feat(zkevm): add log opcode worst case #1745

LouisTsai-Csie · 2025-06-15T13:40:41Z

🗒️ Description

Create Benchmark for LOG Opcode

This PR introduces a benchmark for the LOG opcode in the EVM. The LOG operation includes multiple fields: offset, size, and up to N topics depending on the variant (LOG0 through LOG4).

Gas Cost Calculation

Using LOG0 as an example, the gas cost for the LOG operation is computed as:

Static gas: 375
Dynamic gas:dynamic_gas = 375 * topic_count + 8 * size + memory_expansion_cost

To minimize gas costs:

Use zero size to avoid dynamic gas from memory access.
Use PUSH0 (which costs 2 bytes) for pushing values to the stack, as it’s cheaper than other operation like PUSH1 (3 bytes with an immediate) or DUP1 (3 bytes).

Using PUSH0 is also more space-efficient, since it avoids including immediate values and reduces bytecode size.

Benchmark Bytecode Pattern

A minimal loopable pattern for benchmarking LOGN looks like this:

JUMPDEST
PUSH0      ; offset
PUSH0      ; size
...        ; PUSH0 repeated N times for each topic
LOGN
...        ; repeatable pattern
PUSH0
JUMP

Using PUSH0 is also more space-efficient, since it avoids including immediate values and reduces bytecode size. This makes it preferable over DUP1 for setting up the stack for topics or memory offset/size.

🔗 Related Issues

Issue #1689

✅ Checklist

All: Set appropriate labels for the changes.
All: Considered squashing commits to improve commit history.
All: Considered updating the online docs in the ./docs/ directory.
Tests: Included the type and version of evm t8n tool used to locally execute test cases: e.g., ref with commit hash or geth 1.13.1-stable-3f40e65.
Tests: Ran mkdocs serve locally and verified the auto-generated docs for new tests in the Test Case Reference are correctly formatted.

jochem-brouwer

This test looks fine to me, it tests all LOG opcodes and runs those all as much as possible. However, I think we should also add these situations:

Log data length is nonzero
Log data length + data is nonzero (here we have to initialize the memory to something non-zero)
The topics are nonzero

I am not sure because I am not deep into zkEVM, is calculating the logsBloom part of the scope of the EVM (and would thus add extra logic/cycles) @jsign? If this is indeed part of it (either logsBloom or calculating receiptTrie or both) then those tests should be added for zkEVM (can be done in a different PR).

Those tests are also nice benchmark tests for non-zkEVM also 😄 👍

Will not explicitly approve for now.

LouisTsai-Csie · 2025-06-16T09:53:58Z

Thanks for comment, I do not know logsBloom and receiptTrie before, i will review some documentation for them

jsign · 2025-06-16T15:18:50Z

This test looks fine to me, it tests all LOG opcodes and runs those all as much as possible. However, I think we should also add these situations:

Log data length is nonzero

Log data length + data is nonzero (here we have to initialize the memory to something non-zero)

The topics are nonzero

I am not sure because I am not deep into zkEVM, is calculating the logsBloom part of the scope of the EVM (and would thus add extra logic/cycles) @jsign? If this is indeed part of it (either logsBloom or calculating receiptTrie or both) then those tests should be added for zkEVM (can be done in a different PR).

Those tests are also nice benchmark tests for non-zkEVM also 😄 👍

Will not explicitly approve for now.

Yep, calculating logsBloom/receipts is part of proving -- so I agree that it is useful to cover the cases you mentioned.

jsign

Looking good! Reg @jochem-brouwer suggestion to cover non-zero scenarios (which I agree are useful), I'll let you decide if those might be worked in this PR or in a separate one -- for now leaving an approval if it is the latter.

Just left one comment for your consideration.

jsign · 2025-06-16T15:20:13Z

tests/zkevm/test_worst_compute.py

+
+
+@pytest.mark.parametrize(
+    "opcode,topic_count",


Maybe we can remove the topic_count parameter, and derive it from opcode - Op.LOG0?
Mainly because combinations like Op.LOG0, 1 or Op.LOG3, 4 would never make sense?

