Da: parallel encoder (#665)

* Added basic parallelization to encoder

* Added encoder benchmarks

nomos-da/kzgrs-backend/Cargo.toml

@@ -25,6 +25,7 @@ rayon = { version = "1.10.0", optional = true }
 [dev-dependencies]
 rand = "0.8"
 ark-bls12-381 = "0.4.0"
+divan = "0.1"
 
 [features]
 default = []
@@ -34,3 +35,7 @@ parallel = [
     "kzgrs/parallel"
 ]
 
+[[bench]]
+name = "encoder"
+harness = false
+

nomos-da/kzgrs-backend/benches/encoder.rs

@@ -0,0 +1,26 @@
+use divan::counter::BytesCount;
+use divan::Bencher;
+use kzgrs_backend::encoder::{DaEncoder, DaEncoderParams};
+use rand::RngCore;
+use std::hint::black_box;
+
+fn main() {
+    divan::main()
+}
+
+const PARAMS: DaEncoderParams = DaEncoderParams::default_with(4096);
+const ENCODER: DaEncoder = DaEncoder::new(PARAMS);
+const KB: usize = 1024;
+
+pub fn rand_data(elements_count: usize) -> Vec<u8> {
+    let mut buff = vec![0; elements_count * DaEncoderParams::MAX_BLS12_381_ENCODING_CHUNK_SIZE];
+    rand::thread_rng().fill_bytes(&mut buff);
+    buff
+}
+#[divan::bench(args = [32, 64, 128, 256, 512, 1024, 2048], sample_count = 1, sample_size = 1)]
+fn encode(bencher: Bencher, size: usize) {
+    bencher
+        .with_inputs(|| rand_data(size * KB / DaEncoderParams::MAX_BLS12_381_ENCODING_CHUNK_SIZE))
+        .input_counter(|buff| BytesCount::new(buff.len()))
+        .bench_refs(|buff| black_box(ENCODER.encode(buff)));
+}

nomos-da/kzgrs-backend/src/common/mod.rs

@@ -7,6 +7,8 @@ use serde::{Deserialize, Serialize};
 use std::io::Cursor;
 // crates
 use blake2::digest::{Update, VariableOutput};
+#[cfg(feature = "parallel")]
+use rayon::iter::{IntoParallelRefIterator, ParallelIterator};
 use sha3::{Digest, Sha3_256};
 // internal
 use kzgrs::Commitment;
@@ -107,6 +109,10 @@ impl ChunksMatrix {
     pub fn rows(&self) -> impl Iterator<Item = &Row> + '_ {
         self.0.iter()
     }
+    #[cfg(feature = "parallel")]
+    pub fn par_rows(&self) -> impl ParallelIterator<Item = &Row> + '_ {
+        self.0.par_iter()
+    }
     pub fn columns(&self) -> impl Iterator<Item = Column> + '_ {
         let size = self.0.first().map(|r| r.0.len()).unwrap_or(0);
         (0..size).map(|i| {

nomos-da/kzgrs-backend/src/encoder.rs

@@ -10,6 +10,8 @@ use kzgrs::{
     bytes_to_polynomial, commit_polynomial, encode, Commitment, Evaluations, KzgRsError,
     Polynomial, PolynomialEvaluationDomain, Proof, BYTES_PER_FIELD_ELEMENT,
 };
+#[cfg(feature = "parallel")]
+use rayon::iter::{IntoParallelRefIterator, ParallelIterator};
 
 // internal
 use crate::common::{hash_column_and_commitment, Chunk, ChunksMatrix, Row};
@@ -71,8 +73,16 @@ impl DaEncoder {
         matrix: &ChunksMatrix,
         polynomial_evaluation_domain: PolynomialEvaluationDomain,
     ) -> Result<Vec<((Evaluations, Polynomial), Commitment)>, KzgRsError> {
-        matrix
+        {
-            .rows()
+            #[cfg(not(feature = "parallel"))]
+            {
+                matrix.rows()
+            }
+            #[cfg(feature = "parallel")]
+            {
+                matrix.par_rows()
+            }
+        }
         .map(|r| {
             // Using the unchecked version here. Because during the process of chunkifiying
             // we already make sure to have the chunks of proper elements.
@@ -98,14 +108,31 @@ impl DaEncoder {
         rows: &[Polynomial],
         polynomial_evaluation_domain: PolynomialEvaluationDomain,
     ) -> Vec<Evaluations> {
+        {
+            #[cfg(not(feature = "parallel"))]
+            {
                 rows.iter()
+            }
+            #[cfg(feature = "parallel")]
+            {
+                rows.par_iter()
+            }
+        }
         .map(|poly| Self::rs_encode_row(poly, polynomial_evaluation_domain))
         .collect()
     }
 
     fn compute_rows_proofs(polynomials: &[Polynomial]) -> Result<Vec<Vec<Proof>>, KzgRsError> {
-        Ok(polynomials
+        Ok({
-            .iter()
+            #[cfg(not(feature = "parallel"))]
+            {
+                polynomials.iter()
+            }
+            #[cfg(feature = "parallel")]
+            {
+                polynomials.par_iter()
+            }
+        }
         .map(|poly| fk20_batch_generate_elements_proofs(poly, &GLOBAL_PARAMETERS))
         .collect())
     }