src/mem/cache/compressors/multi.cc

/*
 * Copyright (c) 2019-2020 Inria
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are
 * met: redistributions of source code must retain the above copyright
 * notice, this list of conditions and the following disclaimer;
 * redistributions in binary form must reproduce the above copyright
 * notice, this list of conditions and the following disclaimer in the
 * documentation and/or other materials provided with the distribution;
 * neither the name of the copyright holders nor the names of its
 * contributors may be used to endorse or promote products derived from
 * this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

/** @file
 * Implementation of the a multi compressor that choses the best compression
 * among multiple compressors.
 */

#include "mem/cache/compressors/multi.hh"

#include <cmath>
#include <queue>

#include "base/bitfield.hh"
#include "base/logging.hh"
#include "base/trace.hh"
#include "debug/CacheComp.hh"
#include "params/MultiCompressor.hh"

namespace gem5
{

GEM5_DEPRECATED_NAMESPACE(Compressor, compression);
namespace compression
{

Multi::MultiCompData::MultiCompData(unsigned index,
    std::unique_ptr<Base::CompressionData> comp_data)
    : CompressionData(), index(index), compData(std::move(comp_data))
{
    setSizeBits(compData->getSizeBits());
}

uint8_t
Multi::MultiCompData::getIndex() const
{
    return index;
}

Multi::Multi(const Params &p)
  : Base(p), compressors(p.compressors),
    numEncodingBits(p.encoding_in_tags ? 0 :
        std::log2(alignToPowerOfTwo(compressors.size()))),
    multiStats(stats, *this)
{
    fatal_if(compressors.size() == 0, "There must be at least one compressor");
}

Multi::~Multi()
{
    for (auto& compressor : compressors) {
        delete compressor;
    }
}

void
Multi::setCache(BaseCache *_cache)
{
    Base::setCache(_cache);
    for (auto& compressor : compressors) {
        compressor->setCache(_cache);
    }
}

std::unique_ptr<Base::CompressionData>
Multi::compress(const std::vector<Chunk>& chunks, Cycles& comp_lat,
    Cycles& decomp_lat)
{
    struct Results
    {
        unsigned index;
        std::unique_ptr<Base::CompressionData> compData;
        Cycles decompLat;
        uint8_t compressionFactor;

        Results(unsigned index,
            std::unique_ptr<Base::CompressionData> comp_data,
            Cycles decomp_lat, std::size_t blk_size)
            : index(index), compData(std::move(comp_data)),
              decompLat(decomp_lat)
        {
            const std::size_t size = compData->getSize();
            // If the compressed size is worse than the uncompressed size,
            // we assume the size is the uncompressed size, and thus the
            // compression factor is 1.
            //
            // Some compressors (notably the zero compressor) may rely on
            // extra information being stored in the tags, or added in
            // another compression layer. Their size can be 0, so it is
            // assigned the highest possible compression factor (the original
            // block's size).
            compressionFactor = (size > blk_size) ? 1 :
                ((size == 0) ? blk_size :
                alignToPowerOfTwo(std::floor(blk_size / (double) size)));
        }
    };
    struct ResultsComparator
    {
        bool
        operator()(const std::shared_ptr<Results>& lhs,
            const std::shared_ptr<Results>& rhs) const
        {
            const std::size_t lhs_cf = lhs->compressionFactor;
            const std::size_t rhs_cf = rhs->compressionFactor;

            if (lhs_cf == rhs_cf) {
                // When they have similar compressed sizes, give the one
                // with fastest decompression privilege
                return lhs->decompLat > rhs->decompLat;
            }
            return lhs_cf < rhs_cf;
        }
    };

    // Each sub-compressor can have its own chunk size; therefore, revert
    // the chunks to raw data, so that they handle the conversion internally
    uint64_t data[blkSize / sizeof(uint64_t)];
    std::memset(data, 0, blkSize);
    fromChunks(chunks, data);

    // Find the ranking of the compressor outputs
    std::priority_queue<std::shared_ptr<Results>,
        std::vector<std::shared_ptr<Results>>, ResultsComparator> results;
    Cycles max_comp_lat;
    for (unsigned i = 0; i < compressors.size(); i++) {
        Cycles temp_decomp_lat;
        auto temp_comp_data =
            compressors[i]->compress(data, comp_lat, temp_decomp_lat);
        temp_comp_data->setSizeBits(temp_comp_data->getSizeBits() +
            numEncodingBits);
        results.push(std::make_shared<Results>(i, std::move(temp_comp_data),
            temp_decomp_lat, blkSize));
        max_comp_lat = std::max(max_comp_lat, comp_lat);
    }

    // Assign best compressor to compression data
    const unsigned best_index = results.top()->index;
    std::unique_ptr<CompressionData> multi_comp_data =
        std::unique_ptr<MultiCompData>(
            new MultiCompData(best_index, std::move(results.top()->compData)));
    DPRINTF(CacheComp, "Best compressor: %d\n", best_index);

    // Set decompression latency of the best compressor
    decomp_lat = results.top()->decompLat + decompExtraLatency;

    // Update compressor ranking stats
    for (int rank = 0; rank < compressors.size(); rank++) {
        multiStats.ranks[results.top()->index][rank]++;
        results.pop();
    }

    // Set compression latency (compression latency of the slowest compressor
    // and 1 cycle to pack)
    comp_lat = Cycles(max_comp_lat + compExtraLatency);

    return multi_comp_data;
}

void
Multi::decompress(const CompressionData* comp_data,
    uint64_t* cache_line)
{
    const MultiCompData* casted_comp_data =
        static_cast<const MultiCompData*>(comp_data);
    compressors[casted_comp_data->getIndex()]->decompress(
        casted_comp_data->compData.get(), cache_line);
}

Multi::MultiStats::MultiStats(BaseStats& base_group, Multi& _compressor)
  : statistics::Group(&base_group), compressor(_compressor),
    ADD_STAT(ranks, statistics::units::Count::get(),
             "Number of times each compressor had the nth best compression")
{
}

void
Multi::MultiStats::regStats()
{
    statistics::Group::regStats();

    const std::size_t num_compressors = compressor.compressors.size();
    ranks.init(num_compressors, num_compressors);
    for (unsigned compressor = 0; compressor < num_compressors; compressor++) {
        ranks.subname(compressor, std::to_string(compressor));
        ranks.subdesc(compressor, "Number of times compressor " +
            std::to_string(compressor) + " had the nth best compression.");
        for (unsigned rank = 0; rank < num_compressors; rank++) {
            ranks.ysubname(rank, std::to_string(rank));
        }
    }
}

} // namespace compression
} // namespace gem5