phase-check related issues #780
Conversation
|
The false alarm can be avoided by scroll-tech/halo2#61. |
lispc
changed the title
|
|
This is a very interesting bug and worth some documentation. in halo2-lib there is a function This function is not deterministic. For multiple phases, this function will return different points. When multi-phase is enabled, the cells will be filled in with different points during different phases. And therefore during the phase check we get errors like |
| fn decompose_ec_add_op_phase_2( | ||
| &self, | ||
| ctx: &mut Context<F>, | ||
| ecc_chip: &EccChip<F, FpConfig<F, Fq>>, | ||
| ec_decomposed: &[EcAddDecomposed<F>], | ||
| challenges: &Challenges<Value<F>>, | ||
| ) { | ||
| log::trace!("[ECC] ==> EcAdd Assignment START:"); | ||
| log_context_cursor!(ctx); |
There was a problem hiding this comment.
If you move the ec addition execution to phase 2, then the above configuration of phase1 & phase2 cols [33, 1] is not right as it needs much more phase cols now.
| let mut buf = F::default(); | ||
| challenges.keccak_input().map(|x| buf = x); | ||
| let mut rng = ChaCha20Rng::from_seed(buf.to_repr()); | ||
| let p = G1Affine::random(&mut rng); | ||
| let G1Affine { x, y } = p; | ||
| let point = ecc_chip.load_private(ctx, (Value::known(x), Value::known(y))); |
There was a problem hiding this comment.
When the mock prover is doing 1st synthesis, the rng is seeded with 0. But for 2nd synthesis, it is then seeded with keccak_input().
cargo test --release ecc_circuit 2>&1 | tee ecc.log