Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat: Prototype native merkle trees #4457

Merged
merged 42 commits into from
Feb 9, 2024
Merged

Conversation

PhilWindle
Copy link
Collaborator

@PhilWindle PhilWindle commented Feb 6, 2024

This PR contains work around understanding the performance of our hash functions and merkle trees.

  1. Parameterise our existing MerkleTree and NulliferTree to accept any compatible hash function
  2. Creates a new native AppendOnlyTree only allowing addition of leaves at the end
  3. Creates a new multi-threaded native IndexedMerkleTree.
  4. Add tests to validate new trees against previous ones
  5. Adds benchmarks for hash functions and new trees.
  6. Exposes poseidon hash functions over the WASM and benchmarks it.

Benchmarks

Hash Functions

Pedersen - Native

------------------------------------------------------------------------------------------
Benchmark                                                Time             CPU   Iterations
------------------------------------------------------------------------------------------
native_pedersen_hash_pair_bench/min_time:3.000       0.178 ms        0.178 ms        23450

Pedersen - Single hash over WASM 1000 times

Executed 1000 hashes at an average 593.7273874282837us / hash

Pedersen - 1024 hashes over WASM 10 times

Executed 10240 hashes at an average 581.7138370126486us / hash

Poseidon - Native

----------------------------------------------------------------------------------
Benchmark                                        Time             CPU   Iterations
----------------------------------------------------------------------------------
poseiden_hash_bench                          0.012 ms        0.012 ms        57022

Poseidon - Single hash over WASM 1000 times

Executed 1000 hashes at an average 48.4412202835083us / hash

Poseidon - 1024 hashes over WASM 10 times

Executed 10240 hashes at an average 37.44762847200036us / hash

Merkle Trees

The following benchmarks follow the format of {benchmark-type}/{batch-insertion-size}/{iteration-count}.

Append Only Tree

------------------------------------------------------------------------------------------------
Benchmark                                                      Time             CPU   Iterations
------------------------------------------------------------------------------------------------
append_only_tree_bench<Pedersen>/2/iterations:100           5.75 ms         5.75 ms          100
append_only_tree_bench<Pedersen>/4/iterations:100           6.02 ms         6.02 ms          100
append_only_tree_bench<Pedersen>/8/iterations:100           6.65 ms         6.65 ms          100
append_only_tree_bench<Pedersen>/16/iterations:100          7.89 ms         7.89 ms          100
append_only_tree_bench<Pedersen>/32/iterations:100          10.7 ms         10.7 ms          100
append_only_tree_bench<Pedersen>/64/iterations:100          16.3 ms         16.3 ms          100
append_only_tree_bench<Pedersen>/128/iterations:100         27.8 ms         27.8 ms          100

append_only_tree_bench<Poseidon2>/2/iterations:1000        0.395 ms        0.395 ms         1000
append_only_tree_bench<Poseidon2>/4/iterations:1000        0.407 ms        0.408 ms         1000
append_only_tree_bench<Poseidon2>/8/iterations:1000        0.454 ms        0.454 ms         1000
append_only_tree_bench<Poseidon2>/16/iterations:1000       0.531 ms        0.531 ms         1000
append_only_tree_bench<Poseidon2>/32/iterations:1000       0.726 ms        0.726 ms         1000
append_only_tree_bench<Poseidon2>/64/iterations:1000        1.11 ms         1.11 ms         1000
append_only_tree_bench<Poseidon2>/128/iterations:1000       1.89 ms         1.89 ms         1000

Indexed Tree

----------------------------------------------------------------------------------------------------------
Benchmark                                                                Time             CPU   Iterations
----------------------------------------------------------------------------------------------------------
single_thread_indexed_tree_bench<Pedersen>/2/iterations:100           18.2 ms         18.2 ms          100
single_thread_indexed_tree_bench<Pedersen>/4/iterations:100           31.3 ms         31.3 ms          100
single_thread_indexed_tree_bench<Pedersen>/8/iterations:100           57.2 ms         57.2 ms          100
single_thread_indexed_tree_bench<Pedersen>/16/iterations:100           108 ms          108 ms          100
single_thread_indexed_tree_bench<Pedersen>/32/iterations:100           210 ms          210 ms          100
single_thread_indexed_tree_bench<Pedersen>/64/iterations:100           415 ms          415 ms          100
single_thread_indexed_tree_bench<Pedersen>/128/iterations:100          824 ms          824 ms          100

multi_thread_indexed_tree_bench<Pedersen>/2/iterations:100            12.8 ms         6.65 ms          100
multi_thread_indexed_tree_bench<Pedersen>/4/iterations:100            14.8 ms         7.31 ms          100
multi_thread_indexed_tree_bench<Pedersen>/8/iterations:100            18.5 ms         8.80 ms          100
multi_thread_indexed_tree_bench<Pedersen>/16/iterations:100           23.9 ms         12.0 ms          100
multi_thread_indexed_tree_bench<Pedersen>/32/iterations:100           35.1 ms         18.5 ms          100
multi_thread_indexed_tree_bench<Pedersen>/64/iterations:100           59.3 ms         31.6 ms          100
multi_thread_indexed_tree_bench<Pedersen>/128/iterations:100           109 ms         66.5 ms          100

single_thread_indexed_tree_bench<Poseidon2>/2/iterations:1000         1.26 ms         1.26 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/4/iterations:1000         2.15 ms         2.15 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/8/iterations:1000         3.88 ms         3.88 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/16/iterations:1000        7.41 ms         7.41 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/32/iterations:1000        14.6 ms         14.6 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/64/iterations:1000        28.7 ms         28.7 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/128/iterations:1000       56.8 ms         56.8 ms         1000

multi_thread_indexed_tree_bench<Poseidon2>/2/iterations:1000          1.01 ms        0.812 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/4/iterations:1000          1.17 ms        0.865 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/8/iterations:1000          1.46 ms        0.979 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/16/iterations:1000         2.05 ms         1.23 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/32/iterations:1000         3.24 ms         1.70 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/64/iterations:1000         5.84 ms         2.54 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/128/iterations:1000        13.3 ms         6.36 ms         1000

@PhilWindle PhilWindle changed the title Pw/indexed tree performance feat: Prototype native merkle trees Feb 6, 2024
@AztecBot
Copy link
Collaborator

AztecBot commented Feb 6, 2024

Benchmark results

No metrics with a significant change found.

Detailed results

All benchmarks are run on txs on the Benchmarking contract on the repository. Each tx consists of a batch call to create_note and increment_balance, which guarantees that each tx has a private call, a nested private call, a public call, and a nested public call, as well as an emitted private note, an unencrypted log, and public storage read and write.

This benchmark source data is available in JSON format on S3 here.

Values are compared against data from master at commit e58eda80 and shown if the difference exceeds 1%.

L2 block published to L1

Each column represents the number of txs on an L2 block published to L1.

