/* BigRational.java -- dynamically sized big rational numbers. ** ** Copyright (C) 2002-2010 Eric Laroche. All rights reserved. ** ** @author Eric Laroche <laroche@lrdev.com> ** @version @(#)$Id: BigRational.java,v 1.3 2010/03/24 20:11:34 laroche Exp $ ** ** This program is free software; ** you can redistribute it and/or modify it. ** ** This program 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. ** */ package vitry.runtime; import java.math.BigInteger; /** * Implements * dynamically sized arbitrary precision immutable * rational numbers. * * <P> * Dynamically sized means that (provided enough memory) the BigRational * numbers can't overflow (nor underflow); this characteristic is * different from Java's data types int and long, but common with * BigInteger (which however implements only integer numbers, i.e. no * fractions) and BigDecimal (which however only implements precisely * rational numbers with denominators that are products of 2 and 5 [and * implied 10]). * Arbitrary precision means that there is no loss of precision with * common arithmetic operations such as addition, subtraction, * multiplication, and division (BigDecimal loses precision with * division, if factors other than 2 or 5 are involved). * [Doubles and floats can overflow and underflow, have a limited * precision, and only implement precisely rationals with a denominator * that is a power of 2.] * * <P> * BigRational provides most operations needed in rational number space * calculations, including addition, subtraction, multiplication, division, * integer power, remainder/modulus, comparison, and different roundings. * * <P> * BigRational provides conversion methods to and from the native types * long, int, short, and byte (including support for unsigned values), * double (binary64), float (binary32), quad (binary128, quadruple), * half (binary16), and BigInteger. * BigRational can parse and print many string representations: rational, * dot notations, with exponent, even mixtures thereof, and supports * different radixes/bases (2 to typically 36 [limited by BigInteger * parsing implementation]). * * <P> * BigRational uses java.math.BigInteger (JDK 1.1 and later). * * <P> * Binary operations (e.g. add, multiply) calculate their results from a * BigRational object ('this') and one [BigRational or long] argument, * returning a new immutable BigRational object. * Both the original object and the argument are left unchanged (hence * immutable). * Unary operations (e.g. negate, invert) calculate their result from the * BigRational object ('this'), returning a new immutable BigRational object. * The original object is left unchanged. * * <P> * Most operations are precise (i.e. without loss of precision); * exceptions are the conversion methods to limited precision types * (doubleValue, floatValue), rounding (round), * truncation (bigIntegerValue, floor, ceiling, truncate), * as well as obviously the printing methods that include a precision * parameter (toStringDot, toStringDotRelative, toStringExponent). * * <P> * BigRational doesn't provide a notion of "infinity" ([+-]Infinity) * and "not a number" (NaN); * IEEE 754 floating point Infinity and NaN are rejected (throwing a * NumberFormatException). * Operations such as 0/0 result in an ArithmeticException. * * <P> * Some commonly used short function names (abs, ceil, div, inv, max, min, * mod, mul, neg, pow, rem, sign, sub, trunc) are additionally defined as * aliases to to the full function names (absolute, ceiling, divide, invert, * maximum, minimum, modulus, multiply, negate, power, remainder, signum, * subtract, truncate). * [This makes the interface somewhat fatter.] * * <P> * BigRational internally uses private proxy functions for BigInteger * functionality, including scanning and multiplying, to enhance speed * and to realize fast checks for common values (1, 0, etc.). * * <P> * Constructor samples: * normal rational form, * abbreviated form, * fixed point form, * abbreviated fixed point form, * [exponential] floating point form, * different radixes/bases different from 10, * doubles/floats: * <SMALL><PRE> * BigRational("-21/35"): rational -3/5 * BigRational("/3"): rational 1/3 * BigRational("3.4"): rational 17/5 * BigRational(".7"): 0.7, rational 7/10 * BigRational("-65.4E-3"): -327/5000 * BigRational("f/37", 0x10): 3/11 * BigRational("f.37", 0x10): 3895/256 * BigRational("-dcba.efgh", 23): -46112938320/279841 * BigRational("1011101011010110", 2): 47830 * BigRational(StrictMath.E): 6121026514868073/2251799813685248 * BigRational((float)StrictMath.E): 2850325/1048576 * </PRE></SMALL> * * <P> * Also accepted are denormalized representations such as: * <SMALL><PRE> * BigRational("2.5/-3.5"): -5/7 * BigRational("12.34E-1/-56.78E-1"): -617/2839 * </PRE></SMALL> * * <P> * Printing: * rational form, * fixed point (dot) forms with different absolute precisions * (including negative precision), * with relative precisions, * exponential form, * different radix: * <SMALL><PRE> * BigRational("1234.5678"): "6172839/5000" * BigRational("1234.5678").toStringDot(6): "1234.567800" * BigRational("1234.5678").toStringDot(2): "1234.57" * BigRational("1234.5678").toStringDot(-2): "1200" * BigRational("1234.5678").toStringDotRelative(6): "1234.57" * BigRational("0.00012345678").toStringDotRelative(3): "0.000123" * BigRational("1234.5678").toStringExponent(2): "1.2E3" * BigRational("1011101011010110", 2).toString(0x10): "bad6" * </PRE></SMALL> * * <P> * Usage: * BigRational operations can be conveniently chained * (sample from BigRational internal conversion from IEEE 754 bits): * <SMALL><PRE> * BigRational.valueOf(2).power(exponent).multiply(fraction.add( * BigRational.valueOfUnsigned(mantissa))).multiply(sign); * </PRE></SMALL> * * <P> * The BigRational source and documentation can typically be found at * the author's (Eric Laroche) site, at * <A HREF="http://www.lrdev.com/lr/java/BigRational.java" * >http://www.lrdev.com/lr/java/BigRational.java</A> (source) and * <A HREF="http://www.lrdev.com/lr/java/BigRational.html" * >http://www.lrdev.com/lr/java/BigRational.html</A> (documentation). * * @author Eric Laroche <laroche@lrdev.com> * @version @(#)$Id: BigRational.java,v 1.3 2010/03/24 20:11:34 laroche Exp $ * */ // @Immutable public class BigRational extends Number implements Cloneable, Comparable<BigRational> { /** * */ private static final long serialVersionUID = 5520472122793569971L; /** Numerator. * Numerator may be negative. * Numerator may be zero, in which case m_q must be one. * [Conditions are put in place by normalize().] */ private BigInteger m_n; /** Denominator (quotient). * Denominator is never negative and never zero. * [Conditions are put in place by normalize().] */ private BigInteger m_q; /** Default radix, used in string printing and scanning, * 10. * <P> * Default radix is decimal [of course]. */ public final static int DEFAULT_RADIX = 10; // note: following constants can't be constructed using bigIntegerValueOf(). // that one _uses_ the constants (avoid circular dependencies). /** Constant internally used, for convenience and speed. * Used as zero numerator. * Used for fast checks. */ private final static BigInteger BIG_INTEGER_ZERO = BigInteger.valueOf(0); /** Constant internally used, for convenience and speed. * Used as neutral denominator. * Used for fast checks. */ private final static BigInteger BIG_INTEGER_ONE = BigInteger.valueOf(1); /** Constant internally used, for convenience and speed. * Used for fast checks. */ private final static BigInteger BIG_INTEGER_MINUS_ONE = BigInteger.valueOf(-1); /** Constant internally used, for convenience and speed. * Used in rounding zero numerator. * _Not_ used for checks. */ private final static BigInteger BIG_INTEGER_TWO = BigInteger.valueOf(2); /** Constant internally used, for convenience and speed. * _Not_ used for checks. */ private final static BigInteger BIG_INTEGER_MINUS_TWO = BigInteger.valueOf(-2); /** Constant internally used, for convenience and speed. * Corresponds to DEFAULT_RADIX, used in reading, scaling and printing. * _Not_ used for checks. */ private final static BigInteger BIG_INTEGER_TEN = BigInteger.valueOf(10); /** Constant internally used, for convenience and speed. * Used in reading, scaling and printing. * _Not_ used for checks. */ private final static BigInteger BIG_INTEGER_SIXTEEN = BigInteger.valueOf(16); // some more constants /** The constant two to the power of 64 (18446744073709551616). * Used is slicing larger [than double size] IEEE 754 values. */ private final static BigInteger BIG_INTEGER_TWO_POWER_64 = BigInteger.valueOf(2).pow( 64); /** Construct a BigRational from numerator and denominator. * <P> * Both n and q may be negative. * n/q may be denormalized (i.e. have common factors, or q being negative). */ public BigRational(BigInteger n, BigInteger q) { // note: check for q==null done later if (q != null && bigIntegerIsZero(q)) { throw new NumberFormatException( "quotient zero"); } normalizeFrom(n, q); } /** Construct a BigRational from a big number integer; * denominator is 1. */ public BigRational(BigInteger n) { this(n, BIG_INTEGER_ONE); } /** Construct a BigRational from long fix number integers * representing numerator and denominator. */ public BigRational(long n, long q) { this(bigIntegerValueOf(n), bigIntegerValueOf(q)); } /** Construct a BigRational from a long fix number integer. */ public BigRational(long n) { this(bigIntegerValueOf(n), BIG_INTEGER_ONE); } // note: byte/short/int implicitly upgraded to long, // so we don't implement BigRational(int,int) et al. /** Clone a BigRational. * <P> * [As BigRationals are immutable, this copy-constructor * is not that useful.] */ public BigRational(BigRational that) { normalizeFrom(that); } /** Construct a BigRational from a string representation. * <P> * The supported string formats are * "[+-]n", "[+-]n/[+-]q", "[+-]i.f", "[+-]i", * "[+-]iE[+-]e", "[+-]i.fE[+-]e" (latter two only with radix<=10, * due to possible ambiguities); n and q can be any of the latter * (i.e. mixed representations such as "-1.2E-3/-4.5E-6" are supported). * <P> * Samples: "-21/35", "3.4", "-65.4E-3", * "f/37" (base 16), "1011101011010110" (base 2). * <P> * [Exponential representation wasn't supported in an earlier version.] */ public BigRational(String s, int radix) { if (s == null) { throw new NumberFormatException("null"); } // '/': least precedence, and left-to-right associative // (therefore lastIndexOf and not indexOf: last slash has least precedence) final int slash = s.lastIndexOf('/'); if (slash != -1) { // "[+-]n/[+-]q" String sn = s.substring(0, slash); String sq = s.substring(slash + 1); // suppress recursion: make stack-overflow attacks infeasible if (sn.indexOf('/') != -1) { throw new NumberFormatException("can't nest '/'"); } // handle "/x" as "1/x" // [note: "1" and "-1" are treated specially and optimized // in bigIntegerValueOf(String,int).] if (sn.equals("") || sn.equals("+")) { sn = "1"; } else if (sn.equals("-")) { sn = "-1"; } // handle "x/" as "x" // [note: "1" and "-1" are treated special and optimized // in bigIntegerValueOf(String,int).] if (sq.equals("") || sq.equals("+")) { sq = "1"; } else if (sq.equals("-")) { sq = "-1"; } // [recursion] // [divide()'s outcome is already normalized, // so there would be no need to normalize [again]] normalizeFrom( (new BigRational(sn, radix)).divide(new BigRational(sq, radix))); return; } checkRadix(radix); // catch Java's string representations of doubles/floats unsupported by us checkNaNAndInfinity(s, radix); // [if radix<=10:] 'E': next higher precedence, not associative // or right-to-left associative int exp = -1; // note: a distinction of exponent-'E' from large-base-digit-'e' // would be unintuitive, since case doesn't matter with both uses if (radix <= 10) { // handle both [upper/lower] cases final int exp1 = s.indexOf('E'); final int exp2 = s.indexOf('e'); exp = (exp1 == -1 || (exp2 != -1 && exp2 < exp1) ? exp2 : exp1); } if (exp != -1) { String sm = s.substring(0, exp); String se = s.substring(exp + 1); // suppress recursion: make stack-overflow attacks infeasible if (se.indexOf('E') != -1 || se.indexOf('e') != -1) { throw new NumberFormatException( "can't nest 'E'"); } // skip '+' if (se.length() > 0 && se.charAt(0) == '+') { se = se.substring(1); } // handle '-' boolean negtexp = false; if (se.length() > 0 && se.charAt(0) == '-') { negtexp = true; se = se.substring(1); } // handle "xE", "xE+", "xE-", as "xE0" aka "x" if (se.equals("")) { se = "0"; } // [recursion] BigRational exponent = new BigRational(se, radix); final int iexponent; // transform possible [overflow/fraction] exception try { iexponent = exponent.intValueExact(); } catch (ArithmeticException e) { final NumberFormatException e2 = new NumberFormatException(e.getMessage()); // make sure this operation doesn't shadow the exception to be thrown try { e2.initCause(e); } catch (Throwable e3) { // ignored } throw e2; } exponent = valueOf(radix).power(iexponent); if (negtexp) { exponent = exponent.invert(); } // handle "Ex", "+Ex", "-Ex", as "1Ex" if (sm.equals("") || sm.equals("+")) { sm = "1"; } else if (sm.equals("-")) { sm = "-1"; } // [multiply()'s outcome is already normalized, // so there would be no need to normalize [again]] normalizeFrom( (new BigRational(sm, radix)).multiply(exponent)); return; } // '.': next higher precedence, not associative // (can't have more than one dot) String si, sf; final int dot = s.indexOf('.'); if (dot != -1) { // "[+-]i.f" si = s.substring(0, dot); sf = s.substring(dot + 1); } else { // "[+-]i". [not just delegating to BigInteger.] si = s; sf = ""; } // check for multiple signs or embedded signs checkNumberFormat(si); // skip '+' // skip '+'. [BigInteger [likely] doesn't handle these.] if (si.length() > 0 && si.charAt(0) == '+') { si = si.substring(1); } // handle '-' boolean negt = false; if (si.length() > 0 && si.charAt(0) == '-') { negt = true; si = si.substring(1); } // handle ".x" as "0.x" ("." as "0.0") // handle "" as "0" // note: "0" is treated specially and optimized // in bigIntegerValueOf(String,int). if (si.equals("")) { si = "0"; } BigInteger n = bigIntegerValueOf(si, radix); BigInteger q; // includes the cases "x." and "." if (!sf.equals("")) { // check for signs checkFractionFormat(sf); final BigInteger f = bigIntegerValueOf(sf, radix); final int scale = sf.length(); q = bigIntegerPower(bigIntegerValueOf(radix), scale); n = bigIntegerMultiply(n, q).add(f); } else { q = BIG_INTEGER_ONE; } if (negt) { n = n.negate(); } normalizeFrom(n, q); } // javadoc: in-sync with the radix parameter version /** Construct a BigRational from a string representation, * with default radix (10). * <P> * The supported string formats are * "[+-]n", "[+-]n/[+-]q", "[+-]i.f", "[+-]i", * "[+-]iE[+-]e", "[+-]i.fE[+-]e" (latter two only with radix<=10, * due to possible ambiguities); n and q can be any of the latter * (i.e. mixed representations such as "-1.2E-3/-4.5E-6" are supported). * <P> * Samples: "-21/35", "3.4", "-65.4E-3", * "f/37" (base 16), "1011101011010110" (base 2). * <P> * [Exponential representation wasn't supported in an earlier version.] */ public BigRational(String s) { this(s, DEFAULT_RADIX); } /** Construct a BigRational from an unscaled value and a scale value. */ public BigRational(BigInteger unscaledValue, int scale, int radix) { if (unscaledValue == null) { throw new NumberFormatException("null"); } final boolean negt = (scale < 0); if (negt) { scale = -scale; } checkRadix(radix); final BigInteger scaleValue = bigIntegerPower(bigIntegerValueOf(radix), scale); normalizeFrom( (negt ? bigIntegerMultiply(unscaledValue, scaleValue) : unscaledValue), (negt ? BIG_INTEGER_ONE : scaleValue)); } /** Construct a BigRational from an unscaled value and a scale value, * default radix (10). */ public BigRational(BigInteger unscaledValue, int scale) { this(unscaledValue, scale, DEFAULT_RADIX); } /** Construct a BigRational from an unscaled fix number value * and a scale value. */ public BigRational(long unscaledValue, int scale, int radix) { this(bigIntegerValueOf(unscaledValue), scale, radix); } /** Construct a BigRational from a [IEEE 754] double [size/precision] * floating point number. */ public BigRational(double x) { // // [rounding step, possible loss of precision step] // this (String.valueOf(x)); normalizeFrom(valueOfDoubleBits(Double.doubleToLongBits(x))); } /** Construct a BigRational from a [IEEE 754] single [size/precision] * floating point number. */ public BigRational(float x) { // this ((double)x); normalizeFrom(valueOfFloatBits(Float.floatToIntBits(x))); } // can't have public BigRational(long unscaledValue, int scale) // as alias for BigRational(unscaledValue, scale, DEFAULT_RADIX); // it's too ambiguous with public BigRational(long n, long q). /** Normalize BigRational. * Denominator will be positive, numerator and denominator will have * no common divisor. * BigIntegers -1, 0, 1 will be set to constants for later comparison speed. */ private void normalize() { // note: don't call anything that depends on a normalized this. // i.e.: don't call most (or all) of the BigRational methods. if (m_n == null || m_q == null) { throw new NumberFormatException("null"); } // [these are typically cheap.] int ns = m_n.signum(); int qs = m_q.signum(); // note: we don't throw on qs==0. that'll be done elsewhere. // if (qs == 0) { // throw new NumberFormatException("quotient zero"); // } if (ns == 0 && qs == 0) { // [typically not reached, due to earlier tests.] // [both for speed] m_n = BIG_INTEGER_ZERO; m_q = BIG_INTEGER_ZERO; return; } if (ns == 0) { m_q = BIG_INTEGER_ONE; // [for speed] m_n = BIG_INTEGER_ZERO; return; } if (qs == 0) { // [typically not reached, due to earlier tests.] m_n = BIG_INTEGER_ONE; // [for speed] m_q = BIG_INTEGER_ZERO; return; } // optimization // check the frequent case of q==1, for speed. // note: this only covers the normalized-for-speed 1-case. if (m_q == BIG_INTEGER_ONE) { // [for [later] speed] m_n = bigIntegerValueOf(m_n); return; } // optimization // check the symmetric case too, for speed. // note: this only covers the normalized-for-speed 1-case. if ( (m_n == BIG_INTEGER_ONE || m_n == BIG_INTEGER_MINUS_ONE) && qs > 0) { // [for [later] speed] m_q = bigIntegerValueOf(m_q); return; } // setup torn apart for speed BigInteger na = m_n; BigInteger qa = m_q; if (qs < 0) { m_n = m_n.negate(); m_q = m_q.negate(); ns = -ns; qs = -qs; qa = m_q; if (ns > 0) { na = m_n; } } else { if (ns < 0) { na = m_n.negate(); } } final BigInteger g = na.gcd(qa); // test: optimization (body: not) if (!bigIntegerIsOne(g)) { m_n = m_n.divide(g); m_q = m_q.divide(g); } // for [later] speed, and normalization generally m_n = bigIntegerValueOf(m_n); m_q = bigIntegerValueOf(m_q); } /** Normalize BigRational. * [Convenience method to normalize(void).] */ private void normalizeFrom(BigInteger n, BigInteger q) { m_n = n; m_q = q; normalize(); } /** Normalize BigRational. * [Convenience method to normalize(void).] */ private void normalizeFrom(BigRational that) { if (that == null) { throw new NumberFormatException("null"); } normalizeFrom(that.m_n, that.m_q); } /** Check constraints on radixes. * Radix may not be negative or less than two. */ private static void checkRadix(int radix) { if (radix < 0) { throw new NumberFormatException("radix negative"); } if (radix < 2) { throw new NumberFormatException("radix too small"); } // note: we don't check for "radix too large"; // that's left to BigInteger.toString(radix) // [i.e.: we don't really mind whether the underlying // system supports base36, or base62, or even more] } /** Check some of the integer format constraints. */ private static void checkNumberFormat(String s) { // "x", "-x", "+x", "", "-", "+" if (s == null) { throw new NumberFormatException("null"); } // note: 'embedded sign' catches both-signs cases too. final int p = s.indexOf('+'); final int m = s.indexOf('-'); final int pp = (p == -1 ? -1 : s.indexOf('+', p + 1)); final int mm = (m == -1 ? -1 : s.indexOf('-', m + 1)); if ( (p != -1 && p != 0) || (m != -1 && m != 0) || pp != -1 || mm != -1) { // embedded sign. this covers the both-signs case. throw new NumberFormatException("embedded sign"); } } /** Check number format for fraction part. */ private static void checkFractionFormat(String s) { if (s == null) { throw new NumberFormatException("null"); } if (s.indexOf('+') != -1 || s.indexOf('-') != -1) { throw new NumberFormatException( "sign in fraction"); } } /** Check number input for Java's string representations * of doubles/floats that are unsupported: * "NaN" and "Infinity" (with or without sign). */ private static void checkNaNAndInfinity(String s, int radix) { // the strings may actually be valid given a large enough radix // (e.g. base 36), so limit the radix/check if (radix > 16) { return; } // [null and empty string check] final int length = (s == null ? 0 : s.length()); if (length < 1) { return; } // optimization (String.equals and even more String.equalsIgnoreCase // are quite expensive, charAt and switch aren't) // test for last character in strings below, both cases switch (s.charAt(length - 1)) { case 'N': case 'n': case 'y': case 'Y': break; default: return; } if (s.equalsIgnoreCase("NaN") || s.equalsIgnoreCase("Infinity") || s.equalsIgnoreCase("+Infinity") || s.equalsIgnoreCase("-Infinity")) { throw new NumberFormatException( s); } } /** Check constraints on radixes. * [Convenience method to checkRadix(radix).] */ private static void checkRadixArgument(int radix) { try { checkRadix(radix); } catch (Exception e) { throw new IllegalArgumentException(e.getMessage()); } } /** Proxy to BigInteger.valueOf(). * Speeds up comparisons by using constants. */ private static BigInteger bigIntegerValueOf(long n) { // return the internal constants used for checks if possible. // optimization // check whether it's outside int range. // actually check a much narrower range, fitting the switch below. if (n >= -16 && n <= 16) { // note: test above needed to make the cast below safe // jump table, for speed switch ((int) n) { case 0: return BIG_INTEGER_ZERO; case 1: return BIG_INTEGER_ONE; case -1: return BIG_INTEGER_MINUS_ONE; case 2: return BIG_INTEGER_TWO; case -2: return BIG_INTEGER_MINUS_TWO; case 10: return BIG_INTEGER_TEN; case 16: return BIG_INTEGER_SIXTEEN; } } return BigInteger.valueOf(n); } /** Convert BigInteger to its constant if possible. * Speeds up later comparisons by using constants. */ private static BigInteger bigIntegerValueOf(BigInteger n) { // note: these tests are quite expensive, // so they should be minimized to a reasonable amount. // priority in the tests: 1, 0, -1; // two phase testing. // cheap tests first. // optimization if (n == BIG_INTEGER_ONE) { return n; } // optimization if (n == BIG_INTEGER_ZERO) { // [typically not reached, since zero is handled specially.] return n; } // optimization if (n == BIG_INTEGER_MINUS_ONE) { return n; } // more expensive tests later. // optimization if (n.equals(BIG_INTEGER_ONE)) { return BIG_INTEGER_ONE; } // optimization if (n.equals(BIG_INTEGER_ZERO)) { // [typically not reached from normalize().] return BIG_INTEGER_ZERO; } // optimization if (n.equals(BIG_INTEGER_MINUS_ONE)) { return BIG_INTEGER_MINUS_ONE; } // note: BIG_INTEGER_TWO et al. _not_ used for checks // and therefore not replaced by constants_here_. // this speeds up tests. // not a known constant return n; } /** Proxy to (new BigInteger()). * Speeds up comparisons by using constants. */ private static BigInteger bigIntegerValueOf(String s, int radix) { // note: mind the radix. // however, 0/1/-1 are not a problem. // _often_ used strings (e.g. 0 for empty fraction and // 1 for empty denominator), for speed. // optimization if (s.equals("1")) { return BIG_INTEGER_ONE; } // optimization if (s.equals("0")) { return BIG_INTEGER_ZERO; } // optimization if (s.equals("-1")) { // typically not reached, due to [private] usage pattern, // i.e. the sign is cut before return BIG_INTEGER_MINUS_ONE; } // note: BIG_INTEGER_TWO et al. _not_ used for checks // and therefore even less valuable. // there's a tradeoff between speeds of these tests // and being consistent in using all constants // (at least with the common radixes). // optimization if (radix > 2) { if (s.equals("2")) { return BIG_INTEGER_TWO; } if (s.equals("-2")) { // typically not reached, due to [private] usage pattern, // i.e. the sign is cut before return BIG_INTEGER_MINUS_TWO; } } // optimization if (s.equals("10")) { switch (radix) { case 2: return BIG_INTEGER_TWO; case 10: return BIG_INTEGER_TEN; case 16: return BIG_INTEGER_SIXTEEN; } } // optimization if (radix == 10 && s.equals("16")) { return BIG_INTEGER_SIXTEEN; } // note: not directly finding the other [radix'] representations // of 10 and 16 in the code above // use the constants if possible return bigIntegerValueOf(new BigInteger(s, radix)); } /** Proxy to BigInteger.equals(). * For speed. */ private static boolean bigIntegerEquals(BigInteger n, BigInteger m) { // optimization // first test is for speed. if (n == m) { return true; } return n.equals(m); } /** Zero (0) value predicate. * [For convenience and speed.] */ private static boolean bigIntegerIsZero(BigInteger n) { // optimization // first test is for speed. if (n == BIG_INTEGER_ZERO) { return true; } // well, this is also optimized for speed a bit. return (n.signum() == 0); } /** One (1) value predicate. * [For convenience and speed.] */ private static boolean bigIntegerIsOne(BigInteger n) { // optimization // first test is for speed. if (n == BIG_INTEGER_ONE) { return true; } return bigIntegerEquals(n, BIG_INTEGER_ONE); } /** Minus-one (-1) value predicate. * [For convenience and speed.] */ private static boolean bigIntegerIsMinusOne(BigInteger n) { // optimization // first test is for speed. if (n == BIG_INTEGER_MINUS_ONE) { return true; } return bigIntegerEquals(n, BIG_INTEGER_MINUS_ONE); } /* / ** Positive value predicate. * / private static boolean bigIntegerIsPositive(BigInteger n) { return (n.signum() > 0); } */ /** Negative value predicate. */ private static boolean bigIntegerIsNegative(BigInteger n) { return (n.signum() < 0); } /** Proxy to BigInteger.multiply(). * For speed. * The more common cases of integers (q == 1) are optimized. */ private static BigInteger bigIntegerMultiply(BigInteger n, BigInteger m) { // optimization: one or both operands are zero. if (bigIntegerIsZero(n) || bigIntegerIsZero(m)) { return BIG_INTEGER_ZERO; } // optimization: second operand is one (i.e. neutral element). if (bigIntegerIsOne(m)) { return n; } // optimization: first operand is one (i.e. neutral element). if (bigIntegerIsOne(n)) { return m; } // optimization if (bigIntegerIsMinusOne(m)) { // optimization if (bigIntegerIsMinusOne(n)) { // typically not reached due to earlier test(s) return BIG_INTEGER_ONE; } return n.negate(); } // optimization if (bigIntegerIsMinusOne(n)) { // [m is not -1, see test above] return m.negate(); } // default case. [this would handle all cases.] return n.multiply(m); } /** Proxy to BigInteger.pow(). * For speed. */ private static BigInteger bigIntegerPower(BigInteger n, int exponent) { // generally expecting exponent>=0 // (there's nor much use in inverting in the integer domain) // the checks for exponent<0 below are done all the same // optimization // jump table, for speed. switch (exponent) { case 0: if (bigIntegerIsZero(n)) { // typically not reached, due to earlier test / [private] usage pattern throw new ArithmeticException("zero exp zero"); } return BIG_INTEGER_ONE; case 1: return n; } // optimization if (bigIntegerIsZero(n) && exponent > 0) { // note: exponent==0 already handled above // typically not reached, due to earlier test return BIG_INTEGER_ZERO; } // optimization if (bigIntegerIsOne(n)) { return BIG_INTEGER_ONE; } // optimization if (bigIntegerIsMinusOne(n)) { return (exponent % 2 == 0 ? BIG_INTEGER_ONE : BIG_INTEGER_MINUS_ONE); } return n.pow(exponent); } /** Binary logarithm rounded towards floor (towards negative infinity). */ // @PrecisionLoss private static int bigIntegerLogarithm2(BigInteger n) { if (bigIntegerIsZero(n)) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("logarithm of zero"); } if (bigIntegerIsNegative(n)) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("logarithm of negative number"); } // take this as a start // (don't wholly rely on bitLength() having the same meaning as log2) int exponent = n.bitLength() - 1; if (exponent < 0) { exponent = 0; } BigInteger p = BIG_INTEGER_TWO.pow(exponent + 1); while (n.compareTo(p) >= 0) { // typically not reached p = p.multiply(BIG_INTEGER_TWO); exponent++; } p = p.divide(BIG_INTEGER_TWO); while (n.compareTo(p) < 0) { // typically not reached p = p.divide(BIG_INTEGER_TWO); exponent--; } // [possible loss of precision step] return exponent; } /** Proxy to BigInteger.toString(int radix). */ private static String stringValueOf(BigInteger n, int radix) { return n.toString(radix); } /** Proxy to stringValueOf(bigIntegerValueOf(long), radix); * take the same route to format [long/bigint] integer numbers * [despite the overhead]. */ private static String stringValueOf(long n, int radix) { return stringValueOf(bigIntegerValueOf(n), radix); } /** Convert a IEEE 754 floating point number * (of different sizes, as array of longs, big endian) * to a BigRational. */ private static BigRational fromIeee754(long[] value0, int fractionSize, int exponentSize) { if (value0 == null) { throw new NumberFormatException("null"); } // note: the long(s) in the input array are considered unsigned, // so expansion operations (to e.g. BigRational) and [right-] shift operations // (unlike assignment, equality-test, narrowing, and/or operations) // must be appropriately chosen BigRational fraction0 = ZERO; // start at the little end of the [bigendian] input int i = value0.length - 1; while (fractionSize >= 64) { if (i < 0) { throw new NumberFormatException("not enough bits"); } // mind the long (value0[i]) being unsigned fraction0 = fraction0.add(valueOfUnsigned(value0[i])).divide(TWO_POWER_64); fractionSize -= 64; i--; } // the rest must now fit into value0[0] (a long), // i.e. we don't support exponentSize > 63 at the moment; // as the power() method accepts ints (not longs), // the restriction is actually even on <= 31 bits if (i < 0) { throw new NumberFormatException("no bits"); } if (i > 0) { throw new NumberFormatException("excess bits"); } long value = value0[0]; // [fractionSize [now is] < 64 by loop above] final long fractionMask = ((long) 1 << fractionSize) - 1; final long rawFraction = value & fractionMask; value >>>= fractionSize; // [exponentSize < 32 by [private] usage pattern; rawExponent < 2**31] final int exponentMask = (1 << exponentSize) - 1; final int exponentBias = (1 << (exponentSize - 1)) - 1; final int rawExponent = (int) value & exponentMask; value >>>= exponentSize; final int signSize = 1; final int signMask = (1 << signSize) - 1; // 1 final int rawSign = (int) value & signMask; value >>>= signSize; if (value != 0) { throw new NumberFormatException("excess bits"); } // check for Infinity and NaN (IEEE 754 rawExponent at its maximum) if (rawExponent == exponentMask) { // (no fraction bits means one of the Infinities; else NaN) throw new NumberFormatException(rawFraction == 0 && fraction0.isZero() ? (rawSign == 0 ? "Infinity" : "-Infinity") : "NaN"); } // optimization -- avoids power() calculation below // (isZero and zero multiply) are cheap // check for zero (IEEE 754 rawExponent zero and no fraction bits) if (rawExponent == 0 && rawFraction == 0 && fraction0.isZero()) { return ZERO; } // handle subnormal numbers too (with rawExponent==0) // [fractionSize [still is] < 64] final long mantissa1 = rawFraction | (rawExponent == 0 ? (long) 0 : (long) 1 << fractionSize); // mind mantissa1 being unsigned final BigRational mantissa = fraction0.add(valueOfUnsigned(mantissa1)); // (subnormal numbers; exponent is one off) // [rawExponent < 2**31; exponentBias < 2**30] final int exponent = rawExponent - exponentBias + (rawExponent == 0 ? 1 : 0) - fractionSize; final int sign = (rawSign == 0 ? 