I like this suggestion 😄

marioevz

LGTM, just one suggestion that is in line with the comments from previous reviewers, thanks!

marioevz · 2025-06-17T08:31:14Z

tests/zkevm/test_worst_compute.py

+    env = Environment()
+    max_code_size = fork.max_code_size()
+
+    iter_code = Op.PUSH0 * (2 + topic_count) + opcode


Suggested change

iter_code = Op.PUSH0 * (2 + topic_count) + opcode

iter_code = Op.PUSH0 * len(opcode.kwargs) + opcode

And you can remove the topic_count parameter.

marioevz · 2025-06-17T14:41:42Z

For the mypy error, I think if you rebase to main it should be gone, I just fixed that in another PR.

LouisTsai-Csie · 2025-06-25T20:07:31Z

tests/zkevm/test_worst_compute.py

+    calldata = (
+        Op.PUSH0
+        if empty_value
+        else Op.PUSH32(0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF)


I think this could be further optimized for readability

What about:

Suggested change

else Op.PUSH32(0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF)

else Op.PUSH32(2**256-1)

LouisTsai-Csie · 2025-06-25T20:10:46Z

tests/zkevm/test_worst_compute.py

+        if empty_value
+        else Op.PUSH32(0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF)
+    )
+    topic_count = len(opcode.kwargs or []) - 2


Here I want to extract the topic count, it is kwargs - 2 because offset and value should be excluded. And I could not use len(opcode.kwargs) - 2, this would result in data type checking issue.

For kwargs, it is either a List or None type, for the latter one, there is no len method. - Link

Yeah this is annoying, and most of the opcodes have keyword arguments already, we might need put kwargs to all opcodes and remove the | None at some point.

LouisTsai-Csie · 2025-06-25T20:11:32Z

tests/zkevm/test_worst_compute.py

+    )
+    topic_count = len(opcode.kwargs or []) - 2
+
+    offset = 0 if fixed_offset else Op.MOD(Op.GAS, 7)


Thanks @jsign for this randomize trick!

LouisTsai-Csie · 2025-06-26T11:36:54Z

@jsign @jochem-brouwer I’ve updated the test case and added the following scenarios:

Topics are zero and non-zero
Log data has a non-zero length

Could you please help me review again? And i am asking why the CI is failing

marioevz

Thanks! Just some minor comments.

I think we can move this new test function into a new file, perhaps tests/zkevm/test_worst_opcode.py ?

We can start by moving only this new function for now in this PR, I have an idea on how to fix the coverage script and we can move the rest once that's fixed.

marioevz · 2025-06-27T17:48:03Z

tests/zkevm/test_worst_compute.py

+        if empty_value
+        else Op.PUSH32(0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF)
+    )
+    topic_count = len(opcode.kwargs or []) - 2


Yeah this is annoying, and most of the opcodes have keyword arguments already, we might need put kwargs to all opcodes and remove the | None at some point.

marioevz · 2025-06-27T17:48:55Z

tests/zkevm/test_worst_compute.py

+    calldata = (
+        Op.PUSH0
+        if empty_value
+        else Op.PUSH32(0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF)


What about:

Suggested change

else Op.PUSH32(0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF)

else Op.PUSH32(2**256-1)

marioevz · 2025-06-27T18:45:39Z

tests/zkevm/test_worst_compute.py

+        0,  # 0 bytes
+        10,  # 10 bytes
+        100,  # 100 bytes
+        1 * 1024,  # 1KiB
+        10 * 1024,  # 10KiB
+        100 * 1024,  # 100KiB
+        1024 * 1024,  # 1MiB


I feel this yields too many tests, we should reduce it maybe to zero memory and 1MiB, so it's clear during benchmarking whether the size affects performance or not.

marioevz · 2025-06-27T18:49:29Z

tests/zkevm/test_worst_compute.py

+    pre: Alloc,
+    fork: Fork,
+    opcode: Opcode,
+    empty_value: bool,


Suggested change

empty_value: bool,

empty_topic bool,

empty_value reads a bit like value in the transaction.

LouisTsai-Csie · 2025-06-27T20:09:25Z

@marioevz Thanks for review. My original problem for evmone coverage report is that the CI never ends, but after I rebase again, the issue changed.

After checking the coverage report, the root cause is from BLS signature, which has nothing to do with this benchmark. I move the test case to tests/zkevm/test_worst_opcode.py, let's see if this fix the CI issue

@jsign @jochem-brouwer could you please also review this benchmark, let me know if I miss any cases!

LouisTsai-Csie self-assigned this Jun 15, 2025

LouisTsai-Csie added the feature:zkevm label Jun 15, 2025

LouisTsai-Csie marked this pull request as ready for review June 15, 2025 14:19

LouisTsai-Csie added the scope:tests Scope: Changes EL client test cases in `./tests` label Jun 16, 2025

LouisTsai-Csie requested review from jochem-brouwer and jsign and removed request for jochem-brouwer June 16, 2025 09:19

jochem-brouwer reviewed Jun 16, 2025

View reviewed changes

jsign approved these changes Jun 16, 2025

View reviewed changes

marioevz reviewed Jun 17, 2025

View reviewed changes

LouisTsai-Csie marked this pull request as draft June 17, 2025 10:55

LouisTsai-Csie mentioned this pull request Jun 24, 2025

Create Blockchain tests for Contract Lifecycle and Logging opcodes that consume the whole block #1689

Open

9 tasks

LouisTsai-Csie force-pushed the zkevm-log-opcode branch from ff3440e to d542a92 Compare June 25, 2025 19:55

LouisTsai-Csie marked this pull request as ready for review June 25, 2025 19:57

LouisTsai-Csie force-pushed the zkevm-log-opcode branch from d542a92 to 3853835 Compare June 25, 2025 20:03

LouisTsai-Csie commented Jun 25, 2025

View reviewed changes

spencer-tb added feature:benchmark labels Jun 26, 2025

marioevz reviewed Jun 27, 2025

View reviewed changes

LouisTsai-Csie added 3 commits June 27, 2025 22:02

feat(zkevm): add log opcode worst case

6088409

fix(zkevm): logic issue

53e7ec3

refactor(zkevm): update test parameters for log opcodes

980342e

LouisTsai-Csie force-pushed the zkevm-log-opcode branch from c4a0cdc to 980342e Compare June 27, 2025 20:02

refactor(zkevm): move test path for worst log opcodes

54a09c3

	iter_code = Op.PUSH0 * (2 + topic_count) + opcode
	iter_code = Op.PUSH0 * len(opcode.kwargs) + opcode

	else Op.PUSH32(0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF)
	else Op.PUSH32(2**256-1)



		@pytest.mark.parametrize(
		"opcode,topic_count",

feat(zkevm): add log opcode worst case #1745

Are you sure you want to change the base?

feat(zkevm): add log opcode worst case #1745

Conversation

LouisTsai-Csie commented Jun 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🗒️ Description

🔗 Related Issues

✅ Checklist

Uh oh!

jochem-brouwer left a comment

Choose a reason for hiding this comment

Uh oh!

LouisTsai-Csie commented Jun 16, 2025

Uh oh!

jsign commented Jun 16, 2025

Uh oh!

jsign left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

marioevz left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

marioevz commented Jun 17, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

LouisTsai-Csie commented Jun 26, 2025

Uh oh!

marioevz left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

LouisTsai-Csie commented Jun 27, 2025

Uh oh!

Uh oh!

LouisTsai-Csie commented Jun 15, 2025 •

edited

Loading

jsign left a comment •

edited

Loading