Skip to content

Latest commit

 

History

History
129 lines (119 loc) · 5.23 KB

README.md

File metadata and controls

129 lines (119 loc) · 5.23 KB

circuit-tools

A lightweit SDK for Halo2 frontend circuit.

Features

Logic: Constraint Builder

Leverage Rust's macro to construct conditional constraints based on execution branching. Build all constraints in on step with Halo2's gate.

let mut cb: ConstraintBuilder<F, TestCellType> =  ConstraintBuilder::new(4,  None, None);

meta.create_gate("Test", |meta| {
    circuit!([meta, cb], {
        ifx!(f!(q_enable) => {
            ifx!{a!(a) => {
                require!(a!(res) => a!(b) + f!(c)); 
            } elsex {
                require!(a!(res) => a!(b) + c!(r));  
            }};
        });
    });
    cb.build_constraints()
});

Data: Cell Manager

Manage and allocate cells on demand based on user specified CellType. Decrease amount of columns space wasted, enable degree reduction and lookup automation.

let mut cm = CellManager::new(5, 0);
cm.add_columns(meta, &mut cb, TestCellType::StoragePhase1, 1, false, 1);
cm.add_columns(meta, &mut cb, TestCellType::Lookup, 2, false, 1);

// Allocation
let a = cb.query_default();
let b = cb.query_default();
let c = cb.query_cell_with_type(TestCellType::StoragePhase2);

// Computation
require!(c.expr() => a.expr() + b.expr() * challenge.expr());
// Lookup
require!((a.expr(), b.expr()) =>> @TestCellType::Lookup);

Memory: Dynamic Lookup

Abstraction over using dynamic lookup to prove RW access in RAM. Stores to dynamic table to resemble WRITE and add lookups to represent READ. Lookup operation (idx, r0, r1) => (rdx, w0, w1) returns true if and only if the prover access the writen data at anticipated position correctly.

let mut memory = Memory::new();
memory.add_rw(meta, &mut cb.base, &mut state_cm, MyCellType::Mem1, 1);
memory.add_rw(meta, &mut cb.base, &mut state_cm, MyCellType::Mem2, 1);

let register1 = memory[MyCellType::Mem1];
let register2 = memory[MyCellType::Mem2];

register1.store(cb, &[a0, b0]);
register2.store(cb, &[c0, d9, 123.expr()]);

// ... long time later ...
register2.load(cb, &[c1, d1, 123.expr()]);
register1.load(cb, &[a0, b1]);

Assignment: Cached Region

Used to backtrack the intermediate cells queried in by Constrinat Builder during degree reduction. After a * b * (c + d) being split into x = a * b and y = c + d, the system need to account for the assignment of intermediate cells x and y while the prover only need to assign a, b, c, d based on the execution trace. Hence, Cached Region iterates over all stored expressions and recursively find the prover's assignment to calculate the intermediate values.

layouter.get_challenge(self.rand).map(|r| r1 = r);
layouter.assign_region(
    || "Test", 
    |mut region| {
        let mut region = CachedRegion::new(&mut region, 0.scalar());
        region.push_region(0, 0);

        let (a, b, c, d,  e) = &self.cells;
        assign!(&mut region, a, 0 => 1.scalar())?;
        assign!(&mut region, b, 0 => 2.scalar())?;
        assign!(&mut region, c, 0 => 3.scalar())?;
        assign!(&mut region, d, 0 => 4.scalar())?;
        region.assign_stored_expressions(&self.cb, &[r0])?;
        Ok(())
    }
)

Workflow

Must specify cell type that implement trait CellType to satisfy the generic argument of ConstraintBuilder<F, C: CellType>. Can also declear a TableType to tag corresponding table for column-to-table lookups.

pub enum TableTag {
    Fixed,
    Dyn
}
pub enum TestCellType {
    StoragePhase1,
    Lookup,
}
impl CellType for TestCellType{
    type TableType = TableTag;

    fn lookup_table_type(&self) -> Option<Self::TableType> {
        match self {
            TestCellType::Lookup => Some(TableTag::Fixed),
            _ => None,
        }
    }
    fn byte_type() -> Option<Self> {None}
    fn create_type(_id: usize) -> Self {unreachable!()}
    fn storage_for_phase(phase: u8) -> Self {
        match phase {
            1 => Self::StoragePhase1,
            _ => unreachable!()
        }
    }
}
impl Default for TestCellType {
    fn default() -> Self {Self::StoragePhase1}
}

Initialize the Constraint Builder and Cell Manager with optional challenge that's used in RLC to conbime multi-columns lookups. The Cell Manager needs a max height and this is usually the height of your Halo2 gate; a offset is also needed to query cell from columns. Offset should be set to 0 in usual case. Load fixed table in to Constraint Builder with corresponding tag and initialized columns with the cell manager with (cell_type: MyCellType, phase: u8, permuation: bool, num: usize).

let mut cm = CellManager::new(5, 0);
let mut cb: ConstraintBuilder<F, TestCellType> =  ConstraintBuilder::new(4,  Some(cm), Some(challenge));
cb.load_table(meta, TableTag::Fixed, &fixed_table);
cm.add_columns(meta, &mut cb, TestCellType::StoragePhase1, 1, true, 1);
cm.add_columns(meta, &mut cb, TestCellType::Lookup, 2, false, 1);

In Halo2's gate API, use macro to config your circuit! Remember to call build_constraints() to return the constraints expression for the gate, finally calling build_lookups(meta) that turned into `meta.lookup_any(..) in Halo2.

meta.create_gate("Test", |meta| {
    circuit!([meta, cb], {
        // ... circuit ...
    });
    cb.build_constraints()
});
cb.build_lookups(meta);

Finally in assingment, initialize CachedRegion as demonstrated above and use it to assign values.