opencl gauss filter优化(二)

1.buffer使用image的方式：Horizontal 与 Vertical 算法一样, 共需30ms,wait time 19ms.

 

const sampler_t sampler = CLK_NORMALIZED_COORDS_FALSE | CLK_ADDRESS_CLAMP_TO_EDGE | CLK_FILTER_NEAREST;
__kernel void ImageGaussianFilterHorizontal(__read_only image2d_t  source, // Source image
                            __write_only image2d_t   dest,  // Intermediate dest image
                                             const int imgWidth ,                // Image width
                                             const int imgHeight)
{
    const int y = get_global_id(0);
    if(y>=(imgHeight))
        return;
    const float m_nFilter[11] = {1/256.0,4/256.0,8/256.0,16/256.0,32/256.0,134/256.0,32/256.0,16/256.0,8/256.0,4/256.0,1/256.0};

    const int s = 11;
    const int nStart = 5;
    
    float lines[11];
    for(int i=0;i<11;i++)
        lines[i] = read_imagef( source, sampler,  (int2) (i-5, y) ).x;

    for(int j=0;j<imgWidth;){
    float sum = lines[nStart] * m_nFilter[nStart];
#define    GaussianTwoLines(m) \
    sum += ( (lines[m] + lines[s-1-m])*m_nFilter[m] );
        GaussianTwoLines(0)
        GaussianTwoLines(1)
        GaussianTwoLines(2)
        GaussianTwoLines(3)
        GaussianTwoLines(4)    
        
        write_imagef( dest, (int2) (j, y), sum );
        
        for(int i = 0; i<s-1; i++) lines[i] = lines[i+1];
        j++;
        lines[s-1] = read_imagef( source, sampler, (int2) (j+5, y) ).x;
    }
}

__kernel void ImageGaussianFilterVertical(__read_only image2d_t  source, // Source image
                        __write_only image2d_t   dest,   
                         const int imgWidth ,               
                        const int imgHeight)
{
    const int x = get_global_id(0);
    if(x>=(imgWidth))
        return;
    const float m_nFilter[11] = {1/256.0,4/256.0,8/256.0,16/256.0,32/256.0,134/256.0,32/256.0,16/256.0,8/256.0,4/256.0,1/256.0};

    const int s = 11;
    const int nStart = 5;

    float lines[11];
    for(int i=0;i<11;i++)
        lines[i] = read_imagef( source, sampler,  (int2) (x ,i-5) ).x;

    for(int j=0;j<imgHeight;){
    float sum = lines[nStart] * m_nFilter[nStart];
#define    GaussianTwoLines(m) \
    sum += ( (lines[m] + lines[s-1-m])*m_nFilter[m] );
        GaussianTwoLines(0)
        GaussianTwoLines(1)
        GaussianTwoLines(2)
        GaussianTwoLines(3)
        GaussianTwoLines(4)

        write_imagef( dest, (int2) (x, j), sum );

        for(int i = 0; i<s-1; i++) lines[i] = lines[i+1];
        j++;
        lines[s-1] = read_imagef( source, sampler, (int2) (x,j+5) ).x;
    }
}

View Code