You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
currently the algorithm is for each tx, we build an witness block, then calculate row usage of each subcircuit of this tx, then add the usage vector and find max inside the acc vector.
This algorithm will count some data redundantly. For example, if 2 tx shares same bytecode, then the len of bytecode circuit is counted twice. Overestimated.
We may impl a function "acc_witness_block += tx_witness_block"? Inside this function we do some basic de-duplicate and make more accurate estimation.
Pricinple: row estimation should be fast, a bit over estimation is allowd. Under estimation is seen as bug and should be avoided.
The text was updated successfully, but these errors were encountered:
currently the algorithm is for each tx, we build an witness block, then calculate row usage of each subcircuit of this tx, then add the usage vector and find max inside the acc vector.
This algorithm will count some data redundantly. For example, if 2 tx shares same bytecode, then the len of bytecode circuit is counted twice. Overestimated.
We may impl a function "acc_witness_block += tx_witness_block"? Inside this function we do some basic de-duplicate and make more accurate estimation.
Pricinple: row estimation should be fast, a bit over estimation is allowd. Under estimation is seen as bug and should be avoided.
The text was updated successfully, but these errors were encountered: