EVMv2 AddOperationV2

[siladu]

Part of #10131

Use UInt256.add(UInt256) instead of bytes[] version. Can compare performance of byte version in subsequent PR.

Benchmark                                               (caseName)  Mode  Cnt    Score   Error  Units
AddOperationBenchmark.executeOperation           ADD_RANDOM_RANDOM  avgt   15   79.222 ± 0.998  ns/op
v2.AddOperationBenchmarkV2.executeOperation      ADD_RANDOM_RANDOM  avgt   15   11.103 ± 0.197  ns/op

(Note AddOperationBenchmark uses AddOperationOptimized)

and also results with long[] -> byte[] to use the other algorithm: UInt256.add(byte[],byte[])
main...siladu:besu:evmv2-add-uint256-bytes

Benchmark                                    Mode  Cnt   Score   Error  Units
v2.AddOperationBenchmarkV2.executeOperation  avgt   15  65.264 ± 0.188  ns/op

---

For reference here's the stack_long_array bench

Benchmark                               Mode  Cnt   Score   Error  Units
AddOperationBenchmark.executeOperation  avgt   15  15.230 ± 0.052  ns/op

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Implements EVM v2 ADD using UInt256.add(UInt256) and adds targeted tests/bench updates to validate correctness and measure the performance impact.

Changes:

• Implemented AddOperationV2.staticOperation(MessageFrame) using UInt256 arithmetic and v2 stack limb wiring.
• Added a dedicated AddOperationV2Test plus a v2 stack-reading helper for assertions.
• Updated call sites (EVM run loop + JMH benchmark) and normalized a constant name in MulModOperationV2.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
evm/src/main/java/org/hyperledger/besu/evm/v2/operation/AddOperationV2.java	Implements v2 ADD logic using UInt256.add and writes result back to v2 long[] stack.
evm/src/main/java/org/hyperledger/besu/evm/EVM.java	Updates v2 dispatch to call the new AddOperationV2.staticOperation(frame) signature.
ethereum/core/src/jmh/java/org/hyperledger/besu/ethereum/vm/operations/v2/AddOperationBenchmarkV2.java	Aligns benchmark invocation with the updated staticOperation signature.
evm/src/main/java/org/hyperledger/besu/evm/v2/operation/MulModOperationV2.java	Renames success constant to uppercase to match constant naming conventions.
evm/src/test/java/org/hyperledger/besu/evm/v2/testutils/TestMessageFrameBuilderV2.java	Adds a helper to read a UInt256 from the v2 stack for test assertions.
evm/src/test/java/org/hyperledger/besu/evm/v2/operation/AddOperationV2Test.java	Adds parameterized and edge-case tests for v2 ADD stack behavior and gas cost.

[lu-pinto]

nit: what are you asserting here? test code?

[siladu]

Stack remains unmodified by error codepath

[lu-pinto]

can remove these if check entirely, your inputs are never of this form

[siladu]

I have removed ZERO ternary here aa2af3e

Here we want to assert the actual long[] form with expected bytes in String form.
I think UInt256 is a nice convenience to do this and gives a useful output showing the limbs in the case the test fails:

Expected :UInt256[u3=2305843009213693954, u2=3458764513820540931, u1=4611686018427387908, u0=5764607523034234885]
Actual   :UInt256[u3=5260204364768739339, u2=2810246167479189513, u1=-2017612633061982201, u0=-9223372036854775804]

This is also conventional we are following in existing V2 tests.

If not this, what would you suggest for the assert?

[lu-pinto]

nit: on a halt condition you don't care about the state of the stack since your are halting the frame so this assertion is pointless

[siladu]

I dunno, I think it's reasonable to confirm the operation hasn't modified the stack during its error path, considering it modifies in the happy path.

The fact it isn't used is business logic outside the scope of this test IMO.

[lu-pinto]

same as https://github.com/besu-eth/besu/pull/10255/changes#r3110757045

[lu-pinto]

What about some property tests like the division arithmetics or byte shifting opcodes have? Though I understand if you want to work on it separately to unblock this PR

[siladu]

What about some property tests like the division arithmetics or byte shifting opcodes have? Though I understand if you want to work on it separately to unblock this PR

I addressed this in the javadoc for AddOperationV2Test:

• Structural test data for add(a, b) = expected. Arithmetic correctness is covered by
• UInt256PropertyBasedTest; these cases verify stack arity, limb-level read/write wiring, and
• 256-bit wrap handling.

