magnum/src/Magnum/MeshTools/RemoveDuplicates.cpp

/*
    This file is part of Magnum.

    Copyright © 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019
              Vladimír Vondruš <mosra@centrum.cz>

    Permission is hereby granted, free of charge, to any person obtaining a
    copy of this software and associated documentation files (the "Software"),
    to deal in the Software without restriction, including without limitation
    the rights to use, copy, modify, merge, publish, distribute, sublicense,
    and/or sell copies of the Software, and to permit persons to whom the
    Software is furnished to do so, subject to the following conditions:

    The above copyright notice and this permission notice shall be included
    in all copies or substantial portions of the Software.

    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
    THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
    FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
    DEALINGS IN THE SOFTWARE.
*/

#include "RemoveDuplicates.h"

#include <cstring>
#include <Corrade/Containers/StridedArrayView.h>
#include <Corrade/Utility/Algorithms.h>

#include "Magnum/MeshTools/Concatenate.h"
#include "Magnum/MeshTools/Interleave.h"
#include "Magnum/Trade/MeshData.h"

namespace Magnum { namespace MeshTools {

struct ArrayEqual {
    bool operator()(Containers::ArrayView<const char> a, Containers::ArrayView<const char> b) const {
        CORRADE_INTERNAL_ASSERT(a.size() == b.size());
        return std::memcmp(a, b, a.size()) == 0;
    }
};

struct ArrayHash {
    std::size_t operator()(Containers::ArrayView<const char> a) const {
        return *reinterpret_cast<const std::size_t*>(Utility::MurmurHash2{}(a, a.size()).byteArray());
    }
};

std::size_t removeDuplicatesInto(const Containers::StridedArrayView2D<const char>& data, const Containers::StridedArrayView1D<UnsignedInt>& indices) {
    /* Assuming the second dimension is contiguous so we can calculate the
       hashes easily */
    CORRADE_ASSERT(data.empty()[0] || data.isContiguous<1>(),
        "MeshTools::removeDuplicatesInto(): second data view dimension is not contiguous", {});

    const std::size_t dataSize = data.size()[0];
    CORRADE_ASSERT(indices.size() == dataSize,
        "MeshTools::removeDuplicatesInto(): output index array has" << indices.size() << "elements but expected" << dataSize, {});

    /* Table containing index of first occurence for each unique entry.
       Reserving more buckets than necessary (i.e. as if each entry was
       unique). */
    std::unordered_map<Containers::ArrayView<const char>, UnsignedInt, ArrayHash, ArrayEqual> table{dataSize};

    /* Go through all entries */
    for(std::size_t i = 0; i != dataSize; ++i) {
        /* Try to insert new entry into the table. The inserted index points
           into the original unchanged data array. */
        const Containers::ArrayView<const char> entry = data[i].asContiguous();
        const auto result = table.emplace(entry, i);

        /* Put the (either new or already existing) index into the output
           index array */
        indices[i] = result.first->second;
    }

    CORRADE_INTERNAL_ASSERT(dataSize >= table.size());
    return table.size();
}

std::pair<Containers::Array<UnsignedInt>, std::size_t> removeDuplicates(const Containers::StridedArrayView2D<const char>& data) {
    Containers::Array<UnsignedInt> indices{Containers::NoInit, data.size()[0]};
    const std::size_t size = removeDuplicatesInto(data, indices);
    return {std::move(indices), size};
}

std::size_t removeDuplicatesInPlaceInto(const Containers::StridedArrayView2D<char>& data, const Containers::StridedArrayView1D<UnsignedInt>& indices) {
    /* Assuming the second dimension is contiguous so we can calculate the
       hashes easily */
    CORRADE_ASSERT(data.empty()[0] || data.isContiguous<1>(),
        "MeshTools::removeDuplicatesInPlaceInto(): second data view dimension is not contiguous", {});

    const std::size_t dataSize = data.size()[0];
    CORRADE_ASSERT(indices.size() == dataSize,
        "MeshTools::removeDuplicatesInPlaceInto(): output index array has" << indices.size() << "elements but expected" << dataSize, {});

    /* Table containing index of first occurence for each unique entry.
       Reserving more buckets than necessary (i.e. as if each entry was
       unique). */
    std::unordered_map<Containers::ArrayView<const char>, UnsignedInt, ArrayHash, ArrayEqual> table{dataSize};

    /* Go through all entries */
    for(std::size_t i = 0; i != dataSize; ++i) {
        /* Try to insert new entry into the table. The inserted index points
           into the new data array that has all duplicates removed. */
        const Containers::ArrayView<const char> entry = data[i].asContiguous();
        const auto result = table.emplace(entry, table.size());

        /* Put the (either new or already existing) index into the output index
           array */
        indices[i] = result.first->second;

        /* If this is a new combination, copy the data to new (earlier)
           position in the array. Data in [table.size()-1, i) are already
           present in the [0, table.size()-1) range from previous iterations so
           we aren't overwriting anything. */
        if(result.second && i != table.size() - 1)
            Utility::copy(entry, data[table.size() - 1].asContiguous());
    }

    CORRADE_INTERNAL_ASSERT(dataSize >= table.size());
    return table.size();
}

std::pair<Containers::Array<UnsignedInt>, std::size_t> removeDuplicatesInPlace(const Containers::StridedArrayView2D<char>& data) {
    Containers::Array<UnsignedInt> indices{Containers::NoInit, data.size()[0]};
    const std::size_t size = removeDuplicatesInPlaceInto(data, indices);
    return {std::move(indices), size};
}

namespace {

template<class IndexType> std::size_t removeDuplicatesIndexedInPlaceImplementation(const Containers::StridedArrayView1D<IndexType>& indices, const Containers::StridedArrayView2D<char>& data) {
    /* Somehow ~IndexType{} doesn't work for < 4byte types, as the result is
       int(-1) instead of the type I want */
    CORRADE_ASSERT(data.size()[0] <= IndexType(-1),
        "MeshTools::removeDuplicatesIndexedInPlace(): a" << sizeof(IndexType) << Debug::nospace << "-byte index type is too small for" << data.size()[0] << "vertices", {});

    /* There's no way to avoid the additional allocation, unfortunately ---
       iterating over the indices instead of data would not preserve the
       original order, which is an useful property. The float version has this
       inverted (having the *Indexed() variant as the main implementation)
       because the remapping there has to be done once for every dimension. */
    std::pair<Containers::Array<UnsignedInt>, std::size_t> result = removeDuplicatesInPlace(data);
    for(auto& i: indices) i = result.first[i];
    return result.second;
}

}

std::size_t removeDuplicatesIndexedInPlace(const Containers::StridedArrayView1D<UnsignedInt>& indices, const Containers::StridedArrayView2D<char>& data) {
    return removeDuplicatesIndexedInPlaceImplementation(indices, data);
}

std::size_t removeDuplicatesIndexedInPlace(const Containers::StridedArrayView1D<UnsignedShort>& indices, const Containers::StridedArrayView2D<char>& data) {
    return removeDuplicatesIndexedInPlaceImplementation(indices, data);
}

std::size_t removeDuplicatesIndexedInPlace(const Containers::StridedArrayView1D<UnsignedByte>& indices, const Containers::StridedArrayView2D<char>& data) {
    return removeDuplicatesIndexedInPlaceImplementation(indices, data);
}

std::size_t removeDuplicatesIndexedInPlace(const Containers::StridedArrayView2D<char>& indices, const Containers::StridedArrayView2D<char>& data) {
    CORRADE_ASSERT(indices.isContiguous<1>(), "MeshTools::removeDuplicatesIndexedInPlace(): second index view dimension is not contiguous", {});
    if(indices.size()[1] == 4)
        return removeDuplicatesIndexedInPlace(Containers::arrayCast<1, UnsignedInt>(indices), data);
    else if(indices.size()[1] == 2)
        return removeDuplicatesIndexedInPlace(Containers::arrayCast<1, UnsignedShort>(indices), data);
    else {
        CORRADE_ASSERT(indices.size()[1] == 1, "MeshTools::removeDuplicatesIndexedInPlace(): expected index type size 1, 2 or 4 but got" << indices.size()[1], {});
        return removeDuplicatesIndexedInPlace(Containers::arrayCast<1, UnsignedByte>(indices), data);
    }
}

Trade::MeshData removeDuplicates(const Trade::MeshData& data) {
    return removeDuplicates(Trade::MeshData{data.primitive(),
        {}, data.indexData(), Trade::MeshIndexData{data.indices()},
        {}, data.vertexData(), Trade::meshAttributeDataNonOwningArray(data.attributeData()),
        data.vertexCount()});
}

Trade::MeshData removeDuplicates(Trade::MeshData&& data) {
    CORRADE_ASSERT(data.attributeCount(),
        "MeshTools::removeDuplicates(): can't remove duplicates in an attributeless mesh",
        (Trade::MeshData{MeshPrimitive::Points, 0}));

    /* Turn the passed data into an interleaved owned mutable instance we can
       operate on -- concatenate() alone only makes the data owned,
       interleave() alone only makes the data interleaved (but those can stay
       non-owned). There's a chance the original data are already like this, in
       which case this will be just a passthrough. */
    /** @todo concatenate() causes the resulting index type to be UnsignedInt
        always, replace with owned() or some such when that's done */
    Trade::MeshData ownedInterleaved = interleave(concatenate(std::move(data)));

    const Containers::StridedArrayView2D<char> vertexData = MeshTools::interleavedMutableData(ownedInterleaved);

    UnsignedInt uniqueVertexCount;
    Containers::Array<char> indexData;
    MeshIndexType indexType;
    if(ownedInterleaved.isIndexed()) {
        uniqueVertexCount = removeDuplicatesIndexedInPlace(ownedInterleaved.mutableIndices(), vertexData);
        indexData = ownedInterleaved.releaseIndexData();
        indexType = ownedInterleaved.indexType();
    } else {
        indexData = Containers::Array<char>{Containers::NoInit, ownedInterleaved.vertexCount()*sizeof(UnsignedInt)};
        uniqueVertexCount = removeDuplicatesInPlaceInto(vertexData, Containers::arrayCast<UnsignedInt>(indexData));
        indexType = MeshIndexType::UnsignedInt;
    }

    /* Allocate a new, shorter vertex data and copy the prefix */
    /** @todo better idea? even if we would use growable arrays in duplicate()
        or interleave() above, arrayResize() wouldn't release the excessive
        memory in any way. This is basically equivalent to STL's
        shrink_to_fit(), which also copies */
    Containers::Array<char> uniqueVertexData{Containers::NoInit, uniqueVertexCount*vertexData.size()[1]};
    Utility::copy(vertexData.prefix(uniqueVertexCount),
        Containers::StridedArrayView2D<char>{uniqueVertexData, {uniqueVertexCount, vertexData.size()[1]}});

    /* Route all attributes to the new vertex data */
    Containers::Array<Trade::MeshAttributeData> attributeData{ownedInterleaved.attributeCount()};
    for(UnsignedInt i = 0; i != ownedInterleaved.attributeCount(); ++i)
        attributeData[i] = Trade::MeshAttributeData{ownedInterleaved.attributeName(i),
            ownedInterleaved.attributeFormat(i),
            ownedInterleaved.attributeArraySize(i),
            Containers::StridedArrayView1D<void>{uniqueVertexData,
                uniqueVertexData.data() + ownedInterleaved.attributeOffset(i),
                uniqueVertexCount,
                ownedInterleaved.attributeStride(i)}};

    Trade::MeshIndexData indices{indexType, indexData};
    return Trade::MeshData{ownedInterleaved.primitive(),
        std::move(indexData), indices,
        std::move(uniqueVertexData), std::move(attributeData),
        uniqueVertexCount};
}

}}