integrate triple_xor gate

[Stavbe]

No description provided.

[reviewable-StarkWare]

This change is 

[Stavbe]

• add blake in circuit + tests #462
• integrate blake g gate #461
• integrate triple_xor gate #452 👈 (View in Graphite)
• integrate m31_to_u32 gate in stwo-circuit #437
• integrate verify triple sum #432
• auto-gen files #425
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[ilyalesokhin-starkware]

_typos.toml line 8 at r3 (raw file):

[default.extend-words]
consts = "consts"
ba = "ba"

?

Code quote:

ba = "ba"

[az-starkware]

@az-starkware reviewed 22 files and all commit messages, and made 6 comments.
Reviewable status: all files reviewed, 6 unresolved discussions (waiting on alon-f, ilyalesokhin-starkware, and Stavbe).

---

crates/circuits/src/ivalue.rs line 55 at r3 (raw file):

    /// Packs a `u32` value into a QM31 limb representation `(low_u16, high_u16, 0, 0)`.
    fn pack_u32(value: u32) -> Self;

Not sure if "pack" is the best word here. Pack usually means we take a few things and put them back-to-back in some container, while here we take one thing and split it to fit the constraints of the container (that each limb can only store 31 bits).

Maybe "serialize" and "deserialize"?

Non-blocking.

---

crates/circuits/src/ivalue.rs line 100 at r3 (raw file):

    fn unpack_u32(&self) -> u32 {
        let [low, high, _, _] = self.to_m31_array().map(|m| m.0);

Assert that the two "unused" felts are zeros. If it hurts performace, you can only assert when running in debug mode.

---

crates/circuit_common/src/preprocessed.rs line 142 at r3 (raw file):

        mul,
        pointwise_mul,
        eq: _,

Replace ignored fields with ..

---

crates/circuit_common/src/preprocessed.rs line 330 at r3 (raw file):

        pointwise_mul: _,
        eq: _,
        blake,

Here too

---

crates/circuit_common/src/preprocessed.rs line 190 at r3 (raw file):

        mul: _,
        pointwise_mul: _,
        eq,

Here too

---

crates/circuit_common/src/preprocessed.rs line 371 at r3 (raw file):

/// Adds TripleXor gates to preprocessed trace columns.
/// | input_a_address | input_b_address | input_c_address | output_address | multiplicity |
fn fill_triple_xor_columns(

Don't use numeric column indices (e.g. columns[3]). One possible solution is to change this function to create the columns, and return a dictionary of id -> values. This way won't need the list of column names in add_triple_xor_to_preprocessed_trace.
Other solutions are also OK as long as we don't have a list of string IDs and a list of numeric IDs that have to be kept in sync.

[Stavbe]

@Stavbe made 7 comments.
Reviewable status: 16 of 22 files reviewed, 6 unresolved discussions (waiting on alon-f, az-starkware, and ilyalesokhin-starkware).

---

_typos.toml line 8 at r3 (raw file):

Previously, ilyalesokhin-starkware wrote…

?

Typos fail on the function name (which is auto-gen hash) in: crates/circuit_air/src/circuit_eval_components/triple_xor.rs

---

crates/circuit_common/src/preprocessed.rs line 142 at r3 (raw file):

Previously, az-starkware wrote…

Replace ignored fields with ..

Done.

---

crates/circuit_common/src/preprocessed.rs line 190 at r3 (raw file):

Previously, az-starkware wrote…

Here too

Done.

---

crates/circuit_common/src/preprocessed.rs line 330 at r3 (raw file):

Previously, az-starkware wrote…

Here too

Done.

---

crates/circuit_common/src/preprocessed.rs line 371 at r3 (raw file):

Previously, az-starkware wrote…

Don't use numeric column indices (e.g. columns[3]). One possible solution is to change this function to create the columns, and return a dictionary of id -> values. This way won't need the list of column names in add_triple_xor_to_preprocessed_trace.
Other solutions are also OK as long as we don't have a list of string IDs and a list of numeric IDs that have to be kept in sync.

Since literally all the other gates do this, I think this should be a TODO, and one should update all of them together once we decide on the chosen solution.

---

crates/circuits/src/ivalue.rs line 55 at r3 (raw file):

Previously, az-starkware wrote…

Not sure if "pack" is the best word here. Pack usually means we take a few things and put them back-to-back in some container, while here we take one thing and split it to fit the constraints of the container (that each limb can only store 31 bits).

Maybe "serialize" and "deserialize"?

Non-blocking.

I prefer 'pack' and this is the term I used in the DR.
@ilyalesokhin-starkware WDYT?

---

crates/circuits/src/ivalue.rs line 100 at r3 (raw file):

Previously, az-starkware wrote…

Assert that the two "unused" felts are zeros. If it hurts performace, you can only assert when running in debug mode.

Done.

[ilyalesokhin-starkware]

crates/circuits/src/ivalue.rs line 55 at r3 (raw file):

Previously, Stavbe wrote…

I prefer 'pack' and this is the term I used in the DR.
@ilyalesokhin-starkware WDYT?

maybe from_u32:
"Takes a u32 and encodes it in an IVALUE, for QM31 the representation is `(low_u16, high_u16, 0, 0)."

[ilyalesokhin-starkware]

crates/circuit_common/src/preprocessed.rs line 371 at r3 (raw file):

Previously, Stavbe wrote…

Since literally all the other gates do this, I think this should be a TODO, and one should update all of them together once we decide on the chosen solution.

We just want to write the first 5 columns here, and unfortunately, Rust doesn't have a nice syntax for that.