panthema / 2013 / parallel-string-sorting / parallel-string-sorting-0.6.5 / src / sequential / bingmann-sample_sortBSC.h (Download File)
/******************************************************************************
 * src/sequential/bingmann-sample_sortBSC.h
 *
 * Experiments with sequential Super Scalar String Sample-Sort (S^5).
 *
 * Binary search on splitters with bucket cache. Also with equality branch.
 *
 ******************************************************************************
 * Copyright (C) 2013 Timo Bingmann <tb@panthema.net>
 *
 * 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 3 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, see <http://www.gnu.org/licenses/>.
 *****************************************************************************/

namespace bingmann_sample_sortBSC {

using namespace bingmann_sample_sort;

// ----------------------------------------------------------------------------

/// binary search on splitter array for bucket number
inline unsigned int
find_bkt_assembler(const key_type& key, const key_type* splitter, size_t leaves)
{
/*
    // straightforward binary search
    unsigned int lo = 0, hi = leaves;

    while ( lo < hi )
    {
        unsigned int mid = (lo + hi) >> 1;
        assert(mid < leaves);

        if (key <= splitter[mid])
            hi = mid;
        else // (key > splitter[mid])
            lo = mid + 1;
    }
*/
    unsigned int lo;

#if 0
    // assember code of binary search generated by GCC
    asm("xorl   %%edx, %%edx \n"          // edx = lo
        "movl	%[leaves], %%ecx \n"      // ecx = hi
        // body of while loop
        ".myL2985: \n"
        "leal   (%%rcx,%%rdx), %%eax \n"
        "shrl	%%eax \n"                 // eax = mid = (lo + hi) >> 1;
        "movl	%%eax, %%edi \n"
        "cmpq	%[key], (%[splitter],%%rax,8) \n"
        "jb	.myL2987 \n"
        ".myL3019: \n"
        "cmpl	%%eax, %%edx \n"          // lo < hi
        "movl	%%eax, %%ecx \n"          // ecx = lo
        "jae	.myL2989 \n"              // if (lo < hi) -> end
        "addl	%%edx, %%eax \n"          // eax = ecx + edx = lo + hi
        "shrl	%%eax \n"                 // eax = (lo+hi) >> 1;
        "movl	%%eax, %%edi \n"
        "cmpq	(%[splitter],%%rdi,8), %[key] \n"
        "jbe	.myL3019 \n"
        ".myL2987: \n"
        "leal	1(%%rax), %%edx \n"
        "cmpl	%%edx, %%ecx \n"
        "ja	.myL2985 \n"
        ".myL2989: \n"
        : "=d" (lo)
        : [leaves] "g" (leaves), [key] "r" (key), [splitter] "r" (splitter)
        : "eax", "ecx", "edi");
#endif

    // hand-coded assembler binary search with conditional moves
    asm("xorl   %%ecx, %%ecx \n"          // ecx = lo
        "movl	%[leaves], %%edx \n"      // edx = hi
        // body of while loop
        "1: \n"
        "leal   (%%rcx,%%rdx), %%eax \n"
        "shrl	%%eax \n"                 // eax = mid = (lo + hi) >> 1;
        "cmpq	(%[splitter],%%rax,8), %[key] \n"
        "cmovbe %%eax, %%edx \n"
        "leal	1(%%rax), %%eax \n"
        "cmova  %%eax, %%ecx \n"
        "cmpl	%%edx, %%ecx \n"          // lo < hi
        "jb	1b \n"                    // if (lo < hi) -> loop
        : "=&c" (lo)
        : [leaves] "g" (leaves), [key] "r" (key), [splitter] "r" (splitter)
        : "eax", "edx");

#if 0
    // Verify result of binary search:
    int pos = leaves-1;
    while ( pos >= 0 && key <= splitter[pos] ) --pos;
    pos++;

    std::cout << "lo " << lo << " pos " << pos << "\n";
    if (lo != pos) abort();
#endif

    size_t b = lo * 2;                              // < bucket
    if (lo < leaves && splitter[lo] == key) b += 1; // equal bucket

    return b;
}

/// Variant of string sample-sort: use binary search on splitters, no caching.
template <unsigned int (*find_bkt)(const key_type&, const key_type*, size_t)>
void sample_sortBSC(string* strings, size_t n, size_t depth)
{
#if 0
    static const size_t leaves = 32;
#else
    //static const size_t l2cache = 256*1024;

