FN_multi_function_builder.hh

Go to the documentation of this file.

/* SPDX-FileCopyrightText: 2023 Blender Authors
 *
 * SPDX-License-Identifier: GPL-2.0-or-later */
 
#pragma once

 
#include "FN_multi_function.hh"
 
namespace blender::fn::multi_function::build {

namespace exec_presets {

enum class FallbackMode {
  Simple,
  Materialized,
};

struct Simple {
  static constexpr bool use_devirtualization = false;
  static constexpr FallbackMode fallback_mode = FallbackMode::Simple;
};

struct Materialized {
  static constexpr bool use_devirtualization = false;
  static constexpr FallbackMode fallback_mode = FallbackMode::Materialized;
};

struct AllSpanOrSingle {
  static constexpr bool use_devirtualization = true;
  static constexpr FallbackMode fallback_mode = FallbackMode::Materialized;
 
  template<typename... ParamTags, typename... LoadedParams, size_t... I>
  auto create_devirtualizers(TypeSequence<ParamTags...> /*param_tags*/,
                             std::index_sequence<I...> /*indices*/,
                             const std::tuple<LoadedParams...> &loaded_params) const
  {
    return std::make_tuple([&]() {
      using ParamTag = ParamTags;
      using T = typename ParamTag::base_type;
      if constexpr (ParamTag::category == ParamCategory::SingleInput) {
        const GVArrayImpl &varray_impl = *std::get<I>(loaded_params);
        return GVArrayDevirtualizer<T, true, true>{varray_impl};
      }
      else if constexpr (ELEM(ParamTag::category,
                              ParamCategory::SingleOutput,
                              ParamCategory::SingleMutable))
      {
        T *ptr = std::get<I>(loaded_params);
        return BasicDevirtualizer<T *>{ptr};
      }
    }()...);
  }
};

template<size_t... Indices> struct SomeSpanOrSingle {
  static constexpr bool use_devirtualization = true;
  static constexpr FallbackMode fallback_mode = FallbackMode::Materialized;
 
  template<typename... ParamTags, typename... LoadedParams, size_t... I>
  auto create_devirtualizers(TypeSequence<ParamTags...> /*param_tags*/,
                             std::index_sequence<I...> /*indices*/,
                             const std::tuple<LoadedParams...> &loaded_params) const
  {
    return std::make_tuple([&]() {
      using ParamTag = ParamTags;
      using T = typename ParamTag::base_type;
 
      if constexpr (ParamTag::category == ParamCategory::SingleInput) {
        constexpr bool UseSpan = ValueSequence<size_t, Indices...>::template contains<I>();
        const GVArrayImpl &varray_impl = *std::get<I>(loaded_params);
        return GVArrayDevirtualizer<T, true, UseSpan>{varray_impl};
      }
      else if constexpr (ELEM(ParamTag::category,
                              ParamCategory::SingleOutput,
                              ParamCategory::SingleMutable))
      {
        T *ptr = std::get<I>(loaded_params);
        return BasicDevirtualizer<T *>{ptr};
      }
    }()...);
  }
};
 
}  // namespace exec_presets
 
namespace detail {

template<typename MaskT, typename... Args, typename... ParamTags, size_t... I, typename ElementFn>
/* Perform additional optimizations on this loop because it is a very hot loop. For example, the
 * math node in geometry nodes is processed here. */
#if (defined(__GNUC__) && !defined(__clang__))
[[gnu::optimize("-funroll-loops")]] [[gnu::optimize("O3")]]
#endif
inline void execute_array(TypeSequence<ParamTags...> /*param_tags*/,
                          std::index_sequence<I...> /*indices*/,
                          ElementFn element_fn,
                          MaskT mask,
                          /* Use restrict to tell the compiler that pointer inputs do not alias
                           * each other. This is important for some compiler optimizations. */
                          Args &&__restrict... args)
{
  if constexpr (std::is_same_v<std::decay_t<MaskT>, IndexRange>) {
    /* Having this explicit loop is necessary for MSVC to be able to vectorize this. */
    const int64_t start = mask.start();
    const int64_t end = mask.one_after_last();
    for (int64_t i = start; i < end; i++) {
      element_fn(args[i]...);
    }
  }
  else {
    for (const int64_t i : mask) {
      element_fn(args[i]...);
    }
  }
}
 
