Sha256 parallelization

[krlosMata]

Overview

Sha256 circom component does not parallelize at the time to compute the witness. The other rollup-main components (DecodeTx, RollupTx & FeeTx) are already parallelized, being the Sha256 component the one which takes mucht ime to be calculated.
There are two approaches that could be followed when trying to paralleize Sha256:

1 - External intermediary signals

This approach pretends to calculate Sha256 internal blocks outside of the circuit. Therefore, intermediary signals should be added to the rollup-main circuit. Besides, circomlib Sha256 should be adapted to accept those intermediary signalsa nd check them properly.

• Following this approach:
  • small changes on circomlib Sha256
  • small changes on rollup-main circuit
  • Javascript and Go implementation to get Sha256 intermediary signals
  • it will reduce witness time computation

2 - Internal intermediary signals

This approach computes Sha256 internal blocks directly on the circom Sha256 circomlib circuit. No additional signals should be added to the rollup-main circuit but a new Sha256Parallel should be build on circomlib.

• Following this approach:
  • new circomlib Sha256Parallel circuit implementing block computation with no comnstraints
  • no changes on rollup-main circuit (just changing the template name)
  • general circomlib implementation to be used for anyone an not for an specific circuit
  • should be benchmarrking if this approach really redice the witness computation