs 31)
|
|
int lshl_con = 0;
|
|
if (lshl->Opcode() == Op_LShiftI &&
|
|
const_shift_count(phase, lshl, &lshl_con) &&
|
|
(lshl_con & (BitsPerJavaInteger - 1)) == con) {
|
|
Node *y_z = phase->transform( new URShiftINode(add->in(2),in(2)) );
|
|
Node *sum = phase->transform( new AddINode( lshl->in(1), y_z ) );
|
|
return new AndINode( sum, phase->intcon(mask) );
|
|
}
|
|
}
|
|
|
|
// Check for (x & mask) >>> z. Replace with (x >>> z) & (mask >>> z)
|
|
// This shortens the mask. Also, if we are extracting a high byte and
|
|
// storing it to a buffer, the mask will be removed completely.
|
|
Node *andi = in(1);
|
|
if( in1_op == Op_AndI ) {
|
|
const TypeInt *t3 = phase->type( andi->in(2) )->isa_int();
|
|
if( t3 && t3->is_con() ) { // Right input is a constant
|
|
jint mask2 = t3->get_con();
|
|
mask2 >>= con; // *signed* shift downward (high-order zeroes do not help)
|
|
Node *newshr = phase->transform( new URShiftINode(andi->in(1), in(2)) );
|
|
return new AndINode(newshr, phase->intcon(mask2));
|
|
// The negative values are easier to materialize than positive ones.
|
|
// A typical case from address arithmetic is ((x & ~15) >> 4).
|
|
// It's better to change that to ((x >> 4) & ~0) versus
|
|
// ((x >> 4) & 0x0FFFFFFF). The difference is greatest in LP64.
|
|
}
|
|
}
|
|
|
|
// Check for "(X << z ) >>> z" which simply zero-extends
|
|
Node *shl = in(1);
|
|
// Compare shift counts by value, not by node pointer, to also match a not-yet-normalized
|
|
// negative constant (e.g. -1 vs 31)
|
|
int shl_con = 0;
|
|
if (in1_op == Op_LShiftI &&
|
|
const_shift_count(phase, shl, &shl_con) &&
|
|
(shl_con & (BitsPerJavaInteger - 1)) == con)
|
|
return new AndINode(shl->in(1), phase->intcon(mask));
|
|
|
|
// Check for (x >> n) >>> 31. Replace with (x >>> 31)
|
|
const TypeInt* t2 = phase->type(in(2))->isa_int();
|
|
Node *shr = in(1);
|
|
if ( in1_op == Op_RShiftI ) {
|
|
Node *in11 = shr->in(1);
|
|
Node *in12 = shr->in(2);
|
|
const TypeInt *t11 = phase->type(in11)->isa_int();
|
|
const TypeInt *t12 = phase->type(in12)->isa_int();
|
|
if ( t11 && t2 && t2->is_con(31) && t12 && t12->is_con() ) {
|
|
return new URShiftINode(in11, phase->intcon(31));
|
|
}
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
//------------------------------Value------------------------------------------
|
|
// A URShiftINode shifts its input2 right by input1 amount.
|
|
const Type* URShiftINode::Value(PhaseGVN* phase) const {
|
|
// (This is a near clone of RShiftINode::Value.)
|
|
const Type *t1 = phase->type( in(1) );
|
|
const Type *t2 = phase->type( in(2) );
|
|
// Either input is TOP ==> the result is TOP
|
|
if( t1 == Type::TOP ) return Type::TOP;
|
|
if( t2 == Type::TOP ) return Type::TOP;
|
|
|
|
// Left input is ZERO ==> the result is ZERO.
|
|
if( t1 == TypeInt::ZERO ) return TypeInt::ZERO;
|
|
// Shift by zero does nothing
|
|
if( t2 == TypeInt::ZERO ) return t1;
|
|
|
|
// Either input is BOTTOM ==> the result is BOTTOM
|
|
if (t1 == Type::BOTTOM || t2 == Type::BOTTOM)
|
|
return TypeInt::INT;
|
|
|
|
if (t2 == TypeInt::INT)
|
|
return TypeInt::INT;
|
|
|
|
const TypeInt *r1 = t1->is_int(); // Handy access
|
|
const TypeInt *r2 = t2->is_int(); // Handy access
|
|
|
|
if (r2->is_con()) {
|
|
uint shift = r2->get_con();
|
|
shift &= BitsPerJavaInteger-1; // semantics of Java shifts
|
|
// Shift by a multiple of 32 does nothing:
|
|
if (shift == 0) return t1;
|
|
// Calculate reasonably aggressive bounds for the result.
|
|
jint lo = (juint)r1->_lo >> (juint)shift;
|
|
jint hi = (juint)r1->_hi >> (juint)shift;
|
|
if (r1->_hi >= 0 && r1->_lo < 0) {
|
|
// If the type has both negative and positive values,
|
|
// there are two separate sub-domains to worry about:
|
|
// The positive half and the negative half.
|
|
jint neg_lo = lo;
|
|
jint neg_hi = (juint)-1 >> (juint)shift;
|
|
jint pos_lo = (juint) 0 >> (juint)shift;
|
|
jint pos_hi = hi;
|
|
lo = MIN2(neg_lo, pos_lo); // == 0
|
|
hi = MAX2(neg_hi, pos_hi); // == -1 >>> shift;
|
|
}
|
|
assert(lo <= hi, "must have valid bounds");
|
|
const TypeInt* ti = TypeInt::make(lo, hi, MAX2(r1->_widen,r2->_widen));
|
|
#ifdef ASSERT
|
|
// Make sure we get the sign-capture idiom correct.
|
|
if (shift == BitsPerJavaInteger-1) {
|
|
if (r1->_lo >= 0) assert(ti == TypeInt::ZERO, ">>>31 of + is 0");
|
|
if (r1->_hi < 0) assert(ti == TypeInt::ONE, ">>>31 of - is +1");
|
|
}
|
|
#endif
|
|
return ti;
|
|
}
|
|
|
|
//
|
|
// Do not support shifted oops in info for GC
|
|
//
|
|
// else if( t1->base() == Type::InstPtr ) {
|
|
//
|
|
// const TypeInstPtr *o = t1->is_instptr();
|
|
// if( t1->singleton() )
|
|
// return TypeInt::make( ((uint32_t)o->const_oop() + o->_offset) >> shift );
|
|
// }
|
|
// else if( t1->base() == Type::KlassPtr ) {
|
|
// const TypeKlassPtr *o = t1->is_klassptr();
|
|
// if( t1->singleton() )
|
|
// return TypeInt::make( ((uint32_t)o->const_oop() + o->_offset) >> shift );
|
|
// }
|
|
|
|
return TypeInt::INT;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Identity---------------------------------------
|
|
Node* URShiftLNode::Identity(PhaseGVN* phase) {
|
|
int count = 0;
|
|
if (const_shift_count(phase, this, &count) && (count & (BitsPerJavaLong - 1)) == 0) {
|
|
// Shift by a multiple of 64 does nothing
|
|
return in(1);
|
|
}
|
|
return this;
|
|
}
|
|
|
|
//------------------------------Ideal------------------------------------------
|
|
Node* URShiftLNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
int con = mask_and_replace_shift_amount(phase, this, BitsPerJavaLong);
|
|
if (con == 0) {
|
|
return nullptr;
|
|
}
|
|
|
|
// We'll be wanting the right-shift amount as a mask of that many bits
|
|
const jlong mask = jlong(max_julong >> con);
|
|
|
|
// Check for ((x << z) + Y) >>> z. Replace with x + con>>>z
|
|
// The idiom for rounding to a power of 2 is "(Q+(2^z-1)) >>> z".