enum class MaterializeArgMode {
  Unknown,
  Single,
  Span,
  Materialized,
};
 
template<typename ParamTag> struct MaterializeArgInfo {
  MaterializeArgMode mode = MaterializeArgMode::Unknown;
  const typename ParamTag::base_type *internal_span_data;
};

template<typename... ParamTags, typename ElementFn, typename... Chunks>
#if (defined(__GNUC__) && !defined(__clang__))
[[gnu::optimize("-funroll-loops")]] [[gnu::optimize("O3")]]
#endif
inline void execute_materialized_impl(TypeSequence<ParamTags...> /*param_tags*/,
                                      const ElementFn element_fn,
                                      const int64_t size,
                                      Chunks &&__restrict... chunks)
{
  for (int64_t i = 0; i < size; i++) {
    element_fn(chunks[i]...);
  }
}

template<typename... ParamTags, size_t... I, typename ElementFn, typename... LoadedParams>
inline void execute_materialized(TypeSequence<ParamTags...> /*param_tags*/,
                                 std::index_sequence<I...> /*indices*/,
                                 const ElementFn element_fn,
                                 const IndexMaskSegment mask,
                                 const std::tuple<LoadedParams...> &loaded_params)
{
 
  /* In theory, all elements could be processed in one chunk. However, that has the disadvantage
   * that large temporary arrays are needed. Using small chunks allows using small arrays, which
   * are reused multiple times, which improves cache efficiency. The chunk size also shouldn't be
   * too small, because then overhead of the outer loop over chunks becomes significant again. */
  static constexpr int64_t MaxChunkSize = 64;
  const int64_t mask_size = mask.size();
  const int64_t tmp_buffer_size = std::min(mask_size, MaxChunkSize);
 
  /* Local buffers that are used to temporarily store values for processing. */
  std::tuple<TypedBuffer<typename ParamTags::base_type, MaxChunkSize>...> temporary_buffers;
 
  /* Information about every parameter. */
  std::tuple<MaterializeArgInfo<ParamTags>...> args_info;
 
  (
      /* Setup information for all parameters. */
      [&] {
        /* Use `typedef` instead of `using` to work around a compiler bug. */
        using ParamTag = ParamTags;
        using T = typename ParamTag::base_type;
        [[maybe_unused]] MaterializeArgInfo<ParamTags> &arg_info = std::get<I>(args_info);
        if constexpr (ParamTag::category == ParamCategory::SingleInput) {
          const GVArrayImpl &varray_impl = *std::get<I>(loaded_params);
          const CommonVArrayInfo common_info = varray_impl.common_info();
          if (common_info.type == CommonVArrayInfo::Type::Single) {
            /* If an input #VArray is a single value, we have to fill the buffer with that value
             * only once. The same unchanged buffer can then be reused in every chunk. */
            const T &in_single = *static_cast<const T *>(common_info.data);
            T *tmp_buffer = std::get<I>(temporary_buffers).ptr();
            uninitialized_fill_n(tmp_buffer, tmp_buffer_size, in_single);
            arg_info.mode = MaterializeArgMode::Single;
          }
          else if (common_info.type == CommonVArrayInfo::Type::Span) {
            /* Remember the span so that it doesn't have to be retrieved in every iteration. */
            arg_info.internal_span_data = static_cast<const T *>(common_info.data);
          }
          else {
            arg_info.internal_span_data = nullptr;
          }
        }
      }(),
      ...);
 
  IndexMaskFromSegment index_mask_from_segment;
  const int64_t segment_offset = mask.offset();
 
  /* Outer loop over all chunks. */
  for (int64_t chunk_start = 0; chunk_start < mask_size; chunk_start += MaxChunkSize) {
    const int64_t chunk_end = std::min<int64_t>(chunk_start + MaxChunkSize, mask_size);
    const int64_t chunk_size = chunk_end - chunk_start;
    const IndexMaskSegment sliced_mask = mask.slice(chunk_start, chunk_size);
    const int64_t mask_start = sliced_mask[0];
    const bool sliced_mask_is_range = unique_sorted_indices::non_empty_is_range(
        sliced_mask.base_span());
 
