Function to calculate Euclidean distance of two N-Dim Vectors on C64X

Darleen,

As far as I know, Euclidean distance is not present in any of the libraries.

Optimization of this function for C64x depends on the data-type of the vectors X and Y, whether they are chars or shorts or integers.

As it is a fairly simple function, the compiler should be able to do a good job with the following C code.

for(i=0;i<N;i++)

{

    z = x[i] - y[i];

   acc += z*z;

}

Also, Euclidean distance requires a sqrt at the end of accumulation. As it is outside the loop, you can use the default sqrt() function.

Please let me know if you need any clarification.

Regards

Senthil

The above logic can also be implemented using some optimized kernels provided with the std libraries (http://processors.wiki.ti.com/index.php?title=Software_libraries). For ex, the IMGLIB documentation includes the function: IMG_sub_16s that can be used to implement the first step that calculates the diff. The second step can be implemented by using the function DSP_vecsumsq from the DSPLIB. The SQRT function is provided in the IQMath lib

Senthil,

Thank you for confirming that Euclidean distance is not present in any of the libraries and your suggestion.

Thanks for the advice. Eulidean distance is a simple and basic function, I am surprised it is not included in any of TI libraries. I wish that TI would consider to have similar functions like ippsNormDiff_L2 of Intel IPP libraries in the future release.

Dunling,

We have other distance measures like Manhattan distance and Bhattacharya distance in VLIB.

I would take your request for Euclidean distance and implement it for the next VLIB release positively.

Regards

Senthil