How to Optimize "Matrix Transposition" In C6000 ?

Hi,

Thanks for your post.

I would rather suggest If your matrix participate in doing some calculation, then you may substitute A^T[i,j] === A[j,i], i.e. get elements of transposed matrix without actual transposing. Many linear algebra algorithms like Cholesky decomposition use this way.

This is simply memory location manipulation, I'd suggest doing this with the EDMA instead of using the DSP.  It's going to be fairly inefficient doing it with the DSP, but should be able to set up an EDMA transfer to do it in a single shot.

There are also DSP library API functions available to perform optimized DSP matirx transpose (DSP_mat_trans) & matrix multiplication (DSP_mat_mul) and you could try this by download DSP libarary available for c64x/c64x+ from the below download link:

http://www.ti.com/tool/sprc265

There are c64x core optimization workshops and online materials to enhance performance. Kindly check the below TI wiki resource:

http://processors.wiki.ti.com/index.php/TMS320C6000_DSP_Optimization_Workshop

http://processors.wiki.ti.com/index.php/Optimization_Techniques_for_the_TI_C6000_Compiler

Thanks & regards,

Sivaraj K

-------------------------------------------------------------------------------------------------

Please click the Verify Answer button on this post if it answers your question.

-------------------------------------------------------------------------------------------------------