net.sf.saxon.value.FloatingPointConverter Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of Saxon-HE Show documentation
Show all versions of Saxon-HE Show documentation
The XSLT and XQuery Processor
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Copyright (c) 2018-2022 Saxonica Limited
// This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0.
// If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
// This Source Code Form is "Incompatible With Secondary Licenses", as defined by the Mozilla Public License, v. 2.0.
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
package net.sf.saxon.value;
import net.sf.saxon.str.UnicodeBuilder;
import net.sf.saxon.str.UnicodeString;
import java.math.BigInteger;
/**
* This is a utility class that handles formatting of numbers as strings.
* The algorithm for converting a floating point number to a string is taken from Guy L. Steele and
* Jon L. White, How to Print Floating-Point Numbers Accurately, ACM SIGPLAN 1990. It is algorithm
* (FPP)2 from that paper. There are three separate implementations of the algorithm:
*
* - One using long arithmetic and generating non-exponential output representations
* - One using BigInteger arithmetic and generating non-exponential output representation
* - One using BigInteger arithmetic and generating exponential output representations
*
* The choice of method depends on the value of the number being formatted.
* The module contains some residual code (mainly the routine for formatting integers) from the class
* AppenderHelper by Jack Shirazi in the O'Reilly book Java Performance Tuning. The floating point routines
* in that module were found to be unsuitable, since they used floating point arithmetic which introduces
* rounding errors.
* There are several reasons for doing this conversion within Saxon, rather than leaving it all to Java.
* Firstly, there are differences in the required output format, notably the absence of ".0" when formatting
* whole numbers, and the different rules for the range of numbers where exponential notation is used.
* Secondly, there are bugs in some Java implementations, for example JDK outputs 0.001 as 0.0010, and
* IKVM/GNU gets things very wrong sometimes. Finally, this implementation is faster for "everyday" numbers,
* though it is slower for more extreme numbers. It would probably be reasonable to hand over formatting
* to the Java platform (at least when running the Sun JDK) for exponents outside the range -7 to +7.
*/
public class FloatingPointConverter {
public static FloatingPointConverter THE_INSTANCE = new FloatingPointConverter();
private FloatingPointConverter() {
}
/**
* char array holding the characters for the string "-Infinity".
*/
private static final String NEGATIVE_INFINITY = "-INF";
/**
* char array holding the characters for the string "Infinity".
*/
private static final String POSITIVE_INFINITY = "INF";
/**
* char array holding the characters for the string "NaN".
*/
private static final String NaN = "NaN";
private static final char[] charForDigit = {
'0', '1', '2', '3', '4', '5', '6', '7', '8', '9'
};
public static final long DOUBLE_SIGN_MASK = 0x8000000000000000L;
private static final long doubleExpMask = 0x7ff0000000000000L;
private static final int doubleExpShift = 52;
private static final int doubleExpBias = 1023;
private static final long doubleFractMask = 0xfffffffffffffL;
public static final int FLOAT_SIGN_MASK = 0x80000000;
private static final int floatExpMask = 0x7f800000;
private static final int floatExpShift = 23;
private static final int floatExpBias = 127;
private static final int floatFractMask = 0x7fffff;
private static final BigInteger TEN = BigInteger.valueOf(10);
private static final BigInteger NINE = BigInteger.valueOf(9);
/**
* Format an integer, appending the string representation of the integer to a string buffer
*
* @param s the string buffer
* @param i the integer to be formatted
* @return the supplied string buffer, containing the appended integer
*/
public static UnicodeBuilder appendInt(UnicodeBuilder s, int i) {
// TODO: this elaborate machinery is only being used to output the exponent of a floating point number,
// which never has more than 3 digits...
