TDA4VM: Is it normal that the mathlib powdp() function runs slower on the TDA4VM C66 than on the CPU?

Hi,

Can you specify what processor you are referring to when talking about TDA4VM CPU?

Can you provide more details on how you are measuring performance?

Best,

Asha 

TDA4VM CPU: Arm Cortex-A72 64bits 2.0GHz

Test code:

#include <math.h>
#include <stdio.h>


int main()
{
    double a = 142.224389823827;
    double b = 0;
    for (long i = 0; i < 100000000; i++)
    {
        b = pow(a,2); // a^2
    }
    
    return 0;
}

it takes 4 senconds

Compare with C66x，run code as:

#include <ti/mathlib/mathlib.h>

int main()
{
    double a = 142.224389823827;
    double b = 0;
    for (long i = 0; i < 100000000; i++)
    {
        b = powdp_i(a,2); // a^2
    }
    
    return 0;
}

and it takes 20 seconds

Hi,

I wouldn't expect such performance on C66x when using the mathlib libraries. 

Are you running on a TI TDA4VM EVM with the C66x processor running at 1.35GHz?

Best,

Asha 

Hi，

do you mean the frequency of the c66x is configurable? 

Hi,

I just wanted to check what board you are running on, thank you for clarifying that it is a TI EVM. 

I would expect you would get values closer to what we publish in the MATHLIB test report (MATHLIB_c66x_TestReport.html). The numbers here are reported in cycles however, not time. 

I can get back to your issue and see if I can reproduce the numbers that you are seeing and provide an update by 5/30. 

Best,

Asha

Hi，

Do you have any progress recently？

Hi,

 I have made progress, however I will still need a few days to investigate further. I will try to provide an update on this by 6/4. I apologize for the delay.

Best,

Asha

Hi Asha Bhandarkar,

Any further update on this thread?

Regards,

Brijesh

OK，Thanks!