Metric 8 txs 32 txs 128 txs
l1_rollup_calldata_size_in_bytes 45,700 180,580 720,100
l1_rollup_calldata_gas 222,760 871,324 3,465,820
l1_rollup_execution_gas 315,951 989,772 3,686,198
l2_block_processing_time_in_ms 1,151 4,350 17,262
note_successful_decrypting_time_in_ms 322 (+4%) 985 3,748 (+1%)
note_trial_decrypting_time_in_ms 33.4 (+70%) 116 (+1%) 144 (-3%)
l2_block_building_time_in_ms 18,015 70,847 285,185
l2_block_rollup_simulation_time_in_ms 13,259 52,106 209,985
l2_block_public_tx_process_time_in_ms 4,727 18,663 74,940 (+1%)

L2 chain processing

Each column represents the number of blocks on the L2 chain where each block has 16 txs.

Metric 5 blocks 10 blocks
node_history_sync_time_in_ms 13,786 (-1%) 26,508 (+1%)
note_history_successful_decrypting_time_in_ms 2,549 (+8%) 4,651 (+1%)
note_history_trial_decrypting_time_in_ms 141 (+63%) 225 (-1%)
node_database_size_in_bytes 17,305,680 33,026,128
pxe_database_size_in_bytes 29,923 59,478

Circuits stats

Stats on running time and I/O sizes collected for every circuit run across all benchmarks.

Circuit circuit_simulation_time_in_ms circuit_input_size_in_bytes circuit_output_size_in_bytes
private-kernel-init 307 44,528 25,825
private-kernel-ordering 195 44,145 16,225
base-rollup 1,420 128,654 933
root-rollup 82.3 4,192 729
private-kernel-inner 438 71,331 25,825
public-kernel-private-input 242 32,287 25,825
public-kernel-non-first-iteration 238 32,329 25,825
merge-rollup 7.84 2,712 933

Tree insertion stats

The duration to insert a fixed batch of leaves into each tree type.

Metric 1 leaves 2 leaves 8 leaves 16 leaves 32 leaves 128 leaves 64 leaves 512 leaves 1024 leaves 2048 leaves 8192 leaves
batch_insert_into_append_only_tree_16_depth_ms 10.2 11.0 (-7%) 13.0 (+3%) 16.9 (-3%) 22.9 (+1%) 63.7 (-1%) N/A N/A N/A N/A N/A
batch_insert_into_append_only_tree_16_depth_hash_count 16.9 17.5 23.0 31.6 47.0 143 N/A N/A N/A N/A N/A
batch_insert_into_append_only_tree_16_depth_hash_ms 0.592 0.614 (-7%) 0.555 (+4%) 0.523 (-3%) 0.480 (+1%) 0.440 N/A N/A N/A N/A N/A
batch_insert_into_append_only_tree_32_depth_ms N/A N/A N/A N/A N/A 74.3 (-7%) 47.6 239 461 898 3,552
batch_insert_into_append_only_tree_32_depth_hash_count N/A N/A N/A N/A N/A 159 96.0 543 1,055 2,079 8,223
batch_insert_into_append_only_tree_32_depth_hash_ms N/A N/A N/A N/A N/A 0.459 (-7%) 0.488 0.436 0.431 0.427 0.427
batch_insert_into_indexed_tree_20_depth_ms N/A N/A N/A N/A N/A 103 (-2%) 56.7 350 (+1%) 686 1,363 5,405
batch_insert_into_indexed_tree_20_depth_hash_count N/A N/A N/A N/A N/A 197 104 691 1,363 2,707 10,771
batch_insert_into_indexed_tree_20_depth_hash_ms N/A N/A N/A N/A N/A 0.491 (-1%) 0.497 0.474 0.473 0.473 0.471
batch_insert_into_indexed_tree_40_depth_ms N/A N/A N/A 56.5 N/A N/A N/A N/A N/A N/A N/A
batch_insert_into_indexed_tree_40_depth_hash_count N/A N/A N/A 94.1 N/A N/A N/A N/A N/A N/A N/A
batch_insert_into_indexed_tree_40_depth_hash_ms N/A N/A N/A 0.578 N/A N/A N/A N/A N/A N/A N/A

Miscellaneous

Transaction sizes based on how many contracts are deployed in the tx.

Metric 0 deployed contracts 1 deployed contracts
tx_size_in_bytes 16,859 40,311

Transaction processing duration by data writes.

Metric 0 new commitments 1 new commitments
tx_pxe_processing_time_ms 556 (+1%) 1,309
Metric 0 public data writes 1 public data writes
tx_sequencer_processing_time_ms 0.511 (+2%) 580

@PhilWindle PhilWindle marked this pull request as ready for review February 8, 2024 00:45
@PhilWindle PhilWindle merged commit 7d5e056 into master Feb 9, 2024
85 checks passed
@PhilWindle PhilWindle deleted the pw/indexed-tree-performance branch February 9, 2024 14:29
rahul-kothari pushed a commit that referenced this pull request Feb 13, 2024
🤖 I have created a release *beep* *boop*
---


<details><summary>aztec-packages: 0.24.0</summary>

##
[0.24.0](aztec-packages-v0.23.0...aztec-packages-v0.24.0)
(2024-02-13)


### ⚠ BREAKING CHANGES

* move noir out of yarn-project
([#4479](#4479))
* note type ids
([#4500](#4500))

### Features

* Add fee payment methods
([#4504](#4504))
([d107746](d107746))
* Add hashing to stdlib transcript
([#4161](#4161))
([e78b86f](e78b86f))
* Added cast opcode and cast calldata
([#4423](#4423))
([e58eda8](e58eda8))
* Added cast opcode and cast calldata
(#4423)
([db803bd](db803bd))
* **avm-transpiler:** Implement tags for SET and others
([#4545](#4545))
([3063bf3](3063bf3))
* **avm:** Implement addressing modes for MOV
([#4490](#4490))
([ab4eaf0](ab4eaf0))
* **avm:** Introduce small e2e test
([#4470](#4470))
([7b4c6e7](7b4c6e7))
* Aztec.js API for registering a contract class
([#4469](#4469))
([d566c74](d566c74))
* **docs:** DIP1 - Extracting how-tos
([#4251](#4251))
([9d50e24](9d50e24))
* Enable gmock and upgrade gtest to 1.13
([#4480](#4480))
([5fc02e7](5fc02e7))
* IVC bench
([#4515](#4515))
([d8ae42b](d8ae42b))
* Nicer API for instance deployment
([#4493](#4493))
([99c3fba](99c3fba))
* Note type ids
([#4500](#4500))
([e1da2fd](e1da2fd))
* Op count timers
([#4471](#4471))
([26918de](26918de))
* PG + Goblin
([#4399](#4399))
([295cd55](295cd55))
* Prototype native merkle trees
([#4457](#4457))
([7d5e056](7d5e056))
* Update rollup circuits and contracts in yp
([#4536](#4536))
([6e89d53](6e89d53))


