tornavis/source/blender/functions/intern/field.cc

/*
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version 2
 * of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
 */

#include "BLI_map.hh"
#include "BLI_multi_value_map.hh"
#include "BLI_set.hh"
#include "BLI_stack.hh"
#include "BLI_vector_set.hh"

#include "FN_field.hh"
#include "FN_multi_function_parallel.hh"

namespace blender::fn {

/* --------------------------------------------------------------------
 * Field Evaluation.
 */

struct FieldTreeInfo {
  /**
   * When fields are built, they only have references to the fields that they depend on. This map
   * allows traversal of fields in the opposite direction. So for every field it stores the other
   * fields that depend on it directly.
   */
  MultiValueMap<GFieldRef, GFieldRef> field_users;
  /**
   * The same field input may exist in the field tree as as separate nodes due to the way
   * the tree is constructed. This set contains every different input only once.
   */
  VectorSet<std::reference_wrapper<const FieldInput>> deduplicated_field_inputs;
};

/**
 * Collects some information from the field tree that is required by later steps.
 */
static FieldTreeInfo preprocess_field_tree(Span<GFieldRef> entry_fields)
{
  FieldTreeInfo field_tree_info;

  Stack<GFieldRef> fields_to_check;
  Set<GFieldRef> handled_fields;

  for (GFieldRef field : entry_fields) {
    if (handled_fields.add(field)) {
      fields_to_check.push(field);
    }
  }

  while (!fields_to_check.is_empty()) {
    GFieldRef field = fields_to_check.pop();
    if (field.node().is_input()) {
      const FieldInput &field_input = static_cast<const FieldInput &>(field.node());
      field_tree_info.deduplicated_field_inputs.add(field_input);
      continue;
    }
    BLI_assert(field.node().is_operation());
    const FieldOperation &operation = static_cast<const FieldOperation &>(field.node());
    for (const GFieldRef operation_input : operation.inputs()) {
      field_tree_info.field_users.add(operation_input, field);
      if (handled_fields.add(operation_input)) {
        fields_to_check.push(operation_input);
      }
    }
  }
  return field_tree_info;
}

/**
 * Retrieves the data from the context that is passed as input into the field.
 */
static Vector<const GVArray *> get_field_context_inputs(
    ResourceScope &scope,
    const IndexMask mask,
    const FieldContext &context,
    const Span<std::reference_wrapper<const FieldInput>> field_inputs)
{
  Vector<const GVArray *> field_context_inputs;
  for (const FieldInput &field_input : field_inputs) {
    const GVArray *varray = context.get_varray_for_input(field_input, mask, scope);
    if (varray == nullptr) {
      const CPPType &type = field_input.cpp_type();
      varray = &scope.construct<GVArray_For_SingleValueRef>(
          type, mask.min_array_size(), type.default_value());
    }
    field_context_inputs.append(varray);
  }
  return field_context_inputs;
}

/**
 * \return A set that contains all fields from the field tree that depend on an input that varies
 * for different indices.
 */
static Set<GFieldRef> find_varying_fields(const FieldTreeInfo &field_tree_info,
                                          Span<const GVArray *> field_context_inputs)
{
  Set<GFieldRef> found_fields;
  Stack<GFieldRef> fields_to_check;

  /* The varying fields are the ones that depend on inputs that are not constant. Therefore we
   * start the tree search at the non-constant input fields and traverse through all fields that
   * depend on them. */
  for (const int i : field_context_inputs.index_range()) {
    const GVArray *varray = field_context_inputs[i];
    if (varray->is_single()) {
      continue;
    }
    const FieldInput &field_input = field_tree_info.deduplicated_field_inputs[i];
    const GFieldRef field_input_field{field_input, 0};
    const Span<GFieldRef> users = field_tree_info.field_users.lookup(field_input_field);
    for (const GFieldRef &field : users) {
      if (found_fields.add(field)) {
        fields_to_check.push(field);
      }
    }
  }
  while (!fields_to_check.is_empty()) {
    GFieldRef field = fields_to_check.pop();
    const Span<GFieldRef> users = field_tree_info.field_users.lookup(field);
    for (GFieldRef field : users) {
      if (found_fields.add(field)) {
        fields_to_check.push(field);
      }
    }
  }
  return found_fields;
}

/**
 * Builds the #procedure so that it computes the the fields.
 */
static void build_multi_function_procedure_for_fields(MFProcedure &procedure,
                                                      ResourceScope &scope,
                                                      const FieldTreeInfo &field_tree_info,
                                                      Span<GFieldRef> output_fields)
{
  MFProcedureBuilder builder{procedure};
  /* Every input, intermediate and output field corresponds to a variable in the procedure. */
  Map<GFieldRef, MFVariable *> variable_by_field;

  /* Start by adding the field inputs as parameters to the procedure. */
  for (const FieldInput &field_input : field_tree_info.deduplicated_field_inputs) {
    MFVariable &variable = builder.add_input_parameter(
        MFDataType::ForSingle(field_input.cpp_type()), field_input.debug_name());
    variable_by_field.add_new({field_input, 0}, &variable);
  }

  /* Utility struct that is used to do proper depth first search traversal of the tree below. */
  struct FieldWithIndex {
    GFieldRef field;
    int current_input_index = 0;
  };

  for (GFieldRef field : output_fields) {
    /* We start a new stack for each output field to make sure that a field pushed later to the
     * stack does never depend on a field that was pushed before. */
    Stack<FieldWithIndex> fields_to_check;
    fields_to_check.push({field, 0});
    while (!fields_to_check.is_empty()) {
      FieldWithIndex &field_with_index = fields_to_check.peek();
      const GFieldRef &field = field_with_index.field;
      if (variable_by_field.contains(field)) {
        /* The field has been handled already. */
        fields_to_check.pop();
        continue;
      }
      /* Field inputs should already be handled above. */
      BLI_assert(field.node().is_operation());

      const FieldOperation &operation = static_cast<const FieldOperation &>(field.node());
      const Span<GField> operation_inputs = operation.inputs();

      if (field_with_index.current_input_index < operation_inputs.size()) {
        /* Not all inputs are handled yet. Push the next input field to the stack and increment the
         * input index. */
        fields_to_check.push({operation_inputs[field_with_index.current_input_index]});
        field_with_index.current_input_index++;
      }
      else {
        /* All inputs variables are ready, now gather all variables that are used by the function
         * and call it. */
        const MultiFunction &multi_function = operation.multi_function();
        Vector<MFVariable *> variables(multi_function.param_amount());

        int param_input_index = 0;
        int param_output_index = 0;
        for (const int param_index : multi_function.param_indices()) {
          const MFParamType param_type = multi_function.param_type(param_index);
          const MFParamType::InterfaceType interface_type = param_type.interface_type();
          if (interface_type == MFParamType::Input) {
            const GField &input_field = operation_inputs[param_input_index];
            variables[param_index] = variable_by_field.lookup(input_field);
            param_input_index++;
          }
          else if (interface_type == MFParamType::Output) {
            const GFieldRef output_field{operation, param_output_index};
            const bool output_is_ignored =
                field_tree_info.field_users.lookup(output_field).is_empty() &&
                !output_fields.contains(output_field);
            if (output_is_ignored) {
              /* Ignored outputs don't need a variable. */
              variables[param_index] = nullptr;
            }
            else {
              /* Create a new variable for used outputs. */
              MFVariable &new_variable = procedure.new_variable(param_type.data_type());
              variables[param_index] = &new_variable;
              variable_by_field.add_new(output_field, &new_variable);
            }
            param_output_index++;
          }
          else {
            BLI_assert_unreachable();
          }
        }
        builder.add_call_with_all_variables(multi_function, variables);
      }
    }
  }

  /* Add output parameters to the procedure. */
  Set<MFVariable *> already_output_variables;
  for (const GFieldRef &field : output_fields) {
    MFVariable *variable = variable_by_field.lookup(field);
    if (!already_output_variables.add(variable)) {
      /* One variable can be output at most once. To output the same value twice, we have to make
       * a copy first. */
      const MultiFunction &copy_fn = scope.construct<CustomMF_GenericCopy>("copy",
                                                                           variable->data_type());
      variable = builder.add_call<1>(copy_fn, {variable})[0];
    }
    builder.add_output_parameter(*variable);
  }

