It's likely optimisations can be made on the prover and verifier side. For instance, we realised that it is cheaper for the verifier to compute all evaluations of a certain polynomial over the entire domain using an FFT and then select the ones of interest, than to individually evaluate that polynomial at the point of interest (we should implement a threshold to switch from one technique to the other).
Flamegraph is very useful in determining the bottlenecks.
I imagine similar optimisations to the above can be applied to more places.
It's likely optimisations can be made on the prover and verifier side. For instance, we realised that it is cheaper for the verifier to compute all evaluations of a certain polynomial over the entire domain using an FFT and then select the ones of interest, than to individually evaluate that polynomial at the point of interest (we should implement a threshold to switch from one technique to the other).
Flamegraph is very useful in determining the bottlenecks.
I imagine similar optimisations to the above can be applied to more places.