Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

perf: dynamic dory commitment computation on the GPU should have parallel option #288

Merged
merged 6 commits into from
Oct 22, 2024
Merged

perf: dynamic dory commitment computation on the GPU should have parallel option #288

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
b8637fd
perf: dynamic dory commitment computation should be able to be done i…
jacobtrombetta Oct 21, 2024
65a246d
refactor: improve usage of if_rayon macro in dynamic_dory_commitment_…
jacobtrombetta Oct 21, 2024
cf406ed
refactor: add panic if the row index returned exceeds u32::MAX in the…
jacobtrombetta Oct 22, 2024
025c55a
fix: fix rebase conflict in dynamic_dory_commitment_helper_gpu and bl…
jacobtrombetta Oct 22, 2024
c8c9780
Merge branch 'main' into perf/parallel-dynamic-dory
jacobtrombetta Oct 22, 2024
5ab48f5
Merge branch 'main' into perf/parallel-dynamic-dory
jacobtrombetta Oct 22, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
118 changes: 65 additions & 53 deletions crates/proof-of-sql/src/proof_primitive/dory/blitzar_metadata_table.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ use super::{
G1Affine, F,
};
use crate::{
base::{commitment::CommittableColumn, database::ColumnType},
base::{commitment::CommittableColumn, database::ColumnType, if_rayon},
proof_primitive::dory::offset_to_bytes::OffsetToBytes,
};
use alloc::{vec, vec::Vec};
Expand All @@ -14,6 +14,11 @@ use ark_ff::MontFp;
use ark_std::ops::Mul;
use core::iter;
use itertools::Itertools;
#[cfg(feature = "rayon")]
use rayon::{
iter::{IndexedParallelIterator, IntoParallelIterator, ParallelIterator},
prelude::ParallelSliceMut,
};
use tracing::{span, Level};

const BYTE_SIZE: u32 = 8;
Expand Down Expand Up @@ -72,11 +77,16 @@ pub fn signed_commits(
}
}

unsigned_sub_commits
.into_iter()
.zip(min_sub_commits.into_iter())
.map(|(unsigned, min)| (unsigned + min).into())
.collect()
if_rayon!(
unsigned_sub_commits
.into_par_iter()
.zip(min_sub_commits.into_par_iter()),
unsigned_sub_commits
.into_iter()
.zip(min_sub_commits.into_iter())
)
.map(|(unsigned, min)| (unsigned + min).into())
.collect()
}

/// Copies the column data to the scalar row slice.
Expand Down Expand Up @@ -123,7 +133,6 @@ fn copy_column_data_to_slice(
}
}

#[allow(clippy::cast_possible_truncation)]
/// Creates the metadata tables for Blitzar's `vlen_msm` algorithm.
///
/// # Arguments
Expand All @@ -135,6 +144,10 @@ fn copy_column_data_to_slice(
///
/// A tuple containing the output bit table, output length table,
/// and scalars required to call Blitzar's `vlen_msm` function.
///
/// # Panics
///
/// Panics if the row of a column exceeds `u32::MAX`.
#[tracing::instrument(name = "create_blitzar_metadata_tables", level = "debug", skip_all)]
pub fn create_blitzar_metadata_tables(
committable_columns: &[CommittableColumn],
Expand Down Expand Up @@ -185,7 +198,8 @@ pub fn create_blitzar_metadata_tables(
/ single_entry_in_blitzar_output_bit_table.len())
.flat_map(|i| {
itertools::repeat_n(
full_width_of_row(i + offset_row) as u32,
u32::try_from(full_width_of_row(i + offset_row))
.expect("row lengths should never exceed u32::MAX"),
single_entry_in_blitzar_output_bit_table.len(),
)
})
Expand All @@ -208,54 +222,52 @@ pub fn create_blitzar_metadata_tables(
// Populate the scalars array.
let span = span!(Level::INFO, "pack_blitzar_scalars").entered();
if !blitzar_scalars.is_empty() {
blitzar_scalars
.chunks_exact_mut(num_scalar_columns)
.enumerate()
.for_each(|(scalar_row, scalar_row_slice)| {
// Iterate over the columns and populate the scalars array.
for scalar_col in 0..offset_height {
// Find index in the committable columns. Note, the scalar is in
// column major order, that is why the (row, col) arguments are flipped.
if let Some(index) =
index_from_row_and_column(scalar_col + offset_row, scalar_row).and_then(
|committable_column_idx| committable_column_idx.checked_sub(offset),
)
if_rayon!(
blitzar_scalars.par_chunks_exact_mut(num_scalar_columns),
blitzar_scalars.chunks_exact_mut(num_scalar_columns)
)
.enumerate()
.for_each(|(scalar_row, scalar_row_slice)| {
for scalar_col in 0..offset_height {
// Find index in the committable columns. Note, the scalar is in
// column major order, that is why the (row, col) arguments are flipped.
if let Some(index) = index_from_row_and_column(scalar_col + offset_row, scalar_row)
.and_then(|committable_column_idx| committable_column_idx.checked_sub(offset))
{
for (i, committable_column) in committable_columns
.iter()
.enumerate()
.filter(|(_, committable_column)| index < committable_column.len())
{
for (i, committable_column) in committable_columns
.iter()
.enumerate()
.filter(|(_, committable_column)| index < committable_column.len())
{
let start = cumulative_byte_length_table
[i + scalar_col * single_entry_in_blitzar_output_bit_table.len()];
let end = start
+ (single_entry_in_blitzar_output_bit_table[i] / BYTE_SIZE)
as usize;

copy_column_data_to_slice(
committable_column,
scalar_row_slice,
start,
end,
index,
);
}

ones_columns_lengths
.iter()
.positions(|ones_columns_length| index < *ones_columns_length)
.for_each(|i| {
let ones_index = i
+ scalar_col
* (num_of_bytes_in_committable_columns
+ ones_columns_lengths.len())
+ num_of_bytes_in_committable_columns;

scalar_row_slice[ones_index] = 1_u8;
});
let start = cumulative_byte_length_table
[i + scalar_col * single_entry_in_blitzar_output_bit_table.len()];
let end = start
+ (single_entry_in_blitzar_output_bit_table[i] / BYTE_SIZE) as usize;

copy_column_data_to_slice(
committable_column,
scalar_row_slice,
start,
end,
index,
);
}

ones_columns_lengths
.iter()
.positions(|ones_columns_length| index < *ones_columns_length)
.for_each(|i| {
let ones_index = i
+ scalar_col
* (num_of_bytes_in_committable_columns
+ ones_columns_lengths.len())
+ num_of_bytes_in_committable_columns;

scalar_row_slice[ones_index] = 1_u8;
});
}
});
}
});
}
span.exit();

Expand Down
31 changes: 18 additions & 13 deletions crates/proof-of-sql/src/proof_primitive/dory/dynamic_dory_commitment_helper_gpu.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,8 +3,10 @@ use super::{
dynamic_dory_structure::row_and_column_from_index,
pairings, DynamicDoryCommitment, G1Affine, ProverSetup,
};
use crate::base::{commitment::CommittableColumn, slice_ops::slice_cast};
use crate::base::{commitment::CommittableColumn, if_rayon, slice_ops::slice_cast};
use blitzar::compute::ElementP2;
#[cfg(feature = "rayon")]
use rayon::iter::{IntoParallelIterator, ParallelIterator};
use tracing::{span, Level};

/// Computes the dynamic Dory commitment using the GPU implementation of the `vlen_msm` algorithm.
Expand Down Expand Up @@ -69,18 +71,21 @@ pub(super) fn compute_dynamic_dory_commitments(
committable_columns.len()
])
.unwrap_or_else(|| {
(0..committable_columns.len())
.map(|i| {
let sub_slice = signed_sub_commits[i..]
.iter()
.step_by(committable_columns.len())
.take(num_commits);
DynamicDoryCommitment(pairings::multi_pairing(
sub_slice,
&Gamma_2[gamma_2_offset..gamma_2_offset + num_commits],
))
})
.collect()
if_rayon!(
(0..committable_columns.len()).into_par_iter(),
(0..committable_columns.len())
)
.map(|i| {
let sub_slice = signed_sub_commits[i..]
.iter()
.step_by(committable_columns.len())
.take(num_commits);
DynamicDoryCommitment(pairings::multi_pairing(
sub_slice,
&Gamma_2[gamma_2_offset..gamma_2_offset + num_commits],
))
})
.collect()
});
span.exit();

Expand Down
Loading