org.apache.solr.util.BCDUtils Maven / Gradle / Ivy
The newest version!
/**
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.solr.util;
/**
* @version $Id: BCDUtils.java 555343 2007-07-11 17:46:25Z hossman $
*/
public class BCDUtils {
// idiv is expensive...
// use fixed point math to multiply by 1/10
// http://www.cs.uiowa.edu/~jones/bcd/divide.html
private static int div10(int a) { return (a * 0xcccd) >>> 19; }
private static int mul10(int a) { return (a*10); }
// private static int mul10(int a) { return ((a<<3)+(a<<1)); }
// private static int mul10(int a) { return (a+(a<<2))<<1; } // attempt to use LEA instr
// (imul32 on AMD64 only has a 3 cycle latency in any case)
// something that won't clash with other base100int
// chars (something >= 100)
private static final char NEG_CHAR=(char)126;
// The zero exponent.
// NOTE: for smaller integer representations, this current implementation
// combines sign and exponent into the first char. sign is negative if
// exponent is less than the zero point (no negative exponents themselves)
private static final int ZERO_EXPONENT='a'; // 97
// WARNING: assumption is that this is a legal int...
// no validation is done. [+-]?digit*
//
// Normalization of zeros *is* done...
// 0004, 004, 04, 4 will all end up being equal
// 0,-0 are normalized to '' (zero length)
//
// The value is written to the output buffer
// from the end to the start. The return value
// is the start of the Base100 int in the output buffer.
//
// As the output will be smaller than the input, arr and
// out may refer to the same array if desired.
//
public static int base10toBase100(char[] arr, int start, int end,
char[] out, int outend
)
{
int wpos=outend; // write position
boolean neg=false;
while (--end >= start) {
int val = arr[end];
if (val=='+') { break; }
else if (val=='-') { neg=!neg; break; }
else {
val = val - '0';
if (end > start) {
int val2 = arr[end-1];
if (val2=='+') { out[--wpos]=(char)val; break; }
if (val2=='-') { out[--wpos]=(char)val; neg=!neg; break; }
end--;
val = val + (val2 - '0')*10;
}
out[--wpos] = (char)val;
}
}
// remove leading base100 zeros
while (wpos=0, but this format
// is meant to be valid for floating point numbers as well...
private static final int ZERO_EXPONENT='a'; // 97
// if exponent is larger than what can be represented
// in a single byte (char), then this is the multibyte
// escape char.
// UCS-2 surrogates start at 0xD800
private static final int POSITIVE_EXPONENT_ESCAPE=0x3fff;
// if exponent is smaller than what can be represented in
// a single byte, then this is the multibyte escape
private static final int NEGATIVE_EXPONENT_ESCAPE=1;
// if number is negative, it starts with this optional value
// this should not overlap with any exponent values
private static final int NEGATIVE_SIGN=0;
**********/
// WARNING: assumption is that this is a legal int...
// no validation is done. [+-]?digit*
//
// Normalization of zeros *is* done...
// 0004, 004, 04, 4 will all end up being equal
// 0,-0 are normalized to '' (zero length)
//
// The value is written to the output buffer
// from the end to the start. The return value
// is the start of the Base100 int in the output buffer.
//
// As the output will be smaller than the input, arr and
// out may refer to the same array if desired.
//
public static int base10toBase100SortableInt(char[] arr, int start, int end,
char[] out, int outend
)
{
int wpos=outend; // write position
boolean neg=false;
--end; // position end pointer *on* the last char
// read signs and leading zeros
while (start <= end) {
char val = arr[start];
if (val=='-') neg=!neg;
else if (val>='1' && val<='9') break;
start++;
}
// eat whitespace on RHS?
outer: while (start <= end) {
switch(arr[end]) {
case ' ':
case '\t':
case '\n':
case '\r': end--; break;
default: break outer;
}
}
int hundreds=0;
/******************************************************
* remove RHS zero normalization since it only helps 1 in 100
* numbers and complicates both encoding and decoding.
// remove pairs of zeros on the RHS and keep track of
// the count.
while (start <= end) {
char val = arr[end];
if (val=='0' && start <= end) {
val=arr[end-1];
if (val=='0') {
hundreds++;
end-=2;
continue;
}
}
break;
}
*************************************************************/
// now start at the end and work our way forward
// encoding two base 10 digits into 1 base 100 digit
while (start <= end) {
int val = arr[end--];
val = val - '0';
if (start <= end) {
int val2 = arr[end--];
val = val + (val2 - '0')*10;
}
out[--wpos] = neg ? (char)(99-val) : (char)val;
}
/****** FUTURE: not needed for this implementation of exponent combined with sign
// normalize all zeros to positive values
if (wpos==outend) neg=false;
******/
// adjust exponent by the number of base 100 chars written
hundreds += outend - wpos;
// write the exponent and sign combined
out[--wpos] = neg ? (char)(ZERO_EXPONENT - hundreds) : (char)(ZERO_EXPONENT + hundreds);
return outend-wpos; // the length of the base100 int
}
// Converts a base100 sortable number to base10 character form
// returns number of chars written.
// At least 1 char is always written.
public static int base100SortableIntToBase10(char[] arr, int start, int end,
char[] out, int offset)
{
// Take care of "0" case first. It's the only number that is represented
// in one char.
if (end-start == 1) {
out[offset]='0';
return 1;
}
int wpos = offset; // write position
boolean neg = false;
int exp = arr[start++];
if (exp < ZERO_EXPONENT) {
neg=true;
exp = ZERO_EXPONENT - exp;
out[wpos++]='-';
}
boolean firstDigit=true;
while (start < end) {
int val = arr[start++];
if (neg) val = 99 - val;
// opt - if we ever want a faster version we can avoid one integer
// divide by using fixed point math to multiply by 1/10
// http://www.cs.uiowa.edu/~jones/bcd/divide.html
// TIP: write a small function in gcc or cl and see what
// the optimized assemply output looks like (and which is fastest).
// In C you can specify "unsigned" which gives the compiler more
// info than the Java compiler has.
char tens = (char)(val / 10 + '0');
if (!firstDigit || tens!='0') { // skip leading 0
out[wpos++] = tens; // write tens position
}
out[wpos++] = (char)(val % 10 + '0'); // write ones position
firstDigit=false;
}
// OPTIONAL: if trailing zeros were truncated, then this is where
// we would restore them (compare number of chars read vs exponent)
return wpos-offset;
}
public static int base10toBase10kSortableInt(char[] arr, int start, int end,
char[] out, int outend
)
{
int wpos=outend; // write position
boolean neg=false;
--end; // position end pointer *on* the last char
// read signs and leading zeros
while (start <= end) {
char val = arr[start];
if (val=='-') neg=!neg;
else if (val>='1' && val<='9') break;
start++;
}
// eat whitespace on RHS?
outer: while (start <= end) {
switch(arr[end]) {
case ' ': // fallthrough
case '\t': // fallthrough
case '\n': // fallthrough
case '\r': end--; break;
default: break outer;
}
}
int exp=0;
/******************************************************
* remove RHS zero normalization since it only helps 1 in 100
* numbers and complicates both encoding and decoding.
// remove pairs of zeros on the RHS and keep track of
// the count.
while (start <= end) {
char val = arr[end];
if (val=='0' && start <= end) {
val=arr[end-1];
if (val=='0') {
hundreds++;
end-=2;
continue;
}
}
break;
}
*************************************************************/
// now start at the end and work our way forward
// encoding two base 10 digits into 1 base 100 digit
while (start <= end) {
int val = arr[end--] - '0'; // ones
if (start <= end) {
val += (arr[end--] - '0')*10; // tens
if (start <= end) {
val += (arr[end--] - '0')*100; // hundreds
if (start <= end) {
val += (arr[end--] - '0')*1000; // thousands
}
}
}
out[--wpos] = neg ? (char)(9999-val) : (char)val;
}
/****** FUTURE: not needed for this implementation of exponent combined with sign
// normalize all zeros to positive values
if (wpos==outend) neg=false;
******/
// adjust exponent by the number of base 100 chars written
exp += outend - wpos;
// write the exponent and sign combined
out[--wpos] = neg ? (char)(ZERO_EXPONENT - exp) : (char)(ZERO_EXPONENT + exp);
return outend-wpos; // the length of the base100 int
}
// Converts a base100 sortable number to base10 character form
// returns number of chars written.
