FFT Performance

[M G_2]

Posted on July 03, 2018 at 13:48

I’ve setup a DSP application including a F32 FFT that takes on STM32F407 about 5 ms and on a STM32F429 approximately  .5 ms. The clock settings are in both cases the highest possible values (STM32F407 -> 168 MHz and STM32F429 -> 180 MHz) Did I something wrong in case of the STM32F407 or is the performance difference between both the µC that high ?

[AvaTar · Accepted solution]

Posted on July 03, 2018 at 14:01

The block sizes and compiler settings are identical ?

The 0,5ms for the F429 sound unreasonable for large block sizes.

[M G_2]

Posted on July 03, 2018 at 14:35

Hi and thankyou for responding. After double checking the compiler settings and also the lib which I’ve included it looks like that I've included the wrong lib:

arm_cortexM4l_math.lib . After including the arm_cortexM4lf_math.lib instead the performance increased significantly. On a stm32f407 a 1024 pt F32 FFT now takes about 668 µs which is close to the result achieved in AN4841 for the STM32F429 of about 547 µs.