perf(levm): store valid jump targets with code #4961
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Lines of code reportTotal lines added: Detailed view |
Oppen
changed the title
Oppen
reviewed
Oct 21, 2025
Oppen
reviewed
Oct 21, 2025
Oppen
reviewed
Oct 21, 2025
Oppen
reviewed
Oct 21, 2025
Oppen
reviewed
Oct 21, 2025
Oppen
reviewed
Oct 21, 2025
Oppen
reviewed
Oct 21, 2025
Benchmark Block Execution Results Comparison Against Main
|
edg-l
reviewed
Oct 21, 2025
crates/common/trie/trie.rs
Outdated
| // Hash value for an empty trie, equal to keccak(RLP_NULL) | ||
| pub static ref EMPTY_TRIE_HASH: H256 = H256::from_slice( | ||
| Keccak256::new() | ||
| &Keccak256::new() |
There was a problem hiding this comment.
This and other changes like this are to silence a deprecated notice from sha
|
The size diff appears large due to a cargo.lock update in the tooling/bench code adding +3,819 lines, the pr is not big |
Benchmark for bdbc6fcClick to view benchmark
|
Benchmark for 16909a5Click to view benchmark
|
MegaRedHand
reviewed
Oct 23, 2025
Comment on lines
80
to
85
| # Run hyperfine benchmark, removing the database before each run. | ||
| # Pipe `yes` into the remove command to automatically confirm. | ||
| hyperfine -w 5 -N -r 10 --show-output --export-markdown "bench_pr_comparison.md" \ | ||
| -L bin "$BINS" -n "{bin}" \ | ||
| --prepare "yes | ./bin/ethrex-{bin} removedb" \ | ||
| "./bin/ethrex-{bin} --network fixtures/genesis/perf-ci.json --force import ./fixtures/blockchain/l2-1k-erc20.rlp" |
There was a problem hiding this comment.
The command was using removedb as a setup for each batch of runs, but the command was cancelled due to having no input. I changed it to use --prepare instead, which runs before each single run, and piped yes into it to automatically confirm.
There was a problem hiding this comment.
This was reverted, but we should fix this if we intend to keep this benchmark.
Benchmark for 96dee34Click to view benchmark
|
Benchmark for bda8527Click to view benchmark
|
Benchmark for 947e30aClick to view benchmark
|
Oppen
approved these changes
Oct 23, 2025
MegaRedHand
reviewed
Oct 23, 2025
Comment on lines
+283
to
+287
| let codes_hashed = guest_program_state | ||
| .codes_hashed | ||
| .iter() | ||
| .map(|(h, c)| (*h, c.bytecode.to_vec())) | ||
| .collect(); |
There was a problem hiding this comment.
Many of these conversions seem unneeded. We should check that they don't affect the performance of the L2.
MegaRedHand
reviewed
Oct 23, 2025
| let code_value = AccountCodeRLP::from(code).bytes().clone(); | ||
| batch.put_cf(&cf_codes, code_key, code_value); | ||
| let mut buf = | ||
| Vec::with_capacity(6 + code.bytecode.len() + 2 * code.jump_targets.len()); |
There was a problem hiding this comment.
Also, can't we implement RLPEncode for Code instead? Or put this in a helper.
Benchmark for 9d3432dClick to view benchmark
|
MegaRedHand
reviewed
Oct 23, 2025
| .collect::<Vec<u8>>() | ||
| .as_slice() | ||
| .encode(&mut buf); | ||
| self.write_async(CF_ACCOUNT_CODES, hash_key, buf).await |
There was a problem hiding this comment.
We need to update the documentation of CF_ACCOUNT_CODES
Benchmark for f562e67Click to view benchmark
|
MegaRedHand
reviewed
Oct 23, 2025
| let Some(bytes) = self.read_sync(CF_ACCOUNT_CODES, hash_key)? else { | ||
| return Ok(None); | ||
| }; | ||
| let bytes = Bytes::from_owner(bytes); |
There was a problem hiding this comment.
No need for this variable.
| .map(|bytes| AccountCodeRLP::from_bytes(bytes).to()) | ||
| .transpose() | ||
| .map_err(StoreError::from) | ||
| let Some(bytes) = self.read_sync(CF_ACCOUNT_CODES, hash_key)? else { |
There was a problem hiding this comment.
We could use the get_pinned_cf rocksdb method here to avoid copies before deserializing.
Comment on lines
+1160
to
+1172
| let (bytecode, targets) = decode_bytes(&bytes)?; | ||
| let (targets, rest) = decode_bytes(targets)?; | ||
| if !rest.is_empty() || !targets.len().is_multiple_of(2) { | ||
| return Err(StoreError::DecodeError); | ||
| } | ||
| let code = Code { | ||
| hash: code_hash, | ||
| bytecode: Bytes::copy_from_slice(bytecode), | ||
| jump_targets: targets | ||
| .chunks_exact(2) | ||
| .map(|c| u16::from_le_bytes([c[0], c[1]])) | ||
| .collect(), | ||
| }; |
There was a problem hiding this comment.
This could be an RLPDecode implementation.
| // Store updated code in DB | ||
| if let Some(code) = &update.code { | ||
| self.add_account_code(info.code_hash, code.clone()).await?; | ||
| self.add_account_code(code.clone()).await?; |
There was a problem hiding this comment.
Not introduced by this PR, but this clone isn't needed in the underlying Store implementation, since it only needs a reference for serialization.
Comment on lines
+262
to
+263
| .bytecode | ||
| .bytecode |
There was a problem hiding this comment.
This looks kind of weird
| let address = word_to_address(self.current_call_frame.stack.pop1()?); | ||
| let address_was_cold = !self.substate.add_accessed_address(address); | ||
| let account_code_length = self.db.get_account_code(address)?.len().into(); | ||
| // FIXME: a bit wasteful to fetch the whole code just to get the length. |
There was a problem hiding this comment.
The only use of this opcode I've seen is to check if an account is an EOA (post-7702). If it's an EOA, then SENDER and CALLER are the same, and EXTCODESIZE is 23, but later we also need to fetch the code anyways, so any cache would negate the cost of this use.
Comment on lines
+295
to
+299
| call_frame | ||
| .bytecode | ||
| .jump_targets | ||
| .binary_search(&jump_address) | ||
| .is_ok() |
There was a problem hiding this comment.
I think this should go in Code
MegaRedHand
reviewed
Oct 23, 2025
| pub hash: H256, | ||
| pub bytecode: Bytes, | ||
| // TODO: Consider using Arc<[u16]> (needs to enable serde rc feature) | ||
| pub jump_targets: Vec<u16>, |
There was a problem hiding this comment.
This needs a comment explaining the reason for using u16 instead of bigger numbers. Linking to https://eips.ethereum.org/EIPS/eip-170 and https://eips.ethereum.org/EIPS/eip-7907. Also, EIP-7907 might break this since it bumps initcode max size to 96Kb.
| } | ||
|
|
||
| fn compute_jump_targets(code: &[u8]) -> Vec<u16> { | ||
| debug_assert!(code.len() <= u16::MAX as usize); |
MegaRedHand
approved these changes
Oct 23, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation
Jumps are noticeably slow compared to most of the VM. This is due to lazy computation of invalid jump destinations.
Description
Instead of lazy computation of blocklist, do greedy computation of allowlist and store the result, fetch it with the DB.
Closes #4951
Comparison of gigagas/s in hoodi between one node synced on this branch (orange) and three others on main:
Missing:
unsafecode, it can be converted to a simple copy, there's no evidence that overhead is problematic;from_bytecode;