opr_impl.cpp

/**
 * \file dnn/src/naive/argmxx/opr_impl.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 "src/naive/argmxx/opr_impl.h"

#include "src/common/reduce_helper.h"
#include "src/common/utils.h"
#include "src/naive/handle.h"

#include <numeric>

namespace {

using namespace megdnn;

template <bool is_max>
struct traits;

template <>
struct traits<true> {
    static const float init;
    static bool better_than(float lhs, float rhs) {
        return std::isnan(lhs) ? true : lhs > rhs;
    }
};
const float traits<true>::init = std::numeric_limits<float>::lowest();

template <>
struct traits<false> {
    static const float init;
    static float better_than(float lhs, float rhs) {
        return std::isnan(lhs) ? true : lhs < rhs;
    }
};
const float traits<false>::init = std::numeric_limits<float>::max();

template <typename T, bool is_max>
void exec_forward(
        _megdnn_tensor_in src, _megdnn_tensor_out dst, const ArgmxxBase::Param& param) {
    size_t A, B, C;
    reduce::get_ABC(src.layout, A, B, C, param.axis);
    for (size_t a = 0; a < A; ++a)
        for (size_t c = 0; c < C; ++c) {
            float best_val = traits<is_max>::init;
            size_t best_arg = 0;
            for (size_t b = 0; b < B; ++b) {
                float curr_val = float(src.ptr<T>()[(a * B + b) * C + c]);
                if (traits<is_max>::better_than(curr_val, best_val)) {
                    best_val = curr_val;
                    best_arg = b;
                }
            }
            dst.ptr<dt_int32>()[a * C + c] = best_arg;
        }
}

}  // anonymous namespace

namespace megdnn {
namespace naive {

void ArgmaxForwardImpl::exec(
        _megdnn_tensor_in src, _megdnn_tensor_out dst, _megdnn_workspace workspace) {
    check_exec(src.layout, dst.layout, workspace.size);
#define cb(DType)                                                          \
    if (src.layout.dtype.enumv() == DTypeTrait<DType>::enumv) {            \
        using ctype = typename DTypeTrait<DType>::ctype;                   \
        MEGDNN_DISPATCH_CPU_KERN(                                          \
                static_cast<HandleImpl*>(handle()),                        \
                exec_forward<ctype MEGDNN_COMMA true>(src, dst, param())); \
    }
    MEGDNN_FOREACH_COMPUTING_DTYPE(cb)
#undef cb
}

void ArgminForwardImpl::exec(
        _megdnn_tensor_in src, _megdnn_tensor_out dst, _megdnn_workspace workspace) {
    check_exec(src.layout, dst.layout, workspace.size);
#define cb(DType)                                                           \
    if (src.layout.dtype.enumv() == DTypeTrait<DType>::enumv) {             \
        using ctype = typename DTypeTrait<DType>::ctype;                    \
        MEGDNN_DISPATCH_CPU_KERN(                                           \
                static_cast<HandleImpl*>(handle()),                         \
                exec_forward<ctype MEGDNN_COMMA false>(src, dst, param())); \
    }
    MEGDNN_FOREACH_COMPUTING_DTYPE(cb)
#undef cb
}

}  // namespace naive
}  // namespace megdnn

// vim: syntax=cpp.doxygen