reformat_emitter.cpp

/**
 * \file src/gopt/impl/reformat_emitter.cpp
 * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
 *
 * Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
 *
 * Unless required by applicable law or agreed to in writing,
 * software distributed under the License is distributed on an
 * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or
 * implied.
 */

#include "megbrain/gopt/reformat_emitter.h"
#include <numeric>
#include "megbrain/opr/tensor_manip.h"
#include "megbrain/opr/io.h"

using namespace mgb;
using namespace gopt;
using Dimension = megdnn::Dimension;
using NamedTensorShape = megdnn::NamedTensorShape;

// =================== ModifyShapeMixin ====================*/
ModifyShapeMixin::Pattern ModifyShapeMixin::mixin_analyze() const {
    static constexpr uint32_t UNDETERMINED_EXTENT =
            Dimension::UNDETERMINED_EXTENT;
    ThinHashMap<Dimension::Name, int> name2dominant;
    for (size_t i = 0; i < m_src.ndim; ++i) {
        auto name = m_src[i].name();
        if (m_src[i].extent() == UNDETERMINED_EXTENT) {
            auto insert = name2dominant.insert(std::make_pair(name, i));
            mgb_assert(insert.second);
        }
    }

    Pattern pattern(m_dest.ndim);
    for (size_t i = 0; i < m_dest.ndim; ++i) {
        auto name = m_dest[i].name();
        if (m_dest[i].extent() == UNDETERMINED_EXTENT) {
            int src_dim = name2dominant.at(name);
            bool mul = m_src[src_dim] < m_dest[i];
            int factor = mul ? (m_dest[i] / m_src[src_dim]).extent()
                             : (m_src[src_dim] / m_dest[i]).extent();
            pattern[i] = std::make_tuple(src_dim, factor, mul);
        } else {
            pattern[i] = std::make_tuple(-1, m_dest[i].extent(), false);
        }
    }
    return pattern;
}

ModifyShapeMixin::Checker ModifyShapeMixin::mixin_emit_checker(
        const Pattern& pattern) const {
    auto src = m_src;
    auto checker = [src, pattern](const VarNodeArray& input) {
        mgb_assert(input.size() >= 1);
        const auto& var = input.front();
        const auto& shp = var->shape();
        if (shp.ndim != src.ndim)
            return false;
        bool available = true;
        for (size_t i = 0; i < shp.ndim; ++i) {
            if (src[i].extent() != Dimension::UNDETERMINED_EXTENT) {
                available &= (shp[i] == src[i].extent());
            }
        }
        for (auto&& i : pattern) {
            int axis, factor;
            bool mul;
            std::tie(axis, factor, mul) = i;
            if (axis >= 0 && !mul) {
                available &= (shp[axis] % factor == 0);
            }
        }
        return available;
    };
    return checker;
}

// =================== MakeShapeEmitter ====================*/
MakeShapeEmitter::EmitResult MakeShapeEmitter::emit() const {
    auto pattern = mixin_analyze();
    auto builder = [pattern](const VarNodeArray& input) {
        mgb_assert(input.size() == 1,
                   "number of input of MakeShapeBuilder should be 1(got:%zu)",
                   input.size());
        auto sym_var = SymbolVar(input.front());
        auto shp = opr::GetVarShape::make(sym_var);
        auto cv = [&sym_var](int c) { return sym_var.make_scalar(c); };
        auto sub = [&shp, &cv](int ax) {
            return opr::IndexAt::make(shp, {{0, cv(ax)}});
        };
        SymbolVarArray axs;
        for (auto&& i : pattern) {
            int axis, factor;
            bool mul;
            std::tie(axis, factor, mul) = i;
            if (axis >= 0) {
                if (mul)
                    axs.emplace_back(sub(axis) * factor);
                else
                    axs.emplace_back(sub(axis) / factor);
            } else {
                axs.emplace_back(cv(factor));
            }
        }
        auto tshp = opr::Concat::make(axs, 0);
        return tshp.node();
    };
    auto checker = mixin_emit_checker(pattern);
    return std::make_tuple(builder, checker);
}