### Bug Fixes

* **bb:** Publishing bb for mac intel
([#4523](#4523))
([4982e3c](4982e3c))
* Broken links in docs [REDO]
([#4540](#4540))
([ce2a205](ce2a205))
* **build-system:** Image expiring
([#4521](#4521))
([1501afd](1501afd))
* Convert folding recursive verifier ops to batch mul
([#4517](#4517))
([3750b26](3750b26))
* Cycle_group validate_is_on_curve bug
([#4494](#4494))
([fecf3f7](fecf3f7))
* Field divison / journal comparisions
([#4489](#4489))
([15c06c5](15c06c5))
* Master
([#4547](#4547))
([490ca26](490ca26))
* Mirror_noir_subrepo.yml
([#4550](#4550))
([f8d8311](f8d8311))
* Mul with endomorphism
([#4538](#4538))
([1f4c90d](1f4c90d))
* **noir-mirror:** Don't update .gitrepo on push
([#4555](#4555))
([686140a](686140a))
* Recreate jest.config.ts for sequencer-client
([#4553](#4553))
([d172f0b](d172f0b))
* StandardCircuitBuilder create_logic_constraint and uint logic_operator
([#4530](#4530))
([ce51d20](ce51d20))
* Use ordered-binary value encoding for multi maps
([#4565](#4565))
([04ae0d2](04ae0d2))


### Miscellaneous

* Aligning some naming in `BaseOrMergeRollupPublicInputs`
([#4510](#4510))
([47d66f9](47d66f9))
* **avm-circuit:** Tests use OpCode enum's instead of hardcoded values
([#4554](#4554))
([ca4dd60](ca4dd60))
* **avm-simulator:** Reduce boilerplate in AVM memory types
([#4542](#4542))
([da2f5ed](da2f5ed))
* **avm:** Add/improve tests for AvmContext, tagged memory, etc
([#4484](#4484))
([2fccdf2](2fccdf2))
* **avm:** Remove field support for comparators and bitwise ops
([#4516](#4516))
([87a9663](87a9663))
* **avm:** Use some matchers gtest functionalities to improve unit tests
([#4502](#4502))
([bf4fc6c](bf4fc6c)),
closes
[#4495](#4495)
* Cleanup of `abi.nr` in `aztec-nr`
([#4473](#4473))
([6d9c73a](6d9c73a))
* Cleanup of `abi.nr` in `aztec-nr`
(#4473) [skip ci]
([db803bd](db803bd))
* Create constraints for sha256 compression opcode
([#4503](#4503))
([64bef49](64bef49))
* Little cpp style improvements
([#4528](#4528))
([dcc9ba4](dcc9ba4))
* Move noir out of yarn-project
([#4479](#4479))
([1fe674b](1fe674b)),
closes
[#4107](#4107)
* Pull noir
([#4546](#4546))
([acf5cf2](acf5cf2))
* Pull noir (#4546)
([db803bd](db803bd))
* Redo noir subrepo force push
([#4514](#4514))
([7b519a4](7b519a4))
* Remove .oldValue in contract state update request
([#4499](#4499))
([a796bef](a796bef))
* Removing redundant utilities
([#4532](#4532))
([79bf445](79bf445)),
closes
[#3470](#3470)
* Rename kernel circuits and disambiguate inputs
([#4535](#4535))
([ed6d521](ed6d521))
* Replace relative paths to noir-protocol-circuits
([f1accbf](f1accbf))
* Replace relative paths to noir-protocol-circuits
([20ee430](20ee430))
* Replace relative paths to noir-protocol-circuits
([101ab59](101ab59))
* Simulator utils cleanup
([#4507](#4507))
([1dd0ebf](1dd0ebf))
* Sync to noir-lang/noir
([db803bd](db803bd))
* Testing all values in `PublicGlobalVariables` and
`PrivateGlobalVariables`
([#4481](#4481))
([bc25f9b](bc25f9b))
* Testing historical header in contexts
([#4509](#4509))
([c00229a](c00229a))
* Updating field conversion code without pointer hack
([#4537](#4537))
([94f436e](94f436e))
* Uses sha256compression opcode in Noir and implements acvm solver for
it
([#4511](#4511))
([9dc05bc](9dc05bc))


### Documentation

* Describe the new message box model
([#4485](#4485))
([14cc1dd](14cc1dd))
* Review of docs, so far
([#4505](#4505))
([140c508](140c508))
* **yellowpaper:** Avm tree-access operations
([#4552](#4552))
([913f4bd](913f4bd))
* **yellowpaper:** Separate section for AVM state
([#4440](#4440))
([7881f09](7881f09))
</details>

<details><summary>barretenberg.js: 0.24.0</summary>

##
[0.24.0](barretenberg.js-v0.23.0...barretenberg.js-v0.24.0)
(2024-02-13)


### Features

* Prototype native merkle trees
([#4457](#4457))
([7d5e056](7d5e056))
</details>

<details><summary>barretenberg: 0.24.0</summary>

##
[0.24.0](barretenberg-v0.23.0...barretenberg-v0.24.0)
(2024-02-13)


### Features

* Add hashing to stdlib transcript
([#4161](#4161))
([e78b86f](e78b86f))
* Added cast opcode and cast calldata
([#4423](#4423))
([e58eda8](e58eda8))
* Enable gmock and upgrade gtest to 1.13
([#4480](#4480))
([5fc02e7](5fc02e7))
* IVC bench
([#4515](#4515))
([d8ae42b](d8ae42b))
* Op count timers
([#4471](#4471))
([26918de](26918de))
* PG + Goblin
([#4399](#4399))
([295cd55](295cd55))
* Prototype native merkle trees
([#4457](#4457))
([7d5e056](7d5e056))


### Bug Fixes

* Convert folding recursive verifier ops to batch mul
([#4517](#4517))
([3750b26](3750b26))
* Cycle_group validate_is_on_curve bug
([#4494](#4494))
([fecf3f7](fecf3f7))
* Mul with endomorphism
([#4538](#4538))
([1f4c90d](1f4c90d))
* StandardCircuitBuilder create_logic_constraint and uint logic_operator
([#4530](#4530))
([ce51d20](ce51d20))


### Miscellaneous

* **avm-circuit:** Tests use OpCode enum's instead of hardcoded values
([#4554](#4554))
([ca4dd60](ca4dd60))
* **avm:** Use some matchers gtest functionalities to improve unit tests
([#4502](#4502))
([bf4fc6c](bf4fc6c)),
closes
[#4495](#4495)
* Create constraints for sha256 compression opcode
([#4503](#4503))
([64bef49](64bef49))
* Little cpp style improvements
([#4528](#4528))
([dcc9ba4](dcc9ba4))
* Updating field conversion code without pointer hack
([#4537](#4537))
([94f436e](94f436e))


### Documentation

* **yellowpaper:** Avm tree-access operations
([#4552](#4552))
([913f4bd](913f4bd))
</details>

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).
AztecBot added a commit to AztecProtocol/barretenberg that referenced this pull request Feb 14, 2024
🤖 I have created a release *beep* *boop*
---