// At least 1 char is always written.
public static int base10kSortableIntToBase10(char[] arr, int start, int end,
char[] out, int offset)
{
// Take care of "0" case first. It's the only number that is represented
// in one char since we don't chop trailing zeros.
if (end-start == 1) {
out[offset]='0';
return 1;
}
int wpos = offset; // write position
boolean neg;
int exp = arr[start++];
if (exp < ZERO_EXPONENT) {
neg=true;
// We don't currently use exp on decoding...
// exp = ZERO_EXPONENT - exp;
out[wpos++]='-';
} else {
neg=false;
}
// since so many values will fall in one char, pull it
// out of the loop (esp since the first value must
// be special-cased to not print leading zeros.
// integer division is still expensive, so it's best to check
// if you actually need to do it.
//
// TIP: write a small function in gcc or cl and see what
// the optimized assemply output looks like (and which is fastest).
// In C you can specify "unsigned" which gives the compiler more
// info than the Java compiler has.
int val = arr[start++];
if (neg) val = 9999 - val;
/***
if (val < 10) {
out[wpos++] = (char)(val + '0');
} else if (val < 100) {
out[wpos++] = (char)(val/10 + '0');
out[wpos++] = (char)(val%10 + '0');
} else if (val < 1000) {
out[wpos++] = (char)(val/100 + '0');
out[wpos++] = (char)((val/10)%10 + '0');
out[wpos++] = (char)(val%10 + '0');
} else {
out[wpos++] = (char)(val/1000 + '0');
out[wpos++] = (char)((val/100)%10 + '0');
out[wpos++] = (char)((val/10)%10 + '0');
out[wpos++] = (char)(val % 10 + '0');
}
***/
if (val < 10) {
out[wpos++] = (char)(val + '0');
} else if (val < 100) {
int div = div10(val);
int ones = val - mul10(div); // mod 10
out[wpos++] = (char)(div + '0');
out[wpos++] = (char)(ones + '0');
} else if (val < 1000) {
int div = div10(val);
int ones = val - mul10(div); // mod 10
val=div;
div = div10(val);
int tens = val - mul10(div); // mod 10
out[wpos++] = (char)(div + '0');
out[wpos++] = (char)(tens + '0');
out[wpos++] = (char)(ones + '0');
} else {
int div = div10(val);
int ones = val - mul10(div); // mod 10
val=div;
div = div10(val);
int tens = val - mul10(div); // mod 10
val=div;
div = div10(val);
int hundreds = val - mul10(div); // mod 10
out[wpos++] = (char)(div + '0');
out[wpos++] = (char)(hundreds + '0');
out[wpos++] = (char)(tens + '0');
out[wpos++] = (char)(ones + '0');
}
while (start < end) {
val = arr[start++];
if (neg) val = 9999 - val;
int div = div10(val);
int ones = val - mul10(div); // mod 10
val=div;
div = div10(val);
int tens = val - mul10(div); // mod 10
val=div;
div = div10(val);
int hundreds = val - mul10(div); // mod 10
/***
int ones = val % 10;
val /= 10;
int tens = val!=0 ? val % 10 : 0;
val /= 10;
int hundreds = val!=0 ? val % 10 : 0;
val /= 10;
int thousands = val!=0 ? val % 10 : 0;
***/
/***
int thousands = val>=1000 ? val/1000 : 0;
int hundreds = val>=100 ? (val/100)%10 : 0;
int tens = val>=10 ? (val/10)%10 : 0;
int ones = val % 10;
***/
/***
int thousands = val/1000;
int hundreds = (val/100)%10;
int tens = (val/10)%10;
int ones = val % 10;
***/
out[wpos++] = (char)(div + '0');
out[wpos++] = (char)(hundreds + '0');
out[wpos++] = (char)(tens + '0');
out[wpos++] = (char)(ones + '0');
}
// OPTIONAL: if trailing zeros were truncated, then this is where
// we would restore them (compare number of chars read vs exponent)
return wpos-offset;
}
}