|
|
// If Q is "X << z" the rounding is useless. Look for patterns like
|
|
// ((X<<Z) + Y) >>> Z and replace with (X + Y>>>Z) & Z-mask.
|
|
Node *add = in(1);
|
|
const TypeInt *t2 = phase->type(in(2))->isa_int();
|
|
if (add->Opcode() == Op_AddL) {
|
|
Node *lshl = add->in(1);
|
|
// Compare shift counts by value, not by node pointer, to also match a not-yet-normalized
|
|
// negative constant (e.g. -1 vs 63)
|
|
int lshl_con = 0;
|
|
if (lshl->Opcode() == Op_LShiftL &&
|
|
const_shift_count(phase, lshl, &lshl_con) &&
|
|
(lshl_con & (BitsPerJavaLong - 1)) == con) {
|
|
Node* y_z = phase->transform(new URShiftLNode(add->in(2), in(2)));
|
|
Node* sum = phase->transform(new AddLNode(lshl->in(1), y_z));
|
|
return new AndLNode(sum, phase->longcon(mask));
|
|
}
|
|
}
|
|
|
|
// Check for (x & mask) >>> z. Replace with (x >>> z) & (mask >>> z)
|
|
// This shortens the mask. Also, if we are extracting a high byte and
|
|
// storing it to a buffer, the mask will be removed completely.
|
|
Node *andi = in(1);
|
|
if( andi->Opcode() == Op_AndL ) {
|
|
const TypeLong *t3 = phase->type( andi->in(2) )->isa_long();
|
|
if( t3 && t3->is_con() ) { // Right input is a constant
|
|
jlong mask2 = t3->get_con();
|
|
mask2 >>= con; // *signed* shift downward (high-order zeroes do not help)
|
|
Node *newshr = phase->transform( new URShiftLNode(andi->in(1), in(2)) );
|
|
return new AndLNode(newshr, phase->longcon(mask2));
|
|
}
|
|
}
|
|
|
|
// Check for "(X << z ) >>> z" which simply zero-extends
|
|
Node *shl = in(1);
|
|
// Compare shift counts by value, not by node pointer, to also match a not-yet-normalized
|
|
// negative constant (e.g. -1 vs 63)
|
|
int shl_con = 0;
|
|
if (shl->Opcode() == Op_LShiftL &&
|
|
const_shift_count(phase, shl, &shl_con) &&
|
|
(shl_con & (BitsPerJavaLong - 1)) == con) {
|
|
return new AndLNode(shl->in(1), phase->longcon(mask));
|
|
}
|
|
|
|
// Check for (x >> n) >>> 63. Replace with (x >>> 63)
|
|
Node *shr = in(1);
|
|
if ( shr->Opcode() == Op_RShiftL ) {
|
|
Node *in11 = shr->in(1);
|
|
Node *in12 = shr->in(2);
|
|
const TypeLong *t11 = phase->type(in11)->isa_long();
|
|
const TypeInt *t12 = phase->type(in12)->isa_int();
|
|
if ( t11 && t2 && t2->is_con(63) && t12 && t12->is_con() ) {
|
|
return new URShiftLNode(in11, phase->intcon(63));
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
//------------------------------Value------------------------------------------
|
|
// A URShiftINode shifts its input2 right by input1 amount.
|
|
const Type* URShiftLNode::Value(PhaseGVN* phase) const {
|
|
// (This is a near clone of RShiftLNode::Value.)
|
|
const Type *t1 = phase->type( in(1) );
|
|
const Type *t2 = phase->type( in(2) );
|
|
// Either input is TOP ==> the result is TOP
|
|
if( t1 == Type::TOP ) return Type::TOP;
|
|
if( t2 == Type::TOP ) return Type::TOP;
|
|
|
|
// Left input is ZERO ==> the result is ZERO.
|
|
if( t1 == TypeLong::ZERO ) return TypeLong::ZERO;
|
|
// Shift by zero does nothing
|
|
if( t2 == TypeInt::ZERO ) return t1;
|
|
|
|
// Either input is BOTTOM ==> the result is BOTTOM
|
|
if (t1 == Type::BOTTOM || t2 == Type::BOTTOM)
|
|
return TypeLong::LONG;
|
|
|
|
if (t2 == TypeInt::INT)
|
|
return TypeLong::LONG;
|
|
|
|
const TypeLong *r1 = t1->is_long(); // Handy access
|
|
const TypeInt *r2 = t2->is_int (); // Handy access
|
|
|
|
if (r2->is_con()) {
|
|
uint shift = r2->get_con();
|
|
shift &= BitsPerJavaLong - 1; // semantics of Java shifts
|
|
// Shift by a multiple of 64 does nothing:
|
|
if (shift == 0) return t1;
|
|
// Calculate reasonably aggressive bounds for the result.
|
|
jlong lo = (julong)r1->_lo >> (juint)shift;
|
|
jlong hi = (julong)r1->_hi >> (juint)shift;
|
|
if (r1->_hi >= 0 && r1->_lo < 0) {
|
|
// If the type has both negative and positive values,
|
|
// there are two separate sub-domains to worry about:
|
|
// The positive half and the negative half.
|
|
jlong neg_lo = lo;
|
|
jlong neg_hi = (julong)-1 >> (juint)shift;
|
|
jlong pos_lo = (julong) 0 >> (juint)shift;
|
|
jlong pos_hi = hi;
|
|
//lo = MIN2(neg_lo, pos_lo); // == 0
|
|
lo = neg_lo < pos_lo ? neg_lo : pos_lo;
|
|
//hi = MAX2(neg_hi, pos_hi); // == -1 >>> shift;
|
|
hi = neg_hi > pos_hi ? neg_hi : pos_hi;
|
|
}
|
|
assert(lo <= hi, "must have valid bounds");
|
|
const TypeLong* tl = TypeLong::make(lo, hi, MAX2(r1->_widen,r2->_widen));
|
|
#ifdef ASSERT
|
|
// Make sure we get the sign-capture idiom correct.
|
|
if (shift == BitsPerJavaLong - 1) {
|
|
if (r1->_lo >= 0) assert(tl == TypeLong::ZERO, ">>>63 of + is 0");
|
|
if (r1->_hi < 0) assert(tl == TypeLong::ONE, ">>>63 of - is +1");
|
|
}
|
|
#endif
|
|
return tl;
|
|
}
|
|
|
|
return TypeLong::LONG; // Give up
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Ideal------------------------------------------
|
|
Node* FmaNode::Ideal(PhaseGVN* phase, bool can_reshape) {
|
|
// We canonicalize the node by converting "(-a)*b+c" into "b*(-a)+c"
|
|
// This reduces the number of rules in the matcher, as we only need to check
|
|
// for negations on the second argument, and not the symmetric case where
|
|
// the first argument is negated.
