Merge pull request #2547 from subspace/small-optimizations

Small optimizations

Cargo.toml

@@ -138,3 +138,8 @@ sp-inherents = { version = "4.0.0-dev", git = "https://github.com/subspace/polka
 sp-io = { version = "23.0.0", git = "https://github.com/subspace/polkadot-sdk", rev = "d6b500960579d73c43fc4ef550b703acfa61c4c8" }
 sp-runtime = { version = "24.0.0", git = "https://github.com/subspace/polkadot-sdk", rev = "d6b500960579d73c43fc4ef550b703acfa61c4c8" }
 substrate-prometheus-endpoint = { version = "0.10.0-dev", git = "https://github.com/subspace/polkadot-sdk", rev = "d6b500960579d73c43fc4ef550b703acfa61c4c8" }
+
+# TODO: Importing https://github.com/supranational/blst/pull/203 to take advantage of optimizations introduced there,
+#  switch to upstream once merged or once similar performance improvements land upstream
+[patch."https://github.com/supranational/blst.git"]
+blst = { git = "https://github.com/subspace/blst", rev = "ab042e18cb3b62e131423380513964e4b2c7b445" }

crates/pallet-subspace/src/mock.rs

@@ -38,11 +38,11 @@ use sp_runtime::testing::{Digest, DigestItem, Header, TestXt};
 use sp_runtime::traits::{Block as BlockT, Header as _, IdentityLookup};
 use sp_runtime::{BuildStorage, Perbill};
 use sp_weights::Weight;
-use std::iter;
 use std::marker::PhantomData;
 use std::num::{NonZeroU32, NonZeroU64, NonZeroUsize};
 use std::simd::Simd;
 use std::sync::{Once, OnceLock};
+use std::{iter, slice};
 use subspace_archiving::archiver::{Archiver, NewArchivedSegment};
 use subspace_core_primitives::crypto::kzg::{embedded_kzg_settings, Kzg};
 use subspace_core_primitives::crypto::Scalar;
@@ -464,7 +464,7 @@ pub fn create_signed_vote(
             sector_metadata_output: &mut plotted_sector_metadata_bytes,
             downloading_semaphore: None,
             encoding_semaphore: None,
-            table_generator: &mut table_generator,
+            table_generators: slice::from_mut(&mut table_generator),
             abort_early: &Default::default(),
         }))
         .unwrap();

crates/subspace-farmer-components/benches/auditing.rs

@@ -128,7 +128,7 @@ pub fn criterion_benchmark(c: &mut Criterion) {
             sector_metadata_output: &mut plotted_sector_metadata_bytes,
             downloading_semaphore: black_box(None),
             encoding_semaphore: black_box(None),
-            table_generator: &mut table_generator,
+            table_generators: slice::from_mut(&mut table_generator),
             abort_early: &Default::default(),
         }))
         .unwrap();

crates/subspace-farmer-components/benches/plotting.rs

@@ -35,7 +35,16 @@ fn criterion_benchmark(c: &mut Criterion) {
             .expect("Not zero; qed"),
     )
     .unwrap();
-    let mut table_generator = PosTable::generator();
+    let mut table_generators = [
+        PosTable::generator(),
+        PosTable::generator(),
+        PosTable::generator(),
+        PosTable::generator(),
+        PosTable::generator(),
+        PosTable::generator(),
+        PosTable::generator(),
+        PosTable::generator(),
+    ];
     let archived_history_segment = archiver
         .add_block(
             AsRef::<[u8]>::as_ref(input.as_ref()).to_vec(),
@@ -79,7 +88,7 @@ fn criterion_benchmark(c: &mut Criterion) {
                 sector_metadata_output: black_box(&mut sector_metadata_bytes),
                 downloading_semaphore: black_box(None),
                 encoding_semaphore: black_box(None),
-                table_generator: black_box(&mut table_generator),
+                table_generators: black_box(&mut table_generators),
                 abort_early: &Default::default(),
             }))
             .unwrap();

crates/subspace-farmer-components/benches/proving.rs

@@ -135,7 +135,7 @@ pub fn criterion_benchmark(c: &mut Criterion) {
             sector_metadata_output: &mut plotted_sector_metadata_bytes,
             downloading_semaphore: black_box(None),
             encoding_semaphore: black_box(None),
-            table_generator: &mut table_generator,
+            table_generators: slice::from_mut(&mut table_generator),
             abort_early: &Default::default(),
         }))
         .unwrap();

