StorageSorter

StorageSorter PI

Input

GitHub

pub struct StorageDeduplicatorInputData<F: SmallField> {
    pub shard_id_to_process: UInt8<F>,
    pub unsorted_log_queue_state: QueueState<F, QUEUE_STATE_WIDTH>,
    pub intermediate_sorted_queue_state: QueueState<F, QUEUE_STATE_WIDTH>,
}

Output

GitHub

pub struct StorageDeduplicatorOutputData<F: SmallField> {
    pub final_sorted_queue_state: QueueState<F, QUEUE_STATE_WIDTH>,
}

FSM Input and FSM Output

GitHub

pub struct StorageDeduplicatorFSMInputOutput<F: SmallField> {
    pub lhs_accumulator: [Num<F>; DEFAULT_NUM_PERMUTATION_ARGUMENT_REPETITIONS],
    pub rhs_accumulator: [Num<F>; DEFAULT_NUM_PERMUTATION_ARGUMENT_REPETITIONS],
    pub current_unsorted_queue_state: QueueState<F, QUEUE_STATE_WIDTH>,
    pub current_intermediate_sorted_queue_state: QueueState<F, QUEUE_STATE_WIDTH>,
    pub current_final_sorted_queue_state: QueueState<F, QUEUE_STATE_WIDTH>,
    pub cycle_idx: UInt32<F>,
    pub previous_packed_key: [UInt32<F>; PACKED_KEY_LENGTH],
    pub previous_key: UInt256<F>,
    pub previous_address: UInt160<F>,
    pub previous_timestamp: UInt32<F>,
    pub this_cell_has_explicit_read_and_rollback_depth_zero: Boolean<F>,
    pub this_cell_base_value: UInt256<F>,
    pub this_cell_current_value: UInt256<F>,
    pub this_cell_current_depth: UInt32<F>,
}

Main circuit logic

The main logic of this circuit is sorting and deduplicating storage requests from unsorted_log_queue_state. The result storage requests are pushed to final_sorted_queue_state.

First part

GitHub

We start, as usually, with allocating input fields from PI.

let mut structured_input = StorageDeduplicatorInputOutput::alloc_ignoring_outputs(
    cs,
    structured_input_witness.clone(),
);

In this part, we should decide what unsorted_queue_state to use (the one from Input or the other one from FSM Input). We do the same for sorted queue.

let state = QueueState::conditionally_select(
    cs,
    structured_input.start_flag,
    &unsorted_queue_from_passthrough_state,
    &unsorted_queue_from_fsm_input_state,
);

let state = QueueState::conditionally_select(
    cs,
    structured_input.start_flag,
    &intermediate_sorted_queue_from_passthrough.into_state(),
    &intermediate_sorted_queue_from_fsm_input.into_state(),
);

Also, we decide to create a new queue for the output, or continue working with the existing one.

let state = QueueState::conditionally_select(
    cs,
    structured_input.start_flag,
    &empty_final_sorted_queue.into_state(),
    &final_sorted_queue_from_fsm_input.into_state(),
);

Now we need to generate challenges for permutation argument.

let challenges = crate::utils::produce_fs_challenges::<
    F,
    CS,
    R,
    QUEUE_STATE_WIDTH,
    { TIMESTAMPED_STORAGE_LOG_ENCODING_LEN + 1 },
    DEFAULT_NUM_PERMUTATION_ARGUMENT_REPETITIONS,
>(
    cs,
    structured_input
        .observable_input
        .unsorted_log_queue_state
        .tail,
    structured_input
        .observable_input
        .intermediate_sorted_queue_state
        .tail,
    round_function,
);

And decide whether we generate new accumulators for permutation argument or use existing ones.

let initial_lhs =
    <[Num<F>; DEFAULT_NUM_PERMUTATION_ARGUMENT_REPETITIONS]>::conditionally_select(
        cs,
        structured_input.start_flag,
        &[one; DEFAULT_NUM_PERMUTATION_ARGUMENT_REPETITIONS],
        &structured_input.hidden_fsm_input.lhs_accumulator,
    );

let initial_rhs =
    <[Num<F>; DEFAULT_NUM_PERMUTATION_ARGUMENT_REPETITIONS]>::conditionally_select(
        cs,
        structured_input.start_flag,
        &[one; DEFAULT_NUM_PERMUTATION_ARGUMENT_REPETITIONS],
        &structured_input.hidden_fsm_input.rhs_accumulator,
    );

Main part

GitHub

Here we implement the main logic of the circuit. We run a cycle where on each iteration we try to pop a new element.