// =================== ReshapeEmitter ====================*/
ReshapeEmitter::EmitResult ReshapeEmitter::emit() const {
    auto pattern = mixin_analyze();
    auto builder = [pattern](const VarNodeArray& input) {
        mgb_assert(input.size() == 2,
                   "number of input of Reshape should be 2(got:%zu)",
                   input.size());
        auto ovar = opr::Reshape::make(input[0], input[1]);
        return ovar.node();
    };
    auto checker = mixin_emit_checker(pattern);
    return std::make_tuple(builder, checker);
}

// =================== DimshuffleEmitter ====================*/
DimshuffleEmitter::EmitResult DimshuffleEmitter::emit() const {
    auto&& pattern = m_pattern;
    auto builder = [pattern](const VarNodeArray& input) {
        mgb_assert(input.size() == 1,
                   "number of input of Dimshuffle should be 1(got:%zu)",
                   input.size());
        auto sym_var = SymbolVar(input.front());
        return opr::Dimshuffle::make(sym_var, pattern).node();
    };
    auto checker = [pattern](const VarNodeArray& input) {
        mgb_assert(input.size() == 1,
                   "number of input of Dimshuffle should be 1(got:%zu)",
                   input.size());
        return input.front()->shape().ndim == pattern.size();
    };
    return std::make_tuple(builder, checker);
}

// =================== ReformatEmitter ====================*/
ReformatEmitter::EmitResult ReformatEmitter::emit() const {
    auto builders = analyze();
    auto builder = [builders](const VarNodeArray& input) {
        VarNode *var, *ovar;
        var = ovar = input.front();
        if (builders.make_shape1) {
            auto shp1 = builders.make_shape1({var});
            ovar = builders.reshape1({ovar, shp1});
        }
        ovar = builders.dimshuffle({ovar});
        if (builders.make_shape2) {
            auto shp2 = builders.make_shape2({var});
            ovar = builders.reshape2({ovar, shp2});
        }
        return ovar;
    };
    auto pattern = mixin_analyze();
    auto checker = mixin_emit_checker(pattern);
    return std::make_tuple(builder, checker);
}

ReformatEmitter::UnderlyingBuilders ReformatEmitter::analyze() const {
    struct Dim {
        Dimension dim;
        int index;
        Dim(Dimension dim_, int index_) : dim{dim_}, index{index_} {}
    };
    SmallVector<Dim> src_dims;
    SmallVector<Dim> dest_dims;
    for (size_t i = 0; i < m_src.ndim; ++i)
        src_dims.emplace_back(Dim(m_src[i], i));
    for (size_t i = 0; i < m_dest.ndim; ++i)
        dest_dims.emplace_back(Dim(m_dest[i], i));
    auto compare = [](const Dim& lhs, const Dim& rhs) {
        return lhs.dim < rhs.dim;
    };
    std::sort(src_dims.begin(), src_dims.end(), compare);
    std::sort(dest_dims.begin(), dest_dims.end(), compare);
    auto src_iter = src_dims.begin();
    auto dest_iter = dest_dims.begin();
    for (; src_iter != src_dims.end() && dest_iter != dest_dims.end();) {
        if (src_iter->dim == dest_iter->dim) {
            src_iter++;
            dest_iter++;
        } else if (src_iter->dim < dest_iter->dim) {
            auto split = dest_iter->dim / src_iter->dim;
            int dim_idx = dest_iter->index;
            dest_iter =
                    dest_dims.insert(dest_iter, Dim(src_iter->dim, dim_idx));
            dest_iter++;
            dest_iter->dim = split;
            dest_iter->index = dim_idx;
            src_iter++;
        } else {
            auto split = src_iter->dim / dest_iter->dim;
            int dim_idx = src_iter->index;
            src_iter = src_dims.insert(src_iter, Dim(dest_iter->dim, dim_idx));
            src_iter++;
            src_iter->dim = split;
            src_iter->index = dim_idx;
            dest_iter++;
        }
    }
    mgb_assert(src_dims.size() == dest_dims.size());
    std::vector<int> src_perm(src_dims.size());
    std::vector<int> permute(dest_dims.size());
    std::iota(src_perm.begin(), src_perm.end(), 0);
    std::iota(permute.begin(), permute.end(), 0);
    std::sort(src_perm.begin(), src_perm.end(), [&](const int a, const int b) {
        if (src_dims[a].index != src_dims[b].index)
            return src_dims[a].index < src_dims[b].index;
        return src_dims[a].dim < src_dims[b].dim;
    });
    std::sort(permute.begin(), permute.end(), [&](const int a, const int b) {
        int perm_a = src_perm[a];
        int perm_b = src_perm[b];
        if (dest_dims[perm_a].index != dest_dims[perm_b].index)
            return dest_dims[perm_a].index < dest_dims[perm_b].index;
        return dest_dims[perm_a].dim < dest_dims[perm_b].dim;
    });
    NamedTensorShape i1, i2;
    i1.ndim = src_dims.size(), i2.ndim = dest_dims.size();
    for (size_t i = 0; i < src_dims.size(); ++i) {
        i1[i] = src_dims[src_perm[i]].dim;
        i2[i] = src_dims[src_perm[permute[i]]].dim;
    }
    UnderlyingBuilders builders;
    if (!m_src.eq_shape(i1)) {
        builders.make_shape1 = std::get<0>(MakeShapeEmitter(m_src, i1).emit());
        builders.reshape1 = std::get<0>(ReshapeEmitter(m_src, i1).emit());
    }
    builders.dimshuffle = std::get<0>(DimshuffleEmitter(permute).emit());
    if (!m_dest.eq_shape(i2)) {
        builders.make_shape2 =
                std::get<0>(MakeShapeEmitter(m_src, m_dest).emit());
        builders.reshape2 = std::get<0>(ReshapeEmitter(i2, m_dest).emit());
    }
    return builders;
}

