Reimplement Language Modelling weights

[xapian.git] / xapian-core / common / bitstream.cc

/** @file
 * @brief Classes to encode/decode a bitstream.
 */
/* Copyright (C) 2004,2005,2006,2008,2013,2014,2016,2017,2018 Olly Betts
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation; either version 2 of the
 * License, or (at your option) any later version.
 *
 * 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.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301
 * USA
 */

#include <config.h>

#include "bitstream.h"

#include <xapian/types.h>

#include "omassert.h"
#include "pack.h"

using namespace std;

// Find the position of the most significant set bit counting from 1 with
// 0 being returned if no bits are set (similar to how ffs() reports the least
// significant set bit).
template<typename T>
static inline int
highest_order_bit(T mask)
{
#ifdef HAVE_DO_CLZ
    return mask ? sizeof(T) * 8 - do_clz(mask) : 0;
#else
    // Table of results for 8 bit inputs.
    static const unsigned char hob_tab[256] = {
        0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,
        5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
        6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
        6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
        7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
        7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
        7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
        7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
        8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8
    };

    int result = 0;
    if constexpr(sizeof(T) > 4) {
        if (mask >= 0x100000000ul) {
            mask >>= 32;
            result += 32;
        }
    }
    if (mask >= 0x10000u) {
        mask >>= 16;
        result += 16;
    }
    if (mask >= 0x100u) {
        mask >>= 8;
        result += 8;
    }
    return result + hob_tab[mask];
#endif
}

namespace Xapian {

/// Shift left that's safe for shifts wider than the type.
template<typename T, typename U>
static constexpr inline
T safe_shl(T x, U shift)
{
    return (shift >= sizeof(T) * 8 ? 0 : x << shift);
}

void
BitWriter::encode(Xapian::termpos value, Xapian::termpos outof)
{
    Assert(value < outof);
    unsigned bits = highest_order_bit(outof - Xapian::termpos(1));
    // If the top bit of (outof - Xapian::termpos(1)) is set then
    // the shift will shift the bit out and give zero and the
    // subtraction will result in an unsigned overflow.
    const Xapian::termpos spare =
        UNSIGNED_OVERFLOW_OK(safe_shl(Xapian::termpos(1), bits) - outof);
    if (spare) {
        /* If we have spare values, we can use one fewer bit to encode some
         * values.  We shorten the values in the middle of the range, as
         * testing (on positional data) shows this works best.  "Managing
         * Gigabytes" suggests reversing this for the lowest level and encoding
         * the end values of the range shorter, which is contrary to our
         * testing (MG is talking about posting lists, which probably have
         * different characteristics).
         *
         * For example, if outof is 11, the codes emitted are:
         *
         * value        output
         * 0            0000
         * 1            0001
         * 2            0010
         * 3             011
         * 4             100
         * 5             101
         * 6             110
         * 7             111
         * 8            1000
         * 9            1001
         * 10           1010
         *
         * Note the LSB comes first in the bitstream, so these codes need to be
         * suffix-free to be decoded.
         */
        const Xapian::termpos mid_start = (outof - spare) / 2;
        if (value >= mid_start + spare) {
            value = (value - (mid_start + spare)) |
                    (Xapian::termpos(1) << (bits - 1));
        } else if (value >= mid_start) {
            --bits;
        }
    }

    if (bits + n_bits > sizeof(acc) * 8) {
        // We need to write more bits than there's empty room for in
        // the accumulator.  So we arrange to shift out 8 bits, then
        // adjust things so we're adding 8 fewer bits.
        Assert(bits <= sizeof(acc) * 8);
        acc |= (value << n_bits);
        buf += char(acc);
        acc >>= 8;
        value >>= 8;
        bits -= 8;
    }
    acc |= (value << n_bits);
    n_bits += bits;
    while (n_bits >= 8) {
        buf += char(acc);
        acc >>= 8;
        n_bits -= 8;
    }
}

void
BitWriter::encode_interpolative(const Xapian::VecCOW<Xapian::termpos>& pos,
                                int j, int k)
{
    // "Interpolative code" - for an algorithm description, see "Managing
    // Gigabytes" - pages 126-127 in the second edition.  You can probably
    // view those pages in google books.
    while (j + 1 < k) {
        const Xapian::termpos mid = j + (k - j) / 2;
        // Encode one out of (pos[k] - pos[j] + 1) values
        // (less some at either end because we must be able to fit
        // all the intervening pos in)
        const Xapian::termpos outof = pos[k] - pos[j] + j - k + 1;
        const Xapian::termpos lowest = pos[j] + mid - j;
        encode(pos[mid] - lowest, outof);
        encode_interpolative(pos, j, mid);
        j = mid;
    }
}

Xapian::termpos
BitReader::decode(Xapian::termpos outof, bool force)
{
    (void)force;
    Assert(force == di_current.is_initialized());
    Xapian::termpos bits = highest_order_bit(outof - Xapian::termpos(1));
    // If the top bit of (outof - Xapian::termpos(1)) is set then
    // the shift will shift the bit out and give zero and the
    // subtraction will result in an unsigned overflow.
    const Xapian::termpos spare =
        UNSIGNED_OVERFLOW_OK(safe_shl(Xapian::termpos(1), bits) - outof);
    const Xapian::termpos mid_start = (outof - spare) / 2;
    Xapian::termpos pos;
    if (spare) {
        pos = read_bits(bits - 1);
        if (pos < mid_start) {
            if (read_bits(1)) pos += mid_start + spare;
        }
    } else {
        pos = read_bits(bits);
    }
    Assert(pos < outof);
    return pos;
}

Xapian::termpos
BitReader::read_bits(int count)
{
    Xapian::termpos result;
    if (count > int(sizeof(acc) * 8 - 7)) {
        // If we need more than 7 bits less than fit in acc do the read in two
        // goes to ensure that we don't overflow acc.  This is a little more
        // conservative than it needs to be, but such large values will
        // inevitably be rare (because you can't fit very many of them into
        // the full Xapian::termpos range).
        Assert(count <= int(sizeof(acc) * 8));
        const size_t half_the_bits = sizeof(acc) * 4;
        result = read_bits(half_the_bits);
        return result | (read_bits(count - half_the_bits) << half_the_bits);
    }
    while (n_bits < count) {
        Assert(p < end);
        acc |= Xapian::termpos(static_cast<unsigned char>(*p++)) << n_bits;
        n_bits += 8;
    }
    result = acc & ((Xapian::termpos(1) << count) - Xapian::termpos(1));
    acc >>= count;
    n_bits -= count;
    return result;
}

void
BitReader::decode_interpolative(int j, int k,
                                Xapian::termpos pos_j, Xapian::termpos pos_k)
{
    Assert(!di_current.is_initialized());
    di_stack.reserve(highest_order_bit(pos_k - pos_j));
    di_current.set_j(j, pos_j);
    di_current.set_k(k, pos_k);
}

Xapian::termpos
BitReader::decode_interpolative_next()
{
    Assert(di_current.is_initialized());
    while (!di_stack.empty() || di_current.is_next()) {
        if (!di_current.is_next()) {
            Xapian::termpos pos_ret = di_current.pos_k;
            di_current = di_stack.back();
            di_stack.pop_back();
            int mid = (di_current.j + di_current.k) / 2;
            di_current.set_j(mid, pos_ret);
            return pos_ret;
        }
        di_stack.push_back(di_current);
        int mid = (di_current.j + di_current.k) / 2;
        Xapian::termpos pos_mid = decode(di_current.outof(), true) +
                                  (di_current.pos_j + mid - di_current.j);
        di_current.set_k(mid, pos_mid);
    }
#ifdef XAPIAN_ASSERTIONS
    di_current.uninit();
#endif
    return di_current.pos_k;
}

}