Gas consumption of different part of the deposit function call

[NIC619]

Below are cumulative gas consumption of each part of deposit function call, starting from the beginning of the function:

• latest update: 2019/06/16, contract version: https://github.com/ethereum/eth2.0-specs/blob/193b628fb4fb7c3ef08cf160a9bbafade61a839e/deposit_contract/contracts/validator_registration.v.py
  # Compiler version: vyper==0.1.0b9

def deposit(pubkey: bytes[PUBKEY_LENGTH],
            withdrawal_credentials: bytes[WITHDRAWAL_CREDENTIALS_LENGTH],
            signature: bytes[SIGNATURE_LENGTH]):
    # Avoid overflowing the Merkle tree (and prevent edge case in computing `self.branch`)
    assert self.deposit_count < MAX_DEPOSIT_COUNT

    # Validate deposit data
    deposit_amount: uint256 = msg.value / as_wei_value(1, "gwei")
    assert deposit_amount >= MIN_DEPOSIT_AMOUNT
    assert len(pubkey) == PUBKEY_LENGTH
    assert len(withdrawal_credentials) == WITHDRAWAL_CREDENTIALS_LENGTH
    assert len(signature) == SIGNATURE_LENGTH

    # cumulative gas consumption so far: ≈35304
    # NOTE: including basic transaction cost: 21000 + 68*176(assuming all non zero bytes inputs) = 32968

    # Emit `Deposit` log
    amount: bytes[8] = self.to_little_endian_64(deposit_amount)

    # cumulative gas consumption so far: ≈42583
    # NOTE: one `to_little_endian_64()` call consumes ≈6500 gas

   log.Deposit(pubkey, withdrawal_credentials, amount, signature, self.to_little_endian_64(self.deposit_count))

    # cumulative gas consumption so far: ≈67161
    # NOTE: one `to_little_endian_64()` call consumes ≈6500 gas

    # Compute `DepositData` root
    zero_bytes32: bytes32 = 0x0000000000000000000000000000000000000000000000000000000000000000
    pubkey_root: bytes32 = sha256(concat(pubkey, slice(zero_bytes32, start=0, len=64 - PUBKEY_LENGTH)))

    # cumulative gas consumption so far: ≈70434

    signature_root: bytes32 = sha256(concat(
        sha256(slice(signature, start=0, len=64)),
        sha256(concat(slice(signature, start=64, len=SIGNATURE_LENGTH - 64), zero_bytes32)),
    ))

    # cumulative gas consumption so far: ≈75661

    node: bytes32 = sha256(concat(
        sha256(concat(pubkey_root, withdrawal_credentials)),
        sha256(concat(amount, slice(zero_bytes32, start=0, len=32 - AMOUNT_LENGTH), signature_root)),
    ))

    # cumulative gas consumption so far: ≈81622

   # Add `DepositData` root to Merkle tree (update a single `branch` node)
    self.deposit_count += 1

    # cumulative gas consumption so far: ≈87271
    # NOTE: `self.deposit_count += 1` consumes 5000 gas (except for the first deposit which consumes 20000 gas)

    size: uint256 = self.deposit_count

    # cumulative gas consumption so far: ≈87486

   for height in range(DEPOSIT_CONTRACT_TREE_DEPTH):
        if bitwise_and(size, 1) == 1:  # More gas efficient than `size % 2 == 1`
            self.branch[height] = node
            # NOTE: `self.branch[height] = node` consumes 5000 gas or 20000 gas
            # depends on if `self.branch[height]` has been updated before
            break
        node = sha256(concat(self.branch[height], node))
        # NOTE: `node = sha256(concat(self.branch[height], node))`` consumes ≈1200 gas
        size /= 2
        # NOTE: `size /= 2` consumes ≈40 gas

       # cumulative gas consumption so far: ≈92651 and up (extra 15000 gas if `self.branch[height]` hasn't been updated before
       # and extra ≈1200 gas for each iteration in the loop

[JustinDrake]

@NIC619: This kind of gas analysis is helpful :) Would you mind updating the analysis using the code in #1152?

[NIC619]

Will do!

[JustinDrake]

I advertised gas golfing on Twitter https://twitter.com/drakefjustin/status/1139989022577758211

[NIC619]

updated!

[jacqueswww]

Curious to see what the gas improvements are with vyperlang/vyper#1499

[JustinDrake]

Thanks @NIC619, very clear breakdown. Looks like there isn't much fat.

Only two gas improvements were suggested from the golfing bounties:

1. https://twitter.com/nicksdjohnson/status/1140018871513993216
2. https://twitter.com/jadler0/status/1140422368323678209

Both seem minor enough to not bother impeding formal verification.