Search, hash, sort, and process strings faster via SWAR and SIMD

Project description

StringZilla 🦖

StringZilla banner

Strings are the first fundamental data type every programming language implements in software rather than hardware, so dedicated CPU instructions are rare - and the few that exist are hardly ideal. That's why most languages lean on the C standard library (libc) for their string operations, which, despite its name, ships its hottest code in hand-tuned assembly. It does exploit SIMD, but it isn't perfect. 1️⃣ Even on ubiquitous hardware - over a billion 64-bit ARM CPUs - routines such as strstr and memmem top out at roughly one-third of available throughput. 2️⃣ SIMD coverage is uneven: fast forward scans don't guarantee speedy reverse searches, hashing and case-mapping is not even part of the standard. 3️⃣ Many higher-level languages can't rely on libc at all because their strings aren't NUL-terminated - or may even contain embedded zeroes. That's why StringZilla exists: predictable, high performance on every modern platform, OS, and programming language.

StringZilla code size

StringZilla is the GodZilla of string libraries, using SIMD and SWAR to accelerate binary and UTF-8 string operations on modern CPUs and GPUs. It delivers up to 10x higher CPU throughput in C, C++, Rust, Python, and other languages, and can be 100x faster than existing GPU kernels, covering a broad range of functionality. It accelerates exact and fuzzy string matching, hashing, edit distance computations, sorting, provides allocation-free lazily-evaluated smart-iterators, and even random-string generators.

🐂 C: Upgrade LibC's <string.h> to <stringzilla/stringzilla.h> in C 99
🐉 C++: Upgrade STL's <string> to <stringzilla/stringzilla.hpp> in C++ 11
🧮 CUDA: Process in-bulk with <stringzillas/stringzillas.cuh> in CUDA C++ 17
🐍 Python: Upgrade your str to faster Str
🦀 Rust: Use the StringZilla traits crate
🦫 Go: Use the StringZilla cGo module
🍎 Swift: Use the String+StringZilla extension
🟨 JavaScript: Use the StringZilla library
🐚 Shell: Accelerate common CLI tools with sz- prefix
📚 Researcher? Jump to Algorithms & Design Decisions
💡 Thinking to contribute? Look for "good first issues"
🤝 And check the guide to set up the environment
Want more bindings or features? Let me know!

Who is this for?

For data-engineers parsing large datasets, like the CommonCrawl, RedPajama, or LAION.
For software engineers optimizing strings in their apps and services.
For bioinformaticians and search engineers looking for edit-distances for USearch.
For DBMS devs, optimizing LIKE, ORDER BY, and GROUP BY operations.
For hardware designers, needing a SWAR baseline for string-processing functionality.
For students studying SIMD/SWAR applications to non-data-parallel operations.

Performance

C	C++	Python	StringZilla
Unicode case-folding, expanding characters like `ß` → `ss`
⚪	⚪	`.casefold` x86: 0.4 GB/s	`sz.utf8_case_fold` x86: 1.3 GB/s
Unicode case-insensitive substring search
⚪	⚪	`icu.StringSearch` x86: 0.02 GB/s	`utf8_case_insensitive_find` x86: 3.0 GB/s
find the first occurrence of a random word from text, ≅ 5 bytes long
`strstr` ¹ x86: 7.4 · arm: 2.0 GB/s	`.find` x86: 2.9 · arm: 1.6 GB/s	`.find` x86: 1.1 · arm: 0.6 GB/s	`sz_find` x86: 10.6 · arm: 7.1 GB/s
find the last occurrence of a random word from text, ≅ 5 bytes long
⚪	`.rfind` x86: 0.5 · arm: 0.4 GB/s	`.rfind` x86: 0.9 · arm: 0.5 GB/s	`sz_rfind` x86: 10.8 · arm: 6.7 GB/s
split lines separated by `\n` or `\r` ²
`strcspn` ¹ x86: 5.42 · arm: 2.19 GB/s	`.find_first_of` x86: 0.59 · arm: 0.46 GB/s	`re.finditer` x86: 0.06 · arm: 0.02 GB/s	`sz_find_byteset` x86: 4.08 · arm: 3.22 GB/s
find the last occurrence of any of 6 whitespaces ²
⚪	`.find_last_of` x86: 0.25 · arm: 0.25 GB/s	⚪	`sz_rfind_byteset` x86: 0.43 · arm: 0.23 GB/s
Random string from a given alphabet, 20 bytes long ³
`rand() % n` x86: 18.0 · arm: 9.4 MB/s	`uniform_int_distribution` x86: 47.2 · arm: 20.4 MB/s	`join(random.choices(x))` x86: 13.3 · arm: 5.9 MB/s	`sz_fill_random` x86: 56.2 · arm: 25.8 MB/s
Mapping characters with lookup table transforms
⚪	`std::transform` x86: 3.81 · arm: 2.65 GB/s	`str.translate` x86: 260.0 · arm: 140.0 MB/s	`sz_lookup` x86: 21.2 · arm: 8.5 GB/s
Get sorted order, ≅ 8 million English words ⁴
`qsort_r` x86: 3.55 · arm: 5.77 s	`std::sort` x86: 2.79 · arm: 4.02 s	`numpy.argsort` x86: 7.58 · arm: 13.00 s	`sz_sequence_argsort` x86: 1.91 · arm: 2.37 s
Levenshtein edit distance, text lines ≅ 100 bytes long
⚪	⚪	via `NLTK` ⁵ and `CuDF` x86: 1,615,306 · arm: 1,349,980 · cuda: 6,532,411,354 CUPS	`szs_levenshtein_distances_t` x86: 3,434,427,548 · arm: 1,605,340,403 · cuda: 93,662,026,653 CUPS
Needleman-Wunsch alignment scores, proteins ≅ 1 K amino acids long
⚪	⚪	via `biopython` ⁶ x86: 575,981,513 · arm: 436,350,732 CUPS	`szs_needleman_wunsch_scores_t` x86: 452,629,942 · arm: 520,170,239 · cuda: 9,017,327,818 CUPS

Most StringZilla modules ship ready-to-run benchmarks for C, C++, Python, and more. Grab them from ./scripts, and see CONTRIBUTING.md for instructions. On CPUs that permit misaligned loads, even the 64-bit SWAR baseline outruns both libc and the STL. For wider head-to-heads against Rust and Python favorites, browse the StringWars repository. To inspect collision resistance and distribution shapes for our hashers, see HashEvals.

Most benchmarks were conducted on a 1 GB English text corpus, with an average word length of 6 characters. The code was compiled with GCC 12, using glibc v2.35. The benchmarks were performed on Arm-based Graviton3 AWS c7g instances and r7iz Intel Sapphire Rapids. Most modern Arm-based 64-bit CPUs will have similar relative speedups. Variance within x86 CPUs will be larger. For CUDA benchmarks, the Nvidia H100 GPUs were used. ¹ Unlike other libraries, LibC requires strings to be NULL-terminated. ² Six whitespaces in the ASCII set are: \t\n\v\f\r. Python's and other standard libraries have specialized functions for those. ³ All modulo operations were conducted with uint8_t to allow compilers more optimization opportunities. The C++ STL and StringZilla benchmarks used a 64-bit Mersenne Twister as the generator. For C, C++, and StringZilla, an in-place update of the string was used. In Python every string had to be allocated as a new object, which makes it less fair. ⁴ Contrary to the popular opinion, Python's default sorted function works faster than the C and C++ standard libraries. That holds for large lists or tuples of strings, but fails as soon as you need more complex logic, like sorting dictionaries by a string key, or producing the "sorted order" permutation. The latter is very common in database engines and is most similar to numpy.argsort. The current StringZilla solution can be at least 4x faster without loss of generality. ⁵ Most Python libraries for strings are also implemented in C. ⁶ Unlike the rest of BioPython, the alignment score computation is implemented in C.

Functionality

StringZilla is compatible with most modern CPUs, and provides a broad range of functionality. It's split into 2 layers:

StringZilla: single-header C library and C++ wrapper for high-performance string operations.
StringZillas: parallel CPU/GPU backends used for large-batch operations and accelerators.

Having a second C++/CUDA layer greatly simplifies the implementation of similarity scoring and fingerprinting functions, which would otherwise require too much error-prone boilerplate code in pure C. Both layers are designed to be extremely portable:

across both little-endian and big-endian architectures.
across 32-bit and 64-bit hardware architectures.
across operating systems and compilers.
across ASCII and UTF-8 encoded inputs.

Not all features are available across all bindings. Consider contributing if you need a feature that's not yet implemented.

	Maturity	C	C++	Python	Rust	JS	Swift	Go
Substring Search	🌳	✅	✅	✅	✅	✅	✅	✅
Character Set Search	🌳	✅	✅	✅	✅	✅	✅	✅
Sorting & Sequence Operations	🌳	✅	✅	✅	✅	⚪	⚪	⚪
Lazy Ranges, Compressed Arrays	🌳	❌	✅	✅	✅	❌	⚪	⚪
One-Shot & Streaming Hashes	🌳	✅	✅	✅	✅	✅	✅	✅
Cryptographic Hashes	🌳	✅	✅	✅	✅	✅	✅	✅
Small String Class	🧐	✅	✅	❌	⚪	❌	❌	❌
Random String Generation	🌳	✅	✅	✅	✅	⚪	⚪	⚪

Unicode Case Folding	🧐	✅	✅	✅	⚪	⚪	⚪	⚪
Case-Insensitive UTF-8 Search	🚧	✅	✅	✅	⚪	⚪	⚪	⚪
TR29 Word Boundary Detection	🚧	✅	✅	⚪	⚪	⚪	⚪	⚪

Parallel Similarity Scoring	🌳	✅	✅	✅	✅	⚪	⚪	⚪
Parallel Rolling Fingerprints	🌳	✅	✅	✅	✅	⚪	⚪	⚪

🌳 parts are used in production. 🧐 parts are in beta. 🚧 parts are under active development, and are likely to break in subsequent releases. ✅ are implemented. ⚪ are considered. ❌ are not intended.

Quick Start: Python

Python bindings are available on PyPI for Python 3.8+, and can be installed with pip.

pip install stringzilla         # for serial algorithms
pip install stringzillas-cpus   # for parallel multi-CPU backends
pip install stringzillas-cuda   # for parallel Nvidia GPU backend

You can immediately check the installed version and the used hardware capabilities with following commands:

python -c "import stringzilla; print(stringzilla.__version__)"
python -c "import stringzillas; print(stringzillas.__version__)"
python -c "import stringzilla; print(stringzilla.__capabilities__)"     # for serial algorithms
python -c "import stringzillas; print(stringzillas.__capabilities__)"   # for parallel algorithms

Basic Usage

If you've ever used the Python str, bytes, bytearray, or memoryview classes, you'll know what to expect. StringZilla's Str class is a hybrid of the above, providing a str-like interface to byte arrays.

from stringzilla import Str, File

text_from_str = Str('some-string') # no copies, just a view
text_from_bytes = Str(b'some-array') # no copies, just a view
text_from_file = Str(File('some-file.txt')) # memory-mapped file

import numpy as np
alphabet_array = np.arange(ord("a"), ord("z"), dtype=np.uint8)
text_from_array = Str(memoryview(alphabet_array))

The File class memory-maps a file from persistent storage without loading its copy into RAM. The contents of that file would remain immutable, and the mapping can be shared by multiple Python processes simultaneously. A standard dataset pre-processing use case would be to map a sizable textual dataset like Common Crawl into memory, spawn child processes, and split the job between them.

Basic Operations

Length: len(text) -> int
Indexing: text[42] -> str
Slicing: text[42:46] -> Str
Substring check: 'substring' in text -> bool
Hashing: hash(text) -> int
String conversion: str(text) -> str

Advanced Operations

import sys

x: bool = text.contains('substring', start=0, end=sys.maxsize)
x: int = text.find('substring', start=0, end=sys.maxsize)
x: int = text.count('substring', start=0, end=sys.maxsize, allowoverlap=False)
x: str = text.decode(encoding='utf-8', errors='strict')
x: Strs = text.split(separator=' ', maxsplit=sys.maxsize, keepseparator=False)
x: Strs = text.rsplit(separator=' ', maxsplit=sys.maxsize, keepseparator=False)
x: Strs = text.splitlines(keeplinebreaks=False, maxsplit=sys.maxsize)

It's important to note that the last function's behavior is slightly different from Python's str.splitlines. The native version matches \n, \r, \v or \x0b, \f or \x0c, \x1c, \x1d, \x1e, \x85, \r\n, \u2028, \u2029, including 3x two-byte-long runes. The StringZilla version matches only \n, \v, \f, \r, \x1c, \x1d, \x1e, \x85, avoiding two-byte-long runes.

Character Set Operations

Python strings don't natively support character set operations. This forces people to use regular expressions, which are slow and hard to read. To avoid the need for re.finditer, StringZilla provides the following interfaces:

x: int = text.find_first_of('chars', start=0, end=sys.maxsize)
x: int = text.find_last_of('chars', start=0, end=sys.maxsize)
x: int = text.find_first_not_of('chars', start=0, end=sys.maxsize)
x: int = text.find_last_not_of('chars', start=0, end=sys.maxsize)
x: Strs = text.split_byteset(separator='chars', maxsplit=sys.maxsize, keepseparator=False)
x: Strs = text.rsplit_byteset(separator='chars', maxsplit=sys.maxsize, keepseparator=False)

StringZilla also provides string trimming functions and random string generation:

x: str = text.lstrip('chars')  # Strip leading characters
x: str = text.rstrip('chars')  # Strip trailing characters
x: str = text.strip('chars')   # Strip both ends
x: bytes = sz.random(length=100, seed=42, alphabet='ACGT')  # Random string generation
sz.fill_random(buffer, seed=42, alphabet=None)  # Fill mutable buffer with random bytes

You can also transform the string using Look-Up Tables (LUTs), mapping it to a different character set. This would result in a copy - str for str inputs and bytes for other types.

x: str = text.translate('chars', {}, start=0, end=sys.maxsize, inplace=False)
x: bytes = text.translate(b'chars', {}, start=0, end=sys.maxsize, inplace=False)

For efficiency reasons, pass the LUT as a string or bytes object, not as a dictionary. This can be useful in high-throughput applications dealing with binary data, including bioinformatics and image processing. Here is an example:

import stringzilla as sz
look_up_table = bytes(range(256)) # Identity LUT
image = open("/image/path.jpeg", "rb").read()
sz.translate(image, look_up_table, inplace=True)

Hash

Single-shot and incremental hashing are both supported:

import stringzilla as sz

# One-shot - stable 64-bit output across all platforms!
one = sz.hash(b"Hello, world!", seed=42)

# Incremental updates return itself; digest does not consume state
hasher = sz.Hasher(seed=42)
hasher.update(b"Hello, ").update(b"world!")
streamed = hasher.digest() # or `hexdigest()` for a string
assert one == streamed

SHA-256 Checksums

SHA-256 cryptographic checksums are also available for single-shot and incremental hashing:

import stringzilla as sz

# One-shot SHA-256
digest_bytes = sz.sha256(b"Hello, world!")
assert len(digest_bytes) == 32

# Incremental SHA-256
hasher = sz.Sha256()
hasher.update(b"Hello, ").update(b"world!")
digest_bytes = hasher.digest()
digest_hex = hasher.hexdigest()  # 64 character lowercase hex string

# HMAC-SHA256 for message authentication
mac = sz.hmac_sha256(key=b"secret", message=b"Hello, world!")

StringZilla integrates seamlessly with memory-mapped files for efficient large file processing. The traditional approach with hashlib:

import hashlib

with open("xlsum.csv", "rb") as streamed_file:
    hasher = hashlib.sha256()
    while chunk := streamed_file.read(4096):
        hasher.update(chunk)
    checksum = hasher.hexdigest()

Can be simplified with StringZilla:

from stringzilla import Sha256, File

mapped_file = File("xlsum.csv")
checksum = Sha256().update(mapped_file).hexdigest()

Both output the same digest: 7278165ce01a4ac1e8806c97f32feae908036ca3d910f5177d2cf375e20aeae1. OpenSSL (powering hashlib) has faster Assembly kernels, but StringZilla avoids file I/O overhead with memory mapping and skips Python's abstraction layers:

OpenSSL-backed hashlib.sha256: 12.6s
StringZilla end-to-end: 4.0s — 3× faster!

Unicode Case-Folding and Case-Insensitive Search

StringZilla implements both Unicode Case Folding and Case-Insensitive UTF-8 Search. Unlike most libraries only capable of lower-casing ASCII-represented English alphabet, StringZilla covers over 1M+ codepoints. The case-folding API expects the output buffer to be at least 3× larger than the input, to accommodate for the worst-case character expansions scenarios.

import stringzilla as sz

sz.utf8_case_fold('HELLO')      # b'hello'
sz.utf8_case_fold('Straße')     # b'strasse' — ß (1 char) expands to "ss" (2 chars)
sz.utf8_case_fold('eﬃcient')    # b'efficient' — ﬃ ligature (1 char) expands to "ffi" (3 chars)

The case-insensitive search returns the byte offset of the match, handling expansions correctly.

import stringzilla as sz

sz.utf8_case_insensitive_find('Der große Hund', 'GROSSE')   # 4 — finds "große" at codepoint 4
sz.utf8_case_insensitive_find('Straße', 'STRASSE')          # 0 — ß matches "SS"
sz.utf8_case_insensitive_find('eﬃcient', 'EFFICIENT')       # 0 — ﬃ ligature matches "FFI"

# Iterator for finding ALL matches
haystack = 'Straße STRASSE strasse'
for match in sz.utf8_case_insensitive_find_iter(haystack, 'strasse'):
    print(match, match.offset_within(haystack))  # Yields: 'Straße', 'STRASSE', 'strasse'

# With overlapping matches
list(sz.utf8_case_insensitive_find_iter('aaaa', 'aa'))  # ['aa', 'aa'] — 2 non-overlapping
list(sz.utf8_case_insensitive_find_iter('aaaa', 'aa', include_overlapping=True))  # 3 matches

Collection-Level Operations

Once split into a Strs object, you can sort, shuffle, and reorganize the slices with minimal memory footprint. If all the chunks are located in consecutive memory regions, the memory overhead can be as low as 4 bytes per chunk.

lines: Strs = text.split(separator='\n') # 4 bytes per line overhead for under 4 GB of text
batch: Strs = lines.sample(seed=42) # 10x faster than `random.choices`
lines_shuffled: Strs = lines.shuffled(seed=42) # or shuffle all lines and shard with slices
lines_sorted: Strs = lines.sorted() # returns a new Strs in sorted order
order: tuple = lines.argsort() # similar to `numpy.argsort`

Working on RedPajama, addressing 20 billion annotated English documents, one will need only 160 GB of RAM instead of terabytes. Once loaded, the data will be memory-mapped, and can be reused between multiple Python processes without copies. And of course, you can use slices to navigate the dataset and shard it between multiple workers.

lines[::3] # every third line
lines[1::1] # every odd line
lines[:-100:-1] # last 100 lines in reverse order

Iterators and Memory Efficiency

Python's operations like split() and readlines() immediately materialize a list of copied parts. This can be very memory-inefficient for large datasets. StringZilla saves a lot of memory by viewing existing memory regions as substrings, but even more memory can be saved by using lazily evaluated iterators.

x: SplitIterator[Str] = text.split_iter(separator=' ', keepseparator=False)
x: SplitIterator[Str] = text.rsplit_iter(separator=' ', keepseparator=False)
x: SplitIterator[Str] = text.split_byteset_iter(separator='chars', keepseparator=False)
x: SplitIterator[Str] = text.rsplit_byteset_iter(separator='chars', keepseparator=False)

StringZilla can easily be 10x more memory efficient than native Python classes for tokenization. With lazy operations, it practically becomes free.

import stringzilla as sz
%load_ext memory_profiler

text = open("enwik9.txt", "r").read() # 1 GB, mean word length 7.73 bytes
%memit text.split() # increment: 8670.12 MiB (152 ms)
%memit sz.split(text) # increment: 530.75 MiB (25 ms)
%memit sum(1 for _ in sz.split_iter(text)) # increment: 0.00 MiB

Low-Level Python API

Aside from calling the methods on the Str and Strs classes, you can also call the global functions directly on str and bytes instances. Assuming StringZilla CPython bindings are implemented without any intermediate tools like SWIG or PyBind, the call latency should be similar to native classes.

import stringzilla as sz

contains: bool = sz.contains("haystack", "needle", start=0, end=sys.maxsize)
offset: int = sz.find("haystack", "needle", start=0, end=sys.maxsize)
count: int = sz.count("haystack", "needle", start=0, end=sys.maxsize, allowoverlap=False)

Similarity Scores

StringZilla exposes high-performance, batch-oriented similarity via the stringzillas module. Use DeviceScope to pick hardware and optionally limit capabilities per engine.

import stringzilla as sz
import stringzillas as szs

cpu_scope = szs.DeviceScope(cpu_cores=4)    # force CPU-only
gpu_scope = szs.DeviceScope(gpu_device=0)   # pick GPU 0 if available

strings_a = sz.Strs(["kitten", "flaw"])
strings_b = sz.Strs(["sitting", "lawn"])

strings_a = szs.to_device(strings_a) # optional ahead of time transfer
strings_b = szs.to_device(strings_b) # optional ahead of time transfer

engine = szs.LevenshteinDistances(
    match=0, mismatch=2,        # costs don't have to be 1
    open=3, extend=1,           # may be different in Bio
    capabilities=("serial",)    # avoid SIMD 🤭
)
distances = engine(strings_a, strings_b, device=cpu_scope)
assert int(distances[0]) == 3 and int(distances[1]) == 2

Note, that this computes byte-level distances. For UTF-8 codepoints, use a different engine class:

strings_a = sz.Strs(["café", "αβγδ"])
strings_b = sz.Strs(["cafe", "αγδ"])
engine = szs.LevenshteinDistancesUTF8(capabilities=("serial",))
distances = engine(strings_a, strings_b, device=cpu_scope)
assert int(distances[0]) == 1 and int(distances[1]) == 1

For alignment scoring provide a 256×256 substitution matrix using NumPy:

import numpy as np
import stringzilla as sz
import stringzillas as szs

substitution_matrix = np.zeros((256, 256), dtype=np.int8)
substitution_matrix.fill(-1)                # mismatch score
np.fill_diagonal(substitution_matrix, 0)    # match score

engine = szs.NeedlemanWunsch(substitution_matrix=substitution_matrix, open=1, extend=1)
scores = engine(strings_a, strings_b, device=cpu_scope)

Several Python libraries provide edit distance computation. Most are implemented in C but may be slower than StringZilla on large inputs. For proteins ~10k chars, 100 pairs:

JellyFish: 62.3s
EditDistance: 32.9s
StringZilla: 0.8s

Using the same proteins for Needleman-Wunsch alignment scores:

BioPython: 25.8s
StringZilla: 7.8s

§ Example converting from BioPython to StringZilla.

import numpy as np
from Bio import Align
from Bio.Align import substitution_matrices

aligner = Align.PairwiseAligner()
aligner.substitution_matrix = substitution_matrices.load("BLOSUM62")
aligner.open_gap_score = 1
aligner.extend_gap_score = 1

# Convert the matrix to NumPy
subs_packed = np.array(aligner.substitution_matrix).astype(np.int8)
subs_reconstructed = np.zeros((256, 256), dtype=np.int8)

# Initialize all banned characters to a the largest possible penalty
subs_reconstructed.fill(127)
for packed_row, packed_row_aminoacid in enumerate(aligner.substitution_matrix.alphabet):
    for packed_column, packed_column_aminoacid in enumerate(aligner.substitution_matrix.alphabet):
        reconstructed_row = ord(packed_row_aminoacid)
        reconstructed_column = ord(packed_column_aminoacid)
        subs_reconstructed[reconstructed_row, reconstructed_column] = subs_packed[packed_row, packed_column]

# Let's pick two examples of tripeptides (made of 3 amino acids)
glutathione = "ECG" # Need to rebuild human tissue?
thyrotropin_releasing_hormone = "QHP" # Or to regulate your metabolism?

import stringzillas as szs
engine = szs.NeedlemanWunsch(substitution_matrix=subs_reconstructed, open=1, extend=1)
score = int(engine(sz.Strs([glutathione]), sz.Strs([thyrotropin_releasing_hormone]))[0])
assert score == aligner.score(glutathione, thyrotropin_releasing_hormone) # Equal to 6

Rolling Fingerprints

MinHashing is a common technique for Information Retrieval, producing compact representations of large documents. For $D$ hash-functions and a text of length $L$, in the worst case it involves computing $O(D \cdot L)$ hashes.

import numpy as np
import stringzilla as sz
import stringzillas as szs

texts = sz.Strs([
    "quick brown fox jumps over the lazy dog",
    "quick brown fox jumped over a very lazy dog",
])

cpu = szs.DeviceScope(cpu_cores=4)
ndim = 1024
window_widths = np.array([4, 6, 8, 10], dtype=np.uint64)
engine = szs.Fingerprints(
    ndim=ndim,
    window_widths=window_widths,    # optional
    alphabet_size=256,              # default for byte strings
    capabilities=("serial",),       # defaults to all, can also pass a `DeviceScope`
)

hashes, counts = engine(texts, device=cpu)
assert hashes.shape == (len(texts), ndim)
assert counts.shape == (len(texts), ndim)
assert hashes.dtype == np.uint32 and counts.dtype == np.uint32

Serialization

Filesystem

Similar to how File can be used to read a large file, other interfaces can be used to dump strings to disk faster. The Str class has write_to to write the string to a file, and offset_within to obtain integer offsets of substring view in larger string for navigation.

web_archive = Str("<html>...</html><html>...</html>")
_, end_tag, next_doc = web_archive.partition("</html>") # or use `find`
next_doc_offset = next_doc.offset_within(web_archive)
web_archive.write_to("next_doc.html") # no GIL, no copies, just a view

PyArrow

A Str is easy to cast to PyArrow buffers.

from pyarrow import foreign_buffer
from stringzilla import Strs

strs = Strs(["alpha", "beta", "gamma"])
arrow = foreign_buffer(strs.address, strs.nbytes, strs)

And only slightly harder to convert in reverse direction:

arr = pa.Array.from_buffers(
    pa.large_string() if strs.offsets_are_large else pa.string(),
    len(strs),
    [None,
     pa.foreign_buffer(strs.offsets_address, strs.offsets_nbytes, strs),
     pa.foreign_buffer(strs.tape_address, strs.tape_nbytes, strs)],
)

That means you can convert Str to pyarrow.Buffer and Strs to pyarrow.Array without extra copies. For more details on the tape-like layouts, refer to the StringTape repository.

Quick Start: C/C++

The C library is header-only, so you can just copy the stringzilla.h header into your project. Same applies to C++, where you would copy the stringzilla.hpp header. Alternatively, add it as a submodule, and include it in your build system.

git submodule add https://github.com/ashvardanian/StringZilla.git external/stringzilla
git submodule update --init --recursive

Or using a pure CMake approach:

FetchContent_Declare(
    stringzilla
    GIT_REPOSITORY https://github.com/ashvardanian/StringZilla.git
    GIT_TAG main  # or specify a version tag
)
FetchContent_MakeAvailable(stringzilla)

Last, but not the least, you can also install it as a library, and link against it. This approach is worse for inlining, but brings dynamic runtime dispatch for the most advanced CPU features.

Basic Usage with C 99 and Newer

There is a stable C 99 interface, where all function names are prefixed with sz_. Most interfaces are well documented, and come with self-explanatory names and examples. In some cases, hardware specific overloads are available, like sz_find_skylake or sz_find_neon. Both are companions of the sz_find, first for x86 CPUs with AVX-512 support, and second for Arm NEON-capable CPUs.

#include <stringzilla/stringzilla.h>

// Initialize your haystack and needle
sz_string_view_t haystack = {your_text, your_text_length};
sz_string_view_t needle = {your_subtext, your_subtext_length};

// Perform string-level operations auto-picking the backend or dispatching manually
sz_cptr_t ptr = sz_find(haystack.start, haystack.length, needle.start, needle.length);
sz_size_t substring_position = ptr ? (sz_size_t)(ptr - haystack.start) : SZ_SIZE_MAX; // SZ_SIZE_MAX if not found

// Backend-specific variants return pointers as well
sz_cptr_t ptr = sz_find_skylake(haystack.start, haystack.length, needle.start, needle.length);
sz_cptr_t ptr = sz_find_haswell(haystack.start, haystack.length, needle.start, needle.length);
sz_cptr_t ptr = sz_find_westmere(haystack.start, haystack.length, needle.start, needle.length);
sz_cptr_t ptr = sz_find_neon(haystack.start, haystack.length, needle.start, needle.length);

// Hash strings at once
sz_u64_t hash = sz_hash(haystack.start, haystack.length, 42);    // 42 is the seed
sz_u64_t checksum = sz_bytesum(haystack.start, haystack.length); // or accumulate byte values

// Hash strings incrementally with "init", "update", and "digest":
sz_hash_state_t state;
sz_hash_state_init(&state, 42);
sz_hash_state_update(&state, haystack.start, 1);                        // first char
sz_hash_state_update(&state, haystack.start + 1, haystack.length - 1);  // rest of the string
sz_u64_t streamed_hash = sz_hash_state_digest(&state);

// SHA-256 cryptographic checksums
sz_u8_t digest[32];
sz_sha256_state_t sha_state;
sz_sha256_state_init(&sha_state);
sz_sha256_state_update(&sha_state, haystack.start, haystack.length);
sz_sha256_state_digest(&sha_state, digest);

// Perform collection level operations
sz_sequence_t array = {your_handle, your_count, your_get_start, your_get_length};
sz_sorted_idx_t order[your_count];
sz_sequence_argsort(&array, NULL, order); // NULL allocator uses default

§ Mapping from LibC to StringZilla.

By design, StringZilla has a couple of notable differences from LibC:

all strings are expected to have a length, and are not necessarily null-terminated.
every operations has a reverse order counterpart.

That way sz_find and sz_rfind are similar to strstr and strrstr in LibC. Similarly, sz_find_byte and sz_rfind_byte replace memchr and memrchr. The sz_find_byteset maps to strspn and strcspn, while sz_rfind_byteset has no sibling in LibC.

LibC Functionality	StringZilla Equivalents
`memchr(haystack, needle, haystack_length)`, `strchr`	`sz_find_byte(haystack, haystack_length, needle)`
`memrchr(haystack, needle, haystack_length)`	`sz_rfind_byte(haystack, haystack_length, needle)`
`memcmp`, `strcmp`	`sz_order`, `sz_equal`
`strlen(haystack)`	`sz_find_byte(haystack, haystack_length, needle)`
`strcspn(haystack, reject)`	`sz_find_byteset(haystack, haystack_length, reject_bitset)`
`strspn(haystack, accept)`	`sz_find_byte_not_from(haystack, haystack_length, accept, accept_length)`
`memmem(haystack, haystack_length, needle, needle_length)`, `strstr`	`sz_find(haystack, haystack_length, needle, needle_length)`
`memcpy(destination, source, destination_length)`	`sz_copy(destination, source, destination_length)`
`memmove(destination, source, destination_length)`	`sz_move(destination, source, destination_length)`
`memset(destination, value, destination_length)`	`sz_fill(destination, destination_length, value)`

Basic Usage with C++ 11 and Newer

There is a stable C++ 11 interface available in the ashvardanian::stringzilla namespace. It comes with two STL-like classes: string_view and string. The first is a non-owning view of a string, and the second is a mutable string with a Small String Optimization.

#include <stringzilla/stringzilla.hpp>

namespace sz = ashvardanian::stringzilla;

sz::string haystack = "some string";
sz::string_view needle = sz::string_view(haystack).substr(0, 4);

auto substring_position = haystack.find(needle); // Or `rfind`
auto hash = std::hash<sz::string_view>{}(haystack); // Compatible with STL's `std::hash`

haystack.end() - haystack.begin() == haystack.size(); // Or `rbegin`, `rend`
haystack.find_first_of(" \v\t") == 4; // Or `find_last_of`, `find_first_not_of`, `find_last_not_of`
haystack.starts_with(needle) == true; // Or `ends_with`
haystack.remove_prefix(needle.size()); // Why is this operation in-place?!
haystack.contains(needle) == true; // STL has this only from C++ 23 onwards
haystack.compare(needle) == 1; // Or `haystack <=> needle` in C++ 20 and beyond

StringZilla also provides string literals for automatic type resolution, similar to STL:

using sz::literals::operator""_sv;
using std::literals::operator""sv;

auto a = "some string"; // char const *
auto b = "some string"sv; // std::string_view
auto b = "some string"_sv; // sz::string_view

Unicode Case-Folding and Case-Insensitive Search

char source[] = "Straße";  // German: "Street"
char destination[64];      // Must be at least 3x source length
sz_size_t result_len = sz_utf8_case_fold(source, strlen(source), destination);
// destination now contains "strasse" (7 bytes), result_len = 7

The case-insensitive search API returns a pointer to the start of the first relevant glyph in the haystack, or NULL if not found. It outputs the length of the matched haystack substring in bytes, and accepts a metadata structure to speed up repeated searches for the same needle.

sz_utf8_case_insensitive_needle_metadata_t metadata = {};
sz_size_t match_length;
sz_cptr_t match = sz_utf8_case_insensitive_find(
    haystack, haystack_len,
    needle, needle_len,
    &metadata,      // Reuse for queries with the same needle
    &match_length   // Output: bytes consumed in haystack
);

Same functionality is available in C++:

namespace sz = ashvardanian::stringzilla;

sz::string_view text = "Hello World"; // Single search
auto [offset, length] = text.utf8_case_insensitive_find("HELLO");

sz::utf8_case_insensitive_needle pattern("hello"); // Repeated searches with pre-compiled pattern
for (auto const& haystack : haystacks)
    auto match = haystack.utf8_case_insensitive_find(pattern);

Similarity Scores

StringZilla exposes high-performance, batch-oriented similarity via the stringzillas/stringzillas.h header. Use szs_device_scope_t to pick hardware and optionally limit capabilities per engine.

#include <stringzillas/stringzillas.h>

szs_device_scope_t device = NULL;
szs_device_scope_init_default(&device);

szs_levenshtein_distances_t engine = NULL;
szs_levenshtein_distances_init(0, 1, 1, 1, /*alloc*/ NULL, /*caps*/ sz_cap_serial_k, &engine);

sz_sequence_u32tape_t strings_a {data_a, offsets_a, count}; // or `sz_sequence_u64tape_t` for large inputs
sz_sequence_u32tape_t strings_b {data_b, offsets_b, count}; // or `sz_sequence_t` to pass generic containers

sz_size_t distances[count];
szs_levenshtein_distances_u32tape(engine, device, &strings_a, &strings_b, distances, sizeof(distances[0]));

szs_levenshtein_distances_free(engine);
szs_device_scope_free(device);

To target a different device, use the appropriate szs_device_scope_init_{cpu_cores,gpu_device} function. When dealing with GPU backends, make sure to use the "unified memory" allocators exposed as szs_unified_{alloc,free}. Similar stable C ABIs are exposed for other workloads as well.

UTF-8: szs_levenshtein_distances_utf8_{sequence,u32tape,u64tape}
Needleman-Wunsch: szs_needleman_wunsch_scores_{sequence,u32tape,u64tape}
Smith-Waterman: szs_smith_waterman_scores_{sequence,u32tape,u64tape}

Moreover, in C++ codebases one can tap into the raw templates implementing that functionality, customizing them with custom executors, SIMD plugins, etc. For that include stringzillas/similarities.hpp for C++ and stringzillas/similarities.cuh for CUDA.

#include <stringzillas/similarities.hpp>
#include <stringzilla/types.hpp>       // tape of strings
#include <fork_union.hpp>              // optional thread pool

namespace sz = ashvardanian::stringzilla;
namespace szs = ashvardanian::stringzillas;

// Pack strings into an Arrow-like tape
std::vector<std::string> left = {"kitten", "flaw"};
std::vector<std::string> right = {"sitting", "lawn"};
sz::arrow_strings_tape<char, sz::size_t, std::allocator<char>> tape_a, tape_b;
auto _ = tape_a.try_assign(left.begin(), left.end());
auto _ = tape_b.try_assign(right.begin(), right.end());

// Run on the current thread
using levenshtein_t = szs::levenshtein_distances<char, szs::linear_gap_costs_t, std::allocator<char>, sz_cap_serial_k>;
levenshtein_t engine {szs::uniform_substitution_costs_t{0,1}, szs::linear_gap_costs_t{1}};
std::size_t distances[2];
auto _ = engine(tape_a, tape_b, distances);

// Or run in parallel with a pool
fork_union::basic_pool_t pool;
auto _ = pool.try_spawn(std::thread::hardware_concurrency());
auto _ = engine(tape_a, tape_b, distances, pool);

All of the potentially failing StringZillas' interfaces return error codes, and none raise C++ exceptions. Parallelism is enabled at both collection-level and within individual pairs of large inputs.

Rolling Fingerprints

StringZilla exposes parallel fingerprinting (Min-Hashes or Count-Min-Sketches) via the stringzillas/stringzillas.h header. Use szs_device_scope_t to pick hardware and optionally limit capabilities per engine.

#include <stringzillas/stringzillas.h>

szs_device_scope_t device = NULL;
szs_device_scope_init_default(&device);

szs_fingerprints_t engine = NULL;
sz_size_t const dims = 1024; sz_size_t const window_widths[] = {4, 6, 8, 10};
szs_fingerprints_init(dims, /*alphabet*/ 256, window_widths, 4, /*alloc*/ NULL, /*caps*/ sz_cap_serial_k, &engine);

sz_sequence_u32tape_t texts = {data, offsets, count};
sz_u32_t *min_hashes = (sz_u32_t*)szs_unified_alloc(count * dims * sizeof(*min_hashes));
sz_u32_t *min_counts = (sz_u32_t*)szs_unified_alloc(count * dims * sizeof(*min_counts));
szs_fingerprints_u32tape(engine, device, &texts,
    min_hashes, dims * sizeof(*min_hashes),     // support strided matrices
    min_counts, dims * sizeof(*min_counts));    // for both output arguments

szs_fingerprints_free(engine);
szs_device_scope_free(device);

Moreover, in C++ codebases one can tap into the raw templates implementing that functionality, customizing them with custom executors, SIMD plugins, etc. For that include stringzillas/fingerprints.hpp for C++ and stringzillas/fingerprints.cuh for CUDA.

#include <stringzillas/fingerprints.hpp>
#include <stringzilla/types.hpp>       // tape of strings
#include <fork_union.hpp>              // optional thread pool

namespace sz = ashvardanian::stringzilla;
namespace szs = ashvardanian::stringzillas;

// Pack strings into an Arrow-like tape
std::vector<std::string> docs = {"alpha beta", "alpha betta"};
sz::arrow_strings_tape<char, sz::size_t, std::allocator<char>> tape;
auto _ = tape.try_assign(docs.begin(), docs.end());

// Run on the current thread with a Rabin-Karp family hasher
constexpr std::size_t dimensions_k = 256;
constexpr std::size_t window_width_k = 7;
using row_t = std::array<sz_u32_t, 256>;
using fingerprinter_t = szs::floating_rolling_hashers<sz_cap_serial_k, dimensions_k>;
fingerprinter_t engine;
auto _ = engine.try_extend(window_width_k, dimensions_k);
std::vector<row_t> hashes(docs.size()), counts(docs.size());
auto _ = engine(tape, hashes, counts);

// Or run in parallel with a pool
fork_union::basic_pool_t pool;
auto _ = pool.try_spawn(std::thread::hardware_concurrency());
auto _ = engine(tape, hashes, counts, pool);

CUDA

StringZilla provides CUDA C++ templates for composable string batch-processing operations. Different GPUs have varying warp sizes, shared memory capacities, and register counts, affecting algorithm selection, so it's important to query the gpu_specs_t via gpu_specs_fetch. For memory management, ensure that you use GPU-visible' unified memoryexposed in an STL-compatible manner as aunified_alloctemplate class. For error handling,cuda_status_textends the traditionalstatus_twith GPU-specific information. It's implicitly convertible tostatus_t, so you can use it in places expecting a status_t`.

Most algorithms can load-balance both a large number of small strings and a small number of large strings. Still, with large H100-scale GPUs, it's best to submit thousands of inputs at once.

Memory Ownership and Small String Optimization

Most operations in StringZilla don't assume any memory ownership. But in addition to the read-only search-like operations StringZilla provides a minimalistic C and C++ implementations for a memory owning string "class". Like other efficient string implementations, it uses the Small String Optimization (SSO) to avoid heap allocations for short strings.

typedef union sz_string_t {
    struct internal {
        sz_ptr_t start;
        sz_u8_t length;
        char chars[SZ_STRING_INTERNAL_SPACE]; /// Ends with a null-terminator.
    } internal;

    struct external {
        sz_ptr_t start;
        sz_size_t length;        
        sz_size_t space; /// The length of the heap-allocated buffer.
        sz_size_t padding;
    } external;

} sz_string_t;

As one can see, a short string can be kept on the stack, if it fits within internal.chars array. Before 2015 GCC string implementation was just 8 bytes, and could only fit 7 characters. Different STL implementations today have different thresholds for the Small String Optimization. Similar to GCC, StringZilla is 32 bytes in size, and similar to Clang it can fit 22 characters on stack. Our layout might be preferential, if you want to avoid branches. If you use a different compiler, you may want to check its SSO buffer size with a simple Gist.

	`libstdc++` in GCC 13	`libc++` in Clang 17	StringZilla
String `sizeof`	32	24	32
Inner Capacity	15	22	22

This design has been since ported to many high-level programming languages. Swift, for example, can store 15 bytes in the String instance itself. StringZilla implements SSO at the C level, providing the sz_string_t union and a simple API for primary operations.

sz_memory_allocator_t allocator;
sz_string_t string;

// Init and make sure we are on stack
sz_string_init(&string);
sz_string_is_on_stack(&string); // == sz_true_k

// Optionally pre-allocate space on the heap for future insertions.
sz_string_grow(&string, 100, &allocator); // == sz_true_k

// Append, erase, insert into the string.
sz_string_expand(&string, 0, "_Hello_", 7, &allocator); // == sz_true_k
sz_string_expand(&string, SZ_SIZE_MAX, "world", 5, &allocator); // == sz_true_k
sz_string_erase(&string, 0, 1);

// Unpacking & introspection.
sz_ptr_t string_start;
sz_size_t string_length;
sz_size_t string_space;
sz_bool_t string_is_external;
sz_string_unpack(string, &string_start, &string_length, &string_space, &string_is_external);
sz_equal(string_start, "Hello_world", 11); // == sz_true_k

// Reclaim some memory.
sz_string_shrink_to_fit(&string, &allocator); // == sz_true_k
sz_string_free(&string, &allocator);

Unlike the conventional C strings, the sz_string_t is allowed to contain null characters. To safely print those, pass the string_length to printf as well.

printf("%.*s\n", (int)string_length, string_start);

What's Wrong with the C Standard Library?

StringZilla is not a drop-in replacement for the C Standard Library. It's designed to be a safer and more modern alternative. Conceptually:

LibC strings are expected to be null-terminated, so to use the efficient LibC implementations on slices of larger strings, you'd have to copy them, which is more expensive than the original string operation.
LibC functionality is asymmetric - you can find the first and the last occurrence of a character within a string, but you can't find the last occurrence of a substring.
LibC function names are typically very short and cryptic.
LibC lacks crucial functionality like hashing and doesn't provide primitives for less critical but relevant operations like fuzzy matching.

Something has to be said about its support for UTF-8. Aside from a single-byte char type, LibC provides wchar_t:

The size of wchar_t is not consistent across platforms. On Windows, it's typically 16 bits (suitable for UTF-16), while on Unix-like systems, it's usually 32 bits (suitable for UTF-32). This inconsistency can lead to portability issues when writing cross-platform code.
wchar_t is designed to represent wide characters in a fixed-width format (UTF-16 or UTF-32). In contrast, UTF-8 is a variable-length encoding, where each character can take from 1 to 4 bytes. This fundamental difference means that wchar_t and UTF-8 are incompatible.

StringZilla partially addresses those issues.

What's Wrong with the C++ Standard Library?

C++ Code	Evaluation Result	Invoked Signature
`"Loose"s.replace(2, 2, "vath"s, 1)`	`"Loathe"` 🤢	`(pos1, count1, str2, pos2)`
`"Loose"s.replace(2, 2, "vath", 1)`	`"Love"` 🥰	`(pos1, count1, str2, count2)`

StringZilla is designed to be a drop-in replacement for the C++ Standard Templates Library. That said, some of the design decisions of STL strings are highly controversial, error-prone, and expensive. Most notably:

Argument order for replace, insert, erase and similar functions is impossible to guess.
Bounds-checking exceptions for substr-like functions are only thrown for one side of the range.
Returning string copies in substr-like functions results in absurd volume of allocations.
Incremental construction via push_back-like functions goes through too many branches.
Inconsistency between string and string_view methods, like the lack of remove_prefix and remove_suffix.

Check the following set of asserts validating the std::string specification. It's not realistic to expect the average developer to remember the 14 overloads of std::string::replace.

using str = std::string;

assert(str("hello world").substr(6) == "world");
assert(str("hello world").substr(6, 100) == "world"); // 106 is beyond the length of the string, but its OK
assert_throws(str("hello world").substr(100), std::out_of_range);   // 100 is beyond the length of the string
assert_throws(str("hello world").substr(20, 5), std::out_of_range); // 20 is beyond the length of the string
assert_throws(str("hello world").substr(-1, 5), std::out_of_range); // -1 casts to unsigned without any warnings...
assert(str("hello world").substr(0, -1) == "hello world");          // -1 casts to unsigned without any warnings...

assert(str("hello").replace(1, 2, "123") == "h123lo");
assert(str("hello").replace(1, 2, str("123"), 1) == "h23lo");
assert(str("hello").replace(1, 2, "123", 1) == "h1lo");
assert(str("hello").replace(1, 2, "123", 1, 1) == "h2lo");
assert(str("hello").replace(1, 2, str("123"), 1, 1) == "h2lo");
assert(str("hello").replace(1, 2, 3, 'a') == "haaalo");
assert(str("hello").replace(1, 2, {'a', 'b'}) == "hablo");

To avoid those issues, StringZilla provides an alternative consistent interface. It supports signed arguments, and doesn't have more than 3 arguments per function or The standard API and our alternative can be conditionally disabled with SZ_SAFETY_OVER_COMPATIBILITY=1. When it's enabled, the ~~subjectively~~ risky overloads from the Standard will be disabled.

using str = sz::string;

str("a:b").front(1) == "a"; // no checks, unlike `substr`
str("a:b").front(2) == "2"; // take first 2 characters
str("a:b").back(-1) == "b"; // accepting negative indices
str("a:b").back(-2) == ":b"; // similar to Python's `"a:b"[-2:]`
str("a:b").sub(1, -1) == ":"; // similar to Python's `"a:b"[1:-1]`
str("a:b").sub(-2, -1) == ":"; // similar to Python's `"a:b"[-2:-1]`
str("a:b").sub(-2, 1) == ""; // similar to Python's `"a:b"[-2:1]`
"a:b"_sv[{-2, -1}] == ":"; // works on views and overloads `operator[]`

Assuming StringZilla is a header-only library you can use the full API in some translation units and gradually transition to safer restricted API in others. Bonus - all the bound checking is branchless, so it has a constant cost and won't hurt your branch predictor.

Beyond the C++ Standard Library - Learning from Python

Python is arguably the most popular programming language for data science. In part, that's due to the simplicity of its standard interfaces. StringZilla brings some of that functionality to C++.

Content checks: isalnum, isalpha, isascii, isdigit, islower, isspace, isupper.
Trimming character sets: lstrip, rstrip, strip.
Trimming string matches: remove_prefix, remove_suffix.
Ranges of search results: splitlines, split, rsplit.
Number of non-overlapping substring matches: count.
Partitioning: partition, rpartition.

For example, when parsing documents, it is often useful to split it into substrings. Most often, after that, you would compute the length of the skipped part, the offset and the length of the remaining part. This results in a lot of pointer arithmetic and is error-prone. StringZilla provides a convenient partition function, which returns a tuple of three string views, making the code cleaner.

auto parts = haystack.partition(':'); // Matching a character
auto [before, match, after] = haystack.partition(':'); // Structure unpacking
auto [before, match, after] = haystack.partition(sz::byteset(":;")); // Character-set argument
auto [before, match, after] = haystack.partition(" : "); // String argument
auto [before, match, after] = haystack.rpartition(sz::whitespaces_set()); // Split around the last whitespace

Combining those with the split function, one can easily parse a CSV file or HTTP headers.

for (auto line : haystack.split("\r\n")) {
    auto [key, _, value] = line.partition(':');
    headers[key.strip()] = value.strip();
}

Some other extensions are not present in the Python standard library either. Let's go through the C++ functionality category by category.

Splits and Ranges.
Concatenating Strings without Allocations.
Random Generation.
Edit Distances and Fuzzy Search.

Some of the StringZilla interfaces are not available even Python's native str class. Here is a sneak peek of the most useful ones.

text.hash(); // -> 64 bit unsigned integer 
text.ssize(); // -> 64 bit signed length to avoid `static_cast<std::ssize_t>(text.size())`
text.contains_only(" \w\t"); // == text.find_first_not_of(sz::byteset(" \w\t")) == npos;
text.contains(sz::whitespaces_set()); // == text.find(sz::byteset(sz::whitespaces_set())) != npos;

// Simpler slicing than `substr`
text.front(10); // -> sz::string_view
text.back(10); // -> sz::string_view

// Safe variants, which clamp the range into the string bounds
using sz::string::cap;
text.front(10, cap) == text.front(std::min(10, text.size()));
text.back(10, cap) == text.back(std::min(10, text.size()));

// Character set filtering
text.lstrip(sz::whitespaces_set()).rstrip(sz::newlines_set()); // like Python
text.front(sz::whitespaces_set()); // all leading whitespaces
text.back(sz::digits_set()); // all numerical symbols forming the suffix

// Incremental construction
using sz::string::unchecked;
text.push_back('x'); // no surprises here
text.push_back('x', unchecked); // no bounds checking, Rust style
text.try_push_back('x'); // returns `false` if the string is full and the allocation failed

sz::concatenate(text, "@", domain, ".", tld); // No allocations

Splits and Ranges

One of the most common use cases is to split a string into a collection of substrings. Which would often result in StackOverflow lookups and snippets like the one below.

std::vector<std::string> lines = split(haystack, "\r\n"); // string delimiter
std::vector<std::string> words = split(lines, ' '); // character delimiter

Those allocate memory for each string and the temporary vectors. Each allocation can be orders of magnitude more expensive, than even serial for-loop over characters. To avoid those, StringZilla provides lazily-evaluated ranges, compatible with the Range-v3 library.

for (auto line : haystack.split("\r\n"))
    for (auto word : line.split(sz::byteset(" \w\t.,;:!?")))
        std::cout << word << std::endl;

Each of those is available in reverse order as well. It also allows interleaving matches, if you want both inclusions of xx in xxx. Debugging pointer offsets is not a pleasant exercise, so keep the following functions in mind.

haystack.[r]find_all(needle, interleaving)
haystack.[r]find_all(sz::byteset(""))
haystack.[r]split(needle)
haystack.[r]split(sz::byteset(""))

For $N$ matches the split functions will report $N+1$ matches, potentially including empty strings. Ranges have a few convenience methods as well:

range.size(); // -> std::size_t
range.empty(); // -> bool
range.template to<std::set<std::sting>>(); 
range.template to<std::vector<std::sting_view>>();

Concatenating Strings without Allocations

Another common string operation is concatenation. The STL provides std::string::operator+ and std::string::append, but those are not very efficient, if multiple invocations are performed.

std::string name, domain, tld;
auto email = name + "@" + domain + "." + tld; // 4 allocations

The efficient approach would be to pre-allocate the memory and copy the strings into it.

std::string email;
email.reserve(name.size() + domain.size() + tld.size() + 2);
email.append(name), email.append("@"), email.append(domain), email.append("."), email.append(tld);

That's mouthful and error-prone. StringZilla provides a more convenient concatenate function, which takes a variadic number of arguments. It also overrides the operator| to concatenate strings lazily, without any allocations.

auto email = sz::concatenate(name, "@", domain, ".", tld);   // 0 allocations
auto email = name | "@" | domain | "." | tld;                // 0 allocations
sz::string email = name | "@" | domain | "." | tld;          // 1 allocations

Random Generation

Software developers often need to generate random strings for testing purposes. The STL provides std::generate and std::random_device, that can be used with StringZilla.

sz::string random_string(std::size_t length, char const *alphabet, std::size_t cardinality) {
    sz::string result(length, '\0');
    static std::random_device seed_source; // Expensive to construct - due to system calls
    static std::mt19937 generator(seed_source()); // Also expensive - due to the state size
    std::uniform_int_distribution<std::size_t> distribution(0, cardinality);
    std::generate(result.begin(), result.end(), [&]() { return alphabet[distribution(generator)]; });
    return result;
}

Mouthful and slow. StringZilla provides a C native method - sz_fill_random and a convenient C++ wrapper - sz::generate. Similar to Python it also defines the commonly used character sets.

auto protein = sz::string::random(300, "ARNDCQEGHILKMFPSTWYV"); // static method
auto dna = sz::basic_string<custom_allocator>::random(3_000_000_000, "ACGT");

dna.fill_random("ACGT"); // `noexcept` pre-allocated version
dna.fill_random(&std::rand, "ACGT"); // pass any generator, like `std::mt19937`

char uuid[36];
sz::fill_random(sz::string_span(uuid, 36), "0123456789abcdef-"); // Overwrite any buffer

Bulk Replacements

In text processing, it's often necessary to replace all occurrences of a specific substring or set of characters within a string. Standard library functions may not offer the most efficient or convenient methods for performing bulk replacements, especially when dealing with large strings or performance-critical applications.

haystack.replace_all(needle_string, replacement_string)
haystack.replace_all(sz::byteset(""), replacement_string)
haystack.try_replace_all(needle_string, replacement_string)
haystack.try_replace_all(sz::byteset(""), replacement_string)
haystack.lookup(sz::look_up_table::identity())
haystack.lookup(sz::look_up_table::identity(), haystack.data())

Sorting in C and C++

LibC provides qsort and STL provides std::sort. Both have their quirks. The LibC standard has no way to pass a context to the comparison function, that's only possible with platform-specific extensions. Those have different arguments order on every OS.

// Linux: https://linux.die.net/man/3/qsort_r
void qsort_r(void *elements, size_t count, size_t element_width, 
    int (*compare)(void const *left, void const *right, void *context),
    void *context);
// macOS and FreeBSD: https://developer.apple.com/library/archive/documentation/System/Conceptual/ManPages_iPhoneOS/man3/qsort_r.3.html
void qsort_r(void *elements, size_t count, size_t element_width, 
    void *context,
    int (*compare)(void *context, void const *left, void const *right));
// Windows conflicts with ISO `qsort_s`: https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/qsort-s?view=msvc-170
void qsort_s(id *elements, size_t count, size_t element_width, 
    int (*compare)(void *context, void const *left, void const *right),
    void *context);

C++ generic algorithm is not perfect either. There is no guarantee in the standard that std::sort won't allocate any memory. If you are running on embedded, in real-time or on 100+ CPU cores per node, you may want to avoid that. StringZilla doesn't solve the general case, but hopes to improve the performance for strings. Use sz_sequence_argsort, or the high-level sz::argsort, which can be used sort any collection of elements convertible to sz::string_view.

std::vector<std::string> data({"c", "b", "a"});
std::vector<std::size_t> order = sz::argsort(data); //< Simple shortcut

// Or, taking care of memory allocation:
sz::argsort(data.begin(), data.end(), order.data(), [](auto const &x) -> sz::string_view { return x; });

Standard C++ Containers with String Keys

The C++ Standard Templates Library provides several associative containers, often used with string keys.

std::map<std::string, int, std::less<std::string>> sorted_words;
std::unordered_map<std::string, int, std::hash<std::string>, std::equal_to<std::string>> words;

The performance of those containers is often limited by the performance of the string keys, especially on reads. StringZilla can be used to accelerate containers with std::string keys, by overriding the default comparator and hash functions.

std::map<std::string, int, sz::less> sorted_words;
std::unordered_map<std::string, int, sz::hash, sz::equal_to> words;

Alternatively, a better approach would be to use the sz::string class as a key. The right hash function and comparator would be automatically selected and the performance gains would be more noticeable if the keys are short.

std::map<sz::string, int> sorted_words;
std::unordered_map<sz::string, int> words;

Compilation Settings and Debugging

SZ_DEBUG:

For maximal performance, the C library does not perform any bounds checking in Release builds. In C++, bounds checking happens only in places where the STL std::string would do it. If you want to enable more aggressive bounds-checking, define SZ_DEBUG before including the header. If not explicitly set, it will be inferred from the build type.

SZ_USE_GOLDMONT, SZ_USE_WESTMERE, SZ_USE_HASWELL, SZ_USE_SKYLAKE, SZ_USE_ICE, SZ_USE_NEON, SZ_USE_NEON_AES, SZ_USE_NEON_SHA, SZ_USE_SVE, SZ_USE_SVE2, SZ_USE_SVE2_AES:

One can explicitly disable certain families of SIMD instructions for compatibility purposes. Default values are inferred at compile time depending on compiler support (for dynamic dispatch) and the target architecture (for static dispatch).

SZ_USE_CUDA, SZ_USE_KEPLER, SZ_USE_HOPPER:

One can explicitly disable certain families of PTX instructions for compatibility purposes. Default values are inferred at compile time depending on compiler support (for dynamic dispatch) and the target architecture (for static dispatch).

SZ_ENFORCE_SVE_OVER_NEON:

SVE and SVE2 are expected to supersede NEON on ARM architectures. Still, oftentimes the equivalent SVE kernels are slower due to equally small register files and higher complexity of the instructions. By default, when both SVE and NEON are available, SVE is used selectively only for the algorithms that benefit from it. If you want to enforce SVE usage everywhere, define this flag.

SZ_DYNAMIC_DISPATCH:

By default, StringZilla is a header-only library. But if you are running on different generations of devices, it makes sense to pre-compile the library for all supported generations at once, and dispatch at runtime. This flag does just that and is used to produce the stringzilla.so shared library, as well as the Python bindings.

SZ_USE_MISALIGNED_LOADS:

Default is platform-dependent: enabled on x86 (where unaligned accesses are fast), disabled on others by default. When enabled, many byte-level operations use word-sized loads, which can significantly accelerate the serial (SWAR) backend. Consider enabling it explicitly if you are targeting platforms that support fast unaligned loads.

SZ_AVOID_LIBC and SZ_OVERRIDE_LIBC:

When using the C header-only library one can disable the use of LibC. This may affect the type resolution system on obscure hardware platforms. Moreover, one may let stringzilla override the common symbols like the memcpy and memset with its own implementations. In that case you can use the LD_PRELOAD trick to prioritize its symbols over the ones from the LibC and accelerate existing string-heavy applications without recompiling them. It also adds a layer of security, as the stringzilla isn't undefined for NULL inputs like memcpy(NULL, NULL, 0).

SZ_AVOID_STL and SZ_SAFETY_OVER_COMPATIBILITY:

When using the C++ interface one can disable implicit conversions from std::string to sz::string and back. If not needed, the <string> and <string_view> headers will be excluded, reducing compilation time. Moreover, if STL compatibility is a low priority, one can make the API safer by disabling the overloads, which are subjectively error prone.

STRINGZILLA_BUILD_SHARED, STRINGZILLA_BUILD_TEST, STRINGZILLA_BUILD_BENCHMARK, STRINGZILLA_TARGET_ARCH for CMake users:

When compiling the tests and benchmarks, you can explicitly set the target hardware architecture. It's synonymous to GCC's -march flag and is used to enable/disable the appropriate instruction sets. You can also disable the shared library build, if you don't need it.

Quick Start: Rust

StringZilla is available as a Rust crate, with documentation available on docs.rs/stringzilla. You can immediately check the installed version and the used hardware capabilities with following commands:

cargo add stringzilla
cargo run --example version

To use the latest crate release in your project, add the following to your Cargo.toml:

[dependencies]
stringzilla = ">=3"                                     # for serial algorithms
stringzilla = { version = ">=3", features = ["cpus"] }  # for parallel multi-CPU backends
stringzilla = { version = ">=3", features = ["cuda"] }  # for parallel Nvidia GPU backend

Or if you want to use the latest pre-release version from the repository:

[dependencies]
stringzilla = { git = "https://github.com/ashvardanian/stringzilla", branch = "main-dev" }

Once installed, all of the functionality is available through the stringzilla namespace. Many interfaces will look familiar to the users of the memchr Rust crate.

use stringzilla::sz;

// Identical to `memchr::memmem::find` and `memchr::memmem::rfind` functions
sz::find("Hello, world!", "world") // 7
sz::rfind("Hello, world!", "world") // 7

// Generalizations of `memchr::memrchr[123]`
sz::find_byte_from("Hello, world!", "world") // 2
sz::rfind_byte_from("Hello, world!", "world") // 11

It also provides no constraints on the size of the character set, while memchr allows only 1, 2, or 3 characters. In addition to global functions, stringzilla provides a StringZilla extension trait:

use stringzilla::StringZilla;

let my_string: String = String::from("Hello, world!");
let my_str = my_string.as_str();
let my_cow_str = Cow::from(&my_string);

// Use the generic function with a String
assert_eq!(my_string.sz_find("world"), Some(7));
assert_eq!(my_string.sz_rfind("world"), Some(7));
assert_eq!(my_string.sz_find_byte_from("world"), Some(2));
assert_eq!(my_string.sz_rfind_byte_from("world"), Some(11));
assert_eq!(my_string.sz_find_byte_not_from("world"), Some(0));
assert_eq!(my_string.sz_rfind_byte_not_from("world"), Some(12));

// Same works for &str and Cow<'_, str>
assert_eq!(my_str.sz_find("world"), Some(7));
assert_eq!(my_cow_str.as_ref().sz_find("world"), Some(7));

Hash

Single-shot and incremental hashing are both supported:

let mut hasher = sz::Hasher::new(42);
hasher.write(b"Hello, ");
hasher.write(b"world!");
let streamed = hasher.finish();

let mut hasher = sz::Hasher::new(42);
hasher.write(b"Hello, world!");
assert_eq!(streamed, hasher.finish());

To use StringZilla with std::collections:

use std::collections::HashMap;
let mut map: HashMap<&str, i32, sz::BuildSzHasher> =
    HashMap::with_hasher(sz::BuildSzHasher::with_seed(42));
map.insert("a", 1);
assert_eq!(map.get("a"), Some(&1));

SHA-256 Checksums

SHA-256 cryptographic checksums are available:

use stringzilla::sz;

// One-shot SHA-256
let digest = sz::Sha256::hash(b"Hello, world!");
assert_eq!(digest.len(), 32);

// Incremental SHA-256
let mut hasher = sz::Sha256::new();
hasher.update(b"Hello, ");
hasher.update(b"world!");
let digest = hasher.digest();

// HMAC-SHA256 for message authentication
let mac = sz::hmac_sha256(b"secret", b"Hello, world!");

Unicode Case-Folding and Case-Insensitive Search

use stringzilla::stringzilla as sz;

let source = "Straße";           // German: "Street"
let mut dest = [0u8; 64];        // Must be at least 3x source length
let len = sz::utf8_case_fold(source, &mut dest);
assert_eq!(&dest[..len], b"strasse");  // ß (2 bytes) → "ss" (2 bytes)

The case-insensitive search returns Some((offset, matched_length)) or None. The matched_length may differ from needle length due to expansions.

use stringzilla::stringzilla::{utf8_case_insensitive_find, Utf8CaseInsensitiveNeedle};

// Single search — ß (C3 9F) matches "SS"
if let Some((offset, len)) = utf8_case_insensitive_find("Straße", "STRASSE") {
    assert_eq!(offset, 0);
    assert_eq!(len, 7);  // "Straße" is 7 bytes
}

// Repeated searches with pre-compiled needle metadata
let needle = Utf8CaseInsensitiveNeedle::new(b"STRASSE");
for haystack in &["Straße", "STRASSE", "strasse"] {
    if let Some((offset, len)) = utf8_case_insensitive_find(haystack, &needle) {
        println!("Found at byte {} with length {}", offset, len);
    }
}

Similarity Scores

StringZilla exposes high-performance, batch-oriented similarity via the szs module. Use DeviceScope to pick hardware and optionally limit capabilities per engine.

use stringzilla::szs; // re-exported as `szs`

let cpu_scope = szs::DeviceScope::cpu_cores(4).unwrap();    // force CPU-only
let gpu_scope = szs::DeviceScope::gpu_device(0).unwrap();   // pick GPU 0 if available
let strings_a = vec!["kitten", "flaw"];
let strings_b = vec!["sitting", "lawn"];

let engine = szs::LevenshteinDistances::new(
    &cpu_scope,
    0,  // match cost
    2,  // mismatch cost - costs don't have to be 1
    3,  // open cost - may be different in Bio
    1,  // extend cost
).unwrap();
let distances = engine.compute(&cpu_scope, &strings_a, &strings_b).unwrap();
assert_eq!(distances[0], 3);
assert_eq!(distances[1], 2);

Note, that this computes byte-level distances. For UTF-8 codepoints, use a different engine class:

let strings_a = vec!["café", "αβγδ"];
let strings_b = vec!["cafe", "αγδ"];
let engine = szs::LevenshteinDistancesUtf8::new(&cpu_scope, 0, 1, 1, 1).unwrap();
let distances = engine.compute(&cpu_scope, &strings_a, &strings_b).unwrap();
assert_eq!(distances, vec![1, 1]);

Similarly, for variable substitution costs, also pass in a a weights matrix:

let mut substitution_matrix = [-1i8; 256 * 256];
for i in 0..256 { substitution_matrix[i * 256 + i] = 0; }
let engine = szs::NeedlemanWunschScores::new(&cpu_scope, &substitution_matrix, -3, -1).unwrap();
let scores = engine.compute(&cpu_scope, &strings_a, &strings_b).unwrap();

Or for local alignment scores:

let engine = szs::SmithWatermanScores::new(&cpu_scope, &substitution_matrix, -3, -1).unwrap();
let local_scores = engine.compute(&cpu_scope, &strings_a, &strings_b).unwrap();

For high-performance applications, use the StringTape crate to pass strings to compute_into methods without extra memory allocations:

use stringzilla::{szs, StringTape};

// Create StringTape from data (zero-copy compatible format)
let tape_a = StringTape::from_strings(&["kitten", "sitting", "flaw"]);
let tape_b = StringTape::from_strings(&["sitting", "kitten", "lawn"]);

// Use unified memory vector for GPU compatibility, initialized with max values
let mut distances = szs::UnifiedVec::<u32>::from_elem(u32::MAX, tape_a.len());

let engine = szs::LevenshteinDistances::new(&gpu_scope, 0, 1, 1, 1).unwrap();
engine.compute_into(&gpu_scope, &tape_a, &tape_b, &mut distances).unwrap();

// Results computed directly into unified memory, accessible from both CPU/GPU
assert_eq!(distances[0], 3);  // kitten -> sitting
assert_eq!(distances[1], 3);  // sitting -> kitten
assert_eq!(distances[2], 2);  // flaw -> lawn

Rolling Fingerprints

use stringzilla::szs;

let texts = vec![
    "quick brown fox jumps over the lazy dog",
    "quick brown fox jumped over a very lazy dog",
];
let cpu = szs::DeviceScope::cpu_cores(4).unwrap();
let ndim = 1024;
let window_widths = vec![4u64, 6, 8, 10];

let engine = szs::Fingerprints::new(
    ndim,           // number of hash functions & dimensions
    &window_widths, // optional predefined window widths
    256,            // default alphabet size for byte strings
    &cpu            // device scope
).unwrap();

let (hashes, counts) = engine.compute(&cpu, &texts).unwrap();
assert_eq!(hashes.len(), texts.len() * ndim);
assert_eq!(counts.len(), texts.len() * ndim);

For zero-copy processing with StringTape format and unified memory:

use stringzilla::{szs, StringTape};

let tape = StringTape::from_strings(&[
    "quick brown fox jumps over the lazy dog",
    "quick brown fox jumped over a very lazy dog",
]);

// Pre-allocate unified memory buffers
let mut hashes = szs::UnifiedVec::<u32>::from_elem(u32::MAX, tape.len() * ndim);
let mut counts = szs::UnifiedVec::<u32>::from_elem(u32::MAX, tape.len() * ndim);

let engine = szs::Fingerprints::new(ndim, &window_widths, 256, &cpu).unwrap();
engine.compute_into(&cpu, &tape, &mut hashes, &mut counts).unwrap();

// Results computed directly into unified memory buffers
assert!(hashes.iter().any(|&h| h != u32::MAX));  // Verify computation occurred
assert!(counts.iter().any(|&c| c != u32::MAX));

Quick Start: JavaScript

Install the Node.js package and use zero-copy Buffer APIs.

npm install stringzilla
node -p "require('stringzilla').default.capabilities" # for CommonJS
node -e "import('stringzilla').then(m=>console.log(m.default.capabilities)).catch(console.error)" # for ESM

import sz from 'stringzilla';

const haystack = Buffer.from('Hello, world!');
const needle = Buffer.from('world');

// Substring search (BigInt offsets)
const firstIndex = sz.find(haystack, needle);      // 7n
const lastIndex = sz.findLast(haystack, needle);   // 7n

// Character / charset search
const firstOIndex = sz.findByte(haystack, 'o'.charCodeAt(0));                 // 4n
const firstVowelIndex = sz.findByteFrom(haystack, Buffer.from('aeiou'));      // 1n
const lastVowelIndex = sz.findLastByteFrom(haystack, Buffer.from('aeiou'));   // 8n

// Counting (optionally overlapping)
const lCount = sz.count(haystack, Buffer.from('l'));                // 3n
const llOverlapCount = sz.count(haystack, Buffer.from('ll'), true); // 1n

// Equality/ordering utilities
const isEqual = sz.equal(Buffer.from('a'), Buffer.from('a'));
const order = sz.compare(Buffer.from('a'), Buffer.from('b')); // -1, 0, or 1

// Other helpers
const byteSum = sz.byteSum(haystack); // sum of bytes as BigInt

Unicode Case-Folding and Case-Insensitive Search

StringZilla provides full Unicode case folding (including expansions like ß → ss, ligatures like ﬁ → fi, and special folds like µ → μ, K → k) and a case-insensitive substring search that accounts for those expansions.

import sz from "stringzilla";

// Case folding (returns a UTF-8 Buffer)
console.log(sz.utf8CaseFold(Buffer.from("Straße")).toString("utf8")); // "strasse"
console.log(sz.utf8CaseFold(Buffer.from("ofﬁce")).toString("utf8"));  // "office" (U+FB01 ligature)

// Case-insensitive substring search (full Unicode case folding)
const text = Buffer.from(
    "Die Temperaturschwankungen im kosmischen Mikrowellenhintergrund sind ein Maß von etwa 20 µK.\n" +
    "Typografisch sieht man auch: ein Maß von etwa 20 μK."
);
const patternBytes = Buffer.from("EIN MASS VON ETWA 20 μK");

const first = sz.utf8CaseInsensitiveFind(text, patternBytes);
console.log(first); // { index: 69n, length: ... } (byte offsets)

// Reuse the same needle efficiently
const pattern = new sz.Utf8CaseInsensitiveNeedle(patternBytes);
const again = pattern.findIn(text);
console.log(again.index === first.index);

Hash

Single-shot and incremental hashing are both supported:

import sz from 'stringzilla';

// One-shot - stable 64-bit output across all platforms!
const hash = sz.hash(Buffer.from('Hello, world!'), 42); // returns BigInt

// Incremental updates - hasher maintains state
const hasher = new sz.Hasher(42); // seed: 42
hasher.update(Buffer.from('Hello, '));
hasher.update(Buffer.from('world!'));
const streamedHash = hasher.digest(); // returns BigInt
console.assert(hash === streamedHash);

SHA-256 Checksums

SHA-256 cryptographic checksums are available:

import sz from 'stringzilla';

// One-shot SHA-256
const digest = sz.sha256(Buffer.from('Hello, world!')); // returns Buffer (32 bytes)

// Incremental SHA-256
const hasher = new sz.Sha256();
hasher.update(Buffer.from('Hello, '));
hasher.update(Buffer.from('world!'));
const digestBuffer = hasher.digest();     // returns Buffer (32 bytes)
const digestHex = hasher.hexdigest();     // returns string (64 hex chars)

Quick Start: Swift

StringZilla can be added as a dependency in the Swift Package Manager. In your Package.swift file, add the following:

dependencies: [
    .package(url: "https://github.com/ashvardanian/stringzilla")
]

The package currently covers only the most basic functionality, but is planned to be extended to cover the full C++ API.

var s = "Hello, world! Welcome to StringZilla. 👋"
s[s.findFirst(substring: "world")!...] // "world! Welcome to StringZilla. 👋"
s[s.findLast(substring: "o")!...] // "o StringZilla. 👋"
s[s.findFirst(characterFrom: "aeiou")!...] // "ello, world! Welcome to StringZilla. 👋"
s[s.findLast(characterFrom: "aeiou")!...] // "a. 👋")
s[s.findFirst(characterNotFrom: "aeiou")!...] // "Hello, world! Welcome to StringZilla. 👋"

Unicode Case-Folding and Case-Insensitive Search

import StringZilla

let folded = "Straße".utf8CaseFoldedBytes()
print(String(decoding: folded, as: UTF8.self)) // "strasse"

let haystack =
    "Die Temperaturschwankungen im kosmischen Mikrowellenhintergrund sind ein Maß von etwa 20 µK.\n"
    + "Typografisch sieht man auch: ein Maß von etwa 20 μK."
let needle = "EIN MASS VON ETWA 20 μK"

if let range = haystack.utf8CaseInsensitiveFind(substring: needle) {
    print(haystack[range]) // "ein Maß von etwa 20 µK"
}

// Reuse the same needle efficiently
let compiledNeedle = Utf8CaseInsensitiveNeedle(needle)
if let range = compiledNeedle.findFirst(in: haystack) {
    print(haystack[range])
}

Hash

StringZilla provides high-performance hashing for Swift strings:

import StringZilla

// One-shot hashing - stable 64-bit output across all platforms!
let hash = "Hello, world!".hash(seed: 42)

// Incremental hashing for streaming data
var hasher = StringZillaHasher(seed: 42)
hasher.update("Hello, ")
hasher.update("world!")
let streamedHash = hasher.digest()
assert(hash == streamedHash)

SHA-256 Checksums

SHA-256 cryptographic checksums are available:

import StringZilla

// One-shot SHA-256
let digest = "Hello, world!".sha256() // returns [UInt8] (32 bytes)

// Incremental SHA-256
var hasher = StringZillaSha256()
hasher.update("Hello, ")
hasher.update("world!")
let digestBytes = hasher.digest()     // [UInt8] (32 bytes)
let digestHex = hasher.hexdigest()    // String (64 hex chars)

Quick Start: GoLang

Add the Go binding as a module dependency:

go get github.com/ashvardanian/stringzilla/golang@latest

Build the shared C library once, then ensure your runtime can locate it (Linux shown):

cmake -B build_shared -D STRINGZILLA_BUILD_SHARED=1 -D CMAKE_BUILD_TYPE=Release
cmake --build build_shared --target stringzilla_shared --config Release
export LD_LIBRARY_PATH="$PWD/build_shared:$LD_LIBRARY_PATH"

Use finders (substring, bytes, and sets):

package main

import (
    "fmt"
    sz "github.com/ashvardanian/stringzilla/golang"
)

func main() {
    s := "the quick brown fox jumps over the lazy dog"

    // Substrings
    fmt.Println(sz.Contains(s, "brown"))        // true
    fmt.Println(sz.Index(s, "the"))             // 0
    fmt.Println(sz.LastIndex(s, "the"))         // 35

    // Single bytes
    fmt.Println(sz.IndexByte(s, 'o'))            // 12
    fmt.Println(sz.LastIndexByte(s, 'o'))        // 41

    // Byte sets
    fmt.Println(sz.IndexAny(s, "aeiou"))        // 2  (first vowel)
    fmt.Println(sz.LastIndexAny(s, "aeiou"))    // 43 (last vowel)

    // Counting with/without overlaps
    fmt.Println(sz.Count("aaaaa", "aa", false)) // 2
    fmt.Println(sz.Count("aaaaa", "aa", true))  // 4
    fmt.Println(sz.Count("abc", "", false))     // 4
    fmt.Println(sz.Bytesum("ABC"), sz.Bytesum("ABCD"))
}

Unicode Case-Folding and Case-Insensitive Search

package main

import (
    "fmt"
    sz "github.com/ashvardanian/stringzilla/golang"
)

func main() {
    folded, _ := sz.Utf8CaseFold("Straße", true)
    fmt.Println(folded) // "strasse"

    haystack := "Die Temperaturschwankungen im kosmischen Mikrowellenhintergrund sind ein Maß von etwa 20 µK.\n" +
        "Typografisch sieht man auch: ein Maß von etwa 20 μK."
    needle := "EIN MASS VON ETWA 20 μK"

    start64, len64, _ := sz.Utf8CaseInsensitiveFind(haystack, needle, true)
    start, end := int(start64), int(start64+len64)
    fmt.Println(haystack[start:end]) // "ein Maß von etwa 20 µK"

    // Reuse the same needle efficiently
    compiled, _ := sz.NewUtf8CaseInsensitiveNeedle(needle, true)
    start64, len64, _ = compiled.FindIn(haystack, true)
    start, end = int(start64), int(start64+len64)
    fmt.Println(haystack[start:end])
}

Hash

Single-shot and incremental hashing are both supported. The Hasher type implements Go's standard hash.Hash64 and io.Writer interfaces:

import (
    "io"
    sz "github.com/ashvardanian/stringzilla/golang"
)

// One-shot hashing
one := sz.Hash("Hello, world!", 42)

// Streaming hasher (implements hash.Hash64 and io.Writer)
hasher := sz.NewHasher(42)
hasher.Write([]byte("Hello, "))
hasher.Write([]byte("world!"))
streamed := hasher.Digest()         // or hasher.Sum64()
fmt.Println(one == streamed)        // true

// Works with io.Copy and any io.Reader
file, _ := os.Open("data.txt")
hasher.Reset()
io.Copy(hasher, file)
fileHash := hasher.Sum64()

SHA-256 Checksums

SHA-256 cryptographic checksums are available. The Sha256 type implements Go's standard hash.Hash and io.Writer interfaces:

import (
    "io"
    sz "github.com/ashvardanian/stringzilla/golang"
)

// One-shot SHA-256
digest := sz.HashSha256([]byte("Hello, world!"))
fmt.Printf("%x\n", digest)          // prints 32-byte hash in hex

// Streaming SHA-256 (implements hash.Hash and io.Writer)
hasher := sz.NewSha256()
hasher.Write([]byte("Hello, "))
hasher.Write([]byte("world!"))
digestBytes := hasher.Digest()      // [32]byte
digestHex := hasher.Hexdigest()     // string (64 hex chars)

// Works with io.Copy and any io.Reader
file, _ := os.Open("data.bin")
hasher.Reset()
io.Copy(hasher, file)
fileDigest := hasher.Digest()

// Standard hash.Hash interface methods
sum := hasher.Sum(nil)              // []byte with 32 bytes
size := hasher.Size()               // 32
blockSize := hasher.BlockSize()     // 64

Algorithms & Design Decisions

StringZilla aims to optimize some of the slowest string operations. Some popular operations, however, like equality comparisons and relative order checking, almost always complete on some of the very first bytes in either string. In such operations vectorization is almost useless, unless huge and very similar strings are considered. StringZilla implements those operations as well, but won't result in substantial speedups. Where vectorization stops being effective, parallelism takes over with the new layered cake architecture:

StringZilla C library w/out dependencies
StringZillas parallel extensions:
- Parallel C++ algorithms built with Fork Union
- Parallel CUDA algorithms for Nvidia GPUs
- Parallel ROCm algorithms for AMD GPUs 🔜

Exact Substring Search

Substring search algorithms are generally divided into: comparison-based, automaton-based, and bit-parallel. Different families are effective for different alphabet sizes and needle lengths. The more operations are needed per-character - the more effective SIMD would be. The longer the needle - the more effective the skip-tables are. StringZilla uses different exact substring search algorithms for different needle lengths and backends:

When no SIMD is available - SWAR (SIMD Within A Register) algorithms are used on 64-bit words.
Boyer-Moore-Horspool (BMH) algorithm with Raita heuristic variation for longer needles.
SIMD backends compare characters at multiple strategically chosen offsets within the needle to reduce degeneracy.

On very short needles, especially 1-4 characters long, brute force with SIMD is the fastest solution. On mid-length needles, bit-parallel algorithms are effective, as the character masks fit into 32-bit or 64-bit words. Either way, if the needle is under 64-bytes long, on haystack traversal we will still fetch every CPU cache line. So the only way to improve performance is to reduce the number of comparisons.

For 2-byte needles, see sz_find_2byte_serial_ in include/stringzilla/find.h:

https://github.com/ashvardanian/StringZilla/blob/e1966de91600298d3c5cf4fe7be40d434f0f405e/include/stringzilla/find.h#L422-L463

Going beyond that, to long needles, Boyer-Moore (BM) and its variants are often the best choice. It has two tables: the good-suffix shift and the bad-character shift. Common choice is to use the simplified BMH algorithm, which only uses the bad-character shift table, reducing the pre-processing time. We do the same for mid-length needles up to 256 bytes long. That way the stack-allocated shift table remains small.

For mid-length needles (≤256 bytes), see sz_find_horspool_upto_256bytes_serial_ in include/stringzilla/find.h:

https://github.com/ashvardanian/StringZilla/blob/e1966de91600298d3c5cf4fe7be40d434f0f405e/include/stringzilla/find.h#L620-L667

In the C++ Standards Library, the std::string::find function uses the BMH algorithm with Raita's heuristic. Before comparing the entire string, it matches the first, last, and the middle character. Very practical, but can be slow for repetitive characters. Both SWAR and SIMD backends of StringZilla have a cheap pre-processing step, where we locate unique characters. This makes the library a lot more practical when dealing with non-English corpora.

The offset selection heuristic is implemented in sz_locate_needle_anomalies_ in include/stringzilla/find.h:

https://github.com/ashvardanian/StringZilla/blob/e1966de91600298d3c5cf4fe7be40d434f0f405e/include/stringzilla/find.h#L244-L305

All those, still, have $O(hn)$ worst case complexity. To guarantee $O(h)$ worst case time complexity, the Apostolico-Giancarlo (AG) algorithm adds an additional skip-table. Preprocessing phase is $O(n+sigma)$ in time and space. On traversal, performs from $(h/n)$ to $(3h/2)$ comparisons. It however, isn't practical on modern CPUs. A simpler idea, the Galil-rule might be a more relevant optimizations, if many matches must be found.

Other algorithms previously considered and deprecated:

Apostolico-Giancarlo algorithm for longer needles. Control-flow is too complex for efficient vectorization.
Shift-Or-based Bitap algorithm for short needles. Slower than SWAR.
Horspool-style bad-character check in SIMD backends. Effective only for very long needles, and very uneven character distributions between the needle and the haystack. Faster "character-in-set" check needed to generalize.

§ Reading materials. Exact String Matching Algorithms in Java. SIMD-friendly algorithms for substring searching.

Exact Multiple Substring Search

Few algorithms for multiple substring search are known. Most are based on the Aho-Corasick automaton, which is a generalization of the KMP algorithm. The naive implementation, however:

Allocates disjoint memory for each Trie node and Automaton state.
Requires a lot of pointer chasing, limiting speculative execution.
Has a lot of branches and conditional moves, which are hard to predict.
Matches text a character at a time, which is slow on modern CPUs.

There are several ways to improve the original algorithm. One is to use sparse DFA representation, which is more cache-friendly, but would require extra processing to navigate state transitions.

Levenshtein Edit Distance

Levenshtein distance is the best known edit-distance for strings, that checks, how many insertions, deletions, and substitutions are needed to transform one string to another. It's extensively used in approximate string-matching, spell-checking, and bioinformatics.

The computational cost of the Levenshtein distance is $O(n * m)$, where $n$ and $m$ are the lengths of the string arguments. To compute that, the naive approach requires $O(n * m)$ space to store the "Levenshtein matrix", the bottom-right corner of which will contain the Levenshtein distance. The algorithm producing the matrix has been simultaneously studied/discovered by the Soviet mathematicians Vladimir Levenshtein in 1965, Taras Vintsyuk in 1968, and American computer scientists - Robert Wagner, David Sankoff, Michael J. Fischer in the following years. Several optimizations are known:

Space Optimization: The matrix can be computed in $O(min(n,m))$ space, by only storing the last two rows of the matrix.
Divide and Conquer: Hirschberg's algorithm can be applied to decompose the computation into subtasks.
Automata: Levenshtein automata can be effective, if one of the strings doesn't change, and is a subject to many comparisons.
Shift-Or: Bit-parallel algorithms transpose the matrix into a bit-matrix, and perform bitwise operations on it.

The last approach is quite powerful and performant, and is used by the great RapidFuzz library. It's less known, than the others, derived from the Baeza-Yates-Gonnet algorithm, extended to bounded edit-distance search by Manber and Wu in 1990s, and further extended by Gene Myers in 1999 and Heikki Hyyro between 2002 and 2004.

StringZilla focuses on a different approach, extensively used in Unum's internal combinatorial optimization libraries. It doesn't change the number of trivial operations, but performs them in a different order, removing the data dependency, that occurs when computing the insertion costs. StringZilla evaluates diagonals instead of rows, exploiting the fact that all cells within a diagonal are independent, and can be computed in parallel. We'll store 3 diagonals instead of the 2 rows, and each consecutive diagonal will be computed from the previous two. Substitution costs will come from the sooner diagonal, while insertion and deletion costs will come from the later diagonal.

Row-by-Row Algorithm Computing row 4: ∅ A B C D E ∅ 0 1 2 3 4 5 P 1 ░ ░ ░ ░ ░ Q 2 ■ ■ ■ ■ ■ R 3 ■ ■ □ → . S 4 . . . . . T 5 . . . . .	Anti-Diagonal Algorithm Computing diagonal 5: ∅ A B C D E ∅ 0 1 2 3 4 5 P 1 ░ ░ ■ ■ □ Q 2 ░ ■ ■ □ ↘ R 3 ■ ■ □ ↘ . S 4 ■ □ ↘ . . T 5 □ ↘ . . .
Legend: `0,1,2,3...` = initialization constants `░` = cells processed and forgotten `■` = stored cells `□` = computing in parallel `→ ↘` = movement direction `.` = cells to compute later

This results in much better vectorization for intra-core parallelism and potentially multi-core evaluation of a single request. Moreover, it's easy to generalize to weighted edit-distances, where the cost of a substitution between two characters may not be the same for all pairs, often used in bioinformatics.

§ Reading materials. Faster Levenshtein Distances with a SIMD-friendly Traversal Order.

Needleman-Wunsch and Smith-Waterman Scores for Bioinformatics

The field of bioinformatics studies various representations of biological structures. The "primary" representations are generally strings over sparse alphabets:

DNA sequences, where the alphabet is {A, C, G, T}, ranging from ~100 characters for short reads to 3 billion for the human genome.
RNA sequences, where the alphabet is {A, C, G, U}, ranging from ~50 characters for tRNA to thousands for mRNA.
Proteins, where the alphabet is made of 22 amino acids, ranging from 2 characters for dipeptide to 35,000 for Titin, the longest protein.

The shorter the representation, the more often researchers may want to use custom substitution matrices. Meaning that the cost of a substitution between two characters may not be the same for all pairs. In the general case the serial algorithm is supposed to work for arbitrary substitution costs for each of 256×256 possible character pairs. That lookup table, however, is too large to fit into CPU registers, so instead, the upcoming design focuses on 32×32 substitution matrices, which fit into 1 KB with single-byte "error costs". That said, most BLOSUM and PAM substitution matrices only contain 4-bit values, so they can be packed even further.

Next design goals:

Needleman-Wunsch Automata

Memory Copying, Fills, and Moves

A lot has been written about the time computers spend copying memory and how that operation is implemented in LibC. Interestingly, the operation can still be improved, as most Assembly implementations use outdated instructions. Even performance-oriented STL replacements, like Meta's Folly v2024.09.23 focus on AVX2, and don't take advantage of the new masked instructions in AVX-512 or SVE.

In AVX-512, StringZilla uses non-temporal stores to avoid cache pollution, when dealing with very large strings. Moreover, it handles the unaligned head and the tails of the target buffer separately, ensuring that writes in big copies are always aligned to cache-line boundaries. That's true for both AVX2 and AVX-512 backends.

StringZilla also contains "drafts" of smarter, but less efficient algorithms, that minimize the number of unaligned loads, performing shuffles and permutations. That's a topic for future research, as the performance gains are not yet satisfactory.

§ Reading materials. memset benchmarks by Nadav Rotem. Cache Associativity by Sergey Slotin.

Hashing

StringZilla implements a high-performance 64-bit hash function inspired by the "AquaHash", "aHash", and "GxHash" design and optimized for modern CPU architectures. The algorithm utilizes AES encryption rounds combined with shuffle-and-add operations to achieve exceptional mixing properties while maintaining consistent output across platforms. It passes the rigorous SMHasher test suite, including the --extra flag with no collisions.

The core algorithm operates on a dual-state design:

AES State: Initialized with seed XOR-ed against π constants.
Sum State: Accumulates shuffled input data with a permutation.

For strings ≤64 bytes, a minimal state processes data in 16-byte blocks. Longer strings employ a 4× wider state (512 bits) that processes 64-byte chunks, maximizing throughput on modern superscalar CPUs. The algorithm can be expressed in pseudocode as:

function sz_hash(text: u8[], length: usize, seed: u64) -> u64:
    # 1024 bits worth of π constants
    pi: u64[16] = [
        0x243F6A8885A308D3, 0x13198A2E03707344, 0xA4093822299F31D0, 0x082EFA98EC4E6C89,
        0x452821E638D01377, 0xBE5466CF34E90C6C, 0xC0AC29B7C97C50DD, 0x3F84D5B5B5470917,
        0x9216D5D98979FB1B, 0xD1310BA698DFB5AC, 0x2FFD72DBD01ADFB7, 0xB8E1AFED6A267E96,
        0xBA7C9045F12C7F99, 0x24A19947B3916CF7, 0x0801F2E2858EFC16, 0x636920D871574E69]

    # Permutation order for the sum state
    shuffle_pattern: u8[16] = [
        0x04, 0x0b, 0x09, 0x06, 0x08, 0x0d, 0x0f, 0x05,
        0x0e, 0x03, 0x01, 0x0c, 0x00, 0x07, 0x0a, 0x02]

    # Initialize key and states
    keys_u64s: u64[2] = [seed, seed]
    aes_u64s: u64[2] = [seed ⊕ pi[0], seed ⊕ pi[1]]
    sum_u64s: u64[2] = [seed ⊕ pi[8], seed ⊕ pi[9]]

    if length ≤ 64:
        # Small input: process 1-4 zero-padded blocks of 16 bytes each
        blocks_u8s: u8[16][] = split_into_blocks(text, length, 16)
        for each block_u8s: u8[16] in blocks_u8s:
            aes_u64s = AESENC(aes_u64s, block_u8s)
            sum_u64s = SHUFFLE(sum_u64s, shuffle_pattern) + block_u8s
    else:
        # Large input: use 4× wider 512-bits states
        aes_u64s: u64[8] = [
            seed ⊕ pi[0], seed ⊕ pi[1], seed ⊕ pi[2], seed ⊕ pi[3],
            seed ⊕ pi[4], seed ⊕ pi[5], seed ⊕ pi[6], seed ⊕ pi[7]]
        sum_u64s: u64[8] = [
            seed ⊕ pi[8], seed ⊕ pi[9], seed ⊕ pi[10], seed ⊕ pi[11],
            seed ⊕ pi[12], seed ⊕ pi[13], seed ⊕ pi[14], seed ⊕ pi[15]]

        # Process 64-byte chunks (4×16-byte blocks)
        for each chunk_u8s: u8[64] in text:
            blocks_u8s: u8[16][4] = split_chunk_into_4_blocks(chunk_u8s)
            for i in 0..3:
                offset: usize = i * 2  # Each lane stores two u64s
                aes_u64s[offset:offset+1] = AESENC(aes_u64s[offset:offset+1], blocks_u8s[i])
                sum_u64s[offset:offset+1] = SHUFFLE(sum_u64s[offset:offset+1], shuffle_pattern) + blocks_u8s[i]

        # Fold 8×u64 state back to 2×u64 for finalization
        aes_u64s: u64[2] = fold_to_2u64(aes_u64s)
        sum_u64s: u64[2] = fold_to_2u64(sum_u64s)

    # Finalization: mix length into key
    key_with_length: u64[2] = [keys_u64s[0] + length, keys_u64s[1]]

    # Multiple AES rounds for SMHasher compliance
    mixed_u64s: u64[2] = AESENC(sum_u64s, aes_u64s)
    result_u64s: u64[2] = AESENC(AESENC(mixed_u64s, key_with_length), mixed_u64s)

    return result_u64s[0]  # Extract low 64 bits

This allows us to balance several design trade-offs. First, it allows us to achieve a high port-level parallelism. Looking at AVX-512 capable CPUs and their ZMM instructions, on each cycle, we'll have at least 2 ports busy when dealing with long strings:

VAESENC: 5 cycles on port 0 on Intel Ice Lake, 4 cycles on ports 0/1 on AMD Zen4.
VPSHUFB_Z: 3 cycles on port 5 on Intel Ice Lake, 2 cycles on ports 1/2 on AMD Zen4.
VPADDQ: 1 cycle on ports 0/5 on Intel Ice Lake, 1 cycle on ports 0/1/2/3 on AMD Zen4.

When dealing with smaller strings, we design our approach to avoid large registers and maintain the CPU at the same energy state, thereby avoiding downclocking and expensive power-state transitions.

Unlike some AES-accelerated alternatives, the length of the input is not mixed into the AES block at the start to allow incremental construction, when the final length is not known in advance. Also, unlike some alternatives, with "masked" AVX-512 and "predicated" SVE loads, we avoid expensive block-shuffling procedures on non-divisible-by-16 lengths.

§ Reading materials. Stress-testing hash functions for avalance behaviour, collision bias, and distribution.

SHA-256 Checksums

In addition to the fast AES-based hash, StringZilla implements hardware-accelerated SHA-256 cryptographic checksums. The implementation follows the FIPS 180-4 specification and provides multiple backends.

Random Generation

StringZilla implements a fast Pseudorandom Number Generator inspired by the "AES-CTR-128" algorithm, reusing the same AES primitives as the hash function. Unlike "NIST SP 800-90A" which uses multiple AES rounds, StringZilla uses only one round of AES mixing for performance while maintaining reproducible output across platforms. The generator operates in counter mode with AESENC(nonce + lane_index, nonce ⊕ pi_constants), rotating through the first 512 bits of π for each 16-byte block. The only state required to reproduce an output is a 64-bit nonce, which is much cheaper than a Mersenne Twister.

Sorting

For lexicographic sorting of string collections, StringZilla exports pointer-sized n‑grams ("pgrams") into a contiguous buffer to improve locality, then recursively QuickSorts those pgrams with a 3‑way partition and dives into equal pgrams to compare deeper characters. Very small inputs fall back to insertion sort.

Average time complexity: O(n log n)
Worst-case time complexity: quadratic (due to QuickSort), mitigated in practice by 3‑way partitioning and the n‑gram staging

Unicode 17, UTF-8, and Wide Characters

Most StringZilla operations are byte-level, so they work well with ASCII and UTF-8 content out of the box. In some cases, like edit-distance computation, the result of byte-level evaluation and character-level evaluation may differ.

szs_levenshtein_distances_utf8("αβγδ", "αγδ") == 1 — one unicode symbol.
szs_levenshtein_distances("αβγδ", "αγδ") == 2 — one unicode symbol is two bytes long.

Java, JavaScript, Python 2, C#, and Objective-C, however, use wide characters (wchar) - two byte long codes, instead of the more reasonable fixed-length UTF-32 or variable-length UTF-8. This leads to all kinds of offset-counting issues when facing four-byte long Unicode characters. StringZilla uses proper 32-bit "runes" to represent unpacked Unicode codepoints, ensuring correct results in all operations. Moreover, it implements the Unicode 17.0 standard, being practically the only library besides ICU and PCRE2 to do so, but with order(s) of magnitude better performance.

Case-Folding and Case-Insensitive Search

StringZilla provides Unicode-aware case-insensitive substring search that handles the full complexity of Unicode case folding. This includes multi-character expansions:

Character	Codepoint	UTF-8 Bytes	Case-Folds To	Result Bytes
`ß`	U+00DF	C3 9F	`ss`	73 73
`ﬃ`	U+FB03	EF AC 83	`ffi`	66 66 69
`İ`	U+0130	C4 B0	`i` + `◌̇`	69 CC 87

The search returns byte offsets and lengths in the original haystack, correctly handling length differences. For example, searching for "STRASSE" (7 bytes) in "Straße" (7 bytes: 53 74 72 61 C3 9F 65) succeeds because both case-fold to "strasse".

Note that Turkish İ and ASCII I are distinct: İstanbul case-folds to i̇stanbul (with combining dot), while ISTANBUL case-folds to istanbul (without). They will not match each other — this is correct Unicode behavior for Turkish locale handling.

For wide-character environments (Java, JavaScript, Python 2, C#), consider transcoding with simdutf.

Dynamic Dispatch

Due to the high-level of fragmentation of SIMD support in different CPUs, StringZilla uses the names of select Intel and ARM CPU generations for its backends. You can query supported backends and use them manually. Use it to guarantee constant performance, or to explore how different algorithms scale on your hardware.

sz_find(text, length, pattern, 3);          // Auto-dispatch
sz_find_westmere(text, length, pattern, 3);  // Intel Westmere+ SSE4.2
sz_find_haswell(text, length, pattern, 3);  // Intel Haswell+ AVX2
sz_find_skylake(text, length, pattern, 3);  // Intel Skylake+ AVX-512
sz_find_neon(text, length, pattern, 3);     // Arm NEON 128-bit
sz_find_sve(text, length, pattern, 3);      // Arm SVE 128/256/512/1024/2048-bit

StringZilla automatically picks the most advanced backend for the given CPU. Similarly, in Python, you can log the auto-detected capabilities:

python -c "import stringzilla; print(stringzilla.__capabilities__)"         # ('serial', 'westmere', 'haswell', 'skylake', 'ice', 'neon', 'sve', 'sve2+aes')
python -c "import stringzilla; print(stringzilla.__capabilities_str__)"     # "haswell, skylake, ice, neon, sve, sve2+aes"

You can also explicitly set the backend to use, or scope the backend to a specific function.

import stringzilla as sz
sz.reset_capabilities(('serial',))          # Force SWAR backend
sz.reset_capabilities(('haswell',))         # Force AVX2 backend
sz.reset_capabilities(('neon',))            # Force NEON backend
sz.reset_capabilities(sz.__capabilities__)  # Reset to auto-dispatch

Contributing 👾

Please check out the contributing guide for more details on how to set up the development environment and contribute to this project. If you like this project, you may also enjoy USearch, UCall, UForm, and SimSIMD. 🤗

If you like strings and value efficiency, you may also enjoy the following projects:

simdutf - transcoding UTF-8, UTF-16, and UTF-32 LE and BE.
hyperscan - regular expressions with SIMD acceleration.
pyahocorasick - Aho-Corasick algorithm in Python.
rapidfuzz - fast string matching in C++ and Python.
memchr - fast string search in Rust.

If you are looking for more reading materials on this topic, consider the following:

License 📜

Feel free to use the project under Apache 2.0 or the Three-clause BSD license at your preference.

Project details

Release history Release notifications | RSS feed

This version

4.6.1

May 5, 2026

4.6.0

Dec 26, 2025

4.5.1

Dec 15, 2025

4.5.0

Dec 15, 2025

4.4.2 yanked

Dec 4, 2025

Reason this release was yanked:

MSVC builds

4.4.1 yanked

Dec 4, 2025

Reason this release was yanked:

MSVC builds

4.4.0

Nov 29, 2025

4.3.0

Nov 26, 2025

4.2.3

Oct 27, 2025

4.2.2

Oct 26, 2025

4.2.1

Oct 12, 2025

4.2.0

Oct 7, 2025

4.1.0

Oct 2, 2025

4.0.15

Sep 30, 2025

4.0.14

Sep 22, 2025

4.0.13

Sