/// <inheritdoc/>
public override unsafe void Forward(Span <Tensor> inputs, out Tensor z, out Tensor a)
{
Descriptor.Set4D(DataType.FLOAT, TensorFormat.CUDNN_TENSOR_NCHW, inputs[0].Entities, inputs[0].Length, 1, 1);
fixed(Tensor *p = inputs)
{
Tensor.New(p->Entities, p->Length, out z);
using (DeviceMemory <float> y_gpu = DnnInstance.Gpu.AllocateDevice(*p))
{
// Sum the inputs
for (int i = 1; i < inputs.Length; i++)
{
using (DeviceMemory <float> x_gpu = DnnInstance.Gpu.AllocateDevice(p[i]))
DnnInstance.AddTensor(1, Descriptor, x_gpu.Ptr, 1, Descriptor, y_gpu.Ptr);
}
y_gpu.CopyToHost(p[0].Entities, p[0].Length, out z);
// Activation
DnnInstance.ActivationForward(p[0].Entities, p[0].Length, y_gpu.Ptr, y_gpu.Ptr, ActivationFunctions.Activation);
y_gpu.CopyToHost(z.Entities, z.Length, out a);
}
}
}