|
|
if (in(1)->is_Neg() && !in(2)->is_Neg()) {
|
|
swap_edges(1, 2);
|
|
return this;
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Value------------------------------------------
|
|
const Type* FmaDNode::Value(PhaseGVN* phase) const {
|
|
const Type *t1 = phase->type(in(1));
|
|
if (t1 == Type::TOP) return Type::TOP;
|
|
if (t1->base() != Type::DoubleCon) return Type::DOUBLE;
|
|
const Type *t2 = phase->type(in(2));
|
|
if (t2 == Type::TOP) return Type::TOP;
|
|
if (t2->base() != Type::DoubleCon) return Type::DOUBLE;
|
|
const Type *t3 = phase->type(in(3));
|
|
if (t3 == Type::TOP) return Type::TOP;
|
|
if (t3->base() != Type::DoubleCon) return Type::DOUBLE;
|
|
#ifndef __STDC_IEC_559__
|
|
return Type::DOUBLE;
|
|
#else
|
|
double d1 = t1->getd();
|
|
double d2 = t2->getd();
|
|
double d3 = t3->getd();
|
|
return TypeD::make(fma(d1, d2, d3));
|
|
#endif
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Value------------------------------------------
|
|
const Type* FmaFNode::Value(PhaseGVN* phase) const {
|
|
const Type *t1 = phase->type(in(1));
|
|
if (t1 == Type::TOP) return Type::TOP;
|
|
if (t1->base() != Type::FloatCon) return Type::FLOAT;
|
|
const Type *t2 = phase->type(in(2));
|
|
if (t2 == Type::TOP) return Type::TOP;
|
|
if (t2->base() != Type::FloatCon) return Type::FLOAT;
|
|
const Type *t3 = phase->type(in(3));
|
|
if (t3 == Type::TOP) return Type::TOP;
|
|
if (t3->base() != Type::FloatCon) return Type::FLOAT;
|
|
#ifndef __STDC_IEC_559__
|
|
return Type::FLOAT;
|
|
#else
|
|
float f1 = t1->getf();
|
|
float f2 = t2->getf();
|
|
float f3 = t3->getf();
|
|
return TypeF::make(fma(f1, f2, f3));
|
|
#endif
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------Value------------------------------------------
|
|
const Type* FmaHFNode::Value(PhaseGVN* phase) const {
|
|
const Type* t1 = phase->type(in(1));
|
|
if (t1 == Type::TOP) { return Type::TOP; }
|
|
if (t1->base() != Type::HalfFloatCon) { return Type::HALF_FLOAT; }
|
|
const Type* t2 = phase->type(in(2));
|
|
if (t2 == Type::TOP) { return Type::TOP; }
|
|
if (t2->base() != Type::HalfFloatCon) { return Type::HALF_FLOAT; }
|
|
const Type* t3 = phase->type(in(3));
|
|
if (t3 == Type::TOP) { return Type::TOP; }
|
|
if (t3->base() != Type::HalfFloatCon) { return Type::HALF_FLOAT; }
|
|
#ifndef __STDC_IEC_559__
|
|
return Type::HALF_FLOAT;
|
|
#else
|
|
float f1 = t1->getf();
|
|
float f2 = t2->getf();
|
|
float f3 = t3->getf();
|
|
return TypeH::make(fma(f1, f2, f3));
|
|
#endif
|
|
}
|
|
|
|
//=============================================================================
|
|
//------------------------------hash-------------------------------------------
|
|
// Hash function for MulAddS2INode. Operation is commutative with commutative pairs.
|
|
// The hash function must return the same value when edge swapping is performed.
|
|
uint MulAddS2INode::hash() const {
|
|
return (uintptr_t)in(1) + (uintptr_t)in(2) + (uintptr_t)in(3) + (uintptr_t)in(4) + Opcode();
|
|
}
|
|
|
|
//------------------------------Rotate Operations ------------------------------
|
|
|
|
Node* RotateLeftNode::Identity(PhaseGVN* phase) {
|
|
const Type* t1 = phase->type(in(1));
|
|
if (t1 == Type::TOP) {
|
|
return this;
|
|
}
|
|
int count = 0;
|
|
assert(t1->isa_int() || t1->isa_long(), "Unexpected type");
|
|
int mask = (t1->isa_int() ? BitsPerJavaInteger : BitsPerJavaLong) - 1;
|
|
if (const_shift_count(phase, this, &count) && (count & mask) == 0) {
|
|
// Rotate by a multiple of 32/64 does nothing
|
|
return in(1);
|
|
}
|
|
return this;
|
|
}
|
|
|
|
const Type* RotateLeftNode::Value(PhaseGVN* phase) const {
|
|
const Type* t1 = phase->type(in(1));
|
|
const Type* t2 = phase->type(in(2));
|
|
// Either input is TOP ==> the result is TOP
|
|
if (t1 == Type::TOP || t2 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
|
|
if (t1->isa_int()) {
|
|
const TypeInt* r1 = t1->is_int();
|
|
const TypeInt* r2 = t2->is_int();
|
|
|
|
// Left input is ZERO ==> the result is ZERO.
