noir.js cannot prove on input a circuit of size 2^19 #5106
Comments
codygunton
added a commit
to AztecProtocol/aztec-packages
that referenced
this issue
May 25, 2024
This PR introdues a stdlib boolean flag into biggroup to track whether
an element is the point at infinity. This is uses to handle edge cases
around the point at infinity in biggroup operations. We now correctly
handle points at infinity under addition and subtraction. The
`batch_mul` method correctly handles points at infinity (at least in
three tested cases) under the Mega arithmetization (though this is not
all that meaningful without a full Goblin proof!). The `wnaf_batch_mul`
method correctly handles points at infinity (at least in three tested
cases) under the Ultra arithmetization, which is the only
arithmetization for which it's implemented.
The PR adds constraints that increase the cost of biggroup operations.
This cases the UltraPlonk recursive verifier circuit size to grow,
crossing a power-of-two boundary. This means that we can no longer
execute two UltraPlonk recursive verifications in WASM due to an
out-of-memory error during provcing key creation. (cf the
`double_verify_proof` tests; note that `double_verify_nested_proof` was
already not available in WASM). Moreover, the PR exposed that noir.js is
not capable of executing proof construction for a circuit of size
$2^{19}$. In response to these two issues, we have disabled tests. We
did this in consulation with @TomAFrench and @vezenovm.
Related issues:
noir-lang/noir#5106,
#6672.
---------
Co-authored-by: codygunton <codygunton@gmail.com>
AztecBot
pushed a commit
to AztecProtocol/barretenberg
that referenced
this issue
May 26, 2024
This PR introdues a stdlib boolean flag into biggroup to track whether
an element is the point at infinity. This is uses to handle edge cases
around the point at infinity in biggroup operations. We now correctly
handle points at infinity under addition and subtraction. The
`batch_mul` method correctly handles points at infinity (at least in
three tested cases) under the Mega arithmetization (though this is not
all that meaningful without a full Goblin proof!). The `wnaf_batch_mul`
method correctly handles points at infinity (at least in three tested
cases) under the Ultra arithmetization, which is the only
arithmetization for which it's implemented.
The PR adds constraints that increase the cost of biggroup operations.
This cases the UltraPlonk recursive verifier circuit size to grow,
crossing a power-of-two boundary. This means that we can no longer
execute two UltraPlonk recursive verifications in WASM due to an
out-of-memory error during provcing key creation. (cf the
`double_verify_proof` tests; note that `double_verify_nested_proof` was
already not available in WASM). Moreover, the PR exposed that noir.js is
not capable of executing proof construction for a circuit of size
$2^{19}$. In response to these two issues, we have disabled tests. We
did this in consulation with @TomAFrench and @vezenovm.
Related issues:
noir-lang/noir#5106,
AztecProtocol/aztec-packages#6672.
---------
Co-authored-by: codygunton <codygunton@gmail.com>
|
Was this happening with NoirJS specifically? As in bb.js works, but not when going through NoirJS? |
|
Closing as this is a BB.js concern |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This was uncovered by AztecProtocol/aztec-packages#6391, which pushed the size of the UltraPlonk verifier circuit to size 2^19. The failure is that the prover runs out of memory while creating the proving key. It was subsequently verified that proving fails for other circuits of size 2^19 (e.g., the test program called hashmap) while it succeeds for a circuit of size 2^18 (the test program ecdsa_secp256r1_3x).
The text was updated successfully, but these errors were encountered: