Simple succinct data structures

These structures are comparable to those in SDSL in performance and scalability. As the focus is on (relative) simplicity, ugly low-level optimizations are generally avoided.

Implemented functionality

Integer vectors

• RawVector: A bit array that supports reading, writing, and appending 1-64 bits at a time. Implemented on top of Vec<u64>.
  • RawVectorWriter: An append-only version of RawVector that writes the structure directly to a file.
  • RawVectorMapper: An immutable memory-mapped RawVector.
• IntVector: A bit-packed vector of fixed-width integers implemented on top of RawVector. Like sdsl::int_vector but also supports stack functionality.
  • IntVectorWriter: An append-only version of IntVector that writes the structure directly to a file. Like a subset of sdsl::int_vector_buffer.
  • IntVectorMapper: An immutable memory-mapped IntVector.
• WaveletMatrix: An immutable vector of fixed-width integers. Similar to sdsl::wm_int.
  • Supports rank(), inverse_select(), select(), predecessor(), and successor() with each item value.
  • Iterators over all items and over items with a specified value.
  • Implemented using a BitVector for each level.

Bitvectors

• BitVector: A plain immutable bitvector.
  • Supports rank(), rank_zero(), select(), select_zero(), predecessor(), and successor() queries using optional support structures.
  • Iterators over set bits, unset bits, and all bits.
  • Implemented on top of RawVector.
• RLVector: A run-length encoded bitvector.
  • Supports rank(), rank_zero(), select(), select_zero(), predecessor(), and successor() queries.
  • Iterators over set bits and all bits.
  • Space-efficient construction with RLBuilder.
• SparseVector: An Elias-Fano encoded bitvector.
  • Supports rank(), rank_zero(), select(), select_zero(), predecessor(), and successor() queries .
  • Iterators over set bits and all bits.
  • Space-efficient construction with SparseBuilder.

Planned functionality

Integer vectors

• Mutable memory-mapped vectors.

Bitvectors

• Versions of predecessor() and successor() that return values instead of iterators?
• Slice-like functionality based on iterators?

Notes

• The included .cargo/config.toml sets the target CPU to native.
• This crate is designed for the x86_64 architecture with the BMI2 instruction set (Intel Haswell / AMD Excavator or later). Some operations may be slow without the POPCNT, LZCNT, TZCNT, and PDEP instructions.
• 64-bit ARM is also supported.
• Unix-like operating system is required for mmap().
• Things may not work if the system is not little-endian or if usize is not 64-bit.