<details><summary>aztec-packages: 0.24.0</summary>

##
[0.24.0](AztecProtocol/aztec-packages@aztec-packages-v0.23.0...aztec-packages-v0.24.0)
(2024-02-13)


### ⚠ BREAKING CHANGES

* move noir out of yarn-project
([#4479](AztecProtocol/aztec-packages#4479))
* note type ids
([#4500](AztecProtocol/aztec-packages#4500))

### Features

* Add fee payment methods
([#4504](AztecProtocol/aztec-packages#4504))
([d107746](AztecProtocol/aztec-packages@d107746))
* Add hashing to stdlib transcript
([#4161](AztecProtocol/aztec-packages#4161))
([e78b86f](AztecProtocol/aztec-packages@e78b86f))
* Added cast opcode and cast calldata
([#4423](AztecProtocol/aztec-packages#4423))
([e58eda8](AztecProtocol/aztec-packages@e58eda8))
* Added cast opcode and cast calldata
(AztecProtocol/aztec-packages#4423)
([db803bd](AztecProtocol/aztec-packages@db803bd))
* **avm-transpiler:** Implement tags for SET and others
([#4545](AztecProtocol/aztec-packages#4545))
([3063bf3](AztecProtocol/aztec-packages@3063bf3))
* **avm:** Implement addressing modes for MOV
([#4490](AztecProtocol/aztec-packages#4490))
([ab4eaf0](AztecProtocol/aztec-packages@ab4eaf0))
* **avm:** Introduce small e2e test
([#4470](AztecProtocol/aztec-packages#4470))
([7b4c6e7](AztecProtocol/aztec-packages@7b4c6e7))
* Aztec.js API for registering a contract class
([#4469](AztecProtocol/aztec-packages#4469))
([d566c74](AztecProtocol/aztec-packages@d566c74))
* **docs:** DIP1 - Extracting how-tos
([#4251](AztecProtocol/aztec-packages#4251))
([9d50e24](AztecProtocol/aztec-packages@9d50e24))
* Enable gmock and upgrade gtest to 1.13
([#4480](AztecProtocol/aztec-packages#4480))
([5fc02e7](AztecProtocol/aztec-packages@5fc02e7))
* IVC bench
([#4515](AztecProtocol/aztec-packages#4515))
([d8ae42b](AztecProtocol/aztec-packages@d8ae42b))
* Nicer API for instance deployment
([#4493](AztecProtocol/aztec-packages#4493))
([99c3fba](AztecProtocol/aztec-packages@99c3fba))
* Note type ids
([#4500](AztecProtocol/aztec-packages#4500))
([e1da2fd](AztecProtocol/aztec-packages@e1da2fd))
* Op count timers
([#4471](AztecProtocol/aztec-packages#4471))
([26918de](AztecProtocol/aztec-packages@26918de))
* PG + Goblin
([#4399](AztecProtocol/aztec-packages#4399))
([295cd55](AztecProtocol/aztec-packages@295cd55))
* Prototype native merkle trees
([#4457](AztecProtocol/aztec-packages#4457))
([7d5e056](AztecProtocol/aztec-packages@7d5e056))
* Update rollup circuits and contracts in yp
([#4536](AztecProtocol/aztec-packages#4536))
([6e89d53](AztecProtocol/aztec-packages@6e89d53))


### Bug Fixes

* **bb:** Publishing bb for mac intel
([#4523](AztecProtocol/aztec-packages#4523))
([4982e3c](AztecProtocol/aztec-packages@4982e3c))
* Broken links in docs [REDO]
([#4540](AztecProtocol/aztec-packages#4540))
([ce2a205](AztecProtocol/aztec-packages@ce2a205))
* **build-system:** Image expiring
([#4521](AztecProtocol/aztec-packages#4521))
([1501afd](AztecProtocol/aztec-packages@1501afd))
* Convert folding recursive verifier ops to batch mul
([#4517](AztecProtocol/aztec-packages#4517))
([3750b26](AztecProtocol/aztec-packages@3750b26))
* Cycle_group validate_is_on_curve bug
([#4494](AztecProtocol/aztec-packages#4494))
([fecf3f7](AztecProtocol/aztec-packages@fecf3f7))
* Field divison / journal comparisions
([#4489](AztecProtocol/aztec-packages#4489))
([15c06c5](AztecProtocol/aztec-packages@15c06c5))
* Master
([#4547](AztecProtocol/aztec-packages#4547))
([490ca26](AztecProtocol/aztec-packages@490ca26))
* Mirror_noir_subrepo.yml
([#4550](AztecProtocol/aztec-packages#4550))
([f8d8311](AztecProtocol/aztec-packages@f8d8311))
* Mul with endomorphism
([#4538](AztecProtocol/aztec-packages#4538))
([1f4c90d](AztecProtocol/aztec-packages@1f4c90d))
* **noir-mirror:** Don't update .gitrepo on push
([#4555](AztecProtocol/aztec-packages#4555))
([686140a](AztecProtocol/aztec-packages@686140a))
* Recreate jest.config.ts for sequencer-client
([#4553](AztecProtocol/aztec-packages#4553))
([d172f0b](AztecProtocol/aztec-packages@d172f0b))
* StandardCircuitBuilder create_logic_constraint and uint logic_operator
([#4530](AztecProtocol/aztec-packages#4530))
([ce51d20](AztecProtocol/aztec-packages@ce51d20))
* Use ordered-binary value encoding for multi maps
([#4565](AztecProtocol/aztec-packages#4565))
([04ae0d2](AztecProtocol/aztec-packages@04ae0d2))


### Miscellaneous

* Aligning some naming in `BaseOrMergeRollupPublicInputs`
([#4510](AztecProtocol/aztec-packages#4510))
([47d66f9](AztecProtocol/aztec-packages@47d66f9))
* **avm-circuit:** Tests use OpCode enum's instead of hardcoded values
([#4554](AztecProtocol/aztec-packages#4554))
([ca4dd60](AztecProtocol/aztec-packages@ca4dd60))
* **avm-simulator:** Reduce boilerplate in AVM memory types
([#4542](AztecProtocol/aztec-packages#4542))
([da2f5ed](AztecProtocol/aztec-packages@da2f5ed))
* **avm:** Add/improve tests for AvmContext, tagged memory, etc
([#4484](AztecProtocol/aztec-packages#4484))
([2fccdf2](AztecProtocol/aztec-packages@2fccdf2))
* **avm:** Remove field support for comparators and bitwise ops
([#4516](AztecProtocol/aztec-packages#4516))
([87a9663](AztecProtocol/aztec-packages@87a9663))
* **avm:** Use some matchers gtest functionalities to improve unit tests
([#4502](AztecProtocol/aztec-packages#4502))
([bf4fc6c](AztecProtocol/aztec-packages@bf4fc6c)),
closes
[#4495](AztecProtocol/aztec-packages#4495)
* Cleanup of `abi.nr` in `aztec-nr`
([#4473](AztecProtocol/aztec-packages#4473))
([6d9c73a](AztecProtocol/aztec-packages@6d9c73a))
* Cleanup of `abi.nr` in `aztec-nr`
(AztecProtocol/aztec-packages#4473) [skip ci]
([db803bd](AztecProtocol/aztec-packages@db803bd))
* Create constraints for sha256 compression opcode
([#4503](AztecProtocol/aztec-packages#4503))
([64bef49](AztecProtocol/aztec-packages@64bef49))
* Little cpp style improvements
([#4528](AztecProtocol/aztec-packages#4528))
([dcc9ba4](AztecProtocol/aztec-packages@dcc9ba4))
* Move noir out of yarn-project
([#4479](AztecProtocol/aztec-packages#4479))
([1fe674b](AztecProtocol/aztec-packages@1fe674b)),
closes
[#4107](AztecProtocol/aztec-packages#4107)
* Pull noir
([#4546](AztecProtocol/aztec-packages#4546))
([acf5cf2](AztecProtocol/aztec-packages@acf5cf2))
* Pull noir (AztecProtocol/aztec-packages#4546)
([db803bd](AztecProtocol/aztec-packages@db803bd))
* Redo noir subrepo force push
([#4514](AztecProtocol/aztec-packages#4514))
([7b519a4](AztecProtocol/aztec-packages@7b519a4))
* Remove .oldValue in contract state update request
([#4499](AztecProtocol/aztec-packages#4499))
([a796bef](AztecProtocol/aztec-packages@a796bef))
* Removing redundant utilities
([#4532](AztecProtocol/aztec-packages#4532))
([79bf445](AztecProtocol/aztec-packages@79bf445)),
closes
[#3470](AztecProtocol/aztec-packages#3470)
* Rename kernel circuits and disambiguate inputs
([#4535](AztecProtocol/aztec-packages#4535))
([ed6d521](AztecProtocol/aztec-packages@ed6d521))
* Replace relative paths to noir-protocol-circuits
([f1accbf](AztecProtocol/aztec-packages@f1accbf))
* Replace relative paths to noir-protocol-circuits
([20ee430](AztecProtocol/aztec-packages@20ee430))
* Replace relative paths to noir-protocol-circuits
([101ab59](AztecProtocol/aztec-packages@101ab59))
* Simulator utils cleanup
([#4507](AztecProtocol/aztec-packages#4507))
([1dd0ebf](AztecProtocol/aztec-packages@1dd0ebf))
* Sync to noir-lang/noir
([db803bd](AztecProtocol/aztec-packages@db803bd))
* Testing all values in `PublicGlobalVariables` and
`PrivateGlobalVariables`
([#4481](AztecProtocol/aztec-packages#4481))
([bc25f9b](AztecProtocol/aztec-packages@bc25f9b))
* Testing historical header in contexts
([#4509](AztecProtocol/aztec-packages#4509))
([c00229a](AztecProtocol/aztec-packages@c00229a))
* Updating field conversion code without pointer hack
([#4537](AztecProtocol/aztec-packages#4537))
([94f436e](AztecProtocol/aztec-packages@94f436e))
* Uses sha256compression opcode in Noir and implements acvm solver for
it
([#4511](AztecProtocol/aztec-packages#4511))
([9dc05bc](AztecProtocol/aztec-packages@9dc05bc))


### Documentation

* Describe the new message box model
([#4485](AztecProtocol/aztec-packages#4485))
([14cc1dd](AztecProtocol/aztec-packages@14cc1dd))
* Review of docs, so far
([#4505](AztecProtocol/aztec-packages#4505))
([140c508](AztecProtocol/aztec-packages@140c508))
* **yellowpaper:** Avm tree-access operations
([#4552](AztecProtocol/aztec-packages#4552))
([913f4bd](AztecProtocol/aztec-packages@913f4bd))
* **yellowpaper:** Separate section for AVM state
([#4440](AztecProtocol/aztec-packages#4440))
([7881f09](AztecProtocol/aztec-packages@7881f09))
</details>

<details><summary>barretenberg.js: 0.24.0</summary>

##
[0.24.0](AztecProtocol/aztec-packages@barretenberg.js-v0.23.0...barretenberg.js-v0.24.0)
(2024-02-13)


### Features

* Prototype native merkle trees
([#4457](AztecProtocol/aztec-packages#4457))
([7d5e056](AztecProtocol/aztec-packages@7d5e056))
</details>

<details><summary>barretenberg: 0.24.0</summary>

##
[0.24.0](AztecProtocol/aztec-packages@barretenberg-v0.23.0...barretenberg-v0.24.0)
(2024-02-13)


### Features

* Add hashing to stdlib transcript
([#4161](AztecProtocol/aztec-packages#4161))
([e78b86f](AztecProtocol/aztec-packages@e78b86f))
* Added cast opcode and cast calldata
([#4423](AztecProtocol/aztec-packages#4423))
([e58eda8](AztecProtocol/aztec-packages@e58eda8))
* Enable gmock and upgrade gtest to 1.13
([#4480](AztecProtocol/aztec-packages#4480))
([5fc02e7](AztecProtocol/aztec-packages@5fc02e7))
* IVC bench
([#4515](AztecProtocol/aztec-packages#4515))
([d8ae42b](AztecProtocol/aztec-packages@d8ae42b))
* Op count timers
([#4471](AztecProtocol/aztec-packages#4471))
([26918de](AztecProtocol/aztec-packages@26918de))
* PG + Goblin
([#4399](AztecProtocol/aztec-packages#4399))
([295cd55](AztecProtocol/aztec-packages@295cd55))
* Prototype native merkle trees
([#4457](AztecProtocol/aztec-packages#4457))
([7d5e056](AztecProtocol/aztec-packages@7d5e056))


### Bug Fixes

* Convert folding recursive verifier ops to batch mul
([#4517](AztecProtocol/aztec-packages#4517))
([3750b26](AztecProtocol/aztec-packages@3750b26))
* Cycle_group validate_is_on_curve bug
([#4494](AztecProtocol/aztec-packages#4494))
([fecf3f7](AztecProtocol/aztec-packages@fecf3f7))
* Mul with endomorphism
([#4538](AztecProtocol/aztec-packages#4538))
([1f4c90d](AztecProtocol/aztec-packages@1f4c90d))
* StandardCircuitBuilder create_logic_constraint and uint logic_operator
([#4530](AztecProtocol/aztec-packages#4530))
([ce51d20](AztecProtocol/aztec-packages@ce51d20))


### Miscellaneous

* **avm-circuit:** Tests use OpCode enum's instead of hardcoded values
([#4554](AztecProtocol/aztec-packages#4554))
([ca4dd60](AztecProtocol/aztec-packages@ca4dd60))
* **avm:** Use some matchers gtest functionalities to improve unit tests
([#4502](AztecProtocol/aztec-packages#4502))
([bf4fc6c](AztecProtocol/aztec-packages@bf4fc6c)),
closes
[#4495](AztecProtocol/aztec-packages#4495)
* Create constraints for sha256 compression opcode
([#4503](AztecProtocol/aztec-packages#4503))
([64bef49](AztecProtocol/aztec-packages@64bef49))
* Little cpp style improvements
([#4528](AztecProtocol/aztec-packages#4528))
([dcc9ba4](AztecProtocol/aztec-packages@dcc9ba4))
* Updating field conversion code without pointer hack
([#4537](AztecProtocol/aztec-packages#4537))
([94f436e](AztecProtocol/aztec-packages@94f436e))


### Documentation

* **yellowpaper:** Avm tree-access operations
([#4552](AztecProtocol/aztec-packages#4552))
([913f4bd](AztecProtocol/aztec-packages@913f4bd))
</details>

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).
@Maddiaa0 Maddiaa0 mentioned this pull request Feb 19, 2024
michaelelliot pushed a commit to Swoir/noir_rs that referenced this pull request Feb 28, 2024
This PR contains work around understanding the performance of our hash
functions and merkle trees.

1. Parameterise our existing MerkleTree and NulliferTree to accept any
compatible hash function
2. Creates a new native AppendOnlyTree only allowing addition of leaves
at the end
3. Creates a new multi-threaded native IndexedMerkleTree.
4. Add tests to validate new trees against previous ones
5. Adds benchmarks for hash functions and new trees.
6. Exposes poseidon hash functions over the WASM and benchmarks it.

 ### Benchmarks

**Hash Functions**

**Pedersen - Native**

```
------------------------------------------------------------------------------------------
Benchmark                                                Time             CPU   Iterations
------------------------------------------------------------------------------------------
native_pedersen_hash_pair_bench/min_time:3.000       0.178 ms        0.178 ms        23450
```
**Pedersen - Single hash over WASM 1000 times**

Executed 1000 hashes at an average 593.7273874282837us / hash

**Pedersen - 1024 hashes over WASM 10 times**

Executed 10240 hashes at an average 581.7138370126486us / hash

**Poseidon - Native**
```
----------------------------------------------------------------------------------
Benchmark                                        Time             CPU   Iterations
----------------------------------------------------------------------------------
poseiden_hash_bench                          0.012 ms        0.012 ms        57022
```

**Poseidon - Single hash over WASM 1000 times**

Executed 1000 hashes at an average 48.4412202835083us / hash

**Poseidon - 1024 hashes over WASM 10 times**

Executed 10240 hashes at an average 37.44762847200036us / hash

### Merkle Trees

The following benchmarks follow the format of
{benchmark-type}/{batch-insertion-size}/{iteration-count}.

**Append Only Tree**

```
------------------------------------------------------------------------------------------------
Benchmark                                                      Time             CPU   Iterations
------------------------------------------------------------------------------------------------
append_only_tree_bench<Pedersen>/2/iterations:100           5.75 ms         5.75 ms          100
append_only_tree_bench<Pedersen>/4/iterations:100           6.02 ms         6.02 ms          100
append_only_tree_bench<Pedersen>/8/iterations:100           6.65 ms         6.65 ms          100
append_only_tree_bench<Pedersen>/16/iterations:100          7.89 ms         7.89 ms          100
append_only_tree_bench<Pedersen>/32/iterations:100          10.7 ms         10.7 ms          100
append_only_tree_bench<Pedersen>/64/iterations:100          16.3 ms         16.3 ms          100
append_only_tree_bench<Pedersen>/128/iterations:100         27.8 ms         27.8 ms          100

append_only_tree_bench<Poseidon2>/2/iterations:1000        0.395 ms        0.395 ms         1000
append_only_tree_bench<Poseidon2>/4/iterations:1000        0.407 ms        0.408 ms         1000
append_only_tree_bench<Poseidon2>/8/iterations:1000        0.454 ms        0.454 ms         1000
append_only_tree_bench<Poseidon2>/16/iterations:1000       0.531 ms        0.531 ms         1000
append_only_tree_bench<Poseidon2>/32/iterations:1000       0.726 ms        0.726 ms         1000
append_only_tree_bench<Poseidon2>/64/iterations:1000        1.11 ms         1.11 ms         1000
append_only_tree_bench<Poseidon2>/128/iterations:1000       1.89 ms         1.89 ms         1000
```

**Indexed Tree**
```
----------------------------------------------------------------------------------------------------------
Benchmark                                                                Time             CPU   Iterations
----------------------------------------------------------------------------------------------------------
single_thread_indexed_tree_bench<Pedersen>/2/iterations:100           18.2 ms         18.2 ms          100
single_thread_indexed_tree_bench<Pedersen>/4/iterations:100           31.3 ms         31.3 ms          100
single_thread_indexed_tree_bench<Pedersen>/8/iterations:100           57.2 ms         57.2 ms          100
single_thread_indexed_tree_bench<Pedersen>/16/iterations:100           108 ms          108 ms          100
single_thread_indexed_tree_bench<Pedersen>/32/iterations:100           210 ms          210 ms          100
single_thread_indexed_tree_bench<Pedersen>/64/iterations:100           415 ms          415 ms          100
single_thread_indexed_tree_bench<Pedersen>/128/iterations:100          824 ms          824 ms          100

multi_thread_indexed_tree_bench<Pedersen>/2/iterations:100            12.8 ms         6.65 ms          100
multi_thread_indexed_tree_bench<Pedersen>/4/iterations:100            14.8 ms         7.31 ms          100
multi_thread_indexed_tree_bench<Pedersen>/8/iterations:100            18.5 ms         8.80 ms          100
multi_thread_indexed_tree_bench<Pedersen>/16/iterations:100           23.9 ms         12.0 ms          100
multi_thread_indexed_tree_bench<Pedersen>/32/iterations:100           35.1 ms         18.5 ms          100
multi_thread_indexed_tree_bench<Pedersen>/64/iterations:100           59.3 ms         31.6 ms          100
multi_thread_indexed_tree_bench<Pedersen>/128/iterations:100           109 ms         66.5 ms          100

single_thread_indexed_tree_bench<Poseidon2>/2/iterations:1000         1.26 ms         1.26 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/4/iterations:1000         2.15 ms         2.15 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/8/iterations:1000         3.88 ms         3.88 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/16/iterations:1000        7.41 ms         7.41 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/32/iterations:1000        14.6 ms         14.6 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/64/iterations:1000        28.7 ms         28.7 ms         1000
single_thread_indexed_tree_bench<Poseidon2>/128/iterations:1000       56.8 ms         56.8 ms         1000

multi_thread_indexed_tree_bench<Poseidon2>/2/iterations:1000          1.01 ms        0.812 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/4/iterations:1000          1.17 ms        0.865 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/8/iterations:1000          1.46 ms        0.979 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/16/iterations:1000         2.05 ms         1.23 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/32/iterations:1000         3.24 ms         1.70 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/64/iterations:1000         5.84 ms         2.54 ms         1000
multi_thread_indexed_tree_bench<Poseidon2>/128/iterations:1000        13.3 ms         6.36 ms         1000
```

---------

Co-authored-by: ludamad <adam@aztecprotocol.com>
michaelelliot pushed a commit to Swoir/noir_rs that referenced this pull request Feb 28, 2024
🤖 I have created a release *beep* *boop*
---


<details><summary>aztec-packages: 0.24.0</summary>

##
[0.24.0](AztecProtocol/aztec-packages@aztec-packages-v0.23.0...aztec-packages-v0.24.0)
(2024-02-13)


### ⚠ BREAKING CHANGES

* move noir out of yarn-project
([AztecProtocol#4479](AztecProtocol#4479))
* note type ids
([AztecProtocol#4500](AztecProtocol#4500))

### Features

* Add fee payment methods
([AztecProtocol#4504](AztecProtocol#4504))
([d107746](AztecProtocol@d107746))
* Add hashing to stdlib transcript
([AztecProtocol#4161](AztecProtocol#4161))
([e78b86f](AztecProtocol@e78b86f))
* Added cast opcode and cast calldata
([AztecProtocol#4423](AztecProtocol#4423))
([e58eda8](AztecProtocol@e58eda8))
* Added cast opcode and cast calldata
(AztecProtocol#4423)
([db803bd](AztecProtocol@db803bd))
* **avm-transpiler:** Implement tags for SET and others
([AztecProtocol#4545](AztecProtocol#4545))
([3063bf3](AztecProtocol@3063bf3))
* **avm:** Implement addressing modes for MOV
([AztecProtocol#4490](AztecProtocol#4490))
([ab4eaf0](AztecProtocol@ab4eaf0))
* **avm:** Introduce small e2e test
([AztecProtocol#4470](AztecProtocol#4470))
([7b4c6e7](AztecProtocol@7b4c6e7))
* Aztec.js API for registering a contract class
([AztecProtocol#4469](AztecProtocol#4469))
([d566c74](AztecProtocol@d566c74))
* **docs:** DIP1 - Extracting how-tos
([AztecProtocol#4251](AztecProtocol#4251))
([9d50e24](AztecProtocol@9d50e24))
* Enable gmock and upgrade gtest to 1.13
([AztecProtocol#4480](AztecProtocol#4480))
([5fc02e7](AztecProtocol@5fc02e7))
* IVC bench
([AztecProtocol#4515](AztecProtocol#4515))
([d8ae42b](AztecProtocol@d8ae42b))
* Nicer API for instance deployment
([AztecProtocol#4493](AztecProtocol#4493))
([99c3fba](AztecProtocol@99c3fba))
* Note type ids
([AztecProtocol#4500](AztecProtocol#4500))
([e1da2fd](AztecProtocol@e1da2fd))
* Op count timers
([AztecProtocol#4471](AztecProtocol#4471))
([26918de](AztecProtocol@26918de))
* PG + Goblin
([AztecProtocol#4399](AztecProtocol#4399))
([295cd55](AztecProtocol@295cd55))
* Prototype native merkle trees
([AztecProtocol#4457](AztecProtocol#4457))
([7d5e056](AztecProtocol@7d5e056))
* Update rollup circuits and contracts in yp
([AztecProtocol#4536](AztecProtocol#4536))
([6e89d53](AztecProtocol@6e89d53))


### Bug Fixes

* **bb:** Publishing bb for mac intel
([AztecProtocol#4523](AztecProtocol#4523))
([4982e3c](AztecProtocol@4982e3c))
* Broken links in docs [REDO]
([AztecProtocol#4540](AztecProtocol#4540))
([ce2a205](AztecProtocol@ce2a205))
* **build-system:** Image expiring
([AztecProtocol#4521](AztecProtocol#4521))
([1501afd](AztecProtocol@1501afd))
* Convert folding recursive verifier ops to batch mul
([AztecProtocol#4517](AztecProtocol#4517))
([3750b26](AztecProtocol@3750b26))
* Cycle_group validate_is_on_curve bug
([AztecProtocol#4494](AztecProtocol#4494))
([fecf3f7](AztecProtocol@fecf3f7))
* Field divison / journal comparisions
([AztecProtocol#4489](AztecProtocol#4489))
([15c06c5](AztecProtocol@15c06c5))
* Master
([AztecProtocol#4547](AztecProtocol#4547))
([490ca26](AztecProtocol@490ca26))
* Mirror_noir_subrepo.yml
([AztecProtocol#4550](AztecProtocol#4550))
([f8d8311](AztecProtocol@f8d8311))
* Mul with endomorphism
([AztecProtocol#4538](AztecProtocol#4538))
([1f4c90d](AztecProtocol@1f4c90d))
* **noir-mirror:** Don't update .gitrepo on push
([AztecProtocol#4555](AztecProtocol#4555))
([686140a](AztecProtocol@686140a))
* Recreate jest.config.ts for sequencer-client
([AztecProtocol#4553](AztecProtocol#4553))
([d172f0b](AztecProtocol@d172f0b))
* StandardCircuitBuilder create_logic_constraint and uint logic_operator
([AztecProtocol#4530](AztecProtocol#4530))
([ce51d20](AztecProtocol@ce51d20))
* Use ordered-binary value encoding for multi maps
([AztecProtocol#4565](AztecProtocol#4565))
([04ae0d2](AztecProtocol@04ae0d2))