    /* Move mutable data into temporary array. */
    if (!sliced_mask_is_range) {
      (
          [&] {
            /* Use `typedef` instead of `using` to work around a compiler bug. */
            using ParamTag = ParamTags;
            using T = typename ParamTag::base_type;
            if constexpr (ParamTag::category == ParamCategory::SingleMutable) {
              T *tmp_buffer = std::get<I>(temporary_buffers).ptr();
              T *param_buffer = std::get<I>(loaded_params);
              for (int64_t i = 0; i < chunk_size; i++) {
                new (tmp_buffer + i) T(std::move(param_buffer[sliced_mask[i]]));
              }
            }
          }(),
          ...);
    }
 
    const IndexMask *current_segment_mask = nullptr;
 
    execute_materialized_impl(
        TypeSequence<ParamTags...>(),
        element_fn,
        chunk_size,
        /* Prepare every parameter for this chunk. */
        [&] {
          using ParamTag = ParamTags;
          using T = typename ParamTag::base_type;
          [[maybe_unused]] MaterializeArgInfo<ParamTags> &arg_info = std::get<I>(args_info);
          T *tmp_buffer = std::get<I>(temporary_buffers);
          if constexpr (ParamTag::category == ParamCategory::SingleInput) {
            if (arg_info.mode == MaterializeArgMode::Single) {
              /* The single value has been filled into a buffer already reused for every chunk. */
              return const_cast<const T *>(tmp_buffer);
            }
            if (sliced_mask_is_range && arg_info.internal_span_data != nullptr) {
              /* In this case we can just use an existing span instead of "compressing" it into
               * a new temporary buffer. */
              arg_info.mode = MaterializeArgMode::Span;
              return arg_info.internal_span_data + mask_start;
            }
            const GVArrayImpl &varray_impl = *std::get<I>(loaded_params);
            if (current_segment_mask == nullptr) {
              current_segment_mask = &index_mask_from_segment.update(
                  {segment_offset, sliced_mask.base_span()});
            }
            /* As a fallback, do a virtual function call to retrieve all elements in the current
             * chunk. The elements are stored in a temporary buffer reused for every chunk. */
            varray_impl.materialize_compressed(*current_segment_mask, tmp_buffer, true);
            /* Remember that this parameter has been materialized, so that the values are
             * destructed properly when the chunk is done. */
            arg_info.mode = MaterializeArgMode::Materialized;
            return const_cast<const T *>(tmp_buffer);
          }
          else if constexpr (ELEM(ParamTag::category,
                                  ParamCategory::SingleOutput,
                                  ParamCategory::SingleMutable))
          {
            /* For outputs, just pass a pointer. This is important so that `__restrict` works. */
            if (sliced_mask_is_range) {
              /* Can write into the caller-provided buffer directly. */
              T *param_buffer = std::get<I>(loaded_params);
              return param_buffer + mask_start;
            }
            /* Use the temporary buffer. The values will have to be copied out of that
             * buffer into the caller-provided buffer afterwards. */
            return tmp_buffer;
          }
        }()...);
 
    /* Relocate outputs from temporary buffers to buffers provided by caller. */
    if (!sliced_mask_is_range) {
      (
          [&] {
            /* Use `typedef` instead of `using` to work around a compiler bug. */
            using ParamTag = ParamTags;
            using T = typename ParamTag::base_type;
            if constexpr (ELEM(ParamTag::category,
                               ParamCategory::SingleOutput,
                               ParamCategory::SingleMutable))
            {
              T *tmp_buffer = std::get<I>(temporary_buffers).ptr();
              T *param_buffer = std::get<I>(loaded_params);
              for (int64_t i = 0; i < chunk_size; i++) {
                new (param_buffer + sliced_mask[i]) T(std::move(tmp_buffer[i]));
                std::destroy_at(tmp_buffer + i);
              }
            }
          }(),
          ...);
    }
 
    (
        /* Destruct values that have been materialized before. */
        [&] {
          /* Use `typedef` instead of `using` to work around a compiler bug. */
          using ParamTag = ParamTags;
          using T = typename ParamTag::base_type;
          [[maybe_unused]] MaterializeArgInfo<ParamTags> &arg_info = std::get<I>(args_info);
          if constexpr (ParamTag::category == ParamCategory::SingleInput) {
            if (arg_info.mode == MaterializeArgMode::Materialized) {
              T *tmp_buffer = std::get<I>(temporary_buffers).ptr();
              destruct_n(tmp_buffer, chunk_size);
            }
          }
        }(),
        ...);
  }
 
  (
      /* Destruct buffers for single value inputs. */
      [&] {
        /* Use `typedef` instead of `using` to work around a compiler bug. */
        using ParamTag = ParamTags;
        using T = typename ParamTag::base_type;
        [[maybe_unused]] MaterializeArgInfo<ParamTags> &arg_info = std::get<I>(args_info);
        if constexpr (ParamTag::category == ParamCategory::SingleInput) {
          if (arg_info.mode == MaterializeArgMode::Single) {
            T *tmp_buffer = std::get<I>(temporary_buffers).ptr();
            destruct_n(tmp_buffer, tmp_buffer_size);
          }
        }
      }(),
      ...);
}
 
template<typename ElementFn, typename ExecPreset, typename... ParamTags, size_t... I>
inline void execute_element_fn_as_multi_function(const ElementFn element_fn,
                                                 const ExecPreset exec_preset,
                                                 const IndexMask &mask,
                                                 Params params,
                                                 TypeSequence<ParamTags...> /*param_tags*/,
                                                 std::index_sequence<I...> /*indices*/)
{
 
  /* Load parameters from #Params. */
  /* Contains `const GVArrayImpl *` for inputs and `T *` for outputs. */
  const auto loaded_params = std::make_tuple([&]() {
    /* Use `typedef` instead of `using` to work around a compiler bug. */
    using ParamTag = ParamTags;
    using T = typename ParamTag::base_type;
 
    if constexpr (ParamTag::category == ParamCategory::SingleInput) {
      return params.readonly_single_input(I).get_implementation();
    }
    else if constexpr (ParamTag::category == ParamCategory::SingleOutput) {
      return static_cast<T *>(params.uninitialized_single_output(I).data());
    }
    else if constexpr (ParamTag::category == ParamCategory::SingleMutable) {
      return static_cast<T *>(params.single_mutable(I).data());
    }
  }()...);
 
  /* Try execute devirtualized if enabled and the input types allow it. */
  bool executed_devirtualized = false;
  if constexpr (ExecPreset::use_devirtualization) {
    /* Get segments before devirtualization to avoid generating this code multiple times. */
    const Vector<std::variant<IndexRange, IndexMaskSegment>, 16> mask_segments =
        mask.to_spans_and_ranges<16>();
 
    const auto devirtualizers = exec_preset.create_devirtualizers(
        TypeSequence<ParamTags...>(), std::index_sequence<I...>(), loaded_params);
    executed_devirtualized = call_with_devirtualized_parameters(
        devirtualizers, [&](auto &&...args) {
          for (const std::variant<IndexRange, IndexMaskSegment> &segment : mask_segments) {
            if (std::holds_alternative<IndexRange>(segment)) {
              const auto segment_range = std::get<IndexRange>(segment);
              execute_array(TypeSequence<ParamTags...>(),
                            std::index_sequence<I...>(),
                            element_fn,
                            segment_range,
                            std::forward<decltype(args)>(args)...);
            }
            else {
              const auto segment_indices = std::get<IndexMaskSegment>(segment);
              execute_array(TypeSequence<ParamTags...>(),
                            std::index_sequence<I...>(),
                            element_fn,
                            segment_indices,
                            std::forward<decltype(args)>(args)...);
            }
          }
        });
  }
  else {
    UNUSED_VARS(exec_preset);
  }
 
  /* If devirtualized execution was disabled or not possible, use a fallback method which is
   * slower but always works. */
  if (!executed_devirtualized) {
    /* The materialized method is most common because it avoids most virtual function overhead but
     * still instantiates the function only once. */
    if constexpr (ExecPreset::fallback_mode == exec_presets::FallbackMode::Materialized) {
      mask.foreach_segment([&](const IndexMaskSegment segment) {
        execute_materialized(TypeSequence<ParamTags...>(),
                             std::index_sequence<I...>(),
                             element_fn,
                             segment,
                             loaded_params);
      });
    }
    else {
      /* This fallback is slower because it uses virtual method calls for every element. */
      mask.foreach_segment([&](const IndexMaskSegment segment) {
        execute_array(
            TypeSequence<ParamTags...>(), std::index_sequence<I...>(), element_fn, segment, [&]() {
              /* Use `typedef` instead of `using` to work around a compiler bug. */
              using ParamTag = ParamTags;
              using T = typename ParamTag::base_type;
              if constexpr (ParamTag::category == ParamCategory::SingleInput) {
                const GVArrayImpl &varray_impl = *std::get<I>(loaded_params);
                return GVArray(&varray_impl).typed<T>();
              }
              else if constexpr (ELEM(ParamTag::category,
                                      ParamCategory::SingleOutput,
                                      ParamCategory::SingleMutable))
              {
                T *ptr = std::get<I>(loaded_params);
                return ptr;
              }
            }()...);
      });
    }
  }
}

template<typename ElementFn, typename ExecPreset, typename... ParamTags>
inline auto build_multi_function_call_from_element_fn(const ElementFn element_fn,
                                                      const ExecPreset exec_preset,
                                                      TypeSequence<ParamTags...> /*param_tags*/)
{
  return [element_fn, exec_preset](const IndexMask &mask, Params params) {
    execute_element_fn_as_multi_function(element_fn,
                                         exec_preset,
                                         mask,
                                         params,
                                         TypeSequence<ParamTags...>(),
                                         std::make_index_sequence<sizeof...(ParamTags)>());
  };
}

template<typename CallFn, typename... ParamTags> class CustomMF : public MultiFunction {
 private:
  Signature signature_;
  CallFn call_fn_;
 
 public:
  CustomMF(const char *name, CallFn call_fn, TypeSequence<ParamTags...> /*param_tags*/)
      : call_fn_(std::move(call_fn))
  {
    SignatureBuilder builder{name, signature_};
    /* Loop over all parameter types and add an entry for each in the signature. */
    ([&] { builder.add(ParamTags(), ""); }(), ...);
    this->set_signature(&signature_);
  }
 
  void call(const IndexMask &mask, Params params, Context /*context*/) const override
  {
    call_fn_(mask, params);
  }
};
 
template<typename Out, typename... In, typename ElementFn, typename ExecPreset>
inline auto build_multi_function_with_n_inputs_one_output(const char *name,
                                                          const ElementFn element_fn,
                                                          const ExecPreset exec_preset,
                                                          TypeSequence<In...> /*in_types*/)
{
  constexpr auto param_tags = TypeSequence<ParamTag<ParamCategory::SingleInput, In>...,
                                           ParamTag<ParamCategory::SingleOutput, Out>>();
  auto call_fn = build_multi_function_call_from_element_fn(
      [element_fn](const In &...in, Out &out) { new (&out) Out(element_fn(in...)); },
      exec_preset,
      param_tags);
  return CustomMF(name, call_fn, param_tags);
}
 
template<typename Out1, typename Out2, typename... In, typename ElementFn, typename ExecPreset>
inline auto build_multi_function_with_n_inputs_two_outputs(const char *name,
                                                           const ElementFn element_fn,
                                                           const ExecPreset exec_preset,
                                                           TypeSequence<In...> /*in_types*/)
{
  constexpr auto param_tags = TypeSequence<ParamTag<ParamCategory::SingleInput, In>...,
                                           ParamTag<ParamCategory::SingleOutput, Out1>,
                                           ParamTag<ParamCategory::SingleOutput, Out2>>();
  auto call_fn = build_multi_function_call_from_element_fn(element_fn, exec_preset, param_tags);
  return CustomMF(name, call_fn, param_tags);
}
 
}  // namespace detail

template<typename In1,
         typename Out1,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI1_SO(const char *name,
                   const ElementFn element_fn,
                   const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_one_output<Out1>(
      name, element_fn, exec_preset, TypeSequence<In1>());
}

template<typename In1,
         typename In2,
         typename Out1,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI2_SO(const char *name,
                   const ElementFn element_fn,
                   const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_one_output<Out1>(
      name, element_fn, exec_preset, TypeSequence<In1, In2>());
}

template<typename In1,
         typename In2,
         typename In3,
         typename Out1,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI3_SO(const char *name,
                   const ElementFn element_fn,
                   const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_one_output<Out1>(
      name, element_fn, exec_preset, TypeSequence<In1, In2, In3>());
}

template<typename In1,
         typename In2,
         typename In3,
         typename In4,
         typename Out1,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI4_SO(const char *name,
                   const ElementFn element_fn,
                   const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_one_output<Out1>(
      name, element_fn, exec_preset, TypeSequence<In1, In2, In3, In4>());
}

template<typename In1,
         typename In2,
         typename In3,
         typename In4,
         typename In5,
         typename Out1,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI5_SO(const char *name,
                   const ElementFn element_fn,
                   const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_one_output<Out1>(
      name, element_fn, exec_preset, TypeSequence<In1, In2, In3, In4, In5>());
}

template<typename In1,
         typename In2,
         typename In3,
         typename In4,
         typename In5,
         typename In6,
         typename Out1,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI6_SO(const char *name,
                   const ElementFn element_fn,
                   const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_one_output<Out1>(
      name, element_fn, exec_preset, TypeSequence<In1, In2, In3, In4, In5, In6>());
}

template<typename In1,
         typename In2,
         typename In3,
         typename In4,
         typename In5,
         typename In6,
         typename In7,
         typename In8,
         typename Out1,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI8_SO(const char *name,
                   const ElementFn element_fn,
                   const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_one_output<Out1>(
      name, element_fn, exec_preset, TypeSequence<In1, In2, In3, In4, In5, In6, In7, In8>());
}

template<typename Mut1, typename ElementFn, typename ExecPreset = exec_presets::AllSpanOrSingle>
inline auto SM(const char *name,
               const ElementFn element_fn,
               const ExecPreset exec_preset = exec_presets::AllSpanOrSingle())
{
  constexpr auto param_tags = TypeSequence<ParamTag<ParamCategory::SingleMutable, Mut1>>();
  auto call_fn = detail::build_multi_function_call_from_element_fn(
      element_fn, exec_preset, param_tags);
  return detail::CustomMF(name, call_fn, param_tags);
}

template<typename In1,
         typename Out1,
         typename Out2,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI1_SO2(const char *name,
                    const ElementFn element_fn,
                    const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_two_outputs<Out1, Out2>(
      name, element_fn, exec_preset, TypeSequence<In1>());
}

template<typename In1,
         typename In2,
         typename Out1,
         typename Out2,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI2_SO2(const char *name,
                    const ElementFn element_fn,
                    const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_two_outputs<Out1, Out2>(
      name, element_fn, exec_preset, TypeSequence<In1, In2>());
}

template<typename In1,
         typename In2,
         typename In3,
         typename Out1,
         typename Out2,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI3_SO2(const char *name,
                    const ElementFn element_fn,
                    const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_two_outputs<Out1, Out2>(
      name, element_fn, exec_preset, TypeSequence<In1, In2, In3>());
}

template<typename In1,
         typename In2,
         typename In3,
         typename In4,
         typename Out1,
         typename Out2,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI4_SO2(const char *name,
                    const ElementFn element_fn,
                    const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_two_outputs<Out1, Out2>(
      name, element_fn, exec_preset, TypeSequence<In1, In2, In3, In4>());
}

template<typename In1,
         typename In2,
         typename In3,
         typename In4,
         typename In5,
         typename Out1,
         typename Out2,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI5_SO2(const char *name,
                    const ElementFn element_fn,
                    const ExecPreset exec_preset = exec_presets::Materialized())
{
  return detail::build_multi_function_with_n_inputs_two_outputs<Out1, Out2>(
      name, element_fn, exec_preset, TypeSequence<In1, In2, In3, In4, In5>());
}

