feat: add kmer iterators and optimize layered map performance

Replace `ph` with `ptr_hash` and introduce `epserde` and `rayon` dependencies. Refactor MPHF construction to leverage parallel iteration, eliminating intermediate `Vec<u64>` allocations and reducing memory footprint. Add a `n_kmers` field to track and serialize total kmer counts, alongside three zero-allocation iterators for efficient chunk traversal. Include comprehensive unit tests for the new iterators and update CLAUDE.md to enforce explicit dependency validation policies.
This commit is contained in:
Eric Coissac
2026-05-12 22:28:01 +08:00
parent 9c41891cc8
commit ff75c9198d
7 changed files with 359 additions and 61 deletions
+56 -7
View File
@@ -3,7 +3,7 @@ use std::io::{BufWriter, Write as _};
use std::path::{Path, PathBuf};
use memmap2::Mmap;
use obikseq::{CanonicalKmer, Unitig};
use obikseq::{CanonicalKmer, Kmer, Unitig};
pub use obikseq::MAX_KMERS_PER_CHUNK;
@@ -13,6 +13,7 @@ use crate::error::{SKError, SKResult};
//
// magic: [u8; 4] = b"UIDX"
// n_unitigs: u32 LE
// n_kmers: u64 LE total kmer count across all chunks
// seqls: [u8; n_unitigs] max kmer index per chunk (= n_kmers 1)
// packed_offsets: [u32; n_unitigs + 1] byte offsets to packed bytes in the
// sequence file; last entry is sentinel
@@ -44,6 +45,7 @@ pub struct UnitigFileWriter {
seqls: Vec<u8>,
packed_offsets: Vec<u32>,
next_offset: u32,
n_kmers: usize,
k: usize,
}
@@ -56,6 +58,7 @@ impl UnitigFileWriter {
seqls: Vec::new(),
packed_offsets: Vec::new(),
next_offset: 0,
n_kmers: 0,
k: obikseq::params::k(),
})
}
@@ -98,6 +101,7 @@ impl UnitigFileWriter {
debug_assert!(seql - self.k <= u8::MAX as usize, "chunk exceeds MAX_KMERS_PER_CHUNK");
self.packed_offsets.push(self.next_offset + 1);
self.seqls.push((seql - self.k) as u8);
self.n_kmers += seql - self.k + 1;
unitig
.write_to_binary(&mut self.file)
@@ -122,7 +126,7 @@ impl UnitigFileWriter {
};
self.packed_offsets.push(sentinel);
write_idx(&idx_path(&self.path), &self.seqls, &self.packed_offsets)
write_idx(&idx_path(&self.path), &self.seqls, &self.packed_offsets, self.n_kmers)
}
pub fn len(&self) -> usize {
@@ -134,10 +138,11 @@ impl UnitigFileWriter {
}
}
fn write_idx(path: &Path, seqls: &[u8], packed_offsets: &[u32]) -> SKResult<()> {
fn write_idx(path: &Path, seqls: &[u8], packed_offsets: &[u32], n_kmers: usize) -> SKResult<()> {
let mut w = BufWriter::new(File::create(path).map_err(SKError::Io)?);
w.write_all(&MAGIC).map_err(SKError::Io)?;
w.write_all(&(seqls.len() as u32).to_le_bytes()).map_err(SKError::Io)?;
w.write_all(&(n_kmers as u64).to_le_bytes()).map_err(SKError::Io)?;
w.write_all(seqls).map_err(SKError::Io)?;
for &off in packed_offsets {
w.write_all(&off.to_le_bytes()).map_err(SKError::Io)?;
@@ -155,6 +160,7 @@ pub struct UnitigFileReader {
mmap: Mmap,
seqls: Vec<u8>,
packed_offsets: Vec<u32>,
n_kmers: usize,
k: usize,
}
@@ -162,9 +168,9 @@ impl UnitigFileReader {
pub fn open(path: &Path) -> SKResult<Self> {
let file = File::open(path).map_err(SKError::Io)?;
let mmap = unsafe { Mmap::map(&file).map_err(SKError::Io)? };
let (seqls, packed_offsets) = read_idx(&idx_path(path))?;
let (seqls, packed_offsets, n_kmers) = read_idx(&idx_path(path))?;
let k = obikseq::params::k();
Ok(Self { mmap, seqls, packed_offsets, k })
Ok(Self { mmap, seqls, packed_offsets, n_kmers, k })
}
pub fn len(&self) -> usize {
@@ -175,6 +181,11 @@ impl UnitigFileReader {
self.seqls.is_empty()
}
/// Total number of kmers across all chunks.
pub fn n_kmers(&self) -> usize {
self.n_kmers
}
/// Return the nucleotide length of chunk `i`.
#[inline]
pub fn seql(&self, i: usize) -> usize {
@@ -206,9 +217,42 @@ impl UnitigFileReader {
pub fn verify_canonical_kmer(&self, i: usize, j: usize, query: CanonicalKmer) -> bool {
canonical_raw(self.raw_kmer(i, j), self.k) == query.raw()
}
/// Iterate over all kmers in file order (all positions of chunk 0, then chunk 1, …).
///
/// Each chunk is copied from the mmap once; iteration within the chunk is
/// zero-allocation (sliding-window via [`OwnedPackedSeqKmerIter`]).
pub fn iter_kmers(&self) -> impl Iterator<Item = Kmer> + '_ {
(0..self.len()).flat_map(move |i| self.unitig(i).into_kmers())
}
/// Iterate over all canonical kmers in file order.
///
/// Equivalent to `iter_kmers().map(|km| km.canonical())` but uses the
/// built-in canonical iterator on each chunk, which avoids a separate
/// canonicalization pass.
pub fn iter_canonical_kmers(&self) -> impl Iterator<Item = CanonicalKmer> + '_ {
(0..self.len()).flat_map(move |i| self.unitig(i).into_canonical_kmers())
}
/// Iterate over `(kmer, chunk_id, rank)` for every canonical kmer in the file.
///
/// `chunk_id` is the index of the chunk within this file; `rank` is the
/// 0-based position of the kmer within that chunk. Used to build the
/// evidence table in `obilayeredmap`.
pub fn iter_indexed_canonical_kmers(
&self,
) -> impl Iterator<Item = (CanonicalKmer, usize, usize)> + '_ {
(0..self.len()).flat_map(move |chunk_id| {
self.unitig(chunk_id)
.into_canonical_kmers()
.enumerate()
.map(move |(rank, kmer)| (kmer, chunk_id, rank))
})
}
}
fn read_idx(path: &Path) -> SKResult<(Vec<u8>, Vec<u32>)> {
fn read_idx(path: &Path) -> SKResult<(Vec<u8>, Vec<u32>, usize)> {
let data = std::fs::read(path).map_err(SKError::Io)?;
let mut pos = 0;
@@ -227,6 +271,11 @@ fn read_idx(path: &Path) -> SKResult<(Vec<u8>, Vec<u32>)> {
let n = u32::from_le_bytes(n_bytes.try_into().unwrap()) as usize;
pos += 4;
let nk_bytes = data.get(pos..pos + 8)
.ok_or(SKError::Truncated { context: "unitig index: n_kmers" })?;
let n_kmers = u64::from_le_bytes(nk_bytes.try_into().unwrap()) as usize;
pos += 8;
let seqls = data.get(pos..pos + n)
.ok_or(SKError::Truncated { context: "unitig index: seqls" })?
.to_vec();
@@ -240,7 +289,7 @@ fn read_idx(path: &Path) -> SKResult<(Vec<u8>, Vec<u32>)> {
pos += 4;
}
Ok((seqls, packed_offsets))
Ok((seqls, packed_offsets, n_kmers))
}
// ── Kmer utilities ────────────────────────────────────────────────────────────