### Miscellaneous

* Aligning some naming in `BaseOrMergeRollupPublicInputs`
([AztecProtocol#4510](AztecProtocol#4510))
([47d66f9](AztecProtocol@47d66f9))
* **avm-circuit:** Tests use OpCode enum's instead of hardcoded values
([AztecProtocol#4554](AztecProtocol#4554))
([ca4dd60](AztecProtocol@ca4dd60))
* **avm-simulator:** Reduce boilerplate in AVM memory types
([AztecProtocol#4542](AztecProtocol#4542))
([da2f5ed](AztecProtocol@da2f5ed))
* **avm:** Add/improve tests for AvmContext, tagged memory, etc
([AztecProtocol#4484](AztecProtocol#4484))
([2fccdf2](AztecProtocol@2fccdf2))
* **avm:** Remove field support for comparators and bitwise ops
([AztecProtocol#4516](AztecProtocol#4516))
([87a9663](AztecProtocol@87a9663))
* **avm:** Use some matchers gtest functionalities to improve unit tests
([AztecProtocol#4502](AztecProtocol#4502))
([bf4fc6c](AztecProtocol@bf4fc6c)),
closes
[AztecProtocol#4495](AztecProtocol#4495)
* Cleanup of `abi.nr` in `aztec-nr`
([AztecProtocol#4473](AztecProtocol#4473))
([6d9c73a](AztecProtocol@6d9c73a))
* Cleanup of `abi.nr` in `aztec-nr`
(AztecProtocol#4473) [skip ci]
([db803bd](AztecProtocol@db803bd))
* Create constraints for sha256 compression opcode
([AztecProtocol#4503](AztecProtocol#4503))
([64bef49](AztecProtocol@64bef49))
* Little cpp style improvements
([AztecProtocol#4528](AztecProtocol#4528))
([dcc9ba4](AztecProtocol@dcc9ba4))
* Move noir out of yarn-project
([AztecProtocol#4479](AztecProtocol#4479))
([1fe674b](AztecProtocol@1fe674b)),
closes
[AztecProtocol#4107](AztecProtocol#4107)
* Pull noir
([AztecProtocol#4546](AztecProtocol#4546))
([acf5cf2](AztecProtocol@acf5cf2))
* Pull noir (AztecProtocol#4546)
([db803bd](AztecProtocol@db803bd))
* Redo noir subrepo force push
([AztecProtocol#4514](AztecProtocol#4514))
([7b519a4](AztecProtocol@7b519a4))
* Remove .oldValue in contract state update request
([AztecProtocol#4499](AztecProtocol#4499))
([a796bef](AztecProtocol@a796bef))
* Removing redundant utilities
([AztecProtocol#4532](AztecProtocol#4532))
([79bf445](AztecProtocol@79bf445)),
closes
[AztecProtocol#3470](AztecProtocol#3470)
* Rename kernel circuits and disambiguate inputs
([AztecProtocol#4535](AztecProtocol#4535))
([ed6d521](AztecProtocol@ed6d521))
* Replace relative paths to noir-protocol-circuits
([f1accbf](AztecProtocol@f1accbf))
* Replace relative paths to noir-protocol-circuits
([20ee430](AztecProtocol@20ee430))
* Replace relative paths to noir-protocol-circuits
([101ab59](AztecProtocol@101ab59))
* Simulator utils cleanup
([AztecProtocol#4507](AztecProtocol#4507))
([1dd0ebf](AztecProtocol@1dd0ebf))
* Sync to noir-lang/noir
([db803bd](AztecProtocol@db803bd))
* Testing all values in `PublicGlobalVariables` and
`PrivateGlobalVariables`
([AztecProtocol#4481](AztecProtocol#4481))
([bc25f9b](AztecProtocol@bc25f9b))
* Testing historical header in contexts
([AztecProtocol#4509](AztecProtocol#4509))
([c00229a](AztecProtocol@c00229a))
* Updating field conversion code without pointer hack
([AztecProtocol#4537](AztecProtocol#4537))
([94f436e](AztecProtocol@94f436e))
* Uses sha256compression opcode in Noir and implements acvm solver for
it
([AztecProtocol#4511](AztecProtocol#4511))
([9dc05bc](AztecProtocol@9dc05bc))


### Documentation

* Describe the new message box model
([AztecProtocol#4485](AztecProtocol#4485))
([14cc1dd](AztecProtocol@14cc1dd))
* Review of docs, so far
([AztecProtocol#4505](AztecProtocol#4505))
([140c508](AztecProtocol@140c508))
* **yellowpaper:** Avm tree-access operations
([AztecProtocol#4552](AztecProtocol#4552))
([913f4bd](AztecProtocol@913f4bd))
* **yellowpaper:** Separate section for AVM state
([AztecProtocol#4440](AztecProtocol#4440))
([7881f09](AztecProtocol@7881f09))
</details>

<details><summary>barretenberg.js: 0.24.0</summary>

##
[0.24.0](AztecProtocol/aztec-packages@barretenberg.js-v0.23.0...barretenberg.js-v0.24.0)
(2024-02-13)


### Features

* Prototype native merkle trees
([AztecProtocol#4457](AztecProtocol#4457))
([7d5e056](AztecProtocol@7d5e056))
</details>

<details><summary>barretenberg: 0.24.0</summary>

##
[0.24.0](AztecProtocol/aztec-packages@barretenberg-v0.23.0...barretenberg-v0.24.0)
(2024-02-13)


### Features

* Add hashing to stdlib transcript
([AztecProtocol#4161](AztecProtocol#4161))
([e78b86f](AztecProtocol@e78b86f))
* Added cast opcode and cast calldata
([AztecProtocol#4423](AztecProtocol#4423))
([e58eda8](AztecProtocol@e58eda8))
* Enable gmock and upgrade gtest to 1.13
([AztecProtocol#4480](AztecProtocol#4480))
([5fc02e7](AztecProtocol@5fc02e7))
* IVC bench
([AztecProtocol#4515](AztecProtocol#4515))
([d8ae42b](AztecProtocol@d8ae42b))
* Op count timers
([AztecProtocol#4471](AztecProtocol#4471))
([26918de](AztecProtocol@26918de))
* PG + Goblin
([AztecProtocol#4399](AztecProtocol#4399))
([295cd55](AztecProtocol@295cd55))
* Prototype native merkle trees
([AztecProtocol#4457](AztecProtocol#4457))
([7d5e056](AztecProtocol@7d5e056))


### Bug Fixes

* Convert folding recursive verifier ops to batch mul
([AztecProtocol#4517](AztecProtocol#4517))
([3750b26](AztecProtocol@3750b26))
* Cycle_group validate_is_on_curve bug
([AztecProtocol#4494](AztecProtocol#4494))
([fecf3f7](AztecProtocol@fecf3f7))
* Mul with endomorphism
([AztecProtocol#4538](AztecProtocol#4538))
([1f4c90d](AztecProtocol@1f4c90d))
* StandardCircuitBuilder create_logic_constraint and uint logic_operator
([AztecProtocol#4530](AztecProtocol#4530))
([ce51d20](AztecProtocol@ce51d20))


### Miscellaneous

* **avm-circuit:** Tests use OpCode enum's instead of hardcoded values
([AztecProtocol#4554](AztecProtocol#4554))
([ca4dd60](AztecProtocol@ca4dd60))
* **avm:** Use some matchers gtest functionalities to improve unit tests
([AztecProtocol#4502](AztecProtocol#4502))
([bf4fc6c](AztecProtocol@bf4fc6c)),
closes
[AztecProtocol#4495](AztecProtocol#4495)
* Create constraints for sha256 compression opcode
([AztecProtocol#4503](AztecProtocol#4503))
([64bef49](AztecProtocol@64bef49))
* Little cpp style improvements
([AztecProtocol#4528](AztecProtocol#4528))
([dcc9ba4](AztecProtocol@dcc9ba4))
* Updating field conversion code without pointer hack
([AztecProtocol#4537](AztecProtocol#4537))
([94f436e](AztecProtocol@94f436e))


### Documentation

* **yellowpaper:** Avm tree-access operations
([AztecProtocol#4552](AztecProtocol#4552))
([913f4bd](AztecProtocol@913f4bd))
</details>

---
This PR was generated with [Release
Please](https://github.com/googleapis/release-please). See
[documentation](https://github.com/googleapis/release-please#release-please).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
Archived in project
Development

Successfully merging this pull request may close these issues.

3 participants