1 : -1); return valueOf(2).power(exponent).multiply(mantissa).multiply(sign); } /** Convert a BigRational to a IEEE 754 floating point number * (of different sizes, as array of longs, big endian). * <P> * Possible loss of precision. */ // @PrecisionLoss private static long[] toIeee754(BigRational value, int fractionSize, int exponentSize) { if (value == null) { throw new NumberFormatException("null"); } // [needed size: fractionSize+exponentSize+1; round up bits to a multiple of 64] final long[] out0 = new long[ (fractionSize + exponentSize + 1 + (64 - 1)) / 64]; if (value.isZero()) { // 0.0 // note: as we don't keep a sign with our ZERO, // we never return IEEE 754 -0.0 here for (int j = 0; j < out0.length; j++) { out0[j] = 0; } return out0; } final boolean negt = value.isNegative(); if (negt) { value = value.negate(); } // need to scale to this to get the full mantissa int exponent = fractionSize; final BigRational lower = valueOf(2).power(fractionSize); final BigRational upper = lower.multiply(2); // optimization, and a good guess (but not exact in all cases) final int scale = lower.divide(value).logarithm2(); value = value.multiply(valueOf(2).power(scale)); exponent -= scale; while (value.compareTo(lower) < 0) { // [typically done zero or one time] value = value.multiply(2); exponent--; } while (value.compareTo(upper) >= 0) { // [typically not reached] value = value.divide(2); exponent++; } // [rounding step, possible loss of precision step] BigInteger mantissa = value.bigIntegerValue(); // adjust after [unfortunate] mantissa rounding if (upper.compareTo(mantissa) <= 0) { mantissa = mantissa.divide(BIG_INTEGER_TWO); exponent++; } // start [to fill] at the little end of the [bigendian] output int i = out0.length - 1; int fractionSize1 = fractionSize; while (fractionSize1 >= 64) { final BigInteger[] divrem = mantissa .divideAndRemainder(BIG_INTEGER_TWO_POWER_64); // [according to BigInteger javadoc] this takes the least significant 64 bits; // i.e. in this case the long is considered unsigned, as we want it out0[i] = divrem[1].longValue(); fractionSize1 -= 64; mantissa = divrem[0]; i--; } // the rest must now fit into out0[0] if (i < 0) { // not reached throw new NumberFormatException("too many bits"); } if (i > 0) { // not reached throw new NumberFormatException("not enough bits"); } long fraction = mantissa.longValue(); final int exponentBias = (1 << (exponentSize - 1)) - 1; exponent += exponentBias; final int maximalExponent = (1 << exponentSize) - 1; if (exponent >= maximalExponent) { // overflow // throw new NumberFormatException("overflow"); // [positive or negative] infinity exponent = maximalExponent; fraction = 0; for (int j = 1; j < out0.length; j++) { out0[j] = 0; } // [keep sign] } else if (exponent <= 0) { // handle subnormal numbers too // [with know loss of precision] // drop one bit, while keeping the exponent int s = 1; // [need not shift more than fractionSize] final int n = (-exponent > fractionSize ? fractionSize : -exponent); s += n; exponent += n; // [possible loss of precision step] fraction = shiftrx(fraction, out0, 1, s); boolean zero = (fraction == 0); for (int j = 1; zero && j < out0.length; j++) { zero = (out0[j] == 0); } if (zero) { // underflow // throw new NumberFormatException("underflow"); // 0.0 or -0.0; i.e.: keep sign exponent = 0; // [nonzero == 0 implies the rest of the fraction is zero as well] } } // cut implied most significant bit // [unless with subnormal numbers] if (exponent != 0) { fraction &= ~ ((long) 1 << fractionSize1); } long out = 0; out |= (negt ? 1 : 0); out <<= exponentSize; out |= exponent; out <<= fractionSize1; out |= fraction; out0[0] = out; return out0; } /** Shift right, while propagating shifted bits (long[] is bigendian). */ private static long shiftrx(long a, long[] b, int boff, int n) { while (n > 0) { final int n2 = (n < 63 ? n : 63); final long m = ((long) 1 << n2) - 1; long c = a & m; a >>>= n2; for (int i = boff; i < b.length; i++) { final long t = b[i] & m; b[i] >>>= n2; b[i] |= (c << (64 - n2)); c = t; } n -= n2; } return a; } // note: don't use valueOf() here; valueOf implementations use the constants /** The constant zero (0). */ // [Constant name: see class BigInteger.] public final static BigRational ZERO = new BigRational(0); /** The constant one (1). */ // [Name: see class BigInteger.] public final static BigRational ONE = new BigRational(1); /** The constant minus-one (-1). */ public final static BigRational MINUS_ONE = new BigRational(-1); // some more constants, often used as radixes/bases /** The constant two (2). */ private final static BigRational TWO = new BigRational(2); /** The constant ten (10). */ private final static BigRational TEN = new BigRational(10); /** The constant sixteen (16). */ private final static BigRational SIXTEEN = new BigRational(16); // some more constants /** The constant two to the power of 64 (18446744073709551616). * Used is slicing larger [than double size] IEEE 754 values. */ private final static BigRational TWO_POWER_64 = new BigRational( BIG_INTEGER_TWO_POWER_64); /** Positive predicate. * <P> * Indicates whether this BigRational is larger than zero. * Zero is not positive. * <P> * [For convenience.] */ public boolean isPositive() { return (signum() > 0); } /** Negative predicate. * <P> * Indicates whether this BigRational is smaller than zero. * Zero isn't negative either. * <P> * [For convenience.] */ public boolean isNegative() { return (signum() < 0); } /** Zero predicate. * <P> * Indicates whether this BigRational is zero. * <P> * [For convenience and speed.] */ public boolean isZero() { // optimization // first test is for speed. if (this == ZERO || m_n == BIG_INTEGER_ZERO) { return true; } // well, this is also optimized for speed a bit. return (signum() == 0); } /** One predicate. * <P> * Indicates whether this BigRational is 1. * <P> * [For convenience and speed.] */ public boolean isOne() { // optimization // first test is for speed. if (this == ONE) { return true; } return equals(ONE); } /** Minus-one predicate. * <P> * Indicates whether this BigRational is -1. * <P> * [For convenience and speed.] */ public boolean isMinusOne() { // optimization // first test is for speed. if (this == MINUS_ONE) { return true; } return equals(MINUS_ONE); } /** Integer predicate. * <P> * Indicates whether this BigRational convertible to a BigInteger * without loss of precision. * True iff quotient is one. */ public boolean isInteger() { return bigIntegerIsOne(m_q); } /** BigRational string representation, * format "[-]n[/q]". * <P> * Sample output: "6172839/5000". */ public String toString(int radix) { checkRadixArgument(radix); final String s = stringValueOf(m_n, radix); if (isInteger()) { return s; } return s + "/" + stringValueOf(m_q, radix); } /** BigRational string representation, * format "[-]n[/q]", * default radix (10). * <P> * Default string representation, * as rational, not using an exponent. * <P> * Sample output: "6172839/5000". * <P> * Overwrites Object.toString(). */ @Override public String toString() { return toString(DEFAULT_RADIX); } /** Fixed dot-format "[-]i.f" string representation, * with a precision. * <P> * Precision may be negative, * in which case the rounding affects digits left of the dot, * i.e. the integer part of the number, as well. * <P> * Sample output: "1234.567800". * <P> * Possible loss of precision. */ // @PrecisionLoss public String toStringDot(int precision, int radix) { return toStringDot(precision, radix, false); } /** Fixed dot-format "[-]i.f" string representation, * with a precision. * <P> * Precision may be negative, * in which case the rounding affects digits left of the dot, * i.e. the integer part of the number, as well. * <P> * The exponentFormat parameter allows for shorter [intermediate] * string representation, an optimization, e.g. used with toStringExponent. * <P> * Possible loss of precision. */ // @PrecisionLoss private String toStringDot(int precision, int radix, boolean exponentFormat) { checkRadixArgument(radix); BigRational scaleValue = new BigRational(bigIntegerPower(bigIntegerValueOf(radix), (precision < 0 ? -precision : precision))); if (precision < 0) { scaleValue = scaleValue.invert(); } // default round mode. // [rounding step, possible loss of precision step] BigRational n = multiply(scaleValue).round(); final boolean negt = n.isNegative(); if (negt) { n = n.negate(); } String s = n.toString(radix); if (exponentFormat) { // note that this is _not_ the scientific notation // (one digit left of the dot exactly), // but some intermediate representation suited for post processing // [leaving away the left/right padding steps // is more performant in time and memory space] s = s + "E" + stringValueOf(-precision, radix); } else { if (precision >= 0) { // left-pad with '0' while (s.length() <= precision) { s = "0" + s; } final int dot = s.length() - precision; final String i = s.substring(0, dot); final String f = s.substring(dot); s = i; if (f.length() > 0) { s = s + "." + f; } } else { if (!s.equals("0")) { // right-pad with '0' for (int i = -precision; i > 0; i--) { s = s + "0"; } } } } // add sign if (negt) { s = "-" + s; } return s; } /** Dot-format "[-]i.f" string representation, * with a precision, * default radix (10). * Precision may be negative. * <P> * Sample output: "1234.567800". * <P> * Possible loss of precision. */ // @PrecisionLoss public String toStringDot(int precision) { // [possible loss of precision step] return toStringDot(precision, DEFAULT_RADIX, false); } // note: there is no 'default' precision. /** Dot-format "[-]i.f" string representation, * with a relative precision. * <P> * If the relative precision is zero or negative, * "0" will be returned (i.e. total loss of precision). * <P> * Possible loss of precision. */ // @PrecisionLoss public String toStringDotRelative(int precision, int radix) { // kind of expensive, due to expensive logarithm implementation // (with unusual radixes), and post processing checkRadixArgument(radix); // zero doesn't have any significant digits if (isZero()) { return "0"; } // relative precision zero [or less means]: no significant digits at all, i.e. 0 // [loss of precision step] if (precision <= 0) { return "0"; } // guessed [see below: rounding issues] length: sign doesn't matter; // one digit more than logarithm final int guessedLength = absolute().logarithm(radix) + 1; // [possible loss of precision step] String s = toStringDot(precision - guessedLength, radix); // [floor of] logarithm and [arithmetic] rounding [half-up] // need post-processing: // find first significant digit and check for dot boolean dot = false; int i; for (i = 0; i < s.length(); i++) { final char c = s.charAt(i); if (c == '.') { dot = true; } // expecting nothing than '-', '.', and digits if (c != '-' && c != '.' && c != '0') { break; } } // count digits / [still] check for dot int digits = 0; for (; i < s.length(); i++) { if (s.charAt(i) == '.') { dot = true; } else { digits++; } } // cut excess zeros // expecting at most 1 excess zero, e.g. for "0.0099999" final int excess = digits - precision; if (dot && excess > 0) { s = s.substring(0, s.length() - excess); } return s; } /** Dot-format "[-]i.f" string representation, * with a relative precision, * default radix (10). * <P> * If the relative precision is zero or negative, * "0" will be returned (i.e. total loss of precision). * <P> * Possible loss of precision. */ // @PrecisionLoss public String toStringDotRelative(int precision) { // [possible loss of precision step] return toStringDotRelative(precision, DEFAULT_RADIX); } /** Exponent-format string representation, * with a relative precision, * "[-]i[.f]E[-]e" (where i is one digit other than 0 exactly; * f has no trailing 0); * <P> * Sample output: "1.2E3". * <P> * Possible loss of precision. */ // @PrecisionLoss public String toStringExponent(int precision, int radix) { checkRadixArgument(radix); // zero doesn't have any significant digits if (isZero()) { return "0"; } // relative precision zero [or less means]: no significant digits at all, i.e. 0 // [loss of precision step] if (precision <= 0) { return "0"; } // guessed [see below: rounding issues] length: sign doesn't matter; // one digit more than logarithm final int guessedLength = absolute().logarithm(radix) + 1; // [possible loss of precision step] final String s = toStringDot(precision - guessedLength, radix, true); return toExponentRepresentation(s, radix); } /** Exponent-format string representation, * with a relative precision, * default radix (10), * "[-]i[.f]E[-]e" (where i is one digit other than 0 exactly; * f has no trailing 0); * <P> * Sample output: "1.2E3". * <P> * Possible loss of precision. */ // @PrecisionLoss public String toStringExponent(int precision) { // [possible loss of precision step] return toStringExponent(precision, DEFAULT_RADIX); } /** Transform a [intermediate] dot representation * to an exponent-format representation. */ private static String toExponentRepresentation(String s, int radix) { // skip '+' if (s.length() > 0 && s.charAt(0) == '+') { // typically not reached, due to [private] usage pattern s = s.substring(1); } // handle '-' boolean negt = false; if (s.length() > 0 && s.charAt(0) == '-') { negt = true; s = s.substring(1); } // skip initial zeros while (s.length() > 0 && s.charAt(0) == '0') { s = s.substring(1); } // check for and handle exponent // handle only upper case 'E' (we know we use that in earlier steps); // this allows any base using lower case characters int exponent0 = 0; final int exp = s.indexOf('E'); if (exp != -1) { final String se = s.substring(exp + 1); s = s.substring(0, exp); exponent0 = (new BigRational(se, radix)).intValueExact(); } String si, sf; int exponent; final int dot = s.indexOf('.'); if (dot != -1) { if (dot == 0) { // possibly more insignificant digits s = s.substring(1); exponent = -1; while (s.length() > 0 && s.charAt(0) == '0') { s = s.substring(1); exponent--; } if (s.equals("")) { // typically not reached, due to [private] usage pattern return "0"; } // first significant digit si = s.substring(0, 1); sf = s.substring(1); } else { // initial [significant] digit si = s.substring(0, 1); sf = s.substring(1, dot); exponent = sf.length(); sf = sf + s.substring(dot + 1); } } else { // [note that we just cut the zeros above] if (s.equals("")) { return "0"; } // initial [significant] digit si = s.substring(0, 1); // rest sf = s.substring(1); exponent = sf.length(); } exponent += exponent0; // drop trailing zeros while (sf.length() > 0 && sf.charAt(sf.length() - 1) == '0') { sf = sf.substring(0, sf.length() - 1); } s = si; if (!sf.equals("")) { s = s + "." + sf; } if (exponent != 0) { s = s + "E" + stringValueOf(exponent, radix); } if (negt) { s = "-" + s; } return s; } /** Constant internally used, for speed. */ // calculated via BigRational((float)(StrictMath.log(10)/StrictMath.log(2))) // note: don't use float/double operations in this code though (except for test()) private final static BigRational LOGARITHM_TEN_GUESS = new BigRational(1741647, 524288); /** Constant internally used, for speed. */ private final static BigRational LOGARITHM_SIXTEEN = new BigRational(4); /** Return binary logarithm rounded towards floor (towards negative infinity). * <P> * Possible loss of precision. */ // @PrecisionLoss private int logarithm2() { if (isZero()) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("logarithm of zero"); } if (isNegative()) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("logarithm of negative number"); } final boolean inverted = (compareTo(ONE) < 0); final BigRational a = (inverted ? invert() : this); // [possible loss of precision step] final int log = bigIntegerLogarithm2(a.bigIntegerValue()); return (inverted ? - (log + 1) : log); } /** Return logarithm rounded towards floor (towards negative infinity). * <P> * Possible loss of precision. */ // @PrecisionLoss private int logarithm(int base) { // optimization if (base == 2) { return logarithm2(); } if (isZero()) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("logarithm of zero"); } if (isNegative()) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("logarithm of negative number"); } // if (base < 2) { // // [typically not reached, due to [private] usage pattern] // throw new ArithmeticException("bad base"); // } if (base < 0) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("negative base"); } if (base < 2) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("base too small"); } final boolean inverted = (compareTo(ONE) < 0); BigRational a = (inverted ? invert() : this); final BigRational bbase = valueOf(base); // optimization -- we could start from n=0 // initial guess // [base 2 handled earlier] // [unusual bases are handled a bit less performant] final BigRational lbase = (base == 10 ? LOGARITHM_TEN_GUESS : base == 16 ? LOGARITHM_SIXTEEN : valueOf(ilog2(base))); int n = valueOf(a.logarithm2()).divide(lbase).intValue(); a = a.divide(bbase.power(n)); // note that these steps are needed anyway: // LOGARITHM_TEN_GUESS above e.g. is (as the name suggests) // a guess only (since most logarithms usually can't be expressed // as rationals generally); odd bases or off even worse while (a.compareTo(bbase) >= 0) { a = a.divide(bbase); n++; } while (a.compareTo(ONE) < 0) { a = a.multiply(bbase); n--; } // [possible loss of precision step] return (inverted ? - (n + 1) : n); } /** Return binary logarithm of an int. */ private static int ilog2(int n) { if (n == 0) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("logarithm of zero"); } if (n < 0) { // [typically not reached, due to [private] usage pattern] throw new ArithmeticException("logarithm of negative number"); } int i = 0; // as this method is used in the context of [small] bases/radixes, // we expect less than 8 iterations at most, so no need to optimize while (n > 1) { n /= 2; i++; } return i; } /** Add a BigRational to this BigRational * and return a new BigRational. * <P> * If one of the operands is zero, * [as an optimization] * the other BigRational is returned. */ // [Name: see class BigInteger.] public BigRational add(BigRational that) { // optimization: second operand is zero (i.e. neutral element). if (that.isZero()) { return this; } // optimization: first operand is zero (i.e. neutral element). if (isZero()) { return that; } // note: not checking for that.equals(negate()), // since that would involve creation of a temporary object // note: the calculated n/q may be denormalized, // implicit normalize() is needed. // optimization: same denominator. if (bigIntegerEquals(m_q, that.m_q)) { return new BigRational(m_n.add(that.m_n), m_q); } // optimization: second operand is an integer. if (that.isInteger()) { return new BigRational(m_n.add(that.m_n.multiply(m_q)), m_q); } // optimization: first operand is an integer. if (isInteger()) { return new BigRational(m_n.multiply(that.m_q).add(that.m_n), that.m_q); } // default case. [this would handle all cases.] return new BigRational(m_n.multiply(that.m_q).add(that.m_n.multiply(m_q)), m_q.multiply(that.m_q)); } /** Add a long fix number integer to this BigRational * and return a new BigRational. */ public BigRational add(long that) { return add(valueOf(that)); } /** Subtract a BigRational from this BigRational * and return a new BigRational. * <P> * If the second operand is zero, * [as an optimization] * this BigRational is returned. */ // [Name: see class BigInteger.] public BigRational subtract(BigRational that) { // optimization: second operand is zero. if (that.isZero()) { return this; } // optimization: first operand is zero if (isZero()) { return that.negate(); } // optimization: operands are equal if (equals(that)) { return ZERO; } // note: the calculated n/q may be denormalized, // implicit normalize() is needed. // optimization: same denominator. if (bigIntegerEquals(m_q, that.m_q)) { return new BigRational( m_n.subtract(that.m_n), m_q); } // optimization: second operand is an integer. if (that.isInteger()) { return new BigRational( m_n.subtract(that.m_n.multiply(m_q)), m_q); } // optimization: first operand is an integer. if (isInteger()) { return new BigRational(m_n.multiply(that.m_q) .subtract(that.m_n), that.m_q); } // default case. [this would handle all cases.] return new BigRational(m_n.multiply(that.m_q).subtract(that.m_n.multiply(m_q)), m_q.multiply(that.m_q)); } /** Subtract a long fix number integer from this BigRational * and return a new BigRational. */ public BigRational subtract(long that) { return subtract(valueOf(that)); } /** An alias to subtract(). */ public BigRational sub(BigRational that) { return subtract(that); } /** An alias to subtract(). */ public BigRational sub(long that) { return subtract(that); } /** Multiply a BigRational to this BigRational * and return a new BigRational. * <P> * If one of the operands is one, * [as an optimization] * the other BigRational is returned. */ // [Name: see class BigInteger.] public BigRational multiply(BigRational that) { // optimization: one or both operands are zero. if (that.isZero() || isZero()) { return ZERO; } // optimization: second operand is 1. if (that.isOne()) { return this; } // optimization: first operand is 1. if (isOne()) { return that; } // optimization: second operand is -1. if (that.isMinusOne()) { return negate(); } // optimization: first operand is -1. if (isMinusOne()) { return that.negate(); } // note: the calculated n/q may be denormalized, // implicit normalize() is needed. return new BigRational(bigIntegerMultiply(m_n, that.m_n), bigIntegerMultiply(m_q, that.m_q)); } /** Multiply a long fix number integer to this BigRational * and return a new BigRational. */ public BigRational multiply(long that) { return multiply(valueOf(that)); } /** An alias to multiply(). */ public BigRational mul(BigRational that) { return multiply(that); } /** An alias to multiply(). */ public BigRational mul(long that) { return multiply(that); } /** Divide this BigRational through another BigRational * and return a new BigRational. * <P> * If the second operand is one, * [as an optimization] * this BigRational is returned. */ public BigRational divide(BigRational that) { if (that.isZero()) { throw new ArithmeticException("division by zero"); } // optimization: first operand is zero. if (isZero()) { return ZERO; } // optimization: second operand is 1. if (that.isOne()) { return this; } // optimization: first operand is 1. if (isOne()) { return that.invert(); } // optimization: second operand is -1. if (that.isMinusOne()) { return negate(); } // optimization: first operand is -1. if (isMinusOne()) { return that.invert().negate(); } // note: the calculated n/q may be denormalized, // implicit normalize() is needed. return new BigRational(bigIntegerMultiply(m_n, that.m_q), bigIntegerMultiply(m_q, that.m_n)); } /** Divide this BigRational through a long fix number integer * and return a new BigRational. */ public BigRational divide(long that) { return divide(valueOf(that)); } /** An alias to divide(). */ public BigRational div(BigRational that) { return divide(that); } /** An alias to divide(). */ public BigRational div(long that) { return divide(that); } /** Calculate this BigRational's integer power * and return a new BigRational. * <P> * The integer exponent may be negative. * <P> * If the exponent is one, * [as an optimization] * this BigRational is returned. */ public BigRational power(int exponent) { final boolean zero = isZero(); if (zero) { if (exponent == 0) { throw new ArithmeticException("zero exp zero"); } if (exponent < 0) { throw new ArithmeticException("division by zero"); } } // optimization if (exponent == 0) { return ONE; } // optimization // test for exponent<=0 already done if (zero) { return ZERO; } // optimization if (exponent == 1) { return this; } // optimization if (exponent == -1) { return invert(); } final boolean negt = (exponent < 0); if (negt) { exponent = -exponent; } final BigInteger n = bigIntegerPower(m_n, exponent); final BigInteger q = bigIntegerPower(m_q, exponent); // note: the calculated n/q are not denormalized in the sense // of having common factors, but n might be negative (and become q below) return new BigRational( (negt ? q : n), (negt ? n : q)); } /** An alias to power(). */ // [Name: see classes Math, BigInteger.] public BigRational pow(int exponent) { return power(exponent); } /** Calculate the remainder of this BigRational and another BigRational * and return a new BigRational. * <P> * The remainder result may be negative. * <P> * The remainder is based on round down (towards zero) / truncate. * 5/3 == 1 + 2/3 (remainder 2), 5/-3 == -1 + 2/-3 (remainder 2), * -5/3 == -1 + -2/3 (remainder -2), -5/-3 == 1 + -2/-3 (remainder -2). */ // [Name: see class BigInteger.] public BigRational remainder(BigRational that) { final int s = signum(); final int ts = that.signum(); if (ts == 0) { throw new ArithmeticException("division by zero"); } BigRational a = this; if (s < 0) { a = a.negate(); } // divisor's sign doesn't matter, as stated above. // this is also BigInteger's behavior, but don't let us be // dependent of a change in that. BigRational b = that; if (ts < 0) { b = b.negate(); } BigRational r = a.remainderOrModulusOfPositive(b); if (s < 0) { r = r.negate(); } return r; } /** Calculate the remainder of this BigRational and a long fix number integer * and return a new BigRational. */ public BigRational remainder(long that) { return remainder(valueOf(that)); } /** An alias to remainder(). */ public BigRational rem(BigRational that) { return remainder(that); } /** An alias to remainder(). */ public BigRational rem(long that) { return remainder(that); } /** Calculate the modulus of this BigRational and another BigRational * and return a new BigRational. * <P> * The modulus result may be negative. * <P> * Modulus is based on round floor (towards negative). * 5/3 == 1 + 2/3 (modulus 2), 5/-3 == -2 + -1/-3 (modulus -1), * -5/3 == -2 + 1/3 (modulus 1), -5/-3 == 1 + -2/-3 (modulus -2). */ public BigRational modulus(BigRational that) { final int s = signum(); final int ts = that.signum(); if (ts == 0) { throw new ArithmeticException("division by zero"); } BigRational a = this; if (s < 0) { a = a.negate(); } BigRational b = that; if (ts < 0) { b = b.negate(); } BigRational r = a.remainderOrModulusOfPositive(b); if (s < 0 && ts < 0) { r = r.negate(); } else if (ts < 0) { r = r.subtract(b); } else if (s < 0) { r = b.subtract(r); } return r; } /** Calculate the modulus of this BigRational and a long fix number integer * and return a new BigRational. */ public BigRational modulus(long that) { return modulus(valueOf(that)); } /** An alias to modulus(). */ // [Name: see class BigInteger.] public BigRational mod(BigRational that) { return modulus(that); } /** An alias to modulus(). */ public BigRational mod(long that) { return modulus(that); } /** Remainder or modulus of non-negative values. * Helper function to remainder() and modulus(). */ private BigRational remainderOrModulusOfPositive(BigRational that) { final int s = signum(); final int ts = that.signum(); if (s < 0 || ts < 0) { // typically not reached, due to [private] usage pattern throw new IllegalArgumentException("negative values(s)"); } if (ts == 0) { // typically not reached, due to [private] usage pattern throw new ArithmeticException("division by zero"); } // optimization if (s == 0) { return ZERO; } return new BigRational(bigIntegerMultiply(m_n, that.m_q).remainder( bigIntegerMultiply(m_q, that.m_n)), bigIntegerMultiply(m_q, that.m_q)); } /** Signum. * -1, 0, or 1. * <P> * If this BigRational is negative, -1 is returned; * if it is zero, 0 is returned; * if it is positive, 1 is returned. */ // [Name: see class BigInteger.] public int signum() { // note: m_q is positive. return m_n.signum(); } /** An alias to signum(). */ public int sign() { return signum(); } /** Return a new BigRational with the absolute value of this BigRational. * <P> * If this BigRational is zero or positive, * [as an optimization] * this BigRational is returned. */ public BigRational absolute() { if (signum() >= 0) { return this; } // optimization if (isMinusOne()) { return ONE; } // note: the calculated n/q are not denormalized, // implicit normalize() would not be needed. return new BigRational(m_n.negate(), m_q); } /** An alias to absolute(). * [Name: see classes Math, BigInteger.] */ public BigRational abs() { return absolute(); } /** Return a new BigRational with the negative value of this. * [Name: see class BigInteger.] */ public BigRational negate() { // optimization if (isZero()) { return ZERO; } // optimization if (isOne()) { return MINUS_ONE; } // optimization if (isMinusOne()) { return ONE; } // note: the calculated n/q are not denormalized, // implicit normalize() would not be needed. return new BigRational(m_n.negate(), m_q); } /** An alias to negate(). */ public BigRational neg() { return negate(); } /** Return a new BigRational with the inverted (reciprocal) value of this. */ public BigRational invert() { if (isZero()) { throw new ArithmeticException("division by zero"); } // optimization if (isOne() || isMinusOne()) { return this; } // note: the calculated n/q are not denormalized in the sense // of having common factors, but n might be negative (and become q below) return new BigRational(m_q, m_n); } /** An alias to invert(). */ public BigRational inv() { return invert(); } /** Return the minimal value of two BigRationals. */ public BigRational minimum(BigRational that) { return (compareTo(that) <= 0 ? this : that); } /** Return the minimal value of a BigRational and a long fix number integer. */ public BigRational minimum(long that) { return minimum(valueOf(that)); } /** An alias to minimum(). * [Name: see classes Math, BigInteger.] */ public BigRational min(BigRational that) { return minimum(that); } /** An alias to minimum(). */ public BigRational min(long that) { return minimum(that); } /** Return the maximal value of two BigRationals. */ public BigRational maximum(BigRational that) { return (compareTo(that) >= 0 ? this : that); } /** Return the maximum value of a BigRational and a long fix number integer. */ public BigRational maximum(long that) { return maximum(valueOf(that)); } /** An alias to maximum(). * [Name: see classes Math, BigInteger.] */ public BigRational max(BigRational that) { return maximum(that); } /** An alias to maximum(). */ public BigRational max(long that) { return maximum(that); } /** Compare object for equality. * Overwrites Object.equals(). * Semantic of equality to non-BigRational * changed from earlier version: * only BigRationals can be equal. * Never throws. * <P> * Overwrites Object.equals(Object). */ @Override public boolean equals(Object object) { // optimization if (object == this) { return true; } // test includes null if (! (object instanceof BigRational)) { return false; } final BigRational that = (BigRational) object; // optimization if (that.m_n == m_n && that.m_q == m_q) { return true; } return (bigIntegerEquals(that.m_n, m_n) && bigIntegerEquals(that.m_q, m_q)); // [old version] // // delegate to compareTo(Object) // try { // return (compareTo(object) == 0); // } catch (ClassCastException e) { // return false; // } } /** Hash code. * Overwrites Object.hashCode(). * <P> * Overwrites Object.hashCode(). */ @Override public int hashCode() { return ( (m_n.hashCode() + 1) * (m_q.hashCode() + 2)); } /** Compare this BigRational to another BigRational. */ public int compareTo(BigRational that) { // optimization if (that == this) { return 0; } final int s = signum(); final int t = that.signum(); if (s != t) { return (s < t ? -1 : 1); } // optimization: both zero. if (s == 0) { return 0; } // note: both m_q are positive. return bigIntegerMultiply(m_n, that.m_q).compareTo( bigIntegerMultiply(that.m_n, m_q)); } /** Compare this BigRational to a BigInteger. */ public int compareTo(BigInteger that) { return compareTo(valueOf(that)); } /** Compare this BigRational to a long. * <P> * Bytes, shorts, and ints can use this by being promoted to long. */ public int compareTo(long that) { return compareTo(valueOf(that)); } /** Compare this BigRational to an Object. * <P> * Object can be BigRational/BigInteger/Long/Integer/Short/Byte. * <P> * Implements Comparable.compareTo(Object) (JDK 1.2 and later). * <P> * A sample use is with a sorted map or set, e.g. TreeSet. * <P> * Only BigRational/BigInteger/Long/Integer objects allowed, * method will throw otherwise. * <P> * For backward compatibility reasons we keep compareTo(Object) * additionally to compareTo(BigRational). * Comparable<Object> is declared to be implemented * rather than Comparable<BigRational>. */ public int compareTo(Number object) { if (object instanceof Byte) { return compareTo( ((Byte) object).longValue()); } if (object instanceof Short) { return compareTo( ((Short) object).longValue()); } if (object instanceof Integer) { return compareTo( ((Integer) object).longValue()); } if (object instanceof Long) { return compareTo( ((Long) object).longValue()); } if (object instanceof BigInteger) { return compareTo((BigInteger) object); } // now assuming that