  /* Remove the variables that should not be destructed from the map. */
  for (const GFieldRef &field : output_fields) {
    variable_by_field.remove(field);
  }
  /* Add destructor calls for the remaining variables. */
  for (MFVariable *variable : variable_by_field.values()) {
    builder.add_destruct(*variable);
  }

  builder.add_return();

  // std::cout << procedure.to_dot() << "\n";
  BLI_assert(procedure.validate());
}

/**
 * Evaluate fields in the given context. If possible, multiple fields should be evaluated together,
 * because that can be more efficient when they share common sub-fields.
 *
 * \param scope: The resource scope that owns data that makes up the output virtual arrays. Make
 *   sure the scope is not destructed when the output virtual arrays are still used.
 * \param fields_to_evaluate: The fields that should be evaluated together.
 * \param mask: Determines which indices are computed. The mask may be referenced by the returned
 *   virtual arrays. So the underlying indices (if applicable) should live longer then #scope.
 * \param context: The context that the field is evaluated in. Used to retrieve data from each
 *   #FieldInput in the field network.
 * \param dst_varrays: If provided, the computed data will be written into those virtual arrays
 *   instead of into newly created ones. That allows making the computed data live longer than
 *   #scope and is more efficient when the data will be written into those virtual arrays
 *   later anyway.
 * \return The computed virtual arrays for each provided field. If #dst_varrays is passed, the
 *   provided virtual arrays are returned.
 */
Vector<const GVArray *> evaluate_fields(ResourceScope &scope,
                                        Span<GFieldRef> fields_to_evaluate,
                                        IndexMask mask,
                                        const FieldContext &context,
                                        Span<GVMutableArray *> dst_varrays)
{
  Vector<const GVArray *> r_varrays(fields_to_evaluate.size(), nullptr);
  const int array_size = mask.min_array_size();

  /* Destination arrays are optional. Create a small utility method to access them. */
  auto get_dst_varray_if_available = [&](int index) -> GVMutableArray * {
    if (dst_varrays.is_empty()) {
      return nullptr;
    }
    BLI_assert(dst_varrays[index] == nullptr || dst_varrays[index]->size() >= array_size);
    return dst_varrays[index];
  };

  /* Traverse the field tree and prepare some data that is used in later steps. */
  FieldTreeInfo field_tree_info = preprocess_field_tree(fields_to_evaluate);

  /* Get inputs that will be passed into the field when evaluated. */
  Vector<const GVArray *> field_context_inputs = get_field_context_inputs(
      scope, mask, context, field_tree_info.deduplicated_field_inputs);

  /* Finish fields that output an input varray directly. For those we don't have to do any further
   * processing. */
  for (const int out_index : fields_to_evaluate.index_range()) {
    const GFieldRef &field = fields_to_evaluate[out_index];
    if (!field.node().is_input()) {
      continue;
    }
    const FieldInput &field_input = static_cast<const FieldInput &>(field.node());
    const int field_input_index = field_tree_info.deduplicated_field_inputs.index_of(field_input);
    const GVArray *varray = field_context_inputs[field_input_index];
    r_varrays[out_index] = varray;
  }

  Set<GFieldRef> varying_fields = find_varying_fields(field_tree_info, field_context_inputs);

  /* Separate fields into two categories. Those that are constant and need to be evaluated only
   * once, and those that need to be evaluated for every index. */
  Vector<GFieldRef> varying_fields_to_evaluate;
  Vector<int> varying_field_indices;
  Vector<GFieldRef> constant_fields_to_evaluate;
  Vector<int> constant_field_indices;
  for (const int i : fields_to_evaluate.index_range()) {
    if (r_varrays[i] != nullptr) {
      /* Already done. */
      continue;
    }
    GFieldRef field = fields_to_evaluate[i];
    if (varying_fields.contains(field)) {
      varying_fields_to_evaluate.append(field);
      varying_field_indices.append(i);
    }
    else {
      constant_fields_to_evaluate.append(field);
      constant_field_indices.append(i);
    }
  }

