src/types.hpp

/*
 * Copyright (c) 2020 NVIDIA Corporation.
 * Copyright (c) 2018-2020 Chris Choy (chrischoy@ai.stanford.edu).
 *
 * 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.
 *
 * Please cite "4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural
 * Networks", CVPR'19 (https://arxiv.org/abs/1904.08755) if you use any part
 * of the code.
 */
#ifndef TYPES_HPP
#define TYPES_HPP

#include <array>
#include <functional>
#include <pybind11/pybind11.h>
#include <robin_hood.h>
#include <string>
#include <tuple>
#include <vector>

namespace minkowski {

namespace py = pybind11;

// clang-format off
template <typename uint_type, typename int_type, typename float_type>
struct type_wrapper {
  using tensor_order_type        = uint_type;
  using index_type               = uint_type;
  using stride_type              = std::vector<uint_type>;
  using size_type                = uint_type;
  using dcoordinate_type         = int_type;
  using ccoordinate_type         = float_type;
  using feature_type             = float_type;
  using coordinate_map_hash_type = uint64_t;
  using index_vector_type        = std::vector<index_type>;
};
// clang-format on

using default_types = type_wrapper<uint32_t, int32_t, float>;

// For hashing kernel sizes, strides, nd dilations.
template <default_types::tensor_order_type D,
          typename int_type = default_types::index_type>
using dim_array = std::array<int_type, D>;

template <typename data_type, typename size_type = default_types::size_type>
struct ptr_vector {

  ptr_vector(data_type *ptr, size_type size) : ptr_(ptr), size_(size) {}
  size_type size() const { return size_; };
  data_type *data() { return ptr_; };
  const data_type *data() const { return ptr_; };

  // members
  data_type *ptr_;
  default_types::size_type size_;
};

using coordinate_map_key_type =
    std::pair<default_types::stride_type, std::string>;

struct coordinate_map_key_hasher {
  using result_type = size_t;

  result_type operator()(coordinate_map_key_type const &key) const {
    auto hash = robin_hood::hash_bytes(
        key.first.data(), sizeof(default_types::size_type) * key.first.size());
    hash ^= std::hash<std::string>{}(key.second);
    return hash;
  }
};

struct coordinate_map_key_comparator {
  bool operator()(coordinate_map_key_type const &lhs,
                  coordinate_map_key_type const &rhs) const {
    auto vec_less = lhs.first < rhs.first;
    if (!vec_less && (lhs.first == rhs.first)) {
      return std::lexicographical_compare(lhs.second.begin(), lhs.second.end(),
                                          rhs.second.begin(), rhs.second.end());
    }
    return vec_less;
  }
};

template <typename vector_type>
std::vector<vector_type>
initialize_maps(default_types::size_type number_of_vectors,
                default_types::size_type vector_size) {
  auto vectors = std::vector<vector_type>();
  for (default_types::size_type i = 0; i < number_of_vectors; ++i) {
    auto vector = vector_type(vector_size);
    vectors.push_back(std::move(vector));
  }
  return vectors;
}

// GPU memory manager backend. No effect with CPU_ONLY build
namespace GPUMemoryAllocatorBackend {
enum Type { PYTORCH = 0, CUDA = 1 };
}

namespace MapManagerType {
enum Type { CPU = 0, CUDA = 1, C10 = 2 };
}

namespace CUDAKernelMapMode {
enum Mode { MEMORY_EFFICIENT = 0, SPEED_OPTIMIZED = 1 };
}

namespace MinkowskiAlgorithm {
enum Mode { DEFAULT = 0, MEMORY_EFFICIENT = 1, SPEED_OPTIMIZED = 2 };
}

namespace CoordinateMapBackend {
enum Type { CPU = 0, CUDA = 1 };
}

namespace RegionType {
enum Type { HYPER_CUBE, HYPER_CROSS, CUSTOM };
}

namespace PoolingMode {
enum Type {
  LOCAL_SUM_POOLING,
  LOCAL_AVG_POOLING,
  LOCAL_MAX_POOLING,
  GLOBAL_SUM_POOLING_DEFAULT,
  GLOBAL_AVG_POOLING_DEFAULT,
  GLOBAL_MAX_POOLING_DEFAULT,
  GLOBAL_SUM_POOLING_KERNEL,
  GLOBAL_AVG_POOLING_KERNEL,
  GLOBAL_MAX_POOLING_KERNEL,
  GLOBAL_SUM_POOLING_PYTORCH_INDEX,
  GLOBAL_AVG_POOLING_PYTORCH_INDEX,
  GLOBAL_MAX_POOLING_PYTORCH_INDEX
};
}

namespace BroadcastMode {
enum Type {
  ELEMENTWISE_ADDITON,
  ELEMENTWISE_MULTIPLICATION,
};
}

namespace ConvolutionMode {
enum Type {
  DEFAULT,
  DIRECT_GEMM,
  COPY_GEMM,
};
}

/* Key for KernelMap
 *
 * A tuple of (CoordinateMapKey (input),
 *             CoordinateMapKey (output),
 *             kernel stride,
 *             kernel size,
 *             kernel dilation,
 *             kernel region type,
 *             is_transpose,
 *             is_pool)
 */
using kernel_map_key_type =
    std::tuple<coordinate_map_key_type,    // in
               coordinate_map_key_type,    // out
               default_types::stride_type, // kernel size
               default_types::stride_type, // kernel stride
               default_types::stride_type, // kernel dilation
               RegionType::Type,           // kernel region type
               bool,                       // is transpose
               bool                        // is pool
               >;

// FNV64-1a
// uint64_t for unsigned long, must use CXX -m64
template <typename T> uint64_t hash_vec(T p) {
  uint64_t hash = UINT64_C(14695981039346656037);
  for (auto x : p) {
    hash ^= x;
    hash *= UINT64_C(1099511628211);
  }
  return hash;
}

template <typename hasher = coordinate_map_key_hasher>
struct kernel_map_key_hasher {
  using stride_type = default_types::stride_type;
  using result_type = size_t;

  result_type hash_stride(stride_type const &stride) const {
    return robin_hood::hash_bytes(
        stride.data(), sizeof(default_types::size_type) * stride.size());
  }

  result_type operator()(kernel_map_key_type const &key) const {
    auto const &in_map_key = std::get<0>(key);
    auto const &out_map_key = std::get<1>(key);

    result_type hash = hasher{}(in_map_key);
    hash ^= hasher{}(out_map_key);
    hash ^= hash_stride(std::get<2>(key));
    hash ^= hash_stride(std::get<3>(key));
    hash ^= hash_stride(std::get<4>(key));
    hash ^= (result_type)std::get<5>(key);
    hash ^= (result_type)std::get<6>(key);
    hash ^= (result_type)std::get<7>(key);
    return hash;
  }
};

template <typename hasher = coordinate_map_key_hasher>
struct field_to_sparse_map_key_hasher {
  using result_type = size_t;

  result_type operator()(std::pair<coordinate_map_key_type,
                                   coordinate_map_key_type> const &key) const {
    result_type hash = hasher{}(key.first);
    hash ^= hasher{}(key.second);
    return hash;
  }
};

} // end namespace minkowski

#endif // TYPES_HPP