Hello kilograham,
Thank you for this information. It was clear to me that float computations were done by software.
In order to reduce the duration of calculations, I changed the local oscillator table type from float to int_32 (float values are multiplied by 10 000). I have to test if this precision is sufficient, but in terms of performance, computations falls to 9,1ms which is fine.
I also (finally) succeeded this afternoon to set gcc compiler float library versus pico float library.
results :
- pico library : 9,16 ms
- compiler library : 22,4 ms
That's the evidence, calculations use optimized library.
So, I now have to test.
Thank you all for your help, tricks and ideas.
Franck
Thank you for this information. It was clear to me that float computations were done by software.
In order to reduce the duration of calculations, I changed the local oscillator table type from float to int_32 (float values are multiplied by 10 000). I have to test if this precision is sufficient, but in terms of performance, computations falls to 9,1ms which is fine.
I also (finally) succeeded this afternoon to set gcc compiler float library versus pico float library.
results :
- pico library : 9,16 ms
- compiler library : 22,4 ms
That's the evidence, calculations use optimized library.
So, I now have to test.
Thank you all for your help, tricks and ideas.
Franck
Statistics: Posted by Franck34 — Sun Feb 04, 2024 5:58 pm