common.hpp

#pragma once
#include <dlprim/tensor.hpp>
#include <dlprim/context.hpp>
namespace dlprim {
namespace core {

    class Scale {
    public:
        Scale(Context &ctx,DataType dtype=float_data);
        void enqueue(float s,Tensor &t,ExecutionContext const &ec);
    private:
        cl::Kernel k_;
    };

    void add_tensors(Tensor &a,Tensor &b,Tensor &sum,ExecutionContext const &ec);
    void scale_tensor(float s,Tensor &t,ExecutionContext const &ec);

    void fill_tensor(Tensor &t,double value,ExecutionContext const &e);

    enum RandomDistribution {
        rnd_uniform = 0,
        rnd_normal  = 1,
        rnd_bernoulli = 2
    };
    void fill_random(Tensor &t,cl_ulong philox_seed,cl_ulong philox_seq,RandomDistribution dist,float p1,float p2,ExecutionContext const &e);

   
    
    class SliceCopy {
    public:
        SliceCopy(Context &ctx,DataType dtype=float_data);
        ~SliceCopy();
        
        void tensor_slice_copy(int dim,size_t slice,
                               Tensor &target,size_t target_offset,
                               Tensor &source,size_t source_offset,
                               float target_scale,ExecutionContext const &e);
    private:
        cl::Kernel kernel_;
        DataType dtype_;
    };



} // core
} // dlprim