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 <limits>
#include <numeric>
#include <unordered_map>
#include <Corrade/Containers/Array.h>
#include <Corrade/Containers/StridedArrayView.h>
#include <Corrade/Utility/Algorithms.h>
#include <Corrade/Utility/MurmurHash2.h>

#include "Magnum/Math/FunctionsBatch.h"
#include "Magnum/Math/Range.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);
    }
}

namespace {

template<class IndexType, class T> std::size_t removeDuplicatesFuzzyIndexedInPlaceImplementation(const Containers::StridedArrayView1D<IndexType>& indices, const Containers::StridedArrayView2D<T>& data, T epsilon) {
    /* Compared to the discrete version, we don't require the second dimension
       to be contiguous, as we calculate the hash from a discretized contiguous
       copy */

    /* 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::removeDuplicatesFuzzyIndexedInPlace(): a" << sizeof(IndexType) << Debug::nospace << "-byte index type is too small for" << data.size()[0] << "vertices", {});

    /* Get bounds across all dimensions. When NaNs appear, those will get
       collapsed together when you're lucky, or cause the whole data to
       disappear when you're not -- it needs a much more specialized handling
       to be robust. */
    const std::size_t vectorSize = data.size()[1];
    T range = T(0.0);
    Containers::Array<T> offsets{Containers::NoInit, vectorSize};
    {
        /** @todo this isn't really cache-efficient, do differently */
        std::size_t i = 0;
        for(Containers::StridedArrayView1D<T> dimension: data.template transposed<0, 1>()) {
            const Math::Range1D<T> minmax = Math::minmax(dimension);
            range = Math::max(minmax.size(), range);
            offsets[i++] = minmax.min();
        }
    }

    /* Make epsilon so large that std::size_t can index all vectors inside the
       bounds. */
    epsilon = Math::max(epsilon, range/T(~std::size_t{}));

    /* Table containing original vector index for each discretized vector.
       Reserving more buckets than necessary (i.e. as if each vector was
       unique). */
    std::size_t dataSize = data.size()[0];
    std::unordered_map<Containers::ArrayView<const char>, UnsignedInt, ArrayHash, ArrayEqual> table{dataSize};

    /* Index array that'll be filled in each pass and then used for remapping
       the `indices` */
    Containers::Array<UnsignedInt> remapping{Containers::NoInit, dataSize};

    /* First go with original coordinates, then move them by epsilon/2 in each
       dimension. */
    T moveAmount = T(0.0);
    Containers::Array<std::size_t> discretized{Containers::NoInit, vectorSize};
    for(std::size_t moving = 0; moving <= vectorSize; ++moving) {
        for(std::size_t i = 0; i != dataSize; ++i) {
            /* Take the original vector and discretize it -- append the move
               amount to given dimension, subtract the minmal offset and divide
               by epsilon. */
            const Containers::StridedArrayView1D<T> entry = data[i];
            for(std::size_t vi = 0; vi != vectorSize; ++vi) {
                T c = entry[vi];
                /* In iteration `0` we're not moving in any dimension, in
                   iteration `vectorSize` we're moving in `vectorSize - 1`
                   dimension */
                if(vi + 1 == moving) c += moveAmount;
                discretized[vi] = (c - offsets[vi])/epsilon;
            }

            /* Try to insert new entry into the table. The inserted index
               points into the new data array that has all duplicates removed.
               This is a similar workflow to removeDuplicatesInPlaceInto() with
               the only difference that we're remapping an existing index array
               several times over instead of creating a new one */
            const auto result = table.emplace(Containers::arrayCast<const char>(discretized), table.size());

            /* Add the (either new or already existing) index into the array */
            remapping[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]);
        }

        /* Remap the resulting index array */
        for(auto& i: indices) i = remapping[i];

        /* Move vertex coordinates by epsilon/2 in the next dimension (which
           is moving + 1 in the next loop iteration) */
        moveAmount = epsilon/2;

        /* Next time go only through the unique prefix; clear the table for the
           next pass */
        dataSize = table.size();
        table.clear();
    }

    CORRADE_INTERNAL_ASSERT(data.size()[0] >= dataSize);
    return dataSize;
}

}

std::size_t removeDuplicatesFuzzyIndexedInPlace(const Containers::StridedArrayView1D<UnsignedInt>& indices, const Containers::StridedArrayView2D<Float>& data, const Float epsilon) {
    return removeDuplicatesFuzzyIndexedInPlaceImplementation(indices, data, epsilon);
}

std::size_t removeDuplicatesFuzzyIndexedInPlace(const Containers::StridedArrayView1D<UnsignedShort>& indices, const Containers::StridedArrayView2D<Float>& data, const Float epsilon) {
    return removeDuplicatesFuzzyIndexedInPlaceImplementation(indices, data, epsilon);
}

std::size_t removeDuplicatesFuzzyIndexedInPlace(const Containers::StridedArrayView1D<UnsignedByte>& indices, const Containers::StridedArrayView2D<Float>& data, const Float epsilon) {
    return removeDuplicatesFuzzyIndexedInPlaceImplementation(indices, data, epsilon);
}

std::size_t removeDuplicatesFuzzyIndexedInPlace(const Containers::StridedArrayView1D<UnsignedInt>& indices, const Containers::StridedArrayView2D<Double>& data, const Double epsilon) {
    return removeDuplicatesFuzzyIndexedInPlaceImplementation(indices, data, epsilon);
}

std::size_t removeDuplicatesFuzzyIndexedInPlace(const Containers::StridedArrayView1D<UnsignedShort>& indices, const Containers::StridedArrayView2D<Double>& data, const Double epsilon) {
    return removeDuplicatesFuzzyIndexedInPlaceImplementation(indices, data, epsilon);
}

std::size_t removeDuplicatesFuzzyIndexedInPlace(const Containers::StridedArrayView1D<UnsignedByte>& indices, const Containers::StridedArrayView2D<Double>& data, const Double epsilon) {
    return removeDuplicatesFuzzyIndexedInPlaceImplementation(indices, data, epsilon);
}

namespace {

template<class T> std::size_t removeDuplicatesFuzzyInPlaceIntoImplementation(const Containers::StridedArrayView2D<T>& data, const Containers::StridedArrayView1D<UnsignedInt>& indices, const T epsilon) {
    CORRADE_ASSERT(indices.size() == data.size()[0],
        "MeshTools::removeDuplicatesFuzzyInPlaceInto(): output index array has" << indices.size() << "elements but expected" << data.size()[0], {});

    /* A trivial index array that'll be remapped */
    std::iota(indices.begin(), indices.end(), 0);
    const std::size_t size = removeDuplicatesFuzzyIndexedInPlaceImplementation(Containers::stridedArrayView(indices), data, epsilon);
    return size;
}

template<class T> std::pair<Containers::Array<UnsignedInt>, std::size_t> removeDuplicatesFuzzyInPlaceImplementation(const Containers::StridedArrayView2D<T>& data, const T epsilon) {
    Containers::Array<UnsignedInt> indices{Containers::NoInit, data.size()[0]};
    const std::size_t size = removeDuplicatesFuzzyInPlaceIntoImplementation(data, indices, epsilon);
    return {std::move(indices), size};
}

}

std::pair<Containers::Array<UnsignedInt>, std::size_t> removeDuplicatesFuzzyInPlace(const Containers::StridedArrayView2D<Float>& data, const Float epsilon) {
    return removeDuplicatesFuzzyInPlaceImplementation(data, epsilon);
}

std::pair<Containers::Array<UnsignedInt>, std::size_t> removeDuplicatesFuzzyInPlace(const Containers::StridedArrayView2D<Double>& data, const Double epsilon) {
    return removeDuplicatesFuzzyInPlaceImplementation(data, epsilon);
}

std::size_t removeDuplicatesFuzzyInPlaceInto(const Containers::StridedArrayView2D<Float>& data, const Containers::StridedArrayView1D<UnsignedInt>& indices, const Float epsilon) {
    return removeDuplicatesFuzzyInPlaceIntoImplementation(data, indices, epsilon);
}

std::size_t removeDuplicatesFuzzyInPlaceInto(const Containers::StridedArrayView2D<Double>& data, const Containers::StridedArrayView1D<UnsignedInt>& indices, const Double epsilon) {
    return removeDuplicatesFuzzyInPlaceIntoImplementation(data, indices, epsilon);
}

namespace {

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

}

std::size_t removeDuplicatesFuzzyIndexedInPlace(const Containers::StridedArrayView2D<char>& indices, const Containers::StridedArrayView2D<Float>& data, const Float epsilon) {
    return removeDuplicatesFuzzyIndexedInPlaceImplementation(indices, data, epsilon);
}

std::size_t removeDuplicatesFuzzyIndexedInPlace(const Containers::StridedArrayView2D<char>& indices, const Containers::StridedArrayView2D<Double>& data, const Double epsilon) {
    return removeDuplicatesFuzzyIndexedInPlaceImplementation(indices, data, epsilon);
}

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};
}

}}