基于OpenCL的mean filter性能

1.对于一个标准的3*3 均值滤波，kernel代码如下：

使用buffer/image缓冲对象

__kernel void filter(__global uchar4* inputImage, __global uchar4* outputImage, uint N)
{
 int x = get_global_id(0);
 int y = get_global_id(1);
 int width = get_global_size(0);
 int height = get_global_size(1);

 int k = (N-1)/2;
 int n = N*N; //n*n

 if(x < k || y < k || x > width - k - 1 || y > height - k - 1)
{ 
outputImage[x + y * width] = inputImage[x + y * width];
 return;
}


uint4 finalcolor = (uint4)(0);
 
 int i,j;
 for(j = y - k; j <= y + k; j++)
{
 for(i = x - k; i <= x + k; i++)
{
finalcolor = finalcolor + convert_uint4(inputImage[i + j * width]);
}
} 
 
outputImage[x + y * width] = convert_uchar4(finalcolor/n);

}

__kernel void filterImg( image2d_t inputImage, __write_only image2d_t outputImage, uint N)
{
 int x = get_global_id(0);
 int y = get_global_id(1);
 int width = get_global_size(0);
 int height = get_global_size(1);


uint4 temp = read_imageui(inputImage, imageSampler, (int2)(x,y));

 int k = (N-1)/2;
 int n = N*N; //n*n


 if(x < k || y < k || x > width - k - 1 || y > height - k - 1)
{
write_imageui(outputImage, (int2)(x,y), temp);
 return;
}

 /* k*k area */
uint4 finalcolor = (uint4)(0);

 int i,j;
 for(j = y - k; j <= y + k; j++)
{
 for(i = x - k; i <= x + k; i++)
{
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(i,j));
}
}

finalcolor = finalcolor/n;

write_imageui(outputImage, (int2)(x,y), finalcolor);


}

对一个2048*2048的图像执行filter操作，

global work size = {2048, 2048, 1}, group work size = {16, 16}, 一般group work size应该为64的倍数，因为对于AMD显卡，wave是基本的硬件线程调度单位。

使用了6个GPRs，没有使用ScratchRegs,ScratchRregs是指用vedio meory来模拟GPR，但是线程执行的速度会大大降低，应尽量减少ScratchRegs的数量。

可以看到，使用image对象kernel执行时间要短，但奇怪的是各项性能参数都是buffer对象领先，除了alu busy和alu指令数目。

改为下面的kernel代码，性能会有所提高

__kernel void filter(__global uchar4* inputImage, __global uchar4* outputImage, uint N)
{
 int x = get_global_id(0);
 int y = get_global_id(1);
 int width = get_global_size(0);
 int height = get_global_size(1);

 if(x < 1 || y < 1 || x > width - 2 || y > height - 2)
{ 
outputImage[x + y * width] = inputImage[x + y * width];
 return;
}

 
uint4 finalcolor = (uint4)(0);

finalcolor = finalcolor + convert_uint4(inputImage[x-1+( y-1) * width]);
finalcolor = finalcolor + convert_uint4(inputImage[x+( y-1) * width]);
finalcolor = finalcolor + convert_uint4(inputImage[x+1+( y-1) * width]);
finalcolor = finalcolor + convert_uint4(inputImage[x-1+y * width]);
finalcolor = finalcolor + convert_uint4(inputImage[x+y * width]);
finalcolor = finalcolor + convert_uint4(inputImage[x+1+y * width]);
finalcolor = finalcolor + convert_uint4(inputImage[x-1+( y+1) * width]);
finalcolor = finalcolor + convert_uint4(inputImage[x+( y+1) * width]);
finalcolor = finalcolor + convert_uint4(inputImage[x+1+( y+1) * width]);

outputImage[x + y * width] = convert_uchar4(finalcolor/9);

}
__kernel void filter1(__global uchar4* inputImage, __global uchar4* outputImage, uint N)
{
 int x = get_global_id(0);
 int y = get_global_id(1);
 int width = get_global_size(0);
 int height = get_global_size(1);

 int k = (N-1)/2;
 int n = N*N; //n*n

 if(x < k || y < k || x > width - k - 1 || y > height - k - 1)
{ 
outputImage[x + y * width inputImage[x + y * width];
 return;
}

 // if(x==209 && y ==243)
 //{
 // printf("final color:%d,%d,%d,%d\n", finalcolor.x, finalcolor.y, finalcolor.z,finalcolor.w);
 // }
 
uint4 finalcolor = (uint4)(0);
 
 int i,j;
 for(j = y - k; j <= y + k; j++)
{
 for(i = x - k; i <= x + k; i++)
{
finalcolor = finalcolor + convert_uint4(inputImage[i + j * width]);
}
} 
 
outputImage[x + y * width] = convert_uchar4(finalcolor/n);

}
__kernel void filterImg( image2d_t inputImage, __write_only image2d_t outputImage, uint N)
{
 int x = get_global_id(0);
 int y = get_global_id(1);
 int width = get_global_size(0);
 int height = get_global_size(1);


uint4 temp = read_imageui(inputImage, imageSampler, (int2)(x,y));

 if(x < 1 || y < 1 || x > width - 2 || y > height - 2)
{
write_imageui(outputImage, (int2)(x,y), temp);
 return;
}

 /* k*k area */
uint4 finalcolor = (uint4)(0);

finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x-1,y-1));
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x,y-1));
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x+1,y-1));
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x-1,y));
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x,y));
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x+1,y));
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x-1,y+1));
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x,y+1));
finalcolor = finalcolor + read_imageui(inputImage, imageSampler, (int2)(x+1,y+1));

finalcolor = finalcolor/9;

write_imageui(outputImage, (int2)(x,y), finalcolor);


}