Prefetch JUMPDESTs through RPC #427

einar-polygon · 2024-07-23T13:59:50Z

resolves #290

evm_arithmetization/Cargo.toml

zero_bin/rpc/Cargo.toml

zero_bin/rpc/src/native/txn.rs

zero_bin/rpc/src/main.rs

zero_bin/rpc/src/native/mod.rs

Nashtare

Thanks Einar! Aside from the inline comments, a miss in this PR is the application of JD pre-fetching for jerigon as well, not only the native tracer.

I haven't tested it yet as I write this, but I think your current branch would crash on any jerigon payload (from the ZeroTracer) because the RPC will provide a Vec<ZeroTxResult> which contains TxnInfo, i.e. with the jumpdest table field already processed, as opposed to the TxnMeta from which you're starting the processing.

I think we're also missing fallback mechanisms, wherever the procedure fails for some reason, then we should default to the current behavior, i.e. no provided JDT and let the prover do the work.

evm_arithmetization/src/cpu/kernel/interpreter.rs

evm_arithmetization/src/cpu/kernel/tests/add11.rs

Cargo.toml

evm_arithmetization/src/generation/prover_input.rs

zero_bin/rpc/src/native/txn.rs

Nashtare

You also need to address the format change on all hardcoded witnesses, either by re-generating them, or adding default empty JDTs to each of them, and leaving the prover handle them internally as before.

einar-polygon · 2024-08-26T10:13:15Z

You also need to address the format change on all hardcoded witnesses, either by re-generating them, or adding default empty JDTs to each of them, and leaving the prover handle them internally as before.

Indeed. Which solution do you prefer and who could I talk to about this CI setup?

Nashtare · 2024-08-26T12:12:56Z

You also need to address the format change on all hardcoded witnesses, either by re-generating them, or adding default empty JDTs to each of them, and leaving the prover handle them internally as before.

Indeed. Which solution do you prefer and who could I talk to about this CI setup?

I'm not sure I understand the relation between the broken tests and the CI. We don't have to update them all (especially because some dev blocks are from an outdated devchain IIRC), but at least have them have proper format so that they can run correctly. If your question was how to test this logic of JDT pre-fetching in the CI, we could do this with the Jerigon Integration related jobs, we're testing both native and jerigon tracers there.

Nashtare · 2024-08-26T14:28:08Z

@einar-polygon Has it been tested against some endpoint? Should I run it against a few dozen blocks to see how it behaves? (ideally after conflicts are addressed)

einar-polygon · 2024-08-26T15:00:47Z

@einar-polygon Has it been tested against some endpoint? Should I run it against a few dozen blocks to see how it behaves? (ideally after conflicts are addressed)

I checked that the JUMPDEST_tables were identical for the small Calcun block and a few others against the paid Quicknode endpoint. But it would be good to have a set of interesting blocks with exceptions, reverts and what not.

Here is a start:

txn type 0-3
20548415
20240058
19665756

I will add a bash script to iterate over them while we test.

Nashtare · 2024-08-26T15:22:34Z

I'll launch a run against the devchain once it's rebased (it has a conflicting Erigon version that requires some latest PRs here)

Nashtare · 2024-08-29T13:19:01Z

I got a panic with your endpoint on block 0x13adb68
at line let lower_bytes = U160::from(callee_raw); in zk_evm/zero_bin/rpc/src/native/txn.rs:

Uint conversion error: Value is too large for Uint<160>

Nashtare · 2024-08-29T14:08:15Z

So some more data with your endpoint. I tried 140 consecutive blocks independently.

140/140 failed
36 valid witnesses, failing when batching transactions (tested with -b 2)
32 failures due to unreachable code, all from failed transactions -> cf discussion on slack
72 failures due to Uint conversion error: Value is too large for Uint<160> as shared above

trace_decoder/src/decoding.rs

evm_arithmetization/src/generation/jumpdest.rs

evm_arithmetization/src/generation/prover_input.rs

zero_bin/rpc/src/jumpdest.rs

zero_bin/rpc/src/jerigon.rs

zero_bin/rpc/src/main.rs

zero_bin/rpc/src/jumpdest.rs

zero/src/rpc/mod.rs

zero/src/bin/rpc.rs

zero/src/rpc/jerigon.rs

atanmarko · 2024-09-25T10:44:51Z

zero/src/rpc/jerigon.rs

+{
+    let tx_traces = tx_trace
+        .iter()
+        .map(|(h, t)| (Address::from(h.to_fixed_bytes()), t.clone()))


Cloning TxnTrace is not a good idea. Lot of overhead.
There could be 2 solutions:

Define generate_jumpdest_table with:

tx_traces: impl Iterator<Item = (&'a Address, &'a TxnTrace)>,

Convert tx_trace H160 to Address when you initially create it.

Confusing here is that we use both type Address = H160 and alloy_primitives::Address. But anyway, conversion should be done so that there is no unnecessary cloning of data.

…nsaction_jumps/pr

0xaatif

Casual review

0xaatif · 2024-09-29T21:18:15Z

evm_arithmetization/src/generation/jumpdest.rs

@@ -0,0 +1,191 @@
+use std::cmp::max;


This is really missing some high-level documentation - what is a jumpdest, why do I care about it, how does it fit into the wider picture?

Strong point. I added some module rustdoc.

Let me know how you like it.

evm_arithmetization/src/generation/jumpdest.rs

0xaatif · 2024-09-29T21:23:57Z

evm_arithmetization/src/generation/jumpdest.rs

+#[derive(PartialEq, Eq, Debug, Clone, Serialize, Deserialize, Default)]
+pub(crate) struct JumpDestTableProcessed(HashMap<usize, Vec<usize>>);
+
+/// Map `CodeAddress -> (Context -> [JumpDests])`


nit: hyperlink

But these are not defined types.

0xaatif · 2024-09-29T21:40:53Z

zero/src/rpc/jumpdest.rs

+
+/// Predicate that determines whether a `StructLog` that includes memory is
+/// required.
+fn trace_contains_create(structlog: &[StructLog]) -> bool {


I think this doc is redundant with the function name and signature

zero/src/rpc/jumpdest.rs

* refactor: StructPogPrime * keep module doc, avoid name `vec`

einar-polygon added performance Performance improvement related changes crate: evm_arithmetization Anything related to the evm_arithmetization crate. crate: zero_bin Anything related to the zero-bin subcrates. labels Jul 23, 2024

einar-polygon self-assigned this Jul 23, 2024

github-actions bot added the crate: trace_decoder Anything related to the trace_decoder crate. label Jul 23, 2024

Nashtare added this to the Performance Tuning milestone Jul 24, 2024

einar-polygon force-pushed the einar/prefetch_transaction_jumps/pr branch 2 times, most recently from 7ad89de to 18d5da4 Compare July 30, 2024 12:36

einar-polygon force-pushed the einar/prefetch_transaction_jumps/pr branch 2 times, most recently from 6255f00 to 6613802 Compare August 7, 2024 15:25

Nashtare reviewed Aug 7, 2024

View reviewed changes

evm_arithmetization/Cargo.toml Outdated Show resolved Hide resolved

evm_arithmetization/Cargo.toml Outdated Show resolved Hide resolved

zero_bin/rpc/Cargo.toml Outdated Show resolved Hide resolved

zero_bin/rpc/Cargo.toml Outdated Show resolved Hide resolved

Nashtare requested a review from 4l0n50 August 8, 2024 13:26

Nashtare reviewed Aug 17, 2024

View reviewed changes

zero_bin/rpc/src/native/txn.rs Outdated Show resolved Hide resolved

zero_bin/rpc/src/native/txn.rs Outdated Show resolved Hide resolved

einar-polygon force-pushed the einar/prefetch_transaction_jumps/pr branch from bdcf935 to 270256f Compare August 18, 2024 23:50

einar-polygon commented Aug 19, 2024

View reviewed changes

zero_bin/rpc/src/main.rs Outdated Show resolved Hide resolved

einar-polygon commented Aug 19, 2024

View reviewed changes

zero_bin/rpc/src/native/mod.rs Outdated Show resolved Hide resolved

Nashtare requested changes Aug 19, 2024

View reviewed changes

Nashtare reviewed Aug 22, 2024

View reviewed changes

einar-polygon force-pushed the einar/prefetch_transaction_jumps/pr branch 2 times, most recently from a0f0438 to 05e065a Compare August 28, 2024 13:58

Nashtare reviewed Aug 30, 2024

View reviewed changes

trace_decoder/src/decoding.rs Outdated Show resolved Hide resolved

trace_decoder/src/decoding.rs Outdated Show resolved Hide resolved

trace_decoder/src/decoding.rs Outdated Show resolved Hide resolved

einar-polygon force-pushed the einar/prefetch_transaction_jumps/pr branch from a9b13a3 to e351ff9 Compare September 2, 2024 17:54

Nashtare requested changes Sep 3, 2024

View reviewed changes

Nashtare reviewed Sep 24, 2024

View reviewed changes

zero/src/rpc/mod.rs Outdated Show resolved Hide resolved

einar-polygon added 2 commits September 24, 2024 20:06

Feedback

66ea811

fixup! Feedback

f230b84

atanmarko reviewed Sep 25, 2024

View reviewed changes

einar-polygon added 2 commits September 25, 2024 23:35

feedback

90722a3

fix missing code for CREATE

a0e0879

einar-polygon requested a review from wborgeaud as a code owner September 25, 2024 21:42

einar-polygon added 3 commits September 26, 2024 17:26

fix

6bff4e4

Merge remote-tracking branch 'origin/develop' into einar/prefetch_tra…

5e4162d

…nsaction_jumps/pr

fix arguments

c26f475

0xaatif reviewed Sep 29, 2024

View reviewed changes

einar-polygon added 7 commits September 30, 2024 17:39

feedback

f367409

fix

464dbc0

debugging 460

2783ccd

debugging 460

abee812

dbg

8bfccdd

bugfix

f9c2f76

dbg

313d78d

0xaatif mentioned this pull request Oct 1, 2024

discuss: deser in #427 #681

Merged

einar-polygon and others added 10 commits October 1, 2024 16:53

fix

09079e6

batching working

7c84a63

cleanups

4202ece

feedback docs

e1124d3

feedback

1f8d476

feedback filtermap

7319a15

review

d4838e0

fmt

27b5719

fix set_jumpdest_analysis_inputs_rpc

eaf3ed7

discuss: deser in #427 (#681)

10b6a22

* refactor: StructPogPrime * keep module doc, avoid name `vec`

einar-polygon force-pushed the einar/prefetch_transaction_jumps/pr branch from 143f05d to 10b6a22 Compare October 3, 2024 21:02

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Prefetch JUMPDESTs through RPC #427

Prefetch JUMPDESTs through RPC #427

einar-polygon commented Jul 23, 2024 •

edited

Loading

Nashtare left a comment

Nashtare left a comment

einar-polygon commented Aug 26, 2024

Nashtare commented Aug 26, 2024

Nashtare commented Aug 26, 2024

einar-polygon commented Aug 26, 2024

Nashtare commented Aug 26, 2024

Nashtare commented Aug 29, 2024

Nashtare commented Aug 29, 2024 •

edited

Loading

atanmarko Sep 25, 2024

atanmarko Sep 25, 2024 •

edited

Loading

0xaatif left a comment

0xaatif Sep 29, 2024

einar-polygon Oct 2, 2024

0xaatif Sep 29, 2024

einar-polygon Oct 2, 2024

0xaatif Sep 29, 2024

Prefetch JUMPDESTs through RPC #427

Are you sure you want to change the base?

Prefetch JUMPDESTs through RPC #427

Conversation

einar-polygon commented Jul 23, 2024 • edited Loading

Nashtare left a comment

Choose a reason for hiding this comment

Nashtare left a comment

Choose a reason for hiding this comment

einar-polygon commented Aug 26, 2024

Nashtare commented Aug 26, 2024

Nashtare commented Aug 26, 2024

einar-polygon commented Aug 26, 2024

Nashtare commented Aug 26, 2024

Nashtare commented Aug 29, 2024

Nashtare commented Aug 29, 2024 • edited Loading

atanmarko Sep 25, 2024

Choose a reason for hiding this comment

atanmarko Sep 25, 2024 • edited Loading

Choose a reason for hiding this comment

0xaatif left a comment

Choose a reason for hiding this comment

0xaatif Sep 29, 2024

Choose a reason for hiding this comment

einar-polygon Oct 2, 2024

Choose a reason for hiding this comment

0xaatif Sep 29, 2024

Choose a reason for hiding this comment

einar-polygon Oct 2, 2024

Choose a reason for hiding this comment

0xaatif Sep 29, 2024

Choose a reason for hiding this comment

einar-polygon commented Jul 23, 2024 •

edited

Loading

Nashtare commented Aug 29, 2024 •

edited

Loading

atanmarko Sep 25, 2024 •

edited

Loading