if (i < 0) {
if (i == Integer.MIN_VALUE) {
//cannot make this positive due to integer overflow
s.append("-2147483648");
return s;
}
s.append('-');
i = -i;
}
int c;
if (i < 10) {
//one digit
s.append(charForDigit[i]);
return s;
} else if (i < 100) {
//two digits
s.append(charForDigit[i / 10]);
s.append(charForDigit[i % 10]);
return s;
} else if (i < 1000) {
//three digits
s.append(charForDigit[i / 100]);
s.append(charForDigit[(c = i % 100) / 10]);
s.append(charForDigit[c % 10]);
return s;
} else if (i < 10000) {
//four digits
s.append(charForDigit[i / 1000]);
s.append(charForDigit[(c = i % 1000) / 100]);
s.append(charForDigit[(c %= 100) / 10]);
s.append(charForDigit[c % 10]);
return s;
} else if (i < 100000) {
//five digits
s.append(charForDigit[i / 10000]);
s.append(charForDigit[(c = i % 10000) / 1000]);
s.append(charForDigit[(c %= 1000) / 100]);
s.append(charForDigit[(c %= 100) / 10]);
s.append(charForDigit[c % 10]);
return s;
} else if (i < 1000000) {
//six digits
s.append(charForDigit[i / 100000]);
s.append(charForDigit[(c = i % 100000) / 10000]);
s.append(charForDigit[(c %= 10000) / 1000]);
s.append(charForDigit[(c %= 1000) / 100]);
s.append(charForDigit[(c %= 100) / 10]);
s.append(charForDigit[c % 10]);
return s;
} else if (i < 10000000) {
//seven digits
s.append(charForDigit[i / 1000000]);
s.append(charForDigit[(c = i % 1000000) / 100000]);
s.append(charForDigit[(c %= 100000) / 10000]);
s.append(charForDigit[(c %= 10000) / 1000]);
s.append(charForDigit[(c %= 1000) / 100]);
s.append(charForDigit[(c %= 100) / 10]);
s.append(charForDigit[c % 10]);
return s;
} else if (i < 100000000) {
//eight digits
s.append(charForDigit[i / 10000000]);
s.append(charForDigit[(c = i % 10000000) / 1000000]);
s.append(charForDigit[(c %= 1000000) / 100000]);
s.append(charForDigit[(c %= 100000) / 10000]);
s.append(charForDigit[(c %= 10000) / 1000]);
s.append(charForDigit[(c %= 1000) / 100]);
s.append(charForDigit[(c %= 100) / 10]);
s.append(charForDigit[c % 10]);
return s;
} else if (i < 1000000000) {
//nine digits
s.append(charForDigit[i / 100000000]);
s.append(charForDigit[(c = i % 100000000) / 10000000]);
s.append(charForDigit[(c %= 10000000) / 1000000]);
s.append(charForDigit[(c %= 1000000) / 100000]);
s.append(charForDigit[(c %= 100000) / 10000]);
s.append(charForDigit[(c %= 10000) / 1000]);
s.append(charForDigit[(c %= 1000) / 100]);
s.append(charForDigit[(c %= 100) / 10]);
s.append(charForDigit[c % 10]);
return s;
} else {
//ten digits
s.append(charForDigit[i / 1000000000]);
s.append(charForDigit[(c = i % 1000000000) / 100000000]);
s.append(charForDigit[(c %= 100000000) / 10000000]);
s.append(charForDigit[(c %= 10000000) / 1000000]);
s.append(charForDigit[(c %= 1000000) / 100000]);
s.append(charForDigit[(c %= 100000) / 10000]);
s.append(charForDigit[(c %= 10000) / 1000]);
s.append(charForDigit[(c %= 1000) / 100]);
s.append(charForDigit[(c %= 100) / 10]);
s.append(charForDigit[c % 10]);
return s;
}
}
/**
* Implementation of the (FPP)2 algorithm from Steele and White, for doubles in the range
* 0.01 to 1000000, and floats in the range 0.000001 to 1000000.