|
|
if (r1 == TypeInt::ZERO) {
|
|
return TypeInt::ZERO;
|
|
}
|
|
// Rotate by zero does nothing
|
|
if (r2 == TypeInt::ZERO) {
|
|
return r1;
|
|
}
|
|
if (r1->is_con() && r2->is_con()) {
|
|
juint r1_con = (juint)r1->get_con();
|
|
juint shift = (juint)(r2->get_con()) & (juint)(BitsPerJavaInteger - 1); // semantics of Java shifts
|
|
return TypeInt::make((r1_con << shift) | (r1_con >> (32 - shift)));
|
|
}
|
|
return TypeInt::INT;
|
|
} else {
|
|
assert(t1->isa_long(), "Type must be a long");
|
|
const TypeLong* r1 = t1->is_long();
|
|
const TypeInt* r2 = t2->is_int();
|
|
|
|
// Left input is ZERO ==> the result is ZERO.
|
|
if (r1 == TypeLong::ZERO) {
|
|
return TypeLong::ZERO;
|
|
}
|
|
// Rotate by zero does nothing
|
|
if (r2 == TypeInt::ZERO) {
|
|
return r1;
|
|
}
|
|
if (r1->is_con() && r2->is_con()) {
|
|
julong r1_con = (julong)r1->get_con();
|
|
julong shift = (julong)(r2->get_con()) & (julong)(BitsPerJavaLong - 1); // semantics of Java shifts
|
|
return TypeLong::make((r1_con << shift) | (r1_con >> (64 - shift)));
|
|
}
|
|
return TypeLong::LONG;
|
|
}
|
|
}
|
|
|
|
Node* RotateLeftNode::Ideal(PhaseGVN *phase, bool can_reshape) {
|
|
const Type* t1 = phase->type(in(1));
|
|
const Type* t2 = phase->type(in(2));
|
|
if (t2->isa_int() && t2->is_int()->is_con()) {
|
|
if (t1->isa_int()) {
|
|
int lshift = t2->is_int()->get_con() & 31;
|
|
return new RotateRightNode(in(1), phase->intcon(32 - (lshift & 31)), TypeInt::INT);
|
|
} else if (t1 != Type::TOP) {
|
|
assert(t1->isa_long(), "Type must be a long");
|
|
int lshift = t2->is_int()->get_con() & 63;
|
|
return new RotateRightNode(in(1), phase->intcon(64 - (lshift & 63)), TypeLong::LONG);
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
Node* RotateRightNode::Identity(PhaseGVN* phase) {
|
|
const Type* t1 = phase->type(in(1));
|
|
if (t1 == Type::TOP) {
|
|
return this;
|
|
}
|
|
int count = 0;
|
|
assert(t1->isa_int() || t1->isa_long(), "Unexpected type");
|
|
int mask = (t1->isa_int() ? BitsPerJavaInteger : BitsPerJavaLong) - 1;
|
|
if (const_shift_count(phase, this, &count) && (count & mask) == 0) {
|
|
// Rotate by a multiple of 32/64 does nothing
|
|
return in(1);
|
|
}
|
|
return this;
|
|
}
|
|
|
|
const Type* RotateRightNode::Value(PhaseGVN* phase) const {
|
|
const Type* t1 = phase->type(in(1));
|
|
const Type* t2 = phase->type(in(2));
|
|
// Either input is TOP ==> the result is TOP
|
|
if (t1 == Type::TOP || t2 == Type::TOP) {
|
|
return Type::TOP;
|
|
}
|
|
|
|
if (t1->isa_int()) {
|
|
const TypeInt* r1 = t1->is_int();
|
|
const TypeInt* r2 = t2->is_int();
|
|
|
|
// Left input is ZERO ==> the result is ZERO.
|
|
if (r1 == TypeInt::ZERO) {
|
|
return TypeInt::ZERO;
|
|
}
|
|
// Rotate by zero does nothing
|
|
if (r2 == TypeInt::ZERO) {
|
|
return r1;
|
|
}
|
|
if (r1->is_con() && r2->is_con()) {
|
|
juint r1_con = (juint)r1->get_con();
|
|
juint shift = (juint)(r2->get_con()) & (juint)(BitsPerJavaInteger - 1); // semantics of Java shifts
|
|
return TypeInt::make((r1_con >> shift) | (r1_con << (32 - shift)));
|
|
}
|
|
return TypeInt::INT;
|
|
} else {
|
|
assert(t1->isa_long(), "Type must be a long");
|
|
const TypeLong* r1 = t1->is_long();
|
|
const TypeInt* r2 = t2->is_int();
|
|
// Left input is ZERO ==> the result is ZERO.
|
|
if (r1 == TypeLong::ZERO) {
|
|
return TypeLong::ZERO;
|
|
}
|
|
// Rotate by zero does nothing
|
|
if (r2 == TypeInt::ZERO) {
|
|
return r1;
|
|
}
|
|
if (r1->is_con() && r2->is_con()) {
|
|
julong r1_con = (julong)r1->get_con();
|
|
julong shift = (julong)(r2->get_con()) & (julong)(BitsPerJavaLong - 1); // semantics of Java shifts
|
|
return TypeLong::make((r1_con >> shift) | (r1_con << (64 - shift)));
|
|
}
|
|
return TypeLong::LONG;
|
|
}
|
|
}
|
|
|
|
//------------------------------ Sum & Mask ------------------------------
|
|
|
|
// Returns a lower bound on the number of trailing zeros in expr.
|
|
static jint AndIL_min_trailing_zeros(const PhaseGVN* phase, const Node* expr, BasicType bt) {
|
|
const TypeInteger* type = phase->type(expr)->isa_integer(bt);
|
|
if (type == nullptr) {
|
|
return 0;
|
|
}
|
|
|
|
expr = expr->uncast();
|
|
type = phase->type(expr)->isa_integer(bt);
|
|
if (type == nullptr) {
|
|
return 0;
|
|
}
|
|
|
|
if (type->is_con()) {
|
|
jlong con = type->get_con_as_long(bt);
|
|
return con == 0L ? (type2aelembytes(bt) * BitsPerByte) : count_trailing_zeros(con);
|
|
}
|
|
|
|
if (expr->Opcode() == Op_ConvI2L) {
|
|
expr = expr->in(1)->uncast();
|
|
bt = T_INT;
|
|
type = phase->type(expr)->isa_int();
|
|
}
|
|
|
|
// Pattern: expr = (x << shift)
|
|
if (expr->Opcode() == Op_LShift(bt)) {
|
|
const TypeInt* shift_t = phase->type(expr->in(2))->isa_int();
|
|
if (shift_t == nullptr || !shift_t->is_con()) {
|
|
return 0;
|
|
}
|
|
// We need to truncate the shift, as it may not have been canonicalized yet.
|
|
// T_INT: 0..31 -> shift_mask = 4 * 8 - 1 = 31
|
|
// T_LONG: 0..63 -> shift_mask = 8 * 8 - 1 = 63
|
|
// (JLS: "Shift Operators")
|
|
jint shift_mask = type2aelembytes(bt) * BitsPerByte - 1;
|
|
return shift_t->get_con() & shift_mask;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
// Checks whether expr is neutral additive element (zero) under mask,
|
|
// i.e. whether an expression of the form:
|
|
// (AndX (AddX (expr addend) mask)
|
|
// (expr + addend) & mask
|
|
// is equivalent to
|
|
// (AndX addend mask)
|
|
// addend & mask
|
|
// for any addend.
|
|
// (The X in AndX must be I or L, depending on bt).