  /* Evaluate varying fields if necessary. */
  if (!varying_fields_to_evaluate.is_empty()) {
    /* Build the procedure for those fields. */
    MFProcedure procedure;
    build_multi_function_procedure_for_fields(
        procedure, scope, field_tree_info, varying_fields_to_evaluate);
    MFProcedureExecutor procedure_executor{"Procedure", procedure};
    /* Add multi threading capabilities to the field evaluation. */
    const int grain_size = 10000;
    fn::ParallelMultiFunction parallel_procedure_executor{procedure_executor, grain_size};
    /* Utility variable to make easy to switch the executor. */
    const MultiFunction &executor_fn = parallel_procedure_executor;

    MFParamsBuilder mf_params{executor_fn, &mask};
    MFContextBuilder mf_context;

    /* Provide inputs to the procedure executor. */
    for (const GVArray *varray : field_context_inputs) {
      mf_params.add_readonly_single_input(*varray);
    }

    for (const int i : varying_fields_to_evaluate.index_range()) {
      const GFieldRef &field = varying_fields_to_evaluate[i];
      const CPPType &type = field.cpp_type();
      const int out_index = varying_field_indices[i];

      /* Try to get an existing virtual array that the result should be written into. */
      GVMutableArray *output_varray = get_dst_varray_if_available(out_index);
      void *buffer;
      if (output_varray == nullptr || !output_varray->is_span()) {
        /* Allocate a new buffer for the computed result. */
        buffer = scope.linear_allocator().allocate(type.size() * array_size, type.alignment());

        if (!type.is_trivially_destructible()) {
          /* Destruct values in the end. */
          scope.add_destruct_call(
              [buffer, mask, &type]() { type.destruct_indices(buffer, mask); });
        }

        r_varrays[out_index] = &scope.construct<GVArray_For_GSpan>(
            GSpan{type, buffer, array_size});
      }
      else {
        /* Write the result into the existing span. */
        buffer = output_varray->get_internal_span().data();

        r_varrays[out_index] = output_varray;
      }

      /* Pass output buffer to the procedure executor. */
      const GMutableSpan span{type, buffer, array_size};
      mf_params.add_uninitialized_single_output(span);
    }

    executor_fn.call(mask, mf_params, mf_context);
  }

  /* Evaluate constant fields if necessary. */
  if (!constant_fields_to_evaluate.is_empty()) {
    /* Build the procedure for those fields. */
    MFProcedure procedure;
    build_multi_function_procedure_for_fields(
        procedure, scope, field_tree_info, constant_fields_to_evaluate);
    MFProcedureExecutor procedure_executor{"Procedure", procedure};
    /* Run the code below even when the mask is empty, so that outputs are properly prepared.
     * Higher level code can detect this as well and just skip evaluating the field. */
    const int mask_size = mask.is_empty() ? 0 : 1;
    MFParamsBuilder mf_params{procedure_executor, mask_size};
    MFContextBuilder mf_context;

    /* Provide inputs to the procedure executor. */
    for (const GVArray *varray : field_context_inputs) {
      mf_params.add_readonly_single_input(*varray);
    }

    for (const int i : constant_fields_to_evaluate.index_range()) {
      const GFieldRef &field = constant_fields_to_evaluate[i];
      const CPPType &type = field.cpp_type();
      /* Allocate memory where the computed value will be stored in. */
      void *buffer = scope.linear_allocator().allocate(type.size(), type.alignment());

      if (!type.is_trivially_destructible() && mask_size > 0) {
        BLI_assert(mask_size == 1);
        /* Destruct value in the end. */
        scope.add_destruct_call([buffer, &type]() { type.destruct(buffer); });
      }

      /* Pass output buffer to the procedure executor. */
      mf_params.add_uninitialized_single_output({type, buffer, mask_size});

