I have some IIR filters that are currently running with double precision math. I'd like to move them to CMSIS DSP assisted f32() functions on a cortexM4 with an FPU but I'm worried what its going to do to the filter performance. Is there a way to simulate this with pyFDA.
I see the fixed point stuff but I don't see anything that looks like its a single precision float.
You are right, single precision float isn't implemented yet. Your question is actually two questions :-)
Coefficients need to be quantized, this part is implemented for various fixpoint formats but not for single precision floats. This is a good idea and could be easily implemented.
Especially for IIR filters, the topology (direct form, second order sections, ...) and internal word length / data format can be even more important for the performance than coefficient format. This part - arithmetic quantization - is much more difficult to implement, as the standard scipy routines like freqz() etc. can no longer be used. This feature is targeted for the next release.
I have some IIR filters that are currently running with double precision math. I'd like to move them to CMSIS DSP assisted f32() functions on a cortexM4 with an FPU but I'm worried what its going to do to the filter performance. Is there a way to simulate this with pyFDA.
I see the fixed point stuff but I don't see anything that looks like its a single precision float.