|
|
//
|
|
// We check for the sufficient condition when the lowest set bit in expr is higher than
|
|
// the highest set bit in mask, i.e.:
|
|
// expr: eeeeee0000000000000
|
|
// mask: 000000mmmmmmmmmmmmm
|
|
// <--w bits--->
|
|
// We do not test for other cases.
|
|
//
|
|
// Correctness:
|
|
// Given "expr" with at least "w" trailing zeros,
|
|
// let "mod = 2^w", "suffix_mask = mod - 1"
|
|
//
|
|
// Since "mask" only has bits set where "suffix_mask" does, we have:
|
|
// mask = suffix_mask & mask (SUFFIX_MASK)
|
|
//
|
|
// And since expr only has bits set above w, and suffix_mask only below:
|
|
// expr & suffix_mask == 0 (NO_BIT_OVERLAP)
|
|
//
|
|
// From unsigned modular arithmetic (with unsigned modulo %), and since mod is
|
|
// a power of 2, and we are computing in a ring of powers of 2, we know that
|
|
// (x + y) % mod = (x % mod + y) % mod
|
|
// (x + y) & suffix_mask = (x & suffix_mask + y) & suffix_mask (MOD_ARITH)
|
|
//
|
|
// We can now prove the equality:
|
|
// (expr + addend) & mask
|
|
// = (expr + addend) & suffix_mask & mask (SUFFIX_MASK)
|
|
// = (expr & suffix_mask + addend) & suffix_mask & mask (MOD_ARITH)
|
|
// = (0 + addend) & suffix_mask & mask (NO_BIT_OVERLAP)
|
|
// = addend & mask (SUFFIX_MASK)
|
|
//
|
|
// Hence, an expr with at least w trailing zeros is a neutral additive element under any mask with bit width w.
|
|
static bool AndIL_is_zero_element_under_mask(const PhaseGVN* phase, const Node* expr, const Node* mask, BasicType bt) {
|
|
// When the mask is negative, it has the most significant bit set.
|
|
const TypeInteger* mask_t = phase->type(mask)->isa_integer(bt);
|
|
if (mask_t == nullptr || mask_t->lo_as_long() < 0) {
|
|
return false;
|
|
}
|
|
|
|
// When the mask is constant zero, we defer to MulNode::Value to eliminate the entire AndX operation.
|
|
if (mask_t->hi_as_long() == 0) {
|
|
assert(mask_t->lo_as_long() == 0, "checked earlier");
|
|
return false;
|
|
}
|
|
|
|
jint mask_bit_width = BitsPerLong - count_leading_zeros(mask_t->hi_as_long());
|
|
jint expr_trailing_zeros = AndIL_min_trailing_zeros(phase, expr, bt);
|
|
return expr_trailing_zeros >= mask_bit_width;
|
|
}
|
|
|
|
// Reduces the pattern:
|
|
// (AndX (AddX add1 add2) mask)
|
|
// to
|
|
// (AndX add1 mask), if add2 is neutral wrt mask (see above), and vice versa.
|
|
Node* MulNode::AndIL_sum_and_mask(PhaseGVN* phase, BasicType bt) {
|
|
Node* add = in(1);
|
|
Node* mask = in(2);
|
|
int addidx = 0;
|
|
if (add->Opcode() == Op_Add(bt)) {
|
|
addidx = 1;
|
|
} else if (mask->Opcode() == Op_Add(bt)) {
|
|
mask = add;
|
|
addidx = 2;
|
|
add = in(addidx);
|
|
}
|
|
if (addidx > 0) {
|
|
Node* add1 = add->in(1);
|
|
Node* add2 = add->in(2);
|
|
if (AndIL_is_zero_element_under_mask(phase, add1, mask, bt)) {
|
|
set_req_X(addidx, add2, phase);
|
|
return this;
|
|
} else if (AndIL_is_zero_element_under_mask(phase, add2, mask, bt)) {
|
|
set_req_X(addidx, add1, phase);
|
|
return this;
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|