      /* Create virtual array that can be used after the procedure has been executed below. */
      const int out_index = constant_field_indices[i];
      r_varrays[out_index] = &scope.construct<GVArray_For_SingleValueRef>(
          type, array_size, buffer);
    }

    procedure_executor.call(IndexRange(mask_size), mf_params, mf_context);
  }

  /* Copy data to supplied destination arrays if necessary. In some cases the evaluation above has
   * written the computed data in the right place already. */
  if (!dst_varrays.is_empty()) {
    for (const int out_index : fields_to_evaluate.index_range()) {
      GVMutableArray *output_varray = get_dst_varray_if_available(out_index);
      if (output_varray == nullptr) {
        /* Caller did not provide a destination for this output. */
        continue;
      }
      const GVArray *computed_varray = r_varrays[out_index];
      BLI_assert(computed_varray->type() == output_varray->type());
      if (output_varray == computed_varray) {
        /* The result has been written into the destination provided by the caller already. */
        continue;
      }
      /* Still have to copy over the data in the destination provided by the caller. */
      if (output_varray->is_span()) {
        /* Materialize into a span. */
        computed_varray->materialize_to_uninitialized(mask,
                                                      output_varray->get_internal_span().data());
      }
      else {
        /* Slower materialize into a different structure. */
        const CPPType &type = computed_varray->type();
        BUFFER_FOR_CPP_TYPE_VALUE(type, buffer);
        for (const int i : mask) {
          computed_varray->get_to_uninitialized(i, buffer);
          output_varray->set_by_relocate(i, buffer);
        }
      }
      r_varrays[out_index] = output_varray;
    }
  }
  return r_varrays;
}

void evaluate_constant_field(const GField &field, void *r_value)
{
  ResourceScope scope;
  FieldContext context;
  Vector<const GVArray *> varrays = evaluate_fields(scope, {field}, IndexRange(1), context);
  varrays[0]->get_to_uninitialized(0, r_value);
}

/**
 * If the field depends on some input, the same field is returned.
 * Otherwise the field is evaluated and a new field is created that just computes this constant.
 *
 * Making the field constant has two benefits:
 * - The field-tree becomes a single node, which is more efficient when the field is evaluated many
 *   times.
 * - Memory of the input fields may be freed.
 */
GField make_field_constant_if_possible(GField field)
{
  if (field.node().depends_on_input()) {
    return field;
  }
  const CPPType &type = field.cpp_type();
  BUFFER_FOR_CPP_TYPE_VALUE(type, buffer);
  evaluate_constant_field(field, buffer);
  auto constant_fn = std::make_unique<CustomMF_GenericConstant>(type, buffer, true);
  type.destruct(buffer);
  auto operation = std::make_shared<FieldOperation>(std::move(constant_fn));
  return GField{operation, 0};
}

const GVArray *FieldContext::get_varray_for_input(const FieldInput &field_input,
                                                  IndexMask mask,
                                                  ResourceScope &scope) const
{
  /* By default ask the field input to create the varray. Another field context might overwrite
   * the context here. */
  return field_input.get_varray_for_context(*this, mask, scope);
}

IndexFieldInput::IndexFieldInput() : FieldInput(CPPType::get<int>(), "Index")
{
}

const GVArray *IndexFieldInput::get_varray_for_context(const fn::FieldContext &UNUSED(context),
                                                       IndexMask mask,
                                                       ResourceScope &scope) const
{
  /* TODO: Investigate a similar method to IndexRange::as_span() */
  auto index_func = [](int i) { return i; };
  return &scope.construct<
      fn::GVArray_For_EmbeddedVArray<int, VArray_For_Func<int, decltype(index_func)>>>(
      mask.min_array_size(), mask.min_array_size(), index_func);
}

/* --------------------------------------------------------------------
 * FieldOperation.
 */

FieldOperation::FieldOperation(std::shared_ptr<const MultiFunction> function,
                               Vector<GField> inputs)
    : FieldOperation(*function, std::move(inputs))
{
  owned_function_ = std::move(function);
}

static bool any_field_depends_on_input(Span<GField> fields)
{
  for (const GField &field : fields) {
    if (field.node().depends_on_input()) {
      return true;
    }
  }
  return false;
}

FieldOperation::FieldOperation(const MultiFunction &function, Vector<GField> inputs)
    : FieldNode(false, any_field_depends_on_input(inputs)),
      function_(&function),
      inputs_(std::move(inputs))
{
}

void FieldOperation::foreach_field_input(FunctionRef<void(const FieldInput &)> foreach_fn) const
{
  for (const GField &field : inputs_) {
    field.node().foreach_field_input(foreach_fn);
  }
}

/* --------------------------------------------------------------------
 * FieldInput.
 */

FieldInput::FieldInput(const CPPType &type, std::string debug_name)
    : FieldNode(true, true), type_(&type), debug_name_(std::move(debug_name))
{
}

void FieldInput::foreach_field_input(FunctionRef<void(const FieldInput &)> foreach_fn) const
{
  foreach_fn(*this);
}

/* --------------------------------------------------------------------
 * FieldEvaluator.
 */

static Vector<int64_t> indices_from_selection(const VArray<bool> &selection)
{
  /* If the selection is just a single value, it's best to avoid calling this
   * function when constructing an IndexMask and use an IndexRange instead. */
  BLI_assert(!selection.is_single());

  Vector<int64_t> indices;
  if (selection.is_span()) {
    Span<bool> span = selection.get_internal_span();
    for (const int64_t i : span.index_range()) {
      if (span[i]) {
        indices.append(i);
      }
    }
  }
  else {
    for (const int i : selection.index_range()) {
      if (selection[i]) {
        indices.append(i);
      }
    }
  }
  return indices;
}

int FieldEvaluator::add_with_destination(GField field, GVMutableArray &dst)
{
  const int field_index = fields_to_evaluate_.append_and_get_index(std::move(field));
  dst_varrays_.append(&dst);
  output_pointer_infos_.append({});
  return field_index;
}

int FieldEvaluator::add_with_destination(GField field, GMutableSpan dst)
{
  GVMutableArray &varray = scope_.construct<GVMutableArray_For_GMutableSpan>(dst);
  return this->add_with_destination(std::move(field), varray);
}

int FieldEvaluator::add(GField field, const GVArray **varray_ptr)
{
  const int field_index = fields_to_evaluate_.append_and_get_index(std::move(field));
  dst_varrays_.append(nullptr);
  output_pointer_infos_.append(OutputPointerInfo{
      varray_ptr, [](void *dst, const GVArray &varray, ResourceScope &UNUSED(scope)) {
        *(const GVArray **)dst = &varray;
      }});
  return field_index;
}

int FieldEvaluator::add(GField field)
{
  const int field_index = fields_to_evaluate_.append_and_get_index(std::move(field));
  dst_varrays_.append(nullptr);
  output_pointer_infos_.append({});
  return field_index;
}

void FieldEvaluator::evaluate()
{
  BLI_assert_msg(!is_evaluated_, "Cannot evaluate fields twice.");
  Array<GFieldRef> fields(fields_to_evaluate_.size());
  for (const int i : fields_to_evaluate_.index_range()) {
    fields[i] = fields_to_evaluate_[i];
  }
  evaluated_varrays_ = evaluate_fields(scope_, fields, mask_, context_, dst_varrays_);
  BLI_assert(fields_to_evaluate_.size() == evaluated_varrays_.size());
  for (const int i : fields_to_evaluate_.index_range()) {
    OutputPointerInfo &info = output_pointer_infos_[i];
    if (info.dst != nullptr) {
      info.set(info.dst, *evaluated_varrays_[i], scope_);
    }
  }
  is_evaluated_ = true;
}

IndexMask FieldEvaluator::get_evaluated_as_mask(const int field_index)
{
  const GVArray &varray = this->get_evaluated(field_index);
  GVArray_Typed<bool> typed_varray{varray};

  if (typed_varray->is_single()) {
    if (typed_varray->get_internal_single()) {
      return IndexRange(typed_varray.size());
    }
    return IndexRange(0);
  }

  return scope_.add_value(indices_from_selection(*typed_varray)).as_span();
}

}  // namespace blender::fn