/* * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. * */ #include "precompiled.hpp" #include "memory/allocation.inline.hpp" #include "opto/addnode.hpp" #include "opto/connode.hpp" #include "opto/divnode.hpp" #include "opto/machnode.hpp" #include "opto/matcher.hpp" #include "opto/mulnode.hpp" #include "opto/phaseX.hpp" #include "opto/subnode.hpp" // Portions of code courtesy of Clifford Click // Optimization - Graph Style #include //----------------------magic_int_divide_constants----------------------------- // Compute magic multiplier and shift constant for converting a 32 bit divide // by constant into a multiply/shift/add series. Return false if calculations // fail. // // Borrowed almost verbatim from Hacker's Delight by Henry S. Warren, Jr. with // minor type name and parameter changes. static bool magic_int_divide_constants(jint d, jint &M, jint &s) { int32_t p; uint32_t ad, anc, delta, q1, r1, q2, r2, t; const uint32_t two31 = 0x80000000L; // 2**31. ad = ABS(d); if (d == 0 || d == 1) return false; t = two31 + ((uint32_t)d >> 31); anc = t - 1 - t%ad; // Absolute value of nc. p = 31; // Init. p. q1 = two31/anc; // Init. q1 = 2**p/|nc|. r1 = two31 - q1*anc; // Init. r1 = rem(2**p, |nc|). q2 = two31/ad; // Init. q2 = 2**p/|d|. r2 = two31 - q2*ad; // Init. r2 = rem(2**p, |d|). do { p = p + 1; q1 = 2*q1; // Update q1 = 2**p/|nc|. r1 = 2*r1; // Update r1 = rem(2**p, |nc|). if (r1 >= anc) { // (Must be an unsigned q1 = q1 + 1; // comparison here). r1 = r1 - anc; } q2 = 2*q2; // Update q2 = 2**p/|d|. r2 = 2*r2; // Update r2 = rem(2**p, |d|). if (r2 >= ad) { // (Must be an unsigned q2 = q2 + 1; // comparison here). r2 = r2 - ad; } delta = ad - r2; } while (q1 < delta || (q1 == delta && r1 == 0)); M = q2 + 1; if (d < 0) M = -M; // Magic number and s = p - 32; // shift amount to return. return true; } //--------------------------transform_int_divide------------------------------- // Convert a division by constant divisor into an alternate Ideal graph. // Return NULL if no transformation occurs. static Node *transform_int_divide( PhaseGVN *phase, Node *dividend, jint divisor ) { // Check for invalid divisors assert( divisor != 0 && divisor != min_jint, "bad divisor for transforming to long multiply" ); bool d_pos = divisor >= 0; jint d = d_pos ? divisor : -divisor; const int N = 32; // Result Node *q = NULL; if (d == 1) { // division by +/- 1 if (!d_pos) { // Just negate the value q = new (phase->C) SubINode(phase->intcon(0), dividend); } } else if ( is_power_of_2(d) ) { // division by +/- a power of 2 // See if we can simply do a shift without rounding bool needs_rounding = true; const Type *dt = phase->type(dividend); const TypeInt *dti = dt->isa_int(); if (dti && dti->_lo >= 0) { // we don't need to round a positive dividend needs_rounding = false; } else if( dividend->Opcode() == Op_AndI ) { // An AND mask of sufficient size clears the low bits and // I can avoid rounding. const TypeInt *andconi_t = phase->type( dividend->in(2) )->isa_int(); if( andconi_t && andconi_t->is_con() ) { jint andconi = andconi_t->get_con(); if( andconi < 0 && is_power_of_2(-andconi) && (-andconi) >= d ) { if( (-andconi) == d ) // Remove AND if it clears bits which will be shifted dividend = dividend->in(1); needs_rounding = false; } } } // Add rounding to the shift to handle the sign bit int l = log2_jint(d-1)+1; if (needs_rounding) { // Divide-by-power-of-2 can be made into a shift, but you have to do // more math for the rounding. You need to add 0 for positive // numbers, and "i-1" for negative numbers. Example: i=4, so the // shift is by 2. You need to add 3 to negative dividends and 0 to // positive ones. So (-7+3)>>2 becomes -1, (-4+3)>>2 becomes -1, // (-2+3)>>2 becomes 0, etc. // Compute 0 or -1, based on sign bit Node *sign = phase->transform(new (phase->C) RShiftINode(dividend, phase->intcon(N - 1))); // Mask sign bit to the low sign bits Node *round = phase->transform(new (phase->C) URShiftINode(sign, phase->intcon(N - l))); // Round up before shifting dividend = phase->transform(new (phase->C) AddINode(dividend, round)); } // Shift for division q = new (phase->C) RShiftINode(dividend, phase->intcon(l)); if (!d_pos) { q = new (phase->C) SubINode(phase->intcon(0), phase->transform(q)); } } else { // Attempt the jint constant divide -> multiply transform found in // "Division by Invariant Integers using Multiplication" // by Granlund and Montgomery // See also "Hacker's Delight", chapter 10 by Warren. jint magic_const; jint shift_const; if (magic_int_divide_constants(d, magic_const, shift_const)) { Node *magic = phase->longcon(magic_const); Node *dividend_long = phase->transform(new (phase->C) ConvI2LNode(dividend)); // Compute the high half of the dividend x magic multiplication Node *mul_hi = phase->transform(new (phase->C) MulLNode(dividend_long, magic)); if (magic_const < 0) { mul_hi = phase->transform(new (phase->C) RShiftLNode(mul_hi, phase->intcon(N))); mul_hi = phase->transform(new (phase->C) ConvL2INode(mul_hi)); // The magic multiplier is too large for a 32 bit constant. We've adjusted // it down by 2^32, but have to add 1 dividend back in after the multiplication. // This handles the "overflow" case described by Granlund and Montgomery. mul_hi = phase->transform(new (phase->C) AddINode(dividend, mul_hi)); // Shift over the (adjusted) mulhi if (shift_const != 0) { mul_hi = phase->transform(new (phase->C) RShiftINode(mul_hi, phase->intcon(shift_const))); } } else { // No add is required, we can merge the shifts together. mul_hi = phase->transform(new (phase->C) RShiftLNode(mul_hi, phase->intcon(N + shift_const))); mul_hi = phase->transform(new (phase->C) ConvL2INode(mul_hi)); } // Get a 0 or -1 from the sign of the dividend. Node *addend0 = mul_hi; Node *addend1 = phase->transform(new (phase->C) RShiftINode(dividend, phase->intcon(N-1))); // If the divisor is negative, swap the order of the input addends; // this has the effect of negating the quotient. if (!d_pos) { Node *temp = addend0; addend0 = addend1; addend1 = temp; } // Adjust the final quotient by subtracting -1 (adding 1) // from the mul_hi. q = new (phase->C) SubINode(addend0, addend1); } } return q; } //---------------------magic_long_divide_constants----------------------------- // Compute magic multiplier and shift constant for converting a 64 bit divide // by constant into a multiply/shift/add series. Return false if calculations // fail. // // Borrowed almost verbatim from Hacker's Delight by Henry S. Warren, Jr. with // minor type name and parameter changes. Adjusted to 64 bit word width. static bool magic_long_divide_constants(jlong d, jlong &M, jint &s) { int64_t p; uint64_t ad, anc, delta, q1, r1, q2, r2, t; const uint64_t two63 = 0x8000000000000000LL; // 2**63. ad = ABS(d); if (d == 0 || d == 1) return false; t = two63 + ((uint64_t)d >> 63); anc = t - 1 - t%ad; // Absolute value of nc. p = 63; // Init. p. q1 = two63/anc; // Init. q1 = 2**p/|nc|. r1 = two63 - q1*anc; // Init. r1 = rem(2**p, |nc|). q2 = two63/ad; // Init. q2 = 2**p/|d|. r2 = two63 - q2*ad; // Init. r2 = rem(2**p, |d|). do { p = p + 1; q1 = 2*q1; // Update q1 = 2**p/|nc|. r1 = 2*r1; // Update r1 = rem(2**p, |nc|). if (r1 >= anc) { // (Must be an unsigned q1 = q1 + 1; // comparison here). r1 = r1 - anc; } q2 = 2*q2; // Update q2 = 2**p/|d|. r2 = 2*r2; // Update r2 = rem(2**p, |d|). if (r2 >= ad) { // (Must be an unsigned q2 = q2 + 1; // comparison here). r2 = r2 - ad; } delta = ad - r2; } while (q1 < delta || (q1 == delta && r1 == 0)); M = q2 + 1; if (d < 0) M = -M; // Magic number and s = p - 64; // shift amount to return. return true; } //---------------------long_by_long_mulhi-------------------------------------- // Generate ideal node graph for upper half of a 64 bit x 64 bit multiplication static Node* long_by_long_mulhi(PhaseGVN* phase, Node* dividend, jlong magic_const) { // If the architecture supports a 64x64 mulhi, there is // no need to synthesize it in ideal nodes. if (Matcher::has_match_rule(Op_MulHiL)) { Node* v = phase->longcon(magic_const); return new (phase->C) MulHiLNode(dividend, v); } // Taken from Hacker's Delight, Fig. 8-2. Multiply high signed. // (http://www.hackersdelight.org/HDcode/mulhs.c) // // int mulhs(int u, int v) { // unsigned u0, v0, w0; // int u1, v1, w1, w2, t; // // u0 = u & 0xFFFF; u1 = u >> 16; // v0 = v & 0xFFFF; v1 = v >> 16; // w0 = u0*v0; // t = u1*v0 + (w0 >> 16); // w1 = t & 0xFFFF; // w2 = t >> 16; // w1 = u0*v1 + w1; // return u1*v1 + w2 + (w1 >> 16); // } // // Note: The version above is for 32x32 multiplications, while the // following inline comments are adapted to 64x64. const int N = 64; // Dummy node to keep intermediate nodes alive during construction Node* hook = new (phase->C) Node(4); // u0 = u & 0xFFFFFFFF; u1 = u >> 32; Node* u0 = phase->transform(new (phase->C) AndLNode(dividend, phase->longcon(0xFFFFFFFF))); Node* u1 = phase->transform(new (phase->C) RShiftLNode(dividend, phase->intcon(N / 2))); hook->init_req(0, u0); hook->init_req(1, u1); // v0 = v & 0xFFFFFFFF; v1 = v >> 32; Node* v0 = phase->longcon(magic_const & 0xFFFFFFFF); Node* v1 = phase->longcon(magic_const >> (N / 2)); // w0 = u0*v0; Node* w0 = phase->transform(new (phase->C) MulLNode(u0, v0)); // t = u1*v0 + (w0 >> 32); Node* u1v0 = phase->transform(new (phase->C) MulLNode(u1, v0)); Node* temp = phase->transform(new (phase->C) URShiftLNode(w0, phase->intcon(N / 2))); Node* t = phase->transform(new (phase->C) AddLNode(u1v0, temp)); hook->init_req(2, t); // w1 = t & 0xFFFFFFFF; Node* w1 = phase->transform(new (phase->C) AndLNode(t, phase->longcon(0xFFFFFFFF))); hook->init_req(3, w1); // w2 = t >> 32; Node* w2 = phase->transform(new (phase->C) RShiftLNode(t, phase->intcon(N / 2))); // w1 = u0*v1 + w1; Node* u0v1 = phase->transform(new (phase->C) MulLNode(u0, v1)); w1 = phase->transform(new (phase->C) AddLNode(u0v1, w1)); // return u1*v1 + w2 + (w1 >> 32); Node* u1v1 = phase->transform(new (phase->C) MulLNode(u1, v1)); Node* temp1 = phase->transform(new (phase->C) AddLNode(u1v1, w2)); Node* temp2 = phase->transform(new (phase->C) RShiftLNode(w1, phase->intcon(N / 2))); // Remove the bogus extra edges used to keep things alive PhaseIterGVN* igvn = phase->is_IterGVN(); if (igvn != NULL) { igvn->remove_dead_node(hook); } else { for (int i = 0; i < 4; i++) { hook->set_req(i, NULL); } } return new (phase->C) AddLNode(temp1, temp2); } //--------------------------transform_long_divide------------------------------ // Convert a division by constant divisor into an alternate Ideal graph. // Return NULL if no transformation occurs. static Node *transform_long_divide( PhaseGVN *phase, Node *dividend, jlong divisor ) { // Check for invalid divisors assert( divisor != 0L && divisor != min_jlong, "bad divisor for transforming to long multiply" ); bool d_pos = divisor >= 0; jlong d = d_pos ? divisor : -divisor; const int N = 64; // Result Node *q = NULL; if (d == 1) { // division by +/- 1 if (!d_pos) { // Just negate the value q = new (phase->C) SubLNode(phase->longcon(0), dividend); } } else if ( is_power_of_2_long(d) ) { // division by +/- a power of 2 // See if we can simply do a shift without rounding bool needs_rounding = true; const Type *dt = phase->type(dividend); const TypeLong *dtl = dt->isa_long(); if (dtl && dtl->_lo > 0) { // we don't need to round a positive dividend needs_rounding = false; } else if( dividend->Opcode() == Op_AndL ) { // An AND mask of sufficient size clears the low bits and // I can avoid rounding. const TypeLong *andconl_t = phase->type( dividend->in(2) )->isa_long(); if( andconl_t && andconl_t->is_con() ) { jlong andconl = andconl_t->get_con(); if( andconl < 0 && is_power_of_2_long(-andconl) && (-andconl) >= d ) { if( (-andconl) == d ) // Remove AND if it clears bits which will be shifted dividend = dividend->in(1); needs_rounding = false; } } } // Add rounding to the shift to handle the sign bit int l = log2_long(d-1)+1; if (needs_rounding) { // Divide-by-power-of-2 can be made into a shift, but you have to do // more math for the rounding. You need to add 0 for positive // numbers, and "i-1" for negative numbers. Example: i=4, so the // shift is by 2. You need to add 3 to negative dividends and 0 to // positive ones. So (-7+3)>>2 becomes -1, (-4+3)>>2 becomes -1, // (-2+3)>>2 becomes 0, etc. // Compute 0 or -1, based on sign bit Node *sign = phase->transform(new (phase->C) RShiftLNode(dividend, phase->intcon(N - 1))); // Mask sign bit to the low sign bits Node *round = phase->transform(new (phase->C) URShiftLNode(sign, phase->intcon(N - l))); // Round up before shifting dividend = phase->transform(new (phase->C) AddLNode(dividend, round)); } // Shift for division q = new (phase->C) RShiftLNode(dividend, phase->intcon(l)); if (!d_pos) { q = new (phase->C) SubLNode(phase->longcon(0), phase->transform(q)); } } else if ( !Matcher::use_asm_for_ldiv_by_con(d) ) { // Use hardware DIV instruction when // it is faster than code generated below. // Attempt the jlong constant divide -> multiply transform found in // "Division by Invariant Integers using Multiplication" // by Granlund and Montgomery // See also "Hacker's Delight", chapter 10 by Warren. jlong magic_const; jint shift_const; if (magic_long_divide_constants(d, magic_const, shift_const)) { // Compute the high half of the dividend x magic multiplication Node *mul_hi = phase->transform(long_by_long_mulhi(phase, dividend, magic_const)); // The high half of the 128-bit multiply is computed. if (magic_const < 0) { // The magic multiplier is too large for a 64 bit constant. We've adjusted // it down by 2^64, but have to add 1 dividend back in after the multiplication. // This handles the "overflow" case described by Granlund and Montgomery. mul_hi = phase->transform(new (phase->C) AddLNode(dividend, mul_hi)); } // Shift over the (adjusted) mulhi if (shift_const != 0) { mul_hi = phase->transform(new (phase->C) RShiftLNode(mul_hi, phase->intcon(shift_const))); } // Get a 0 or -1 from the sign of the dividend. Node *addend0 = mul_hi; Node *addend1 = phase->transform(new (phase->C) RShiftLNode(dividend, phase->intcon(N-1))); // If the divisor is negative, swap the order of the input addends; // this has the effect of negating the quotient. if (!d_pos) { Node *temp = addend0; addend0 = addend1; addend1 = temp; } // Adjust the final quotient by subtracting -1 (adding 1) // from the mul_hi. q = new (phase->C) SubLNode(addend0, addend1); } } return q; } //============================================================================= //------------------------------Identity--------------------------------------- // If the divisor is 1, we are an identity on the dividend. Node *DivINode::Identity( PhaseTransform *phase ) { return (phase->type( in(2) )->higher_equal(TypeInt::ONE)) ? in(1) : this; } //------------------------------Idealize--------------------------------------- // Divides can be changed to multiplies and/or shifts Node *DivINode::Ideal(PhaseGVN *phase, bool can_reshape) { if (in(0) && remove_dead_region(phase, can_reshape)) return this; // Don't bother trying to transform a dead node if( in(0) && in(0)->is_top() ) return NULL; const Type *t = phase->type( in(2) ); if( t == TypeInt::ONE ) // Identity? return NULL; // Skip it const TypeInt *ti = t->isa_int(); if( !ti ) return NULL; if( !ti->is_con() ) return NULL; jint i = ti->get_con(); // Get divisor if (i == 0) return NULL; // Dividing by zero constant does not idealize set_req(0,NULL); // Dividing by a not-zero constant; no faulting // Dividing by MININT does not optimize as a power-of-2 shift. if( i == min_jint ) return NULL; return transform_int_divide( phase, in(1), i ); } //------------------------------Value------------------------------------------ // A DivINode divides its inputs. The third input is a Control input, used to // prevent hoisting the divide above an unsafe test. const Type *DivINode::Value( PhaseTransform *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 ) return Type::TOP; if( t2 == Type::TOP ) return Type::TOP; // x/x == 1 since we always generate the dynamic divisor check for 0. if( phase->eqv( in(1), in(2) ) ) return TypeInt::ONE; // Either input is BOTTOM ==> the result is the local BOTTOM const Type *bot = bottom_type(); if( (t1 == bot) || (t2 == bot) || (t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) ) return bot; // Divide the two numbers. We approximate. // If divisor is a constant and not zero const TypeInt *i1 = t1->is_int(); const TypeInt *i2 = t2->is_int(); int widen = MAX2(i1->_widen, i2->_widen); if( i2->is_con() && i2->get_con() != 0 ) { int32 d = i2->get_con(); // Divisor jint lo, hi; if( d >= 0 ) { lo = i1->_lo/d; hi = i1->_hi/d; } else { if( d == -1 && i1->_lo == min_jint ) { // 'min_jint/-1' throws arithmetic exception during compilation lo = min_jint; // do not support holes, 'hi' must go to either min_jint or max_jint: // [min_jint, -10]/[-1,-1] ==> [min_jint] UNION [10,max_jint] hi = i1->_hi == min_jint ? min_jint : max_jint; } else { lo = i1->_hi/d; hi = i1->_lo/d; } } return TypeInt::make(lo, hi, widen); } // If the dividend is a constant if( i1->is_con() ) { int32 d = i1->get_con(); if( d < 0 ) { if( d == min_jint ) { // (-min_jint) == min_jint == (min_jint / -1) return TypeInt::make(min_jint, max_jint/2 + 1, widen); } else { return TypeInt::make(d, -d, widen); } } return TypeInt::make(-d, d, widen); } // Otherwise we give up all hope return TypeInt::INT; } //============================================================================= //------------------------------Identity--------------------------------------- // If the divisor is 1, we are an identity on the dividend. Node *DivLNode::Identity( PhaseTransform *phase ) { return (phase->type( in(2) )->higher_equal(TypeLong::ONE)) ? in(1) : this; } //------------------------------Idealize--------------------------------------- // Dividing by a power of 2 is a shift. Node *DivLNode::Ideal( PhaseGVN *phase, bool can_reshape) { if (in(0) && remove_dead_region(phase, can_reshape)) return this; // Don't bother trying to transform a dead node if( in(0) && in(0)->is_top() ) return NULL; const Type *t = phase->type( in(2) ); if( t == TypeLong::ONE ) // Identity? return NULL; // Skip it const TypeLong *tl = t->isa_long(); if( !tl ) return NULL; if( !tl->is_con() ) return NULL; jlong l = tl->get_con(); // Get divisor if (l == 0) return NULL; // Dividing by zero constant does not idealize set_req(0,NULL); // Dividing by a not-zero constant; no faulting // Dividing by MINLONG does not optimize as a power-of-2 shift. if( l == min_jlong ) return NULL; return transform_long_divide( phase, in(1), l ); } //------------------------------Value------------------------------------------ // A DivLNode divides its inputs. The third input is a Control input, used to // prevent hoisting the divide above an unsafe test. const Type *DivLNode::Value( PhaseTransform *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 ) return Type::TOP; if( t2 == Type::TOP ) return Type::TOP; // x/x == 1 since we always generate the dynamic divisor check for 0. if( phase->eqv( in(1), in(2) ) ) return TypeLong::ONE; // Either input is BOTTOM ==> the result is the local BOTTOM const Type *bot = bottom_type(); if( (t1 == bot) || (t2 == bot) || (t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) ) return bot; // Divide the two numbers. We approximate. // If divisor is a constant and not zero const TypeLong *i1 = t1->is_long(); const TypeLong *i2 = t2->is_long(); int widen = MAX2(i1->_widen, i2->_widen); if( i2->is_con() && i2->get_con() != 0 ) { jlong d = i2->get_con(); // Divisor jlong lo, hi; if( d >= 0 ) { lo = i1->_lo/d; hi = i1->_hi/d; } else { if( d == CONST64(-1) && i1->_lo == min_jlong ) { // 'min_jlong/-1' throws arithmetic exception during compilation lo = min_jlong; // do not support holes, 'hi' must go to either min_jlong or max_jlong: // [min_jlong, -10]/[-1,-1] ==> [min_jlong] UNION [10,max_jlong] hi = i1->_hi == min_jlong ? min_jlong : max_jlong; } else { lo = i1->_hi/d; hi = i1->_lo/d; } } return TypeLong::make(lo, hi, widen); } // If the dividend is a constant if( i1->is_con() ) { jlong d = i1->get_con(); if( d < 0 ) { if( d == min_jlong ) { // (-min_jlong) == min_jlong == (min_jlong / -1) return TypeLong::make(min_jlong, max_jlong/2 + 1, widen); } else { return TypeLong::make(d, -d, widen); } } return TypeLong::make(-d, d, widen); } // Otherwise we give up all hope return TypeLong::LONG; } //============================================================================= //------------------------------Value------------------------------------------ // An DivFNode divides its inputs. The third input is a Control input, used to // prevent hoisting the divide above an unsafe test. const Type *DivFNode::Value( PhaseTransform *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 ) return Type::TOP; if( t2 == Type::TOP ) return Type::TOP; // Either input is BOTTOM ==> the result is the local BOTTOM const Type *bot = bottom_type(); if( (t1 == bot) || (t2 == bot) || (t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) ) return bot; // x/x == 1, we ignore 0/0. // Note: if t1 and t2 are zero then result is NaN (JVMS page 213) // Does not work for variables because of NaN's if( phase->eqv( in(1), in(2) ) && t1->base() == Type::FloatCon) if (!g_isnan(t1->getf()) && g_isfinite(t1->getf()) && t1->getf() != 0.0) // could be negative ZERO or NaN return TypeF::ONE; if( t2 == TypeF::ONE ) return t1; // If divisor is a constant and not zero, divide them numbers if( t1->base() == Type::FloatCon && t2->base() == Type::FloatCon && t2->getf() != 0.0 ) // could be negative zero return TypeF::make( t1->getf()/t2->getf() ); // If the dividend is a constant zero // Note: if t1 and t2 are zero then result is NaN (JVMS page 213) // Test TypeF::ZERO is not sufficient as it could be negative zero if( t1 == TypeF::ZERO && !g_isnan(t2->getf()) && t2->getf() != 0.0 ) return TypeF::ZERO; // Otherwise we give up all hope return Type::FLOAT; } //------------------------------isA_Copy--------------------------------------- // Dividing by self is 1. // If the divisor is 1, we are an identity on the dividend. Node *DivFNode::Identity( PhaseTransform *phase ) { return (phase->type( in(2) ) == TypeF::ONE) ? in(1) : this; } //------------------------------Idealize--------------------------------------- Node *DivFNode::Ideal(PhaseGVN *phase, bool can_reshape) { if (in(0) && remove_dead_region(phase, can_reshape)) return this; // Don't bother trying to transform a dead node if( in(0) && in(0)->is_top() ) return NULL; const Type *t2 = phase->type( in(2) ); if( t2 == TypeF::ONE ) // Identity? return NULL; // Skip it const TypeF *tf = t2->isa_float_constant(); if( !tf ) return NULL; if( tf->base() != Type::FloatCon ) return NULL; // Check for out of range values if( tf->is_nan() || !tf->is_finite() ) return NULL; // Get the value float f = tf->getf(); int exp; // Only for special case of dividing by a power of 2 if( frexp((double)f, &exp) != 0.5 ) return NULL; // Limit the range of acceptable exponents if( exp < -126 || exp > 126 ) return NULL; // Compute the reciprocal float reciprocal = ((float)1.0) / f; assert( frexp((double)reciprocal, &exp) == 0.5, "reciprocal should be power of 2" ); // return multiplication by the reciprocal return (new (phase->C) MulFNode(in(1), phase->makecon(TypeF::make(reciprocal)))); } //============================================================================= //------------------------------Value------------------------------------------ // An DivDNode divides its inputs. The third input is a Control input, used to // prevent hoisting the divide above an unsafe test. const Type *DivDNode::Value( PhaseTransform *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 ) return Type::TOP; if( t2 == Type::TOP ) return Type::TOP; // Either input is BOTTOM ==> the result is the local BOTTOM const Type *bot = bottom_type(); if( (t1 == bot) || (t2 == bot) || (t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) ) return bot; // x/x == 1, we ignore 0/0. // Note: if t1 and t2 are zero then result is NaN (JVMS page 213) // Does not work for variables because of NaN's if( phase->eqv( in(1), in(2) ) && t1->base() == Type::DoubleCon) if (!g_isnan(t1->getd()) && g_isfinite(t1->getd()) && t1->getd() != 0.0) // could be negative ZERO or NaN return TypeD::ONE; if( t2 == TypeD::ONE ) return t1; #if defined(IA32) if (!phase->C->method()->is_strict()) // Can't trust native compilers to properly fold strict double // division with round-to-zero on this platform. #endif { // If divisor is a constant and not zero, divide them numbers if( t1->base() == Type::DoubleCon && t2->base() == Type::DoubleCon && t2->getd() != 0.0 ) // could be negative zero return TypeD::make( t1->getd()/t2->getd() ); } // If the dividend is a constant zero // Note: if t1 and t2 are zero then result is NaN (JVMS page 213) // Test TypeF::ZERO is not sufficient as it could be negative zero if( t1 == TypeD::ZERO && !g_isnan(t2->getd()) && t2->getd() != 0.0 ) return TypeD::ZERO; // Otherwise we give up all hope return Type::DOUBLE; } //------------------------------isA_Copy--------------------------------------- // Dividing by self is 1. // If the divisor is 1, we are an identity on the dividend. Node *DivDNode::Identity( PhaseTransform *phase ) { return (phase->type( in(2) ) == TypeD::ONE) ? in(1) : this; } //------------------------------Idealize--------------------------------------- Node *DivDNode::Ideal(PhaseGVN *phase, bool can_reshape) { if (in(0) && remove_dead_region(phase, can_reshape)) return this; // Don't bother trying to transform a dead node if( in(0) && in(0)->is_top() ) return NULL; const Type *t2 = phase->type( in(2) ); if( t2 == TypeD::ONE ) // Identity? return NULL; // Skip it const TypeD *td = t2->isa_double_constant(); if( !td ) return NULL; if( td->base() != Type::DoubleCon ) return NULL; // Check for out of range values if( td->is_nan() || !td->is_finite() ) return NULL; // Get the value double d = td->getd(); int exp; // Only for special case of dividing by a power of 2 if( frexp(d, &exp) != 0.5 ) return NULL; // Limit the range of acceptable exponents if( exp < -1021 || exp > 1022 ) return NULL; // Compute the reciprocal double reciprocal = 1.0 / d; assert( frexp(reciprocal, &exp) == 0.5, "reciprocal should be power of 2" ); // return multiplication by the reciprocal return (new (phase->C) MulDNode(in(1), phase->makecon(TypeD::make(reciprocal)))); } //============================================================================= //------------------------------Idealize--------------------------------------- Node *ModINode::Ideal(PhaseGVN *phase, bool can_reshape) { // Check for dead control input if( in(0) && remove_dead_region(phase, can_reshape) ) return this; // Don't bother trying to transform a dead node if( in(0) && in(0)->is_top() ) return NULL; // Get the modulus const Type *t = phase->type( in(2) ); if( t == Type::TOP ) return NULL; const TypeInt *ti = t->is_int(); // Check for useless control input // Check for excluding mod-zero case if( in(0) && (ti->_hi < 0 || ti->_lo > 0) ) { set_req(0, NULL); // Yank control input return this; } // See if we are MOD'ing by 2^k or 2^k-1. if( !ti->is_con() ) return NULL; jint con = ti->get_con(); Node *hook = new (phase->C) Node(1); // First, special check for modulo 2^k-1 if( con >= 0 && con < max_jint && is_power_of_2(con+1) ) { uint k = exact_log2(con+1); // Extract k // Basic algorithm by David Detlefs. See fastmod_int.java for gory details. static int unroll_factor[] = { 999, 999, 29, 14, 9, 7, 5, 4, 4, 3, 3, 2, 2, 2, 2, 2, 1 /*past here we assume 1 forever*/}; int trip_count = 1; if( k < ARRAY_SIZE(unroll_factor)) trip_count = unroll_factor[k]; // If the unroll factor is not too large, and if conditional moves are // ok, then use this case if( trip_count <= 5 && ConditionalMoveLimit != 0 ) { Node *x = in(1); // Value being mod'd Node *divisor = in(2); // Also is mask hook->init_req(0, x); // Add a use to x to prevent him from dying // Generate code to reduce X rapidly to nearly 2^k-1. for( int i = 0; i < trip_count; i++ ) { Node *xl = phase->transform( new (phase->C) AndINode(x,divisor) ); Node *xh = phase->transform( new (phase->C) RShiftINode(x,phase->intcon(k)) ); // Must be signed x = phase->transform( new (phase->C) AddINode(xh,xl) ); hook->set_req(0, x); } // Generate sign-fixup code. Was original value positive? // int hack_res = (i >= 0) ? divisor : 1; Node *cmp1 = phase->transform( new (phase->C) CmpINode( in(1), phase->intcon(0) ) ); Node *bol1 = phase->transform( new (phase->C) BoolNode( cmp1, BoolTest::ge ) ); Node *cmov1= phase->transform( new (phase->C) CMoveINode(bol1, phase->intcon(1), divisor, TypeInt::POS) ); // if( x >= hack_res ) x -= divisor; Node *sub = phase->transform( new (phase->C) SubINode( x, divisor ) ); Node *cmp2 = phase->transform( new (phase->C) CmpINode( x, cmov1 ) ); Node *bol2 = phase->transform( new (phase->C) BoolNode( cmp2, BoolTest::ge ) ); // Convention is to not transform the return value of an Ideal // since Ideal is expected to return a modified 'this' or a new node. Node *cmov2= new (phase->C) CMoveINode(bol2, x, sub, TypeInt::INT); // cmov2 is now the mod // Now remove the bogus extra edges used to keep things alive if (can_reshape) { phase->is_IterGVN()->remove_dead_node(hook); } else { hook->set_req(0, NULL); // Just yank bogus edge during Parse phase } return cmov2; } } // Fell thru, the unroll case is not appropriate. Transform the modulo // into a long multiply/int multiply/subtract case // Cannot handle mod 0, and min_jint isn't handled by the transform if( con == 0 || con == min_jint ) return NULL; // Get the absolute value of the constant; at this point, we can use this jint pos_con = (con >= 0) ? con : -con; // integer Mod 1 is always 0 if( pos_con == 1 ) return new (phase->C) ConINode(TypeInt::ZERO); int log2_con = -1; // If this is a power of two, they maybe we can mask it if( is_power_of_2(pos_con) ) { log2_con = log2_intptr((intptr_t)pos_con); const Type *dt = phase->type(in(1)); const TypeInt *dti = dt->isa_int(); // See if this can be masked, if the dividend is non-negative if( dti && dti->_lo >= 0 ) return ( new (phase->C) AndINode( in(1), phase->intcon( pos_con-1 ) ) ); } // Save in(1) so that it cannot be changed or deleted hook->init_req(0, in(1)); // Divide using the transform from DivI to MulL Node *result = transform_int_divide( phase, in(1), pos_con ); if (result != NULL) { Node *divide = phase->transform(result); // Re-multiply, using a shift if this is a power of two Node *mult = NULL; if( log2_con >= 0 ) mult = phase->transform( new (phase->C) LShiftINode( divide, phase->intcon( log2_con ) ) ); else mult = phase->transform( new (phase->C) MulINode( divide, phase->intcon( pos_con ) ) ); // Finally, subtract the multiplied divided value from the original result = new (phase->C) SubINode( in(1), mult ); } // Now remove the bogus extra edges used to keep things alive if (can_reshape) { phase->is_IterGVN()->remove_dead_node(hook); } else { hook->set_req(0, NULL); // Just yank bogus edge during Parse phase } // return the value return result; } //------------------------------Value------------------------------------------ const Type *ModINode::Value( PhaseTransform *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 ) return Type::TOP; if( t2 == Type::TOP ) return Type::TOP; // We always generate the dynamic check for 0. // 0 MOD X is 0 if( t1 == TypeInt::ZERO ) return TypeInt::ZERO; // X MOD X is 0 if( phase->eqv( in(1), in(2) ) ) return TypeInt::ZERO; // Either input is BOTTOM ==> the result is the local BOTTOM const Type *bot = bottom_type(); if( (t1 == bot) || (t2 == bot) || (t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) ) return bot; const TypeInt *i1 = t1->is_int(); const TypeInt *i2 = t2->is_int(); if( !i1->is_con() || !i2->is_con() ) { if( i1->_lo >= 0 && i2->_lo >= 0 ) return TypeInt::POS; // If both numbers are not constants, we know little. return TypeInt::INT; } // Mod by zero? Throw exception at runtime! if( !i2->get_con() ) return TypeInt::POS; // We must be modulo'ing 2 float constants. // Check for min_jint % '-1', result is defined to be '0'. if( i1->get_con() == min_jint && i2->get_con() == -1 ) return TypeInt::ZERO; return TypeInt::make( i1->get_con() % i2->get_con() ); } //============================================================================= //------------------------------Idealize--------------------------------------- Node *ModLNode::Ideal(PhaseGVN *phase, bool can_reshape) { // Check for dead control input if( in(0) && remove_dead_region(phase, can_reshape) ) return this; // Don't bother trying to transform a dead node if( in(0) && in(0)->is_top() ) return NULL; // Get the modulus const Type *t = phase->type( in(2) ); if( t == Type::TOP ) return NULL; const TypeLong *tl = t->is_long(); // Check for useless control input // Check for excluding mod-zero case if( in(0) && (tl->_hi < 0 || tl->_lo > 0) ) { set_req(0, NULL); // Yank control input return this; } // See if we are MOD'ing by 2^k or 2^k-1. if( !tl->is_con() ) return NULL; jlong con = tl->get_con(); Node *hook = new (phase->C) Node(1); // Expand mod if( con >= 0 && con < max_jlong && is_power_of_2_long(con+1) ) { uint k = exact_log2_long(con+1); // Extract k // Basic algorithm by David Detlefs. See fastmod_long.java for gory details. // Used to help a popular random number generator which does a long-mod // of 2^31-1 and shows up in SpecJBB and SciMark. static int unroll_factor[] = { 999, 999, 61, 30, 20, 15, 12, 10, 8, 7, 6, 6, 5, 5, 4, 4, 4, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1 /*past here we assume 1 forever*/}; int trip_count = 1; if( k < ARRAY_SIZE(unroll_factor)) trip_count = unroll_factor[k]; // If the unroll factor is not too large, and if conditional moves are // ok, then use this case if( trip_count <= 5 && ConditionalMoveLimit != 0 ) { Node *x = in(1); // Value being mod'd Node *divisor = in(2); // Also is mask hook->init_req(0, x); // Add a use to x to prevent him from dying // Generate code to reduce X rapidly to nearly 2^k-1. for( int i = 0; i < trip_count; i++ ) { Node *xl = phase->transform( new (phase->C) AndLNode(x,divisor) ); Node *xh = phase->transform( new (phase->C) RShiftLNode(x,phase->intcon(k)) ); // Must be signed x = phase->transform( new (phase->C) AddLNode(xh,xl) ); hook->set_req(0, x); // Add a use to x to prevent him from dying } // Generate sign-fixup code. Was original value positive? // long hack_res = (i >= 0) ? divisor : CONST64(1); Node *cmp1 = phase->transform( new (phase->C) CmpLNode( in(1), phase->longcon(0) ) ); Node *bol1 = phase->transform( new (phase->C) BoolNode( cmp1, BoolTest::ge ) ); Node *cmov1= phase->transform( new (phase->C) CMoveLNode(bol1, phase->longcon(1), divisor, TypeLong::LONG) ); // if( x >= hack_res ) x -= divisor; Node *sub = phase->transform( new (phase->C) SubLNode( x, divisor ) ); Node *cmp2 = phase->transform( new (phase->C) CmpLNode( x, cmov1 ) ); Node *bol2 = phase->transform( new (phase->C) BoolNode( cmp2, BoolTest::ge ) ); // Convention is to not transform the return value of an Ideal // since Ideal is expected to return a modified 'this' or a new node. Node *cmov2= new (phase->C) CMoveLNode(bol2, x, sub, TypeLong::LONG); // cmov2 is now the mod // Now remove the bogus extra edges used to keep things alive if (can_reshape) { phase->is_IterGVN()->remove_dead_node(hook); } else { hook->set_req(0, NULL); // Just yank bogus edge during Parse phase } return cmov2; } } // Fell thru, the unroll case is not appropriate. Transform the modulo // into a long multiply/int multiply/subtract case // Cannot handle mod 0, and min_jlong isn't handled by the transform if( con == 0 || con == min_jlong ) return NULL; // Get the absolute value of the constant; at this point, we can use this jlong pos_con = (con >= 0) ? con : -con; // integer Mod 1 is always 0 if( pos_con == 1 ) return new (phase->C) ConLNode(TypeLong::ZERO); int log2_con = -1; // If this is a power of two, then maybe we can mask it if( is_power_of_2_long(pos_con) ) { log2_con = exact_log2_long(pos_con); const Type *dt = phase->type(in(1)); const TypeLong *dtl = dt->isa_long(); // See if this can be masked, if the dividend is non-negative if( dtl && dtl->_lo >= 0 ) return ( new (phase->C) AndLNode( in(1), phase->longcon( pos_con-1 ) ) ); } // Save in(1) so that it cannot be changed or deleted hook->init_req(0, in(1)); // Divide using the transform from DivL to MulL Node *result = transform_long_divide( phase, in(1), pos_con ); if (result != NULL) { Node *divide = phase->transform(result); // Re-multiply, using a shift if this is a power of two Node *mult = NULL; if( log2_con >= 0 ) mult = phase->transform( new (phase->C) LShiftLNode( divide, phase->intcon( log2_con ) ) ); else mult = phase->transform( new (phase->C) MulLNode( divide, phase->longcon( pos_con ) ) ); // Finally, subtract the multiplied divided value from the original result = new (phase->C) SubLNode( in(1), mult ); } // Now remove the bogus extra edges used to keep things alive if (can_reshape) { phase->is_IterGVN()->remove_dead_node(hook); } else { hook->set_req(0, NULL); // Just yank bogus edge during Parse phase } // return the value return result; } //------------------------------Value------------------------------------------ const Type *ModLNode::Value( PhaseTransform *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 ) return Type::TOP; if( t2 == Type::TOP ) return Type::TOP; // We always generate the dynamic check for 0. // 0 MOD X is 0 if( t1 == TypeLong::ZERO ) return TypeLong::ZERO; // X MOD X is 0 if( phase->eqv( in(1), in(2) ) ) return TypeLong::ZERO; // Either input is BOTTOM ==> the result is the local BOTTOM const Type *bot = bottom_type(); if( (t1 == bot) || (t2 == bot) || (t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) ) return bot; const TypeLong *i1 = t1->is_long(); const TypeLong *i2 = t2->is_long(); if( !i1->is_con() || !i2->is_con() ) { if( i1->_lo >= CONST64(0) && i2->_lo >= CONST64(0) ) return TypeLong::POS; // If both numbers are not constants, we know little. return TypeLong::LONG; } // Mod by zero? Throw exception at runtime! if( !i2->get_con() ) return TypeLong::POS; // We must be modulo'ing 2 float constants. // Check for min_jint % '-1', result is defined to be '0'. if( i1->get_con() == min_jlong && i2->get_con() == -1 ) return TypeLong::ZERO; return TypeLong::make( i1->get_con() % i2->get_con() ); } //============================================================================= //------------------------------Value------------------------------------------ const Type *ModFNode::Value( PhaseTransform *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 ) return Type::TOP; if( t2 == Type::TOP ) return Type::TOP; // Either input is BOTTOM ==> the result is the local BOTTOM const Type *bot = bottom_type(); if( (t1 == bot) || (t2 == bot) || (t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) ) return bot; // If either number is not a constant, we know nothing. if ((t1->base() != Type::FloatCon) || (t2->base() != Type::FloatCon)) { return Type::FLOAT; // note: x%x can be either NaN or 0 } float f1 = t1->getf(); float f2 = t2->getf(); jint x1 = jint_cast(f1); // note: *(int*)&f1, not just (int)f1 jint x2 = jint_cast(f2); // If either is a NaN, return an input NaN if (g_isnan(f1)) return t1; if (g_isnan(f2)) return t2; // If an operand is infinity or the divisor is +/- zero, punt. if (!g_isfinite(f1) || !g_isfinite(f2) || x2 == 0 || x2 == min_jint) return Type::FLOAT; // We must be modulo'ing 2 float constants. // Make sure that the sign of the fmod is equal to the sign of the dividend jint xr = jint_cast(fmod(f1, f2)); if ((x1 ^ xr) < 0) { xr ^= min_jint; } return TypeF::make(jfloat_cast(xr)); } //============================================================================= //------------------------------Value------------------------------------------ const Type *ModDNode::Value( PhaseTransform *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 ) return Type::TOP; if( t2 == Type::TOP ) return Type::TOP; // Either input is BOTTOM ==> the result is the local BOTTOM const Type *bot = bottom_type(); if( (t1 == bot) || (t2 == bot) || (t1 == Type::BOTTOM) || (t2 == Type::BOTTOM) ) return bot; // If either number is not a constant, we know nothing. if ((t1->base() != Type::DoubleCon) || (t2->base() != Type::DoubleCon)) { return Type::DOUBLE; // note: x%x can be either NaN or 0 } double f1 = t1->getd(); double f2 = t2->getd(); jlong x1 = jlong_cast(f1); // note: *(long*)&f1, not just (long)f1 jlong x2 = jlong_cast(f2); // If either is a NaN, return an input NaN if (g_isnan(f1)) return t1; if (g_isnan(f2)) return t2; // If an operand is infinity or the divisor is +/- zero, punt. if (!g_isfinite(f1) || !g_isfinite(f2) || x2 == 0 || x2 == min_jlong) return Type::DOUBLE; // We must be modulo'ing 2 double constants. // Make sure that the sign of the fmod is equal to the sign of the dividend jlong xr = jlong_cast(fmod(f1, f2)); if ((x1 ^ xr) < 0) { xr ^= min_jlong; } return TypeD::make(jdouble_cast(xr)); } //============================================================================= DivModNode::DivModNode( Node *c, Node *dividend, Node *divisor ) : MultiNode(3) { init_req(0, c); init_req(1, dividend); init_req(2, divisor); } //------------------------------make------------------------------------------ DivModINode* DivModINode::make(Compile* C, Node* div_or_mod) { Node* n = div_or_mod; assert(n->Opcode() == Op_DivI || n->Opcode() == Op_ModI, "only div or mod input pattern accepted"); DivModINode* divmod = new (C) DivModINode(n->in(0), n->in(1), n->in(2)); Node* dproj = new (C) ProjNode(divmod, DivModNode::div_proj_num); Node* mproj = new (C) ProjNode(divmod, DivModNode::mod_proj_num); return divmod; } //------------------------------make------------------------------------------ DivModLNode* DivModLNode::make(Compile* C, Node* div_or_mod) { Node* n = div_or_mod; assert(n->Opcode() == Op_DivL || n->Opcode() == Op_ModL, "only div or mod input pattern accepted"); DivModLNode* divmod = new (C) DivModLNode(n->in(0), n->in(1), n->in(2)); Node* dproj = new (C) ProjNode(divmod, DivModNode::div_proj_num); Node* mproj = new (C) ProjNode(divmod, DivModNode::mod_proj_num); return divmod; } //------------------------------match------------------------------------------ // return result(s) along with their RegMask info Node *DivModINode::match( const ProjNode *proj, const Matcher *match ) { uint ideal_reg = proj->ideal_reg(); RegMask rm; if (proj->_con == div_proj_num) { rm = match->divI_proj_mask(); } else { assert(proj->_con == mod_proj_num, "must be div or mod projection"); rm = match->modI_proj_mask(); } return new (match->C)MachProjNode(this, proj->_con, rm, ideal_reg); } //------------------------------match------------------------------------------ // return result(s) along with their RegMask info Node *DivModLNode::match( const ProjNode *proj, const Matcher *match ) { uint ideal_reg = proj->ideal_reg(); RegMask rm; if (proj->_con == div_proj_num) { rm = match->divL_proj_mask(); } else { assert(proj->_con == mod_proj_num, "must be div or mod projection"); rm = match->modL_proj_mask(); } return new (match->C)MachProjNode(this, proj->_con, rm, ideal_reg); }