Skip to content

Commit

Permalink
perf: Combine constraints for stack push / pop
Browse files Browse the repository at this point in the history
  • Loading branch information
@jan-ferdinand
jan-ferdinand committed Jun 10, 2024
1 parent d74e10a commit 70361ff
Show file tree
Hide file tree
Showing 2 changed files with 52 additions and 39 deletions.
10 changes: 5 additions & 5 deletions specification/src/arithmetization-overview.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,9 +15,9 @@
| [CascadeTable](cascade-table.md) | 6 | 2 | 12 |
| [LookupTable](lookup-table.md) | 4 | 2 | 10 |
| [U32Table](u32-table.md) | 10 | 1 | 13 |
| DegreeLowering | 212 | 36 | 320 |
| DegreeLowering | 200 | 36 | 308 |
| Randomizers | 0 | 1 | 3 |
| **TOTAL** | **361** | **86** | **619** |
| **TOTAL** | **349** | **86** | **607** |
<!-- auto-gen info stop table_overview -->

## Constraints
Expand Down Expand Up @@ -50,7 +50,7 @@ After automatically lowering degree to 4:
| table name | #initial | #consistency | #transition | #terminal |
|:-----------------------------------------------|---------:|-------------:|------------:|----------:|
| [ProgramTable](program-table.md) | 6 | 4 | 10 | 2 |
| [ProcessorTable](processor-table.md) | 31 | 10 | 219 | 1 |
| [ProcessorTable](processor-table.md) | 31 | 10 | 207 | 1 |
| [OpStackTable](operational-stack-table.md) | 3 | 0 | 5 | 0 |
| [RamTable](random-access-memory-table.md) | 7 | 0 | 13 | 1 |
| [JumpStackTable](jump-stack-table.md) | 6 | 0 | 6 | 0 |
Expand All @@ -59,7 +59,7 @@ After automatically lowering degree to 4:
| [LookupTable](lookup-table.md) | 3 | 1 | 4 | 1 |
| [U32Table](u32-table.md) | 1 | 26 | 34 | 2 |
| [Grand Cross-Table Argument](table-linking.md) | 0 | 0 | 0 | 14 |
| **TOTAL** | **81** | **94** | **378** | **23** |
| **TOTAL** | **81** | **94** | **366** | **23** |
<!-- auto-gen info stop constraints_overview -->


Expand All @@ -71,5 +71,5 @@ In order to gauge the runtime cost for this step, the following table provides e
<!-- auto-gen info start tasm_air_evaluation_cost -->
| Processor | Op Stack | RAM |
|----------:|---------:|------:|
| 35207 | 65377 | 23220 |
| 35891 | 66815 | 23667 |
<!-- auto-gen info stop tasm_air_evaluation_cost -->
81 changes: 47 additions & 34 deletions triton-vm/src/table/processor_table.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1511,16 +1511,18 @@ impl ExtProcessorTable {
}

/// Compute the randomly-weighted linear combination of the supplied stack
/// elements.
/// elements using the first `stack.len()` [challenges] as weights.
///
/// # Panics
///
/// Panics if the length of the supplied `stack` vector does not equal
/// [`OpStackElement::COUNT`].
/// Panics if the supplied stack is larger than [`OpStackElement::COUNT`].
///
/// [challenges]: StackWeight0
fn compress_stack(
circuit_builder: &ConstraintCircuitBuilder<DualRowIndicator>,
stack: Vec<ConstraintCircuitMonad<DualRowIndicator>>,
) -> ConstraintCircuitMonad<DualRowIndicator> {
assert!(stack.len() <= OpStackElement::COUNT);
let challenges = [
StackWeight0,
StackWeight1,
Expand All @@ -1543,7 +1545,7 @@ impl ExtProcessorTable {

challenges
.into_iter()
.zip_eq(stack)
.zip(stack)
.map(|(weight, st)| weight * st)
.sum()
}
Expand Down Expand Up @@ -1595,11 +1597,14 @@ impl ExtProcessorTable {

let next_stack = stack.iter().map(|&st| next_row(st)).collect_vec();
let curr_stack_with_swapped_i = |i| stack_with_swapped_i(i).map(curr_row).collect_vec();
let compress = |stack: Vec<_>| {
assert_eq!(OpStackElement::COUNT, stack.len());
Self::compress_stack(circuit_builder, stack)
};

