2018-04-27 03:07 AM
Hi,
I need the fastest implementation of a biquad IIR filter for STM32F4 in assembler. I have to run more than 60 filters at 48kHz. My current best time are 24 cycles per filter with TDF2 (17 at STM32F7).
The order of the instructions is important, I noticed. Wrong instruction orders are penalized with a lot more computation time.
Hope there is an assembler freak with the same passion!
best regards
2018-04-27 05:43 AM
Is the current code available from some public place ? github ?
2018-04-27 07:57 AM
>>Wrong instruction orders are penalized with a lot more computation time.
Best to avoid dependency chains as they put bubbles in the pipeline.. And have sequences that keep all execution units busy.
>>Hope there is an assembler freak with the same passion!
The STM32 forum is perhaps an overly narrow venue, perhaps you should discuss the assembler stuff at an ARM centric venue. DSP and Filters not really my passion.