Conversation
| #[derive(Debug, Clone, Copy, Default, PartialEq, Eq, CanonicalDeserialize, CanonicalSerialize)] | ||
| pub struct PoseidonDigest(pub Fp); | ||
|
|
||
| impl HashMarker for PoseidonDigest {} |
There was a problem hiding this comment.
I think this can be removed
| input1: U256::from(0), | ||
| input2: U256::from(0), | ||
| }; | ||
| let result = InstanceTrace::new(instance); |
There was a problem hiding this comment.
Using InstanceTrace will decrease performance since it is not optimised for speed but rather obtaining the values needed in the STARK trace.
I think this PR should implement a seperate poseidon hash function that's focussed more on performance (i.e. doesn't keep track of the state of rounds in the permutation). Let's move all the added methods and tests in this file to crypto/src/hash/poseidon.rs. I'd suggest using the parameters from builtins/src/poseidon/params.rs and using InstanceTrace::new as a guide. Let's put the implementation in a new function called poseidon_permutation([Fp; 3]).
| index: 0, | ||
| input0: U256::from_limbs(x.0 .0), | ||
| input1: U256::from(0), | ||
| input2: U256::from(0), |
There was a problem hiding this comment.
Might need to set input2 to 1
| index: 0, | ||
| input0: U256::from_limbs(x.0 .0), | ||
| input1: U256::from_limbs(y.0 .0), | ||
| input2: U256::from(0), |
There was a problem hiding this comment.
| } | ||
|
|
||
| fn merge(v0: &PoseidonDigest, v1: &PoseidonDigest) -> PoseidonDigest { | ||
| PoseidonDigest(poseidon_hash_many([**v0, **v1].to_vec())) |
There was a problem hiding this comment.
Due to how frequently this function is called it has a big impact on the overall performance of the prover. Allocations on the heap every invocation makes this function much more expensive than it needs to be. Let's implement this without any Vec
| fn poseidon_hash_many_4_example() { | ||
| let elements = vec![Fp!("0"), Fp!("0"), Fp!("0"), Fp!("0")]; | ||
|
|
||
| let expected = BigInt::new([ |
There was a problem hiding this comment.
How are all the expected values obtained?
| } | ||
|
|
||
| #[test] | ||
| fn poseidon_hash_many_3_example() { |
There was a problem hiding this comment.
test name can just be "hash_3_values". Let's update the other test names as well. Also we should have tests for hash_single and poseidon_hash
| fn merge_with_int(seed: &PoseidonDigest, value: u64) -> PoseidonDigest { | ||
| PoseidonDigest(poseidon_hash_many([**seed, value.into()].to_vec())) | ||
| } |
There was a problem hiding this comment.
| fn merge_with_int(seed: &PoseidonDigest, value: u64) -> PoseidonDigest { | |
| PoseidonDigest(poseidon_hash_many([**seed, value.into()].to_vec())) | |
| } | |
| fn merge_with_int(seed: &PoseidonDigest, value: u64) -> PoseidonDigest { | |
| self.merge(seed.0, value.into()) | |
| } |
| fn hash_elements(elements: impl IntoIterator<Item = Fp>) -> PoseidonDigest { | ||
| PoseidonDigest(poseidon_hash_many(elements.into_iter().collect())) | ||
| } |
There was a problem hiding this comment.
Let's make this function behave the same way as poseidon_hash_many_given_poseidon_perm and use the same padding rule
2 participants
No description provided.