xapian-core/api/sortable-serialise.cc

   1 /** @file sortable-serialise.cc
   2  * @brief Serialise floating point values to string which sort the same way.
   3  */
   4 /* Copyright (C) 2007,2009,2015 Olly Betts
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
  19  */
  20
  21 #include <config.h>
  22
  23 #ifndef XAPIAN_UNITTEST
  24 # include <xapian/queryparser.h>
  25 // Only enable exceptions for the testsuite - we don't want them in a library
  26 // build as these functions are marked not to throw exceptions in the API
  27 // headers.
  28 # define UNITTEST_ASSERT(X)
  29 #else
  30 # include "omassert.h"
  31 # define UNITTEST_ASSERT(X) Assert(X)
  32 #endif
  33
  34 #include <cfloat>
  35 #include <cmath>
  36 #include <cstring>
  37
  38 #include <string>
  39
  40 using namespace std;
  41
  42 #if FLT_RADIX != 2
  43 # error Code currently assumes FLT_RADIX == 2
  44 #endif
  45
  46 #ifdef _MSC_VER
  47 // Disable warning about negating an unsigned type, which we do deliberately.
  48 # pragma warning(disable:4146)
  49 #endif
  50
  51 size_t
  52 Xapian::sortable_serialise_(double value, char * buf) XAPIAN_NOEXCEPT
  53 {
  54     double mantissa;
  55     int exponent;
  56
  57     // Negative infinity.
  58     if (value < -DBL_MAX) return 0;
  59
  60     mantissa = frexp(value, &exponent);
  61
  62     /* Deal with zero specially.
  63      *
  64      * IEEE representation of doubles uses 11 bits for the exponent, with a
  65      * bias of 1023.  We bias this by subtracting 8, and non-IEEE
  66      * representations may allow higher exponents, so allow exponents down to
  67      * -2039 - if smaller exponents are possible anywhere, we underflow such
  68      *  numbers to 0.
  69      */
  70     if (mantissa == 0.0 || exponent < -2039) {
  71         *buf = '\x80';
  72         return 1;
  73     }
  74
  75     bool negative = (mantissa < 0);
  76     if (negative) mantissa = -mantissa;
  77
  78     // Infinity, or extremely large non-IEEE representation.
  79     if (value > DBL_MAX || exponent > 2055) {
  80         if (negative) {
  81             // This can only happen with a non-IEEE representation, because
  82             // we've already tested for value < -DBL_MAX
  83             return 0;
  84         } else {
  85             memset(buf, '\xff', 9);
  86             return 9;
  87         }
  88     }
  89
  90     // Encoding:
  91     //
  92     // [ 7 | 6 | 5 | 4 3 2 1 0]
  93     //   Sm  Se  Le
  94     //
  95     // Sm stores the sign of the mantissa: 1 = positive or zero, 0 = negative.
  96     // Se stores the sign of the exponent: Sm for positive/zero, !Sm for neg.
  97     // Le stores the length of the exponent: !Se for 3 bits, Se for 11 bits.
  98     unsigned char next = (negative ? 0 : 0xe0);
  99
 100     // Bias the exponent by 8 so that more small integers get short encodings.
 101     exponent -= 8;
 102     bool exponent_negative = (exponent < 0);
 103     if (exponent_negative) {
 104         exponent = -exponent;
 105         next ^= 0x60;
 106     }
 107
 108     size_t len = 0;
 109
 110     /* We store the exponent in 3 or 11 bits.  If the number is negative, we
 111      * flip all the bits of the exponent, since larger negative numbers should
 112      * sort first.
 113      *
 114      * If the exponent is negative, we flip the bits of the exponent, since
 115      * larger negative exponents should sort first (unless the number is
 116      * negative, in which case they should sort later).
 117      */
 118     UNITTEST_ASSERT(exponent >= 0);
 119     if (exponent < 8) {
 120         next ^= 0x20;
 121         next |= static_cast<unsigned char>(exponent << 2);
 122         if (negative ^ exponent_negative) next ^= 0x1c;
 123     } else {
 124         UNITTEST_ASSERT((exponent >> 11) == 0);
 125         // Put the top 5 bits of the exponent into the lower 5 bits of the
 126         // first byte:
 127         next |= static_cast<unsigned char>(exponent >> 6);
 128         if (negative ^ exponent_negative) next ^= 0x1f;
 129         buf[len++] = next;
 130         // And the lower 6 bits of the exponent go into the upper 6 bits
 131         // of the second byte:
 132         next = static_cast<unsigned char>(exponent) << 2;
 133         if (negative ^ exponent_negative) next ^= 0xfc;
 134     }
 135
 136     // Convert the 52 (or 53) bits of the mantissa into two 32-bit words.
 137     mantissa *= 1 << (negative ? 26 : 27);
 138     unsigned word1 = static_cast<unsigned>(mantissa);
 139     mantissa -= word1;
 140     unsigned word2 = static_cast<unsigned>(mantissa * 4294967296.0); // 1<<32
 141     // If the number is positive, the first bit will always be set because 0.5
 142     // <= mantissa < 1, unless mantissa is zero, which we handle specially
 143     // above).  If the number is negative, we negate the mantissa instead of
 144     // flipping all the bits, so in the case of 0.5, the first bit isn't set
 145     // so we need to store it explicitly.  But for the cost of one extra
 146     // leading bit, we can save several trailing 0xff bytes in lots of common
 147     // cases.
 148     UNITTEST_ASSERT(negative || (word1 & (1<<26)));
 149     if (negative) {
 150         // We negate the mantissa for negative numbers, so that the sort order
 151         // is reversed (since larger negative numbers should come first).
 152         word1 = -word1;
 153         if (word2 != 0) ++word1;
 154         word2 = -word2;
 155     }
 156
 157     word1 &= 0x03ffffff;
 158     next |= static_cast<unsigned char>(word1 >> 24);
 159     buf[len++] = next;
 160     buf[len++] = char(word1 >> 16);
 161     buf[len++] = char(word1 >> 8);
 162     buf[len++] = char(word1);
 163
 164     buf[len++] = char(word2 >> 24);
 165     buf[len++] = char(word2 >> 16);
 166     buf[len++] = char(word2 >> 8);
 167     buf[len++] = char(word2);
 168
 169     // Finally, we can chop off any trailing zero bytes.
 170     while (len > 0 && buf[len - 1] == '\0') {
 171         --len;
 172     }
 173
 174     return len;
 175 }
 176
 177 /// Get a number from the character at a given position in a string, returning
 178 /// 0 if the string isn't long enough.
 179 static inline unsigned char
 180 numfromstr(const std::string & str, std::string::size_type pos)
 181 {
 182     return (pos < str.size()) ? static_cast<unsigned char>(str[pos]) : '\0';
 183 }
 184
 185 double
 186 Xapian::sortable_unserialise(const std::string & value) XAPIAN_NOEXCEPT
 187 {
 188     // Zero.
 189     if (value.size() == 1 && value[0] == '\x80') return 0.0;
 190
 191     // Positive infinity.
 192     if (value.size() == 9 &&
 193         memcmp(value.data(), "\xff\xff\xff\xff\xff\xff\xff\xff\xff", 9) == 0) {
 194 #ifdef INFINITY
 195         // INFINITY is C99.  Oddly, it's of type "float" so sanity check in
 196         // case it doesn't cast to double as infinity (apparently some
 197         // implementations have this problem).
 198         if (double(INFINITY) > HUGE_VAL) return INFINITY;
 199 #endif
 200         return HUGE_VAL;
 201     }
 202
 203     // Negative infinity.
 204     if (value.empty()) {
 205 #ifdef INFINITY
 206         if (double(INFINITY) > HUGE_VAL) return -INFINITY;
 207 #endif
 208         return -HUGE_VAL;
 209     }
 210
 211     unsigned char first = numfromstr(value, 0);
 212     size_t i = 0;
 213
 214     first ^= static_cast<unsigned char>(first & 0xc0) >> 1;
 215     bool negative = !(first & 0x80);
 216     bool exponent_negative = (first & 0x40);
 217     bool explen = !(first & 0x20);
 218     int exponent = first & 0x1f;
 219     if (!explen) {
 220         exponent >>= 2;
 221         if (negative ^ exponent_negative) exponent ^= 0x07;
 222     } else {
 223         first = numfromstr(value, ++i);
 224         exponent <<= 6;
 225         exponent |= (first >> 2);
 226         if (negative ^ exponent_negative) exponent ^= 0x07ff;
 227     }
 228
 229     unsigned word1;
 230     word1 = (unsigned(first & 0x03) << 24);
 231     word1 |= numfromstr(value, ++i) << 16;
 232     word1 |= numfromstr(value, ++i) << 8;
 233     word1 |= numfromstr(value, ++i);
 234
 235     unsigned word2 = 0;
 236     if (i < value.size()) {
 237         word2 = numfromstr(value, ++i) << 24;
 238         word2 |= numfromstr(value, ++i) << 16;
 239         word2 |= numfromstr(value, ++i) << 8;
 240         word2 |= numfromstr(value, ++i);
 241     }
 242
 243     if (negative) {
 244         word1 = -word1;
 245         if (word2 != 0) ++word1;
 246         word2 = -word2;
 247         UNITTEST_ASSERT((word1 & 0xf0000000) != 0);
 248         word1 &= 0x03ffffff;
 249     }
 250     if (!negative) word1 |= 1<<26;
 251
 252     double mantissa = 0;
 253     if (word2) mantissa = word2 / 4294967296.0; // 1<<32
 254     mantissa += word1;
 255     mantissa /= 1 << (negative ? 26 : 27);
 256
 257     if (exponent_negative) exponent = -exponent;
 258     exponent += 8;
 259
 260     if (negative) mantissa = -mantissa;
 261
 262     // We use scalbn() since it's equivalent to ldexp() when FLT_RADIX == 2
 263     // (which we currently assume), except that ldexp() will set errno if the
 264     // result overflows or underflows, which isn't really desirable here.
 265     return scalbn(mantissa, exponent);
 266 }