feat: Add compiled merkle proof with C verifier #27
Conversation
Similar to sparse-merkle-tree, this commit adds a new stack based compiled merkle proof format. This new format can be used to build highly performant C verifier on CKB-VM. An optimized C implementation of the verifier is also included in this commit, including tests.
| } else { | ||
| (lhs_pos, lhs_height, rhs_pos, rhs_t, lhs, rhs) | ||
| }; | ||
| // calculate sibling |
There was a problem hiding this comment.
I think the height var in stack is only used in Command::Hash, can we remove it from the stack and calculate it via pos here?
| // calculate sibling | |
| // calculate sibling | |
| let height = pos_height_in_tree(pos); |
and a step further, the pos is only used with StackItemType::Node, we may change the enum variant to StackItemType::Node(u64) and stack to <(T, StackItemType)
There was a problem hiding this comment.
I tried this, but noticing that recalculating pos_height_in_tree would have a noticeable impact on verifier performance (~2% in cycle consumptions of Rust code), so I would rather suggest keeping the code as it is. Counting heights only by incrementing as in current code, can be faster.
There was a problem hiding this comment.
I tried this, but noticing that recalculating
pos_height_in_treewould have a noticeable impact on verifier performance (~2% in cycle consumptions of Rust code), so I would rather suggest keeping the code as it is. Counting heights only by incrementing as in current code, can be faster.
I see, can you help to try the cycle difference between StackItemType::Node(u64, u8) // pos, height | Peak(u8) and original solution?
There was a problem hiding this comment.
This additional commit b5b4602 which refactors as you mentioned, takes an additional 2000 cycles out of a total of 1583330 cycles, which is quite neglegible, hence I pull the trigger and include the refactored code.
| } | ||
| let (rhs, rhs_t) = stack.pop().unwrap(); | ||
| let (lhs, lhs_t) = stack.pop().unwrap(); | ||
| let (pos, height, next_pos, next_t, item, sibling_item) = match (lhs_t, &rhs_t) { |
There was a problem hiding this comment.
Just an FYI: I have tried the path of introducing a pos function to StackItemType enum, so we don't need next_pos variable here. But surprisingly, a common pos function here actually will make performance worse(roughly 30000 more cycles on the overall 1583330 cycles), hence I'm keeping this next_pos variable.
Similar to sparse-merkle-tree, this commit adds a new stack based compiled merkle proof format. This new format can be used to build highly performant C verifier on CKB-VM. An optimized C implementation of the verifier is also included in this commit, including tests.
Based on a random generated MMR of 1116 leafs, a proof containing 71 leafs is generated, the following results are then obtained: