Update EIP-7883: Assume minimal base/mod length of 32

[chfast]

From the performance analysis of the modexp precompile after EIP-7883, the increase minimal cost of 500 don't fully cover the worst case of small base/modulus. E.g. the case of the smallest base/modulus of 8 bytes the maximum exponent within the 500 gas cost budget has 880 bits. This case performs significantly worst than the average case (modulus 32 bytes).

We propose to eliminate the pathological cases of base/modulus smaller than 32 bytes by computing the gas cost as if they had 32 byte length. By the Mainnet usage data of the modexp, a modulus smaller than 32 bytes had never been used.

Additional context in the EIP discussion: https://ethereum-magicians.org/t/eip-7883-modexp-gas-cost-increase/22841/6.

The below graphs (of different zoom levels) visualize the proposed changes. They show:

• EIP-2565 common (modulus 32 bytes) gas curve,
• EIP-7883 gas curves for modulus 8, 16, 24, 32 bytes long,
• The proposed gas cost collapsing moduli <= 32 bytes into single curve.

You can notice how slowly rises the EIP-7883 mod=8b curve.

[eth-bot]

✅ All reviewers have approved.

[chfast]

The benchmarks are available in ethereum/execution-spec-tests#1701.

This change moves the worst case: from mod_even_8b_exp_896 to mod_even_32b_exp_96.

• Geth: 31 → 47 Mgps
• Besu: 40 → 90 Mgps
• evmone: 79 → 190 Mgps

[marcindsobczak]

Assuming minimal base/mod length of 32 is a great idea and worth to update an EIP and add it. It is covering some edge-case scenarios while having zero impact on common cases. Win-win.
On the other hand, with min price at 200 there are some cases that are performing significantly worse than EcRecover precompile. I would stick to increasing min price to 500.

[marcindsobczak]

With minimal base/mod length of 32 and min price at 500, our worst case should be mod_even_32b_exp_96.

[chfast]

With minimal base/mod length of 32 and min price at 500, our worst case should be mod_even_32b_exp_96.

I restored the minimal cost of 500. The description and benchmark results are also updated.

[MarekM25]

If we approve this we need to adjust test cases
https://eips.ethereum.org/EIPS/eip-7883#test-cases

[chfast]

If we approve this we need to adjust test cases https://eips.ethereum.org/EIPS/eip-7883#test-cases

Correct. But these test cases are not attached so I cannot do so.

[spencer-tb]

If we approve this we need to adjust test cases https://eips.ethereum.org/EIPS/eip-7883#test-cases

Correct. But these test cases are not attached so I cannot do so.

Here is the pricing changes from our side in EEST if that helps.
https://github.com/ethereum/execution-spec-tests/pull/1729/files#diff-439a8fdd72de2d7e306ccff66f38ba0ae4ee4dba077de1dd90571eee3374d336

[gakonst]

ACK from Reth side.

[rakita]

Makes sense to do it

[barnabasbusa]

Can we get this PR merged in?

[lu-pinto]

lgtm

[eth-bot]

All Reviewers Have Approved; Performing Automatic Merge...