/* ============== PaddingEmitter ================= */
PaddingEmitter::EmitResult PaddingEmitter::emit() const {
    auto&& const_extent = m_const_extent;
    auto&& axis = m_axis;
    auto builder = [const_extent, axis](const VarNodeArray& vars) {
        auto i = vars[0];
        auto padding_shp_var = vars[1];
        TensorShape shape;
        shape.ndim = i->shape().ndim;
        for (size_t ax = 0; ax < shape.ndim; ++ax)
            shape[ax] = 1;
        shape[axis] = const_extent;
        auto host_val =
                std::make_shared<HostTensorND>(i->comp_node(), i->dtype());
        host_val->resize(shape);
        auto ptr = host_val->raw_ptr();
        size_t size_bytes = TensorLayout{shape, i->dtype()}.span().dist_byte();
        std::memset(ptr, 0, size_bytes);
        auto padding =
                opr::ImmutableTensor::make(*i->owner_graph(), *host_val);
        padding = opr::Broadcast::make(padding, padding_shp_var);
        auto o = opr::Concat::make({i, padding}, axis);
        return o.node();
    };
    auto checker = [axis](const VarNodeArray& vars) {
        mgb_assert(vars.size() == 2);
        return vars[0]->shape().ndim > axis;
    };
    return std::make_tuple(builder, checker);
}

/* ============== SubtensorEmitter ================= */
SubtensorEmitter::EmitResult SubtensorEmitter::emit() const {
    auto&& const_extent = m_const_extent;
    auto&& axis = m_axis;
    auto builder = [const_extent, axis](const VarNodeArray& vars) {
        auto i = vars[0];
        auto x = SymbolVar(i);
        auto cv = [&x](int v) { return x.make_scalar(v); };
        using AIdx = opr::Subtensor::AxisIndexer;
        std::vector<AIdx> index(i->shape().ndim);
        for (size_t ax = 0; ax < index.size(); ++ax) {
            if (ax == axis)
                index[ax] =
                        AIdx::make_interval(ax, None, cv(const_extent), None);
            else
                index[ax] = AIdx::make_interval(ax, None, None, cv(1));
        }
        auto o = opr::Subtensor::make(x, index);
        return o.node();
    };
    auto checker = [axis](const VarNodeArray& vars) {
        mgb_assert(vars.size() == 2);
        return vars[0]->shape().ndim > axis;
    };
    return std::make_tuple(builder, checker);
}

// vim: syntax=cpp.doxygen