Question on the functions of the C28x Floating Point Unit DSP Library

Mathieu TARRIUS

Hello,

I have to do some processing that are mainly: matrix operations of complex numbers in floating point format.

So I need very often to multiply 2 matrix.

I understood that the provided library provides "vectors" API. So to manage matrix we just need to call several time some "vectors" API.

I do not have any problem with that.

However, to do the multiplication by vectors, I need always an API such as:

Multiply-and-Accumulate of a Complex Vector (Single Precision Float
and a Complex Vector (Single Precision Float)

but there is only the following one:

Multiply-and-Accumulate of a Real Vector (Signed 16-bit Integer)
and a Complex Vector (Single Precision Float)

I would like to know if I missed something ?

One solution that I saw is to first call the function mpy_SP_CVxCV and then do the sum of the vector elements. But this sum does not exist (even if of course I can do it manually in C with a loop) or do I have top call: mac_SP_i16RVxCV with i16RV = (1, 1, .., 1) ?

I have the version v3.3.8 of this library.

Do you plan to deliver the functions that I indicated or you recommend to me to use an alternative solution?

Thanks and bets regards,

Mathieu

over 8 years ago

0 Vishal_Coelho over 8 years ago

TI__Mastermind 20850 points

Hi Mathieu,

What you are looking for is a mac_SP_CVxCV which we haven't implemented yet. I have put in a "requirement" request to add this and mac_SP_CVxCVC for the next realease.

If you give me about a week i will post the routine in reply to this post.

0 Mathieu TARRIUS over 8 years ago in reply to Vishal_Coelho

Intellectual 900 points

Hi Vishal,

Thanks a lot for your answer.

I will really appreciate if you can provide in advance in the post a first version of this routine. In addition, later if you can also include it in the next version it would be perfect!

Thanks again.

Regards,

Mathieu

0 Vishal_Coelho over 8 years ago in reply to Mathieu TARRIUS

TI__Mastermind 20850 points

Mathieu,

Here is a preliminary version of the routine.mac_SP_CVxCV.asm

Add the following prototype to fpu_vector.h

//! \brief Multiply-and-Accumulate of a Complex Vector and a Complex Vector.
//!
//! This module multiplies and accumulates a complex vector and another
//! complex vector.
//! \f[y_{re} &=& \sum(w_{re}[i]* x_{re}[i] - w_{im}[i] * x_{im}[i]) \f]
//! \f[y_{im} &=& \sum(w_{re}[i]* x_{im}[i] + w_{im}[i] * x_{re}[i]) \f]
//! \param y complex result
//! \param w pointer to the first complex input vector
//! \param x pointer to the second complex input vector
//! \param N length of the w and x vectors
//! \attention N must be a minimum of 3
//
extern complex_float mac_SP_CVxCV(const complex_float *w,
                         const complex_float *x, const uint16_t N);

It will be a while till the next release as it is a major one so i expect this may not be the final version of the routine, although i dont really see further room to optimize it--- but someone else might. Anyway, let me know if this works or if there are issues with it.

0 Mathieu TARRIUS over 8 years ago in reply to Vishal_Coelho

Intellectual 900 points

Thanks a lot Vishal for this function provided in advance. I really appreciated!
I tested it and it seems to work fine.
In addition it saves numerous CPU cycles. It looks like that the Number of Cycles = 6*N + constant
Do you confirm that the latest version for FPU DSP Software Library is: V1.50.00.00: Moderate Update ?
Thanks and best regards,
Mathieu

0 Vishal_Coelho over 8 years ago in reply to Mathieu TARRIUS

TI__Mastermind 20850 points

Mathieu, Yes the last released version of the FPU library was 1.50.00.00

C2000™︎ microcontrollers

C2000 microcontrollers forum

Question on the functions of the C28x Floating Point Unit DSP Library