crates/subspace-farmer-components/benches/reading.rs

@@ -6,7 +6,7 @@ use rand::prelude::*;
 use std::fs::OpenOptions;
 use std::io::Write;
 use std::num::{NonZeroU64, NonZeroUsize};
-use std::{env, fs};
+use std::{env, fs, slice};
 use subspace_archiving::archiver::Archiver;
 use subspace_core_primitives::crypto::kzg;
 use subspace_core_primitives::crypto::kzg::Kzg;
@@ -128,7 +128,7 @@ pub fn criterion_benchmark(c: &mut Criterion) {
             sector_metadata_output: &mut plotted_sector_metadata_bytes,
             downloading_semaphore: black_box(None),
             encoding_semaphore: black_box(None),
-            table_generator: &mut table_generator,
+            table_generators: slice::from_mut(&mut table_generator),
             abort_early: &Default::default(),
         }))
         .unwrap();

crates/subspace-farmer-components/src/plotting.rs

@@ -12,11 +12,11 @@ use futures::StreamExt;
 use parity_scale_codec::{Decode, Encode};
 use parking_lot::Mutex;
 use rayon::prelude::*;
+use std::mem;
 use std::simd::Simd;
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
 use std::time::Duration;
-use std::{mem, slice};
 use subspace_core_primitives::crypto::kzg::Kzg;
 use subspace_core_primitives::crypto::{blake3_hash, blake3_hash_parallel, Scalar};
 use subspace_core_primitives::{
@@ -147,8 +147,8 @@ where
     /// Semaphore for part of the plotting when farmer encodes downloaded sector, should typically
     /// allow one permit at a time for efficient CPU utilization
     pub encoding_semaphore: Option<&'a Semaphore>,
-    /// Proof of space table generator
-    pub table_generator: &'a mut PosTable::Generator,
+    /// Proof of space table generators
+    pub table_generators: &'a mut [PosTable::Generator],
     /// Whether encoding should be aborted early
     pub abort_early: &'a AtomicBool,
 }
@@ -179,7 +179,7 @@ where
         sector_metadata_output,
         downloading_semaphore,
         encoding_semaphore,
-        table_generator,
+        table_generators,
         abort_early,
     } = options;
 
@@ -211,7 +211,7 @@ where
             pieces_in_sector,
             sector_output,
             sector_metadata_output,
-            table_generators: slice::from_mut(table_generator),
+            table_generators,
             abort_early,
         },
     )
@@ -349,7 +349,7 @@ where
     /// Where plotted sector metadata should be written, vector must either be empty (in which case
     /// it'll be resized to correct size automatically) or correctly sized from the beginning
     pub sector_metadata_output: &'a mut Vec<u8>,
-    /// Proof of space table generator
+    /// Proof of space table generators
     pub table_generators: &'a mut [PosTable::Generator],
     /// Whether encoding should be aborted early
     pub abort_early: &'a AtomicBool,
@@ -419,7 +419,7 @@ where
 
                     loop {
                         // This instead of `while` above because otherwise mutex will be held for
-                        // the duration of the loop and wil limit concurrency to 1 table generator
+                        // the duration of the loop and will limit concurrency to 1 table generator
                         let Some(((piece_offset, record), encoded_chunks_used)) =
                             iter.lock().next()
                         else {
@@ -561,6 +561,7 @@ fn record_encoding<PosTable>(
         .extend(&source_record_chunks)
         .expect("Instance was verified to be able to work with this many values earlier; qed");
 
+    chunks_scratch.clear();
     // For every erasure coded chunk check if there is quality present, if so then encode
     // with PoSpace quality bytes and set corresponding `quality_present` bit to `true`
     (u16::from(SBucket::ZERO)..=u16::from(SBucket::MAX))

crates/subspace-proof-of-space/src/chiapos/table.rs

@@ -791,7 +791,7 @@ where
         let mut positions = Vec::with_capacity(t_n.len());
         let mut metadatas = Vec::with_capacity(t_n.len());
 
-        for (y, [left_position, right_position], metadata) in t_n {
+        for (y, [left_position, right_position], metadata) in t_n.drain(..) {
             ys.push(y);
             positions.push([left_position, right_position]);
             // Last table doesn't have metadata
@@ -800,6 +800,11 @@ where
             }
         }
 
+        // Drop from a background thread, which typically helps with overall concurrency
+        rayon::spawn(move || {
+            drop(t_n);
+        });
+
         Self::Other {
             ys,
             positions,

test/subspace-test-client/src/lib.rs

@@ -31,6 +31,7 @@ use sp_api::ProvideRuntimeApi;
 use sp_consensus_subspace::{FarmerPublicKey, FarmerSignature, SubspaceApi};
 use sp_core::{Decode, Encode};
 use std::num::{NonZeroU64, NonZeroUsize};
+use std::slice;
 use std::sync::Arc;
 use subspace_core_primitives::crypto::kzg::{embedded_kzg_settings, Kzg};
 use subspace_core_primitives::objects::BlockObjectMapping;
@@ -245,7 +246,7 @@ where
         sector_metadata_output: &mut sector_metadata,
         downloading_semaphore: None,
         encoding_semaphore: None,
-        table_generator: &mut table_generator,
+        table_generators: slice::from_mut(&mut table_generator),
         abort_early: &Default::default(),
     })
     .await