let (_, original_encoding) = original_queue.pop_front(cs, should_pop);
let (sorted_item, sorted_encoding) = intermediate_sorted_queue.pop_front(cs, should_pop);

Then we accumulate encodings for permutation argument. You can read more about it here.

for (((lhs_dst, rhs_dst), challenges), additive_part) in lhs
    .iter_mut()
    .zip(rhs.iter_mut())
    .zip(fs_challenges.iter())
    .zip(additive_parts.iter())
{
    lhs_lc.clear();
    rhs_lc.clear();

    for ((original_el, sorted_el), challenge) in extended_original_encoding
        .iter()
        .zip(sorted_encoding.iter())
        .zip(challenges.iter())
    {
        let lhs_contribution = original_el.mul(cs, &challenge);
        let rhs_contribution = sorted_el.mul(cs, &challenge);

        lhs_lc.push((lhs_contribution.get_variable(), F::ONE));
        rhs_lc.push((rhs_contribution.get_variable(), F::ONE));
    }

    lhs_lc.push((additive_part.get_variable(), F::ONE));
    rhs_lc.push((additive_part.get_variable(), F::ONE));

    let lhs_lc = Num::linear_combination(cs, &lhs_lc);
    let rhs_lc = Num::linear_combination(cs, &rhs_lc);

    let lhs_candidate = lhs_dst.mul(cs, &lhs_lc);
    let rhs_candidate = rhs_dst.mul(cs, &rhs_lc);

    *lhs_dst = Num::conditionally_select(cs, should_pop, &lhs_candidate, &*lhs_dst);
    *rhs_dst = Num::conditionally_select(cs, should_pop, &rhs_candidate, &*rhs_dst);
}

Now we enforce sorting.

previous_key_is_greater.conditionally_enforce_false(cs, not_item_is_trivial);

Maybe we should push the old query if the new key is different. So we push if at least one of these conditions holds:

  • there was a read at depth 0;
  • the sell is changes;
  • write that was declined, but not by a rollback.
let query = LogQuery {
    address: previous_address,
    key: previous_key,
    read_value: this_cell_base_value,
    written_value: this_cell_current_value,
    rw_flag: should_write,
    aux_byte: UInt8::zero(cs),
    rollback: Boolean::allocated_constant(cs, false),
    is_service: Boolean::allocated_constant(cs, false),
    shard_id: shard_id_to_process,
    tx_number_in_block: UInt32::zero(cs),
    timestamp: UInt32::zero(cs),
};

sorted_queue.push(cs, query, should_push);

After that, we update some inner variables.

let meaningful_value = UInt256::conditionally_select(
    cs,
    record.rw_flag,
    &record.written_value,
    &record.read_value,
);

this_cell_base_value = UInt256::conditionally_select(
    cs,
    new_non_trivial_cell,
    &record.read_value,
    &this_cell_base_value,
);

...

Now we continue working with current query. We check that the read field is correct.

let read_is_equal_to_current =
    UInt256::equals(cs, &this_cell_current_value, &record.read_value);
read_is_equal_to_current.conditionally_enforce_true(cs, check_read_consistency);

After that, we do some other variable updates.

After the main cycle, we do one more iteration if we took the last query from the queue during the last cycle.

let query = LogQuery {
    address: previous_address,
    key: previous_key,
    read_value: this_cell_base_value,
    written_value: this_cell_current_value,
    rw_flag: should_write,
    aux_byte: UInt8::zero(cs),
    rollback: Boolean::allocated_constant(cs, false),
    is_service: Boolean::allocated_constant(cs, false),
    shard_id: shard_id_to_process,
    tx_number_in_block: UInt32::zero(cs),
    timestamp: UInt32::zero(cs),
};

sorted_queue.push(cs, query, should_push);

Final part

GitHub

If the queues are empty, we check the permutation argument accumulators equality.

let completed = unsorted_is_empty.and(cs, sorted_is_empty);
new_lhs.iter().zip(new_rhs).for_each(|(l, r)| {
    Num::conditionally_enforce_equal(cs, completed, &l, &r);
});

Now we update PI output parts and compute a commitment. Then we allocate it as public variables.

let input_commitment =
    commit_variable_length_encodable_item(cs, &compact_form, round_function);
for el in input_commitment.iter() {
    let gate = PublicInputGate::new(el.get_variable());
    gate.add_to_cs(cs);
}

SortDecommitments

LogSorter

Made with ❤️ by the ZKsync Community