TanhLayer.cs

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using MyCaffe.basecode;
using MyCaffe.common;
using MyCaffe.param;
 
namespace MyCaffe.layers
{
    public class TanhLayer<T> : NeuronLayer<T>
    {
        long m_hCudnn = 0;
        long m_hBottomDesc = 0;
        long m_hTopDesc = 0;
 
 
        public TanhLayer(CudaDnn<T> cuda, Log log, LayerParameter p)
            : base(cuda, log, p)
        {
            m_type = LayerParameter.LayerType.TANH;
        }
 
        protected override void dispose()
        {
            if (m_hBottomDesc != 0)
            {
                m_cuda.FreeTensorDesc(m_hBottomDesc);
                m_hBottomDesc = 0;
            }
 
            if (m_hTopDesc != 0)
            {
                m_cuda.FreeTensorDesc(m_hTopDesc);
                m_hTopDesc = 0;
            }
 
            if (m_hCudnn != 0)
            {
                m_cuda.FreeCuDNN(m_hCudnn);
                m_hCudnn = 0;
            }
 
            base.dispose();
        }
 
        public override void LayerSetUp(BlobCollection<T> colBottom, BlobCollection<T> colTop)
        {
            if (!m_param.tanh_param.useCudnn())
                return;
 
            // Setup the convert to half flags used by the Layer just before calling forward and backward.
            m_bUseHalfSize = m_param.use_halfsize;
 
            // Initialize CuDNN
            m_hCudnn = m_cuda.CreateCuDNN();
            m_hBottomDesc = m_cuda.CreateTensorDesc();
            m_hTopDesc = m_cuda.CreateTensorDesc();
        }
 
        public override void Reshape(BlobCollection<T> colBottom, BlobCollection<T> colTop)
        {
            base.Reshape(colBottom, colTop);
            if (!reshapeNeeded(colBottom, colTop, false))
                return;
 
            if (!m_param.tanh_param.useCudnn())
                return;
 
            int nN = colBottom[0].num;
            int nK = colBottom[0].channels;
            int nH = colBottom[0].height;
            int nW = colBottom[0].width;
 
            m_cuda.SetTensorDesc(m_hBottomDesc, nN, nK, nH, nW, m_bUseHalfSize);
            m_cuda.SetTensorDesc(m_hTopDesc, nN, nK, nH, nW, m_bUseHalfSize);
        }
 
        protected override void forward(BlobCollection<T> colBottom, BlobCollection<T> colTop)
        {
            if (!m_param.tanh_param.useCudnn())
                forward_cuda(colBottom, colTop);
            else
                forward_cudnn(colBottom, colTop);
        }
 
        protected override void backward(BlobCollection<T> colTop, List<bool> rgbPropagateDown, BlobCollection<T> colBottom)
        {
            if (!m_param.tanh_param.useCudnn())
                backward_cuda(colTop, rgbPropagateDown, colBottom);
            else
                backward_cudnn(colTop, rgbPropagateDown, colBottom);
        }
 
        protected void forward_cuda(BlobCollection<T> colBottom, BlobCollection<T> colTop)
        {
            long hBottomData = colBottom[0].gpu_data;
            long hTopData = colTop[0].mutable_gpu_data;
            int nCount = colBottom[0].count();
 
            m_cuda.tanh_fwd(nCount, hBottomData, hTopData);
        }
 
        protected void backward_cuda(BlobCollection<T> colTop, List<bool> rgbPropagateDown, BlobCollection<T> colBottom)
        {
            long hTopData = colTop[0].gpu_data;
            long hTopDiff = colTop[0].gpu_diff;
            long hBottomDiff = colBottom[0].mutable_gpu_diff;
            int nCount = colBottom[0].count();
 
            m_cuda.tanh_bwd(nCount, hTopDiff, hTopData, hBottomDiff);
        }
 
        protected void forward_cudnn(BlobCollection<T> colBottom, BlobCollection<T> colTop)
        {
            long hBottomData = colBottom[0].gpu_data;
            long hTopData = colTop[0].mutable_gpu_data;
 
            m_cuda.TanhForward(m_hCudnn, m_tOne, m_hBottomDesc, hBottomData, m_tZero, m_hTopDesc, hTopData);
        }
 
        protected void backward_cudnn(BlobCollection<T> colTop, List<bool> rgbPropagateDown, BlobCollection<T> colBottom)
        {
            if (!rgbPropagateDown[0])
                return;
 
            long hTopData = colTop[0].gpu_data;
            long hTopDiff = colTop[0].gpu_diff;
            long hBottomData = colBottom[0].gpu_data;
            long hBottomDiff = colBottom[0].mutable_gpu_diff;
 
            m_cuda.TanhBackward(m_hCudnn, m_tOne, m_hTopDesc, hTopData, m_hTopDesc, hTopDiff, m_hBottomDesc, hBottomData, m_tZero, m_hBottomDesc, hBottomDiff);
        }
    }
}