let next_stack_is_current_stack_with_swapped_i = |i| {
Self::indicator_polynomial(circuit_builder, i)
* (Self::compress_stack(circuit_builder, next_stack.clone())
- Self::compress_stack(circuit_builder, curr_stack_with_swapped_i(i)))
* (compress(next_stack.clone()) - compress(curr_stack_with_swapped_i(i)))
};
let next_stack_is_current_stack_with_correct_element_swapped = (0..OpStackElement::COUNT)
.map(next_stack_is_current_stack_with_swapped_i)
Expand Down Expand Up @@ -2999,59 +3004,67 @@ impl ExtProcessorTable {
n: usize,
) -> Vec<ConstraintCircuitMonad<DualRowIndicator>> {
let constant = |c: usize| circuit_builder.b_constant(u64::try_from(c).unwrap());
let curr_base_row = |col: ProcessorBaseTableColumn| {
let curr_row = |col: ProcessorBaseTableColumn| {
circuit_builder.input(CurrentBaseRow(col.master_base_table_index()))
};
let next_base_row = |col: ProcessorBaseTableColumn| {
let next_row = |col: ProcessorBaseTableColumn| {
circuit_builder.input(NextBaseRow(col.master_base_table_index()))
};

let stack = || (0..OpStackElement::COUNT).map(ProcessorTable::op_stack_column_by_index);
let new_stack = stack().dropping_back(n).map(next_row).collect_vec();
let old_stack_with_top_n_removed = stack().skip(n).map(curr_row).collect_vec();

let compress = |stack: Vec<_>| {
assert_eq!(OpStackElement::COUNT - n, stack.len());
Self::compress_stack(circuit_builder, stack)
};
let compressed_new_stack = compress(new_stack);
let compressed_old_stack = compress(old_stack_with_top_n_removed);

let op_stack_pointer_shrinks_by_n =
next_base_row(OpStackPointer) - curr_base_row(OpStackPointer) + constant(n);
next_row(OpStackPointer) - curr_row(OpStackPointer) + constant(n);
let new_stack_is_old_stack_with_top_n_removed = compressed_new_stack - compressed_old_stack;

let mut constraints = vec![
vec![
op_stack_pointer_shrinks_by_n,
new_stack_is_old_stack_with_top_n_removed,
Self::running_product_op_stack_accounts_for_shrinking_stack_by(circuit_builder, n),
];

for i in n..OpStackElement::COUNT {
let curr_stack_element = ProcessorTable::op_stack_column_by_index(i);
let next_stack_element = ProcessorTable::op_stack_column_by_index(i - n);
let element_i_is_shifted_by_n =
next_base_row(next_stack_element) - curr_base_row(curr_stack_element);
constraints.push(element_i_is_shifted_by_n);
}
constraints
]
}

fn constraints_for_growing_stack_by(
circuit_builder: &ConstraintCircuitBuilder<DualRowIndicator>,
n: usize,
) -> Vec<ConstraintCircuitMonad<DualRowIndicator>> {
let constant = |c: usize| circuit_builder.b_constant(u32::try_from(c).unwrap());
let curr_base_row = |col: ProcessorBaseTableColumn| {
let curr_row = |col: ProcessorBaseTableColumn| {
circuit_builder.input(CurrentBaseRow(col.master_base_table_index()))
};
let next_base_row = |col: ProcessorBaseTableColumn| {
let next_row = |col: ProcessorBaseTableColumn| {
circuit_builder.input(NextBaseRow(col.master_base_table_index()))
};

let stack = || (0..OpStackElement::COUNT).map(ProcessorTable::op_stack_column_by_index);
let new_stack = stack().skip(n).map(next_row).collect_vec();
let old_stack_with_top_n_added = stack().map(curr_row).dropping_back(n).collect_vec();

let compress = |stack: Vec<_>| {
assert_eq!(OpStackElement::COUNT - n, stack.len());
Self::compress_stack(circuit_builder, stack)
};
let compressed_new_stack = compress(new_stack);
let compressed_old_stack = compress(old_stack_with_top_n_added);

let op_stack_pointer_grows_by_n =
next_base_row(OpStackPointer) - curr_base_row(OpStackPointer) - constant(n);
next_row(OpStackPointer) - curr_row(OpStackPointer) - constant(n);
let new_stack_is_old_stack_with_top_n_added = compressed_new_stack - compressed_old_stack;

let mut constraints = vec![
vec![
op_stack_pointer_grows_by_n,
new_stack_is_old_stack_with_top_n_added,
Self::running_product_op_stack_accounts_for_growing_stack_by(circuit_builder, n),
];

for i in 0..OpStackElement::COUNT - n {
let curr_stack_element = ProcessorTable::op_stack_column_by_index(i);
let next_stack_element = ProcessorTable::op_stack_column_by_index(i + n);
let element_i_is_shifted_by_n =
next_base_row(next_stack_element) - curr_base_row(curr_stack_element);
constraints.push(element_i_is_shifted_by_n);
}
constraints
]
}

fn conditional_constraints_for_shrinking_stack_by(
Expand Down

0 comments on commit 70361ff

Please sign in to comment.