Coverage exists here

besu/evm/src/test/java/org/hyperledger/besu/evm/UInt256PropertyBasedTest.java

Line 1225 in 570bc9f

// region Simple ADD tests

[siladu]

Added some zero operand benchmarks to another branch to test with/without fast path, see 102dc44#diff-1dd3d21e8e8983e4ecdba9fc4f1c0664ebc7ce55392d59a04495f0ff10238d98

Benchmark                                            Mode  Cnt   Score   Error  Units
AddOperationBenchmark.executeOperation               avgt   15  85.950 ± 0.591  ns/op
AddOperationBenchmark.fastPathZeroA                  avgt   15  30.074 ± 0.325  ns/op
AddOperationBenchmark.fastPathZeroB                  avgt   15  32.596 ± 0.034  ns/op
# this PR - fast paths removed
v2.AddOperationBenchmarkV2.executeOperation          avgt   15  13.288 ± 0.688  ns/op
v2.AddOperationBenchmarkV2.fastPathZeroA             avgt   15   6.784 ± 0.005  ns/op
v2.AddOperationBenchmarkV2.fastPathZeroB             avgt   15   7.488 ± 0.015  ns/op
# branch - fast paths re-added
v2.AddOperationFastPathBenchmarkV2.executeOperation  avgt   15  25.626 ± 1.479  ns/op
v2.AddOperationFastPathBenchmarkV2.fastPathZeroA     avgt   15   4.774 ± 0.147  ns/op
v2.AddOperationFastPathBenchmarkV2.fastPathZeroB     avgt   15  14.635 ± 0.031  ns/op

[siladu]

No significant difference between zero fast path or no fast path for ADDMOD

Added some extra cases on separate branch to explore this main...siladu:besu:evmv2-add-bench-fastpath

Benchmark                                                     (caseName)  Mode  Cnt    Score   Error  Units
AddModFastPathOperationBenchmark.executeOperation  ZERO_ADDMOD_128_256_0  avgt   15   36.987 ± 0.830  ns/op
AddModFastPathOperationBenchmark.executeOperation         ADDMOD_0_32_32  avgt   15   37.693 ± 0.174  ns/op
AddModFastPathOperationBenchmark.executeOperation         ADDMOD_32_0_32  avgt   15   37.105 ± 0.469  ns/op
AddModFastPathOperationBenchmark.executeOperation       ADDMOD_0_256_128  avgt   15  102.156 ± 0.682  ns/op
AddModFastPathOperationBenchmark.executeOperation       ADDMOD_256_0_128  avgt   15  100.709 ± 0.732  ns/op
AddModFastPathOperationBenchmark.executeOperation       ADDMOD_0_256_192  avgt   15   92.984 ± 0.615  ns/op
AddModFastPathOperationBenchmark.executeOperation       ADDMOD_256_0_192  avgt   15   95.771 ± 2.254  ns/op
AddModFastPathOperationBenchmark.executeOperation       ADDMOD_0_256_256  avgt   15   59.819 ± 0.387  ns/op
AddModFastPathOperationBenchmark.executeOperation       ADDMOD_256_0_256  avgt   15   59.726 ± 0.670  ns/op
AddModFastPathOperationBenchmark.executeOperation            FULL_RANDOM  avgt   15  117.613 ± 1.958  ns/op
AddModOperationBenchmark.executeOperation          ZERO_ADDMOD_128_256_0  avgt   15   36.788 ± 0.369  ns/op
AddModOperationBenchmark.executeOperation                 ADDMOD_0_32_32  avgt   15   37.832 ± 0.158  ns/op
AddModOperationBenchmark.executeOperation                 ADDMOD_32_0_32  avgt   15   37.250 ± 0.575  ns/op
AddModOperationBenchmark.executeOperation               ADDMOD_0_256_128  avgt   15  102.295 ± 0.646  ns/op
AddModOperationBenchmark.executeOperation               ADDMOD_256_0_128  avgt   15  101.087 ± 0.234  ns/op
AddModOperationBenchmark.executeOperation               ADDMOD_0_256_192  avgt   15   92.784 ± 0.742  ns/op
AddModOperationBenchmark.executeOperation               ADDMOD_256_0_192  avgt   15   93.832 ± 1.016  ns/op
AddModOperationBenchmark.executeOperation               ADDMOD_0_256_256  avgt   15   60.195 ± 0.221  ns/op
AddModOperationBenchmark.executeOperation               ADDMOD_256_0_256  avgt   15   60.214 ± 0.282  ns/op
AddModOperationBenchmark.executeOperation                    FULL_RANDOM  avgt   15  119.087 ± 3.569  ns/op