Edit this page

class ov::pass::DivisionByZeroFP16Resolver

Overview

: clamps eps into fp16 minimal normalized value in input_1/Maximum(input_2, eps); input_1/Add(input_2, eps); and input_1*Pow(Maximum[Add](input_2, eps), -z) patterns to prevent division by zero. More…

#include <division_by_zero_fp16_resolver.hpp>

class DivisionByZeroFP16Resolver: public ov::pass::MatcherPass
{
public:
    // methods

    OPENVINO_RTTI("DivisionByZeroFP16Resolver", "0");
};

Inherited Members

public:
    // typedefs

    typedef DiscreteTypeInfo type_info_t;

    // methods

    bool get_property(const PassPropertyMask& prop_mask) const;
    void set_name(const std::string& name);
    std::string get_name() const;
    void set_callback(const param_callback& callback);
    virtual void set_pass_config(const std::shared_ptr<PassConfig>& pass_config);
    std::shared_ptr<PassConfig> get_pass_config();
    bool m_transformation_callback(const std::shared_ptr<const Node>& node);
    bool transformation_callback(const std::shared_ptr<const Node>& node);
    virtual const type_info_t& get_type_info() const = 0;
    OPENVINO_RTTI("ov::pass::MatcherPass");
    MatcherPass& operator = (const MatcherPass&);
    bool apply(std::shared_ptr<ov::Node> node);

    template <typename T, class... Args>
    std::shared_ptr<T> register_new_node(Args&&... args);

    template <typename T>
    std::shared_ptr<T> register_new_node(const std::shared_ptr<T>& node);

    std::shared_ptr<ov::Node> register_new_node_(const std::shared_ptr<ov::Node>& node);
    const std::vector<std::shared_ptr<ov::Node>>& get_new_nodes();
    void clear_new_nodes();
    std::shared_ptr<pattern::Matcher> get_matcher();

Detailed Documentation

: clamps eps into fp16 minimal normalized value in input_1/Maximum(input_2, eps); input_1/Add(input_2, eps); and input_1*Pow(Maximum[Add](input_2, eps), -z) patterns to prevent division by zero.

eps must be always nonzero to prevent from NaNs in such expressions if input_1 and input_2 simultaneously happened to be zero. We should keep in such patterns eps >= fp16 minimal normalized value so that CompressFloatConstants should not cast them into zero during compression into f16.