it's either 'instanceof BigRational' // or it'll throw a ClassCastException. return compareTo((BigRational) object); } /** Number of explicit fraction bits * in an IEEE 754 double (binary64) float, * 52. */ private final static int DOUBLE_FLOAT_FRACTION_SIZE = 52; /** Number of exponent bits * in an IEEE 754 double (binary64) float, * 11. */ private final static int DOUBLE_FLOAT_EXPONENT_SIZE = 11; /** Number of explicit fraction bits * in an IEEE 754 single (binary32) float, * 23. */ private final static int SINGLE_FLOAT_FRACTION_SIZE = 23; /** Number of exponent bits * in an IEEE 754 single (binary32) float, * 8. */ private final static int SINGLE_FLOAT_EXPONENT_SIZE = 8; /** Number of explicit fraction bits * in an IEEE 754 half (binary16) float, * 10. */ private final static int HALF_FLOAT_FRACTION_SIZE = 10; /** Number of exponent bits * in an IEEE 754 half (binary16) float, * 5. */ private final static int HALF_FLOAT_EXPONENT_SIZE = 5; /** Number of explicit fraction bits * in an IEEE 754 quad (binary128, quadruple) float, * 112. */ private final static int QUAD_FLOAT_FRACTION_SIZE = 112; /** Number of exponent bits * in an IEEE 754 quad (binary128, quadruple) float, * 15. */ private final static int QUAD_FLOAT_EXPONENT_SIZE = 15; /** Convert to BigInteger, by rounding. * <P> * Possible loss of precision. */ // @PrecisionLoss public BigInteger bigIntegerValue() { // [rounding step, possible loss of precision step] return round().m_n; } /** Convert to long, by rounding and delegating to BigInteger. * Implements Number.longValue(). * As described with BigInteger.longValue(), this just returns * the low-order [64] bits (losing information about magnitude * and sign). * <P> * Possible loss of precision. * <P> * Overwrites Number.longValue(). */ @Override // @PrecisionLoss public long longValue() { // delegate to BigInteger. // [rounding step, possible loss of precision step] return bigIntegerValue().longValue(); } /** Convert to int, by rounding and delegating to BigInteger. * Implements Number.intValue(). * As described with BigInteger.longValue(), this just returns * the low-order [32] bits (losing information about magnitude * and sign). * <P> * Possible loss of precision. * <P> * Overwrites Number.intValue(). */ @Override // @PrecisionLoss public int intValue() { // delegate to BigInteger. // [rounding step, possible loss of precision step] return bigIntegerValue().intValue(); } /** Convert to double floating point value. * Implements Number.doubleValue(). * <P> * Possible loss of precision. * <P> * Overwrites Number.doubleValue(). */ @Override // @PrecisionLoss public double doubleValue() { // // [rounding step, possible loss of precision step] // return (m_n.doubleValue() / m_q.doubleValue()); // // [rounding step, possible loss of precision step] // return Double.parseDouble(toStringExponent(24)); return Double.longBitsToDouble( // [rounding step, possible loss of precision step] doubleBitsValue()); } /** Convert to single floating point value. * Implements Number.floatValue(). * <P> * Note that BigRational's [implicit] [default] rounding mode * that applies [too] on indirect double to BigRational to float * rounding (round-half-up) may differ from what's done in a direct * cast/coercion from double to float (e.g. round-half-even). * <P> * Possible loss of precision. * <P> * Overwrites Number.floatValue(). */ @Override // @PrecisionLoss public float floatValue() { // // [rounding step, possible loss of precision step] // return (float)doubleValue(); return Float.intBitsToFloat( // [rounding step, possible loss of precision step] singleBitsValue()); } /** Convert to IEEE 754 double float bits. * The bits can be converted to a double by Double.longBitsToDouble(). * <P> * Possible loss of precision. */ // @PrecisionLoss public long doubleBitsValue() { // [rounding step, possible loss of precision step] return (toIeee754(this, DOUBLE_FLOAT_FRACTION_SIZE, DOUBLE_FLOAT_EXPONENT_SIZE)[0]); } /** Convert to IEEE 754 single float bits. * The bits can be converted to a float by Float.intBitsToFloat(). * <P> * Possible loss of precision. */ // @PrecisionLoss public int floatBitsValue() { // [rounding step, possible loss of precision step] return (int) (toIeee754(this, SINGLE_FLOAT_FRACTION_SIZE, SINGLE_FLOAT_EXPONENT_SIZE)[0]); } /** An alias to floatBitsValue(). * <P> * Possible loss of precision. */ // @PrecisionLoss public int singleBitsValue() { // [rounding step, possible loss of precision step] return floatBitsValue(); } /** Convert this BigRational to IEEE 754 half float (binary16) bits. * <P> * As a short value is returned rather than a int, * care has to be taken no unwanted sign expansion * happens in subsequent operations, * e.g. by masking (x.halfBitsValue()&0xffffl) * or similar (x.halfBitsValue()==(short)0xbc00). * <P> * Possible loss of precision. */ // @PrecisionLoss public short halfBitsValue() { // [rounding step, possible loss of precision step] return (short) (toIeee754(this, HALF_FLOAT_FRACTION_SIZE, HALF_FLOAT_EXPONENT_SIZE)[0]); } /** Convert this BigRational to IEEE 754 quad float (binary128, quadruple) bits. * <P> * The bits are returned in an array of two longs, * big endian (higher significant long first). * <P> * Possible loss of precision. */ // @PrecisionLoss public long[] quadBitsValue() { // [rounding step, possible loss of precision step] return toIeee754(this, QUAD_FLOAT_FRACTION_SIZE, QUAD_FLOAT_EXPONENT_SIZE); } /** Convert this BigRational to a long integer, * either returning an exact result (no rounding or truncation needed), * or throw an ArithmeticException. */ public long longValueExact() { final long i = longValue(); // test is kind-of costly if (!equals(valueOf(i))) { throw new ArithmeticException(isInteger() ? "overflow" : "rounding necessary"); } return i; } /** Convert this BigRational to an int, * either returning an exact result (no rounding or truncation needed), * or throw an ArithmeticException. */ public int intValueExact() { final int i = intValue(); // test is kind-of costly if (!equals(valueOf(i))) { throw new ArithmeticException(isInteger() ? "overflow" : "rounding necessary"); } return i; } /** Convert this BigRational to its constant * (ONE, ZERO, MINUS_ONE) * if possible. */ public static BigRational valueOf(BigRational value) { if (value == null) { throw new NumberFormatException("null"); } // note: these tests are quite expensive, // but they are minimized to a reasonable amount. // priority in the tests: 1, 0, -1; // two phase testing. // cheap tests first. // optimization if (value == ONE) { return value; } // optimization if (value == ZERO) { return value; } // optimization if (value == MINUS_ONE) { return value; } // more expensive tests later. // optimization if (value.equals(ONE)) { return ONE; } // optimization if (value.equals(ZERO)) { return ZERO; } // optimization if (value.equals(MINUS_ONE)) { return MINUS_ONE; } // not a known constant return value; } /** Build a BigRational from a String. * <P> * [Roughly] equivalent to <CODE>new BigRational(value)</CODE>. */ public static BigRational valueOf(String value) { if (value == null) { throw new NumberFormatException("null"); } // optimization if (value.equals("0")) { return ZERO; } // optimization if (value.equals("1")) { return ONE; } // optimization if (value.equals("-1")) { return MINUS_ONE; } return new BigRational(value); } /** Build a BigRational from a BigInteger. * <P> * Equivalent to <CODE>new BigRational(value)</CODE>. */ public static BigRational valueOf(BigInteger value) { return new BigRational(value); } /** Build a BigRational from a long fix number integer. * <P> * [Roughly] equivalent to <CODE>new BigRational(value)</CODE>. * <P> * As an optimization, commonly used numbers are returned * as a reused constant. */ public static BigRational valueOf(long value) { // return the internal constants if possible // optimization // check whether it's outside int range. // actually check a much narrower range, fitting the switch below. if (value >= -16 && value <= 16) { // note: test above needed to make the cast below safe // jump table, for speed switch ((int) value) { case 0: return ZERO; case 1: return ONE; case -1: return MINUS_ONE; case 2: return TWO; case 10: return TEN; case 16: return SIXTEEN; } } return new BigRational(value); } // note: byte/short/int implicitly upgraded to long, // so strictly the additional implementations aren't needed; // with unsigned (below) they however are /** Build a BigRational from an int. */ public static BigRational valueOf(int value) { return valueOf((long) value); } /** Build a BigRational from a short. */ public static BigRational valueOf(short value) { return valueOf((long) value); } /** Build a BigRational from a byte. */ public static BigRational valueOf(byte value) { return valueOf((long) value); } /** Build a BigRational from a [IEEE 754] double [size/precision] * floating point number. */ public static BigRational valueOf(double value) { return new BigRational(value); } /** Build a BigRational from a [IEEE 754] single [size/precision] * floating point number. */ public static BigRational valueOf(float value) { return new BigRational(value); } /** Build a BigRational from an unsigned long fix number integer. * <P> * The resulting BigRational is positive, * i.e. the negative longs are mapped to 2**63..2**64 (exclusive). */ public static BigRational valueOfUnsigned(long value) { final BigRational b = valueOf(value); // mind the long being unsigned with highest significant // bit (bit#63) set (interpreted as negative by valueOf(long)) return (b.isNegative() ? b.add(TWO_POWER_64) : b); } /** Build a BigRational from an unsigned int. * <P> * The resulting BigRational is positive, * i.e. the negative ints are mapped to 2**31..2**32 (exclusive). */ public static BigRational valueOfUnsigned(int value) { // masking: suppress sign expansion return valueOf(value & 0xffffffffl); } /** Build a BigRational from an unsigned short. * <P> * The resulting BigRational is positive, * i.e. the negative shorts are mapped to 2**15..2**16 (exclusive). */ public static BigRational valueOfUnsigned(short value) { // masking: suppress sign expansion return valueOf(value & 0xffffl); } /** Build a BigRational from an unsigned byte. * <P> * The resulting BigRational is positive, * i.e. the negative bytes are mapped to 2**7..2**8 (exclusive). */ public static BigRational valueOfUnsigned(byte value) { // masking: suppress sign expansion return valueOf(value & 0xffl); } /** Build a BigRational from an IEEE 754 double size * (double precision, binary64) floating point number * represented as long. * <P> * An IEEE 754 double size (binary64) number uses * 1 bit for the sign, * 11 bits for the exponent, * and 52 bits (plus 1 implicit bit) for the fraction/mantissa. * The minimal exponent encodes subnormal nubers; * the maximal exponent encodes Infinities and NaNs. * <P> * Infinities and NaNs are not supported as BigRationals. * <P> * The conversion from the bits to a BigRational * is done without loss of precision. */ public static BigRational valueOfDoubleBits(long value) { return fromIeee754(new long[]{ value, }, DOUBLE_FLOAT_FRACTION_SIZE, DOUBLE_FLOAT_EXPONENT_SIZE); } /** Build a BigRational from an IEEE 754 single size * (single precision, binary32) floating point number * represented as int. * <P> * An IEEE 754 single size (binary32) number uses * 1 bit for the sign, * 8 bits for the exponent, * and 23 bits (plus 1 implicit bit) for the fraction/mantissa. * The minimal exponent encodes subnormal nubers; * the maximal exponent encodes Infinities and NaNs. * <P> * Infinities and NaNs are not supported as BigRationals. * <P> * The conversion from the bits to a BigRational * is done without loss of precision. */ public static BigRational valueOfFloatBits(int value) { // [masking: suppress sign expansion, that leads to excess bits, // that's not accepted by fromIeee754()] return fromIeee754(new long[]{ value & 0xffffffffl, }, SINGLE_FLOAT_FRACTION_SIZE, SINGLE_FLOAT_EXPONENT_SIZE); } /** An alias to valueOfFloatBits(). */ public static BigRational valueOfSingleBits(int value) { return valueOfFloatBits(value); } /** Build a BigRational from an IEEE 754 half size * (half precision, binary16) floating point number * represented as short. * <P> * An IEEE 754 half size (binary16) number uses * 1 bit for the sign, * 5 bits for the exponent, * and 10 bits (plus 1 implicit bit) for the fraction/mantissa. * The minimal exponent encodes subnormal nubers; * the maximal exponent encodes Infinities and NaNs. * <P> * Infinities and NaNs are not supported as BigRationals. * <P> * The conversion from the bits to a BigRational * is done without loss of precision. */ public static BigRational valueOfHalfBits(short value) { // [masking: suppress sign expansion, that leads to excess bits, // that's not accepted by fromIeee754()] return fromIeee754(new long[]{ value & 0xffffl, }, HALF_FLOAT_FRACTION_SIZE, HALF_FLOAT_EXPONENT_SIZE); } /** Build a BigRational from an IEEE 754 quad size * (quadruple precision, binary128) floating point number * represented as an array of two longs * (big endian; higher significant long first). * <P> * An IEEE 754 quad size (binary128, quadruple) number uses * 1 bit for the sign, * 15 bits for the exponent, * and 112 bits (plus 1 implicit bit) for the fraction/mantissa. * The minimal exponent encodes subnormal nubers; * the maximal exponent encodes Infinities and NaNs. * <P> * Infinities and NaNs are not supported as BigRationals. * <P> * The conversion from the bits to a BigRational * is done without loss of precision. */ public static BigRational valueOfQuadBits(long[] value) { return fromIeee754(value, QUAD_FLOAT_FRACTION_SIZE, QUAD_FLOAT_EXPONENT_SIZE); } /** Compare two IEEE 754 quad size (quadruple precision, binary128) * floating point numbers (each represented as two longs). * NaNs are not considered; comparison is done by bits. * [Convenience method.] */ // note: especially due the NaN issue commented above // (a NaN maps to many bits representations), // we call this method quadBitsEqual rather than quadEqual public static boolean quadBitsEqual(long[] a, long[] b) { if (a == null || b == null) { throw new NumberFormatException("null"); } if (a.length != 2 || b.length != 2) { throw new NumberFormatException("not a quad"); } return (a[1] == b[1] && a[0] == b[0]); } /** Rounding mode to round away from zero. */ public final static int ROUND_UP = 0; /** Rounding mode to round towards zero. */ public final static int ROUND_DOWN = 1; /** Rounding mode to round towards positive infinity. */ public final static int ROUND_CEILING = 2; /** Rounding mode to round towards negative infinity. */ public final static int ROUND_FLOOR = 3; /** Rounding mode to round towards nearest neighbor unless both * neighbors are equidistant, in which case to round up. */ public final static int ROUND_HALF_UP = 4; /** Rounding mode to round towards nearest neighbor unless both * neighbors are equidistant, in which case to round down. */ public final static int ROUND_HALF_DOWN = 5; /** Rounding mode to round towards the nearest neighbor unless both * neighbors are equidistant, in which case to round towards the even neighbor. */ public final static int ROUND_HALF_EVEN = 6; /** Rounding mode to assert that the requested operation has an exact * result, hence no rounding is necessary. * If this rounding mode is specified on an operation that yields an inexact result, * an ArithmeticException is thrown. */ public final static int ROUND_UNNECESSARY = 7; /** Rounding mode to round towards nearest neighbor unless both * neighbors are equidistant, in which case to round ceiling. */ public final static int ROUND_HALF_CEILING = 8; /** Rounding mode to round towards nearest neighbor unless both * neighbors are equidistant, in which case to round floor. */ public final static int ROUND_HALF_FLOOR = 9; /** Rounding mode to round towards the nearest neighbor unless both * neighbors are equidistant, in which case to round towards the odd neighbor. */ public final static int ROUND_HALF_ODD = 10; /** Default round mode, ROUND_HALF_UP. */ public final static int DEFAULT_ROUND_MODE = ROUND_HALF_UP; /** Round. * <P> * Round mode is one of * {<CODE>ROUND_UP, ROUND_DOWN, ROUND_CEILING, ROUND_FLOOR, * ROUND_HALF_UP, ROUND_HALF_DOWN, ROUND_HALF_EVEN, * ROUND_HALF_CEILING, ROUND_HALF_FLOOR, ROUND_HALF_ODD, * ROUND_UNNECESSARY, DEFAULT_ROUND_MODE (==ROUND_HALF_UP)</CODE>}. * <P> * If rounding isn't necessary, * i.e. this BigRational is an integer, * [as an optimization] * this BigRational is returned. * <P> * Possible loss of precision. */ // @PrecisionLoss public BigRational round(int roundMode) { // optimization // return self if we don't need to round, independent of rounding mode if (isInteger()) { return this; } return new BigRational( // [rounding step, possible loss of precision step] roundToBigInteger(roundMode)); } /** Round to BigInteger helper function. * Internally used. * <P> * Possible loss of precision. */ // @PrecisionLoss private BigInteger roundToBigInteger(int roundMode) { // note: remainder and its duplicate are calculated for all cases. BigInteger n = m_n; final BigInteger q = m_q; final int sgn = n.signum(); // optimization if (sgn == 0) { // [typically not reached due to earlier test for integerp] return BIG_INTEGER_ZERO; } // keep info on the sign final boolean pos = (sgn > 0); // operate on positive values if (!pos) { n = n.negate(); } final BigInteger[] divrem = n.divideAndRemainder(q); BigInteger dv = divrem[0]; final BigInteger r = divrem[1]; // return if we don't need to round, independent of rounding mode if (bigIntegerIsZero(r)) { // [typically not reached since remainder is not zero // with normalized that are not integerp] if (!pos) { dv = dv.negate(); } return dv; } boolean up = false; final int comp = r.multiply(BIG_INTEGER_TWO).compareTo(q); switch (roundMode) { // Rounding mode to round away from zero. case ROUND_UP: up = true; break; // Rounding mode to round towards zero. case ROUND_DOWN: up = false; break; // Rounding mode to round towards positive infinity. case ROUND_CEILING: up = pos; break; // Rounding mode to round towards negative infinity. case ROUND_FLOOR: up = !pos; break; // Rounding mode to round towards "nearest neighbor" unless both // neighbors are equidistant, in which case round up. case ROUND_HALF_UP: up = (comp >= 0); break; // Rounding mode to round towards "nearest neighbor" unless both // neighbors are equidistant, in which case round down. case ROUND_HALF_DOWN: up = (comp > 0); break; case ROUND_HALF_CEILING: up = (comp != 0 ? comp > 0 : pos); break; case ROUND_HALF_FLOOR: up = (comp != 0 ? comp > 0 : !pos); break; // Rounding mode to round towards the "nearest neighbor" unless both // neighbors are equidistant, in which case, round towards the even neighbor. case ROUND_HALF_EVEN: up = (comp != 0 ? comp > 0 : !bigIntegerIsZero(dv .remainder(BIG_INTEGER_TWO))); break; case ROUND_HALF_ODD: up = (comp != 0 ? comp > 0 : bigIntegerIsZero(dv .remainder(BIG_INTEGER_TWO))); break; // Rounding mode to assert that the requested operation has an exact // result, hence no rounding is necessary. If this rounding mode is // specified on an operation that yields an inexact result, an // ArithmeticException is thrown. case ROUND_UNNECESSARY: if (!bigIntegerIsZero(r)) { throw new ArithmeticException( "rounding necessary"); } // [typically not reached due to earlier test for integerp] up = false; break; default: throw new IllegalArgumentException("unsupported rounding mode"); } if (up) { dv = dv.add(BIG_INTEGER_ONE); } if (!pos) { dv = dv.negate(); } // [rounding step, possible loss of precision step] return dv; } /** Round by default mode (ROUND_HALF_UP). * <P> * Possible loss of precision. */ // @PrecisionLoss public BigRational round() { // [rounding step, possible loss of precision step] return round(DEFAULT_ROUND_MODE); } /** Floor, round towards negative infinity. * <P> * Possible loss of precision. */ // @PrecisionLoss public BigRational floor() { // [rounding step, possible loss of precision step] return round(ROUND_FLOOR); } /** Ceiling, round towards positive infinity. * <P> * Possible loss of precision. */ // @PrecisionLoss public BigRational ceiling() { // [rounding step, possible loss of precision step] return round(ROUND_CEILING); } /** An alias to ceiling(). * <P> * Possible loss of precision. */ // [Name: see class Math.] // @PrecisionLoss public BigRational ceil() { // [possible loss of precision step] return ceiling(); } /** Truncate, round towards zero. * <P> * Possible loss of precision. */ // @PrecisionLoss public BigRational truncate() { // [rounding step, possible loss of precision step] return round(ROUND_DOWN); } /** An alias to truncate(). * <P> * Possible loss of precision. */ // @PrecisionLoss public BigRational trunc() { // [possible loss of precision step] return truncate(); } /** Integer part. * <P> * Possible loss of precision. */ // @PrecisionLoss public BigRational integerPart() { // [rounding step, possible loss of precision step] return round(ROUND_DOWN); } /** Fractional part. * <P> * Possible loss of precision. */ // @PrecisionLoss public BigRational fractionalPart() { // this==ip+fp; sign(fp)==sign(this) // [possible loss of precision step] return subtract(integerPart()); } /** Return an array of BigRationals with both integer and fractional part. * <P> * Integer part is returned at offset 0; fractional part at offset 1. */ public BigRational[] integerAndFractionalPart() { // note: this duplicates fractionalPart() code, for speed. final BigRational[] pp = new BigRational[2]; final BigRational ip = integerPart(); pp[0] = ip; pp[1] = subtract(ip); return pp; } }