* In this range (a) XPath requires that the output should not be in exponential
* notation, and (b) the arithmetic can be handled using longs rather than BigIntegers
*
* @param sb the string buffer to which the formatted result is to be appended
* @param e the exponent of the floating point number
* @param f the fraction part of the floating point number, such that the "real" value of the
* number is f * 2^(e-p), with p>=0 and 0 lt f lt 2^p
* @param p the precision
*/
private static void fppfpp(UnicodeBuilder sb, int e, long f, int p) {
long R = f << Math.max(e - p, 0);
long S = 1L << Math.max(0, -(e - p));
long Mminus = 1L << Math.max(e - p, 0);
long Mplus = Mminus;
boolean initial = true;
// simpleFixup
if (f == 1L << (p - 1)) {
Mplus = Mplus << 1;
R = R << 1;
S = S << 1;
}
int k = 0;
while (R < (S + 9) / 10) { // (S+9)/10 == ceiling(S/10)
k--;
R = R * 10;
Mminus = Mminus * 10;
Mplus = Mplus * 10;
}
while (2 * R + Mplus >= 2 * S) {
S = S * 10;
k++;
}
for (int z = k; z < 0; z++) {
if (initial) {
sb.append("0.");
}
initial = false;
sb.append('0');
}
// end simpleFixup
//int H = k-1;
boolean low;
boolean high;
int U;
while (true) {
k--;
long R10 = R * 10;
U = (int) (R10 / S);
R = R10 - (U * S); // = R*10 % S, but faster - saves a division
Mminus = Mminus * 10;
Mplus = Mplus * 10;
low = 2 * R < Mminus;
high = 2 * R > 2 * S - Mplus;
if (low || high) break;
if (k == -1) {
if (initial) {
sb.append('0');
}
sb.append('.');
}
sb.append(charForDigit[U]);
initial = false;
}
if (high && (!low || 2 * R > S)) {
U++;
}
if (k == -1) {
if (initial) {
sb.append('0');
}
sb.append('.');
}
sb.append(charForDigit[U]);
for (int z = 0; z < k; z++) {
sb.append('0');
}
}
/**
* Implementation of the (FPP)2 algorithm from Steele and White, for doubles in the range
* 0.000001 to 0.01. In this range XPath requires that the output should not be in exponential
* notation, but the scale factors are large enough to exceed the capacity of long arithmetic.
*
* @param sb the string buffer to which the formatted result is to be appended
* @param e the exponent of the floating point number
* @param f the fraction part of the floating point number, such that the "real" value of the
* number is f * 2^(e-p), with p>=0 and 0 lt f lt 2^p
* @param p the precision
*/
private static void fppfppBig(UnicodeBuilder sb, int e, long f, int p) {
//long R = f << Math.max(e-p, 0);
BigInteger R = BigInteger.valueOf(f).shiftLeft(Math.max(e - p, 0));
//long S = 1L << Math.max(0, -(e-p));
BigInteger S = BigInteger.ONE.shiftLeft(Math.max(0, -(e - p)));
//long Mminus = 1 << Math.max(e-p, 0);
BigInteger Mminus = BigInteger.ONE.shiftLeft(Math.max(e - p, 0));
//long Mplus = Mminus;
BigInteger Mplus = Mminus;
boolean initial = true;
// simpleFixup
if (f == 1L << (p - 1)) {
Mplus = Mplus.shiftLeft(1);
R = R.shiftLeft(1);
S = S.shiftLeft(1);
}
int k = 0;
while (R.compareTo(S.add(NINE).divide(TEN)) < 0) { // (S+9)/10 == ceiling(S/10)
k--;
R = R.multiply(TEN);
Mminus = Mminus.multiply(TEN);
Mplus = Mplus.multiply(TEN);
}
while (R.shiftLeft(1).add(Mplus).compareTo(S.shiftLeft(1)) >= 0) {
S = S.multiply(TEN);
k++;
}
for (int z = k; z < 0; z++) {
if (initial) {
sb.append("0.");
}
initial = false;
sb.append('0');
}
// end simpleFixup
//int H = k-1;
boolean low;
boolean high;
int U;
while (true) {
k--;
BigInteger R10 = R.multiply(TEN);
U = R10.divide(S).intValue();
R = R10.mod(S);
Mminus = Mminus.multiply(TEN);
Mplus = Mplus.multiply(TEN);
BigInteger R2 = R.shiftLeft(1);
low = R2.compareTo(Mminus) < 0;
high = R2.compareTo(S.shiftLeft(1).subtract(Mplus)) > 0;
if (low || high) break;
if (k == -1) {
if (initial) {
sb.append('0');
}
sb.append('.');
}
sb.append(charForDigit[U]);
initial = false;
}
if (high && (!low || R.shiftLeft(1).compareTo(S) > 0)) {
U++;
}
if (k == -1) {
if (initial) {
sb.append('0');
}
sb.append('.');
}
sb.append(charForDigit[U]);
for (int z = 0; z < k; z++) {
sb.append('0');
}
}
/**
* Implementation of the (FPP)2 algorithm from Steele and White, for numbers outside the range
* 0.000001 to 1000000. In this range XPath requires that the output should be in exponential
* notation
*
* @param sb the string buffer to which the formatted result is to be appended
* @param e the exponent of the floating point number
* @param f the fraction part of the floating point number, such that the "real" value of the
* number is f * 2^(e-p), with p>=0 and 0 lt f lt 2^p
* @param p the precision
*/
private static void fppfppExponential(UnicodeBuilder sb, int e, long f, int p) {
//long R = f << Math.max(e-p, 0);
BigInteger R = BigInteger.valueOf(f).shiftLeft(Math.max(e - p, 0));
//long S = 1L << Math.max(0, -(e-p));
BigInteger S = BigInteger.ONE.shiftLeft(Math.max(0, -(e - p)));
//long Mminus = 1 << Math.max(e-p, 0);
BigInteger Mminus = BigInteger.ONE.shiftLeft(Math.max(e - p, 0));
//long Mplus = Mminus;
BigInteger Mplus = Mminus;
boolean initial = true;
boolean doneDot = false;
// simpleFixup
if (f == 1L << (p - 1)) {
Mplus = Mplus.shiftLeft(1);
R = R.shiftLeft(1);
S = S.shiftLeft(1);
}
int k = 0;
while (R.compareTo(S.add(NINE).divide(TEN)) < 0) { // (S+9)/10 == ceiling(S/10)
k--;
R = R.multiply(TEN);
Mminus = Mminus.multiply(TEN);
Mplus = Mplus.multiply(TEN);
}
while (R.shiftLeft(1).add(Mplus).compareTo(S.shiftLeft(1)) >= 0) {
S = S.multiply(TEN);
k++;
}
// end simpleFixup
int H = k - 1;
boolean low;
boolean high;
int U;
while (true) {
k--;
BigInteger R10 = R.multiply(TEN);
U = R10.divide(S).intValue();
R = R10.mod(S);
Mminus = Mminus.multiply(TEN);
Mplus = Mplus.multiply(TEN);
BigInteger R2 = R.shiftLeft(1);
low = R2.compareTo(Mminus) < 0;
high = R2.compareTo(S.shiftLeft(1).subtract(Mplus)) > 0;
if (low || high) break;
sb.append(charForDigit[U]);
if (initial) {
sb.append('.');
doneDot = true;
}
initial = false;
}
if (high && (!low || R.shiftLeft(1).compareTo(S) > 0)) {
U++;
}
sb.append(charForDigit[U]);
if (!doneDot) {
sb.append(".0");
}
sb.append('E');
appendInt(sb, H);
}
/**
* Append a string representation of a double value to a string buffer
*
* @param d the double to be formatted
* @param forceExponential forces exponential notation if set (if not set, exponential notation
* is used only for values outside the range 1e-6 to 1e+6)
* @return the original string buffer, now containing the string representation of the supplied double
*/
//@CSharpReplaceBody(code = "return net.sf.saxon.str.BMPString.of(d.ToString(System.Globalization.CultureInfo.InvariantCulture));") // for now...
public static UnicodeString convertDouble(double d, boolean forceExponential) {
UnicodeBuilder s = new UnicodeBuilder(32);
if (d == Double.NEGATIVE_INFINITY) {
s.appendLatin(NEGATIVE_INFINITY);
} else if (d == Double.POSITIVE_INFINITY) {
s.appendLatin(POSITIVE_INFINITY);
} else if (Double.isNaN(d)) {
s.appendLatin(NaN);
} else if (d == 0.0) {
if ((Double.doubleToLongBits(d) & DOUBLE_SIGN_MASK) != 0) {
s.append('-');
}
s.append('0');
if (forceExponential) {
s.append(".0E0");
}
} else if (d == Double.MAX_VALUE) {
s.append("1.7976931348623157E308");
} else if (d == -Double.MAX_VALUE) {
s.append("-1.7976931348623157E308");
} else if (d == Double.MIN_VALUE) {
s.append("4.9E-324");
} else if (d == -Double.MIN_VALUE) {
s.append("-4.9E-324");
} else {
if (d < 0) {
s.append('-');
d = -d;
}
long bits = Double.doubleToLongBits(d);
long fraction = (1L << 52) | (bits & doubleFractMask);
long rawExp = (bits & doubleExpMask) >> doubleExpShift;
int exp = (int) rawExp - doubleExpBias;
if (rawExp == 0) {
// don't know how to handle this currently: hand it over to Java to deal with
s.append(Double.toString(d));
return s.toUnicodeString();
}
if (forceExponential || (d >= 1000000 || d < 0.000001)) {
fppfppExponential(s, exp, fraction, 52);
} else {
if (d <= 0.01) {
fppfppBig(s, exp, fraction, 52);
} else {
fppfpp(s, exp, fraction, 52);
}
}
}
return s.toUnicodeString();
}
/**
* Append a string representation of a float value to a string buffer
*
* @param s the string buffer to which the result will be appended
* @param f the float to be formatted
* @param forceExponential forces exponential notation if set (if not set, exponential notation
* is used only for values outside the range 1e-6 to 1e+6)
* @return the original string buffer, now containing the string representation of the supplied float
*/
/*@NotNull*/
//@CSharpReplaceBody(code="return s.append(f.ToString(System.Globalization.CultureInfo.InvariantCulture)).toUnicodeString();") // for now...
public static UnicodeString appendFloat(UnicodeBuilder s, float f, boolean forceExponential) {
if (f == Float.NEGATIVE_INFINITY) {
s.append(NEGATIVE_INFINITY);
} else if (f == Float.POSITIVE_INFINITY) {
s.append(POSITIVE_INFINITY);
} else if (Float.isNaN(f)) {
s.append(NaN);
} else if (f == 0.0) {
if ((Float.floatToIntBits(f) & FLOAT_SIGN_MASK) != 0) {
s.append('-');
}
s.append('0');
} else if (f == Float.MAX_VALUE) {
s.appendLatin("3.4028235E38");
} else if (f == -Float.MAX_VALUE) {
s.appendLatin("-3.4028235E38");
} else if (f == Float.MIN_VALUE) {
s.appendLatin("1.4E-45");
} else if (f == -Float.MIN_VALUE) {
s.appendLatin("-1.4E-45");
} else {
if (f < 0) {
s.append('-');
f = -f;
}
int bits = Float.floatToIntBits(f);
int fraction = (1 << 23) | (bits & floatFractMask);
int rawExp = ((bits & floatExpMask) >> floatExpShift);
int exp = rawExp - floatExpBias;
int precision = 23;
if (rawExp == 0) {
// don't know how to handle this currently: hand it over to Java to deal with
s.append(Float.toString(f));
return s.toUnicodeString();
}
if (forceExponential || (f >= 1000000 || f < 0.000001F)) {
fppfppExponential(s, exp, fraction, precision);
} else {
fppfpp(s, exp, fraction, precision);
}
}
return s.toUnicodeString();
}
// public static void main(String[] args) {
// if (args.length > 0 && args[0].equals("F")) {
// if (args.length == 2) {
// StringTokenizer tok = new StringTokenizer(args[1], ",");
// while (tok.hasMoreElements()) {
// String input = tok.nextToken();
// float f = Float.parseFloat(input);
// StringBuilder sb = new StringBuilder(20);
// appendFloat(sb, f);
// System.err.println("input: " + input + " output: " + sb.toString() + " java: " + f);
// }
// } else {
// Random gen = new Random();
// for (int i=1; i<1000; i++) {
// int p=gen.nextInt(999*i*i);
// int q=gen.nextInt(999*i*i);
// String input = (p + "." + q);
// float f = Float.parseFloat(input);
// StringBuilder sb = new StringBuilder(20);
// appendFloat(sb, f);
// System.err.println("input: " + input + " output: " + sb.toString() + " java: " + f);
// }
// }
// } else {
// if (args.length == 2) {
// StringTokenizer tok = new StringTokenizer(args[1], ",");
// while (tok.hasMoreElements()) {
// String input = tok.nextToken();
// double f = Double.parseDouble(input);
// StringBuilder sb = new StringBuilder(20);
// appendDouble(sb, f);
// System.err.println("input: " + input + " output: " + sb.toString() + " java: " + f);
// }
// } else {
// long start = System.currentTimeMillis();
// Random gen = new Random();
// for (int i=1; i<100000; i++) {
// //int p=gen.nextInt(999*i*i);
// int q=gen.nextInt(999*i);
// //String input = (p + "." + q);
// String input = "0.000" + q;
// double f = Double.parseDouble(input);
// StringBuilder sb = new StringBuilder(20);
// appendDouble(sb, f);
// //System.err.println("input: " + input + " output: " + sb.toString() + " java: " + f);
// }
// System.err.println("** elapsed time " + (System.currentTimeMillis() - start));
// }
// }
// }
}