topic Re: ippiConvFull for iameg convolution in Intel® Integrated Performance Primitives

ippiConvFull for iameg convolution

alberwan — Fri, 02 Jan 2009 22:28:14 GMT

Hi:

I am new to IPP.

Ineed to loop600 floating images to do some calculations. The image size is 1024x768 and the kernel size is 3x3. Each image needs to be convolved. Should I use ippiConvFull_32f_C1R or some other functions to speed up the processing? If using ippiConvFull_32f_C1R, can anybody tell me how to choose srcStep and DstStep?

Thanks in advance.

Re: ippiConvFull for iameg convolution

alberwan — Fri, 02 Jan 2009 22:35:49 GMT

By the way, I created a testing fucntion to do convolution:

IppStatus imageConv_by_ConvFull()

{

//src1

Ipp32f src1[5 * 4] = {

{ 1},{ 2},{ 3},{ 4},{ 5},

{ 6},{ 7},{ 8},{ 9},{10},

{11},{12},{13},{14},{15},

{16},{17},{18},{19},{20},

};

IppiSize src1Size = {5, 4};

//int src1Step = 5;

int src1Step = 5*sizeof

(Ipp32f);

//src2

Ipp32f src2[3*3] = {

{1.0},{1.0},{1.0},

};

IppiSize src2Size = {3, 3};

//int src2Step=3;

int src2Step=3*sizeof

(Ipp32f);

//Dst: result

Ipp32f Dst[7 * 6];

//int dstStep = 7;

int dstStep = 7*sizeof

(Ipp32f);

IppStatus Status;

Status=ippiConvFull_32f_C1R(src1,src1Step,src1Size,src2,src2Step,src2Size,Dst,dstStep);

return Status;

}

But it seems the result is not right. Can someone tell me what's wrong?

Thanks.

Re: ippiConvFull for iameg convolution

Chao_Y_Intel — Sun, 04 Jan 2009 03:11:03 GMT

What is output for the test code?When I run the code here, it gives the following result for Dst:

1.0 3.0 6.0 9.0 12.0 9.0 5.0
7.0 16.0 27.0 33.0 39.0 28.0 15.0
18.0 39.0 63.0 72.0 81.0 57.0 30.0
33.0 69.0 108.0 117.0 126.0 87.0 45.0
27.0 56.0 87.0 93.0 99.0 68.0 35.0
16.0 33.0 51.0 54.0 57.0 39.0 20.0

That looks correct result. You can check the IPP manual on how the function computes the full convolution of the two images.

Thanks,
Chao