    // bounding equations:
    // splitters            + bktsize
    // n * sizeof(key_type) + (2*n+1) * sizeof(size_t) <= l2cache

    static const size_t leaves = ( l2cache - sizeof(size_t) ) / (sizeof(key_type) + 2 * sizeof(size_t));

#endif

    if (n < g_samplesort_smallsort)
    {
        g_rs_steps++;
        return bingmann_radix_sort::msd_CI5(strings, n, depth);
    }
    g_ss_steps++;

    //std::cout << "leaves: " << leaves << "\n";

    // step 1: select splitters with oversampling

    size_t samplesize = oversample_factor * leaves;

    key_type* samples = new key_type[ samplesize ];

    LCGRandom rng(&samples);

    for (unsigned int i = 0; i < samplesize; ++i)
    {
        samples[i] = get_char<key_type>(strings[ rng() % n ], depth);
    }

    std::sort(samples, samples + samplesize);

    key_type splitter[leaves];
    unsigned char splitter_lcp[leaves];

    DBG(debug_splitter, "splitter:");
    splitter_lcp[0] = 0; // sentinel for first < everything bucket
    for (size_t i = 0, j = oversample_factor/2; i < leaves; ++i)
    {
        splitter[i] = samples[j];
        DBG(debug_splitter, "key " << toHex(splitter[i]));

        if (i != 0) {
            key_type xorSplit = splitter[i-1] ^ splitter[i];

            DBG1(debug_splitter, "    XOR -> " << toHex(xorSplit) << " - ");

            DBG3(debug_splitter, count_high_zero_bits(xorSplit) << " bits = "
                << count_high_zero_bits(xorSplit) / 8 << " chars lcp");

            splitter_lcp[i] = count_high_zero_bits(xorSplit) / 8;
        }

        j += oversample_factor;
    }

    delete [] samples;

    // step 2: classify all strings and count bucket sizes

    static const size_t bktnum = 2*leaves+1;

    uint16_t* bktcache = new uint16_t[n];

    for (size_t si = 0; si < n; ++si)
    {
        // binary search in splitter with equal check
        key_type key = get_char<key_type>(strings[si], depth);

        unsigned int b = find_bkt(key, splitter, leaves);

        assert(b < bktnum);

        bktcache[si] = b;
    }

    size_t* bktsize = new size_t[bktnum];
    memset(bktsize, 0, bktnum * sizeof(size_t));

    for (size_t si = 0; si < n; ++si)
        ++bktsize[ bktcache[si] ];

    if (debug_bucketsize)
    {
        DBG1(1, "bktsize: ");
        for (size_t i = 0; i < bktnum; ++i)
        {
            DBG2(1, bktsize[i] << " ");
        }
        DBG3(1, "");
    }

    // step 3: prefix sum

    size_t bktindex[bktnum];
    bktindex[0] = bktsize[0];
    size_t last_bkt_size = bktsize[0];
    for (unsigned int i=1; i < bktnum; ++i) {
        bktindex[i] = bktindex[i-1] + bktsize[i];
        if (bktsize[i]) last_bkt_size = bktsize[i];
    }
    assert(bktindex[bktnum-1] == n);

    // step 4: premute in-place

    for (size_t i=0, j; i < n - last_bkt_size; )
    {
        string perm = strings[i];
        uint16_t permbkt = bktcache[i];

        while ( (j = --bktindex[ permbkt ]) > i )
        {
            std::swap(perm, strings[j]);
            std::swap(permbkt, bktcache[j]);
        }

        strings[i] = perm;
        i += bktsize[ permbkt ];
    }

    delete [] bktcache;

    // step 5: recursion

    size_t i = 0, bsum = 0;
    while (i < bktnum-1)
    {
        // i is even -> bkt[i] is less-than bucket
        if (bktsize[i] > 1)
        {
            DBG(debug_recursion, "Recurse[" << depth << "]: < bkt " << bsum << " size " << bktsize[i] << " lcp " << int(splitter_lcp[i/2]));
            sample_sortBSC<find_bkt>(strings+bsum, bktsize[i], depth + splitter_lcp[i/2]);
        }
        bsum += bktsize[i++];

        // i is odd -> bkt[i] is equal bucket
        if (bktsize[i] > 1)
        {
            if ( (splitter[i/2] & 0xFF) == 0 ) { // equal-bucket has NULL-terminated key, done.
                DBG(debug_recursion, "Recurse[" << depth << "]: = bkt " << bsum << " size " << bktsize[i] << " is done!");
            }
            else {
                DBG(debug_recursion, "Recurse[" << depth << "]: = bkt " << bsum << " size " << bktsize[i] << " lcp keydepth!");
                sample_sortBSC<find_bkt>(strings+bsum, bktsize[i], depth + sizeof(key_type));
            }
        }
        bsum += bktsize[i++];
    }
    if (bktsize[i] > 0)
    {
        DBG(debug_recursion, "Recurse[" << depth << "]: > bkt " << bsum << " size " << bktsize[i] << " no lcp");
        sample_sortBSC<find_bkt>(strings+bsum, bktsize[i], depth);
    }
    bsum += bktsize[i++];
    assert(i == bktnum && bsum == n);

    delete [] bktsize;
}

void bingmann_sample_sortBSC(string* strings, size_t n)
{
    sample_sort_pre();
    sample_sortBSC<bingmann_sample_sortBS::find_bkt_binsearch>(strings,n,0);
    sample_sort_post();
}

CONTESTANT_REGISTER(bingmann_sample_sortBSC, "bingmann/sample_sortBSC",
                    "bingmann/sample_sortBSC (binary search, bkt cache)")

void bingmann_sample_sortBSCA(string* strings, size_t n)
{
    sample_sort_pre();
    sample_sortBSC<find_bkt_assembler>(strings,n,0);
    sample_sort_post();
}

CONTESTANT_REGISTER(bingmann_sample_sortBSCA, "bingmann/sample_sortBSCA",
                    "bingmann/sample_sortBSCA (binary search, assembler CMOV, bkt cache)")

// ----------------------------------------------------------------------------

/// binary search on splitter array for bucket number
inline unsigned int
find_bkt_equal(const key_type& key, const key_type* splitter, size_t leaves)
{
    // straightforward binary search
    unsigned int lo = 0, hi = leaves;

    while ( lo < hi )
    {
        unsigned int mid = (lo + hi) >> 1;
        assert(mid < leaves);

        if (key == splitter[mid])
            return 2 * mid + 1;
        else if (key < splitter[mid])
            hi = mid;
        else // (key > splitter[mid])
            lo = mid + 1;
    }

    return 2 * lo; // < bucket
}

/// binary search on splitter array for bucket number
inline unsigned int
find_bkt_asmequal(const key_type& key, const key_type* splitter, size_t leaves)
{
    unsigned int lo;

    // hand-coded assembler binary search with conditional moves
    asm("xorl   %%ecx, %%ecx \n"          // ecx = lo
        "movl	%[leaves], %%edx \n"      // edx = hi
        // body of while loop
        "1: \n"
        "leal   (%%rcx,%%rdx), %%eax \n"
        "shrl	%%eax \n"                 // eax = mid = (lo + hi) >> 1;
        "cmpq	(%[splitter],%%rax,8), %[key] \n"
        "je     2f \n"
        "cmovb  %%eax, %%edx \n"
        "leal	1(%%rax), %%eax \n"
        "cmova  %%eax, %%ecx \n"
        "cmpl	%%edx, %%ecx \n"          // lo < hi
        "jb	1b \n"                    // if (lo < hi) -> loop
        "leal   (%%ecx,%%ecx), %%eax \n"  // return 2 * lo
        "jmp    3f \n"
        "2: \n"
        "leal   1(%%rax,%%rax), %%eax \n"  // return 2 * lo + 1
        "3: \n"
        : "=&a" (lo)
        : [leaves] "g" (leaves), [key] "r" (key), [splitter] "r" (splitter)
        : "ecx", "edx");

    //assert(find_bkt_equal(key,splitter,leaves) == lo);

    return lo;
}

void bingmann_sample_sortBSCE(string* strings, size_t n)
{
    sample_sort_pre();
    bingmann_sample_sortBSC::sample_sortBSC<find_bkt_equal>(strings,n,0);
    sample_sort_post();
}

CONTESTANT_REGISTER(bingmann_sample_sortBSCE, "bingmann/sample_sortBSCE",
                    "bingmann/sample_sortBSCE (binary search equal, bkt cache)")

void bingmann_sample_sortBSCEA(string* strings, size_t n)
{
    sample_sort_pre();
    bingmann_sample_sortBSC::sample_sortBSC<find_bkt_asmequal>(strings,n,0);
    sample_sort_post();
}

CONTESTANT_REGISTER(bingmann_sample_sortBSCEA, "bingmann/sample_sortBSCEA",
                    "bingmann/sample_sortBSCEA (binary search equal, assembler CMOV, bkt cache)")

// ----------------------------------------------------------------------------

} // namespace bingmann_sample_sortBSC