template<typename In1,
         typename Out1,
         typename Out2,
         typename Out3,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI1_SO3(const char *name,
                    const ElementFn element_fn,
                    const ExecPreset exec_preset = exec_presets::Materialized())
{
  constexpr auto param_tags = TypeSequence<ParamTag<ParamCategory::SingleInput, In1>,
                                           ParamTag<ParamCategory::SingleOutput, Out1>,
                                           ParamTag<ParamCategory::SingleOutput, Out2>,
                                           ParamTag<ParamCategory::SingleOutput, Out3>>();
  auto call_fn = detail::build_multi_function_call_from_element_fn(
      element_fn, exec_preset, param_tags);
  return detail::CustomMF(name, call_fn, param_tags);
}

template<typename In1,
         typename Out1,
         typename Out2,
         typename Out3,
         typename Out4,
         typename ElementFn,
         typename ExecPreset = exec_presets::Materialized>
inline auto SI1_SO4(const char *name,
                    const ElementFn element_fn,
                    const ExecPreset exec_preset = exec_presets::Materialized())
{
  constexpr auto param_tags = TypeSequence<ParamTag<ParamCategory::SingleInput, In1>,
                                           ParamTag<ParamCategory::SingleOutput, Out1>,
                                           ParamTag<ParamCategory::SingleOutput, Out2>,
                                           ParamTag<ParamCategory::SingleOutput, Out3>,
                                           ParamTag<ParamCategory::SingleOutput, Out4>>();
  auto call_fn = detail::build_multi_function_call_from_element_fn(
      element_fn, exec_preset, param_tags);
  return detail::CustomMF(name, call_fn, param_tags);
}
 
}  // namespace blender::fn::multi_function::build
 
namespace blender::fn::multi_function {

class CustomMF_GenericConstant : public MultiFunction {
 private:
  const CPPType &type_;
  const void *value_;
  Signature signature_;
  bool owns_value_;
 
  template<typename T> friend class CustomMF_Constant;
 
 public:
  CustomMF_GenericConstant(const CPPType &type, const void *value, bool make_value_copy);
  ~CustomMF_GenericConstant() override;
  void call(const IndexMask &mask, Params params, Context context) const override;
  uint64_t hash() const override;
  bool equals(const MultiFunction &other) const override;
};

class CustomMF_GenericConstantArray : public MultiFunction {
 private:
  GSpan array_;
  Signature signature_;
 
 public:
  CustomMF_GenericConstantArray(GSpan array);
  void call(const IndexMask &mask, Params params, Context context) const override;
};

template<typename T> class CustomMF_Constant : public MultiFunction {
 private:
  T value_;
  Signature signature_;
 
 public:
  template<typename U> CustomMF_Constant(U &&value) : value_(std::forward<U>(value))
  {
    SignatureBuilder builder{"Constant", signature_};
    builder.single_output<T>("Value");
    this->set_signature(&signature_);
  }
 
  void call(const IndexMask &mask, Params params, Context /*context*/) const override
  {
    MutableSpan<T> output = params.uninitialized_single_output<T>(0);
    mask.foreach_index_optimized<int64_t>([&](const int64_t i) { new (&output[i]) T(value_); });
  }
 
  uint64_t hash() const override
  {
    return get_default_hash(value_);
  }
 
  bool equals(const MultiFunction &other) const override
  {
    const CustomMF_Constant *other1 = dynamic_cast<const CustomMF_Constant *>(&other);
    if (other1 != nullptr) {
      return value_ == other1->value_;
    }
    const CustomMF_GenericConstant *other2 = dynamic_cast<const CustomMF_GenericConstant *>(
        &other);
    if (other2 != nullptr) {
      const CPPType &type = CPPType::get<T>();
      if (type == other2->type_) {
        return type.is_equal_or_false(static_cast<const void *>(&value_), other2->value_);
      }
    }
    return false;
  }
};
 
class CustomMF_DefaultOutput : public MultiFunction {
 private:
  int output_amount_;
  Signature signature_;
 
 public:
  CustomMF_DefaultOutput(Span<DataType> input_types, Span<DataType> output_types);
  void call(const IndexMask &mask, Params params, Context context) const override;
};
 
class CustomMF_GenericCopy : public MultiFunction {
 private:
  Signature signature_;
 
 public:
  CustomMF_GenericCopy(DataType data_type);
  void call(const IndexMask &mask, Params params, Context context) const override;
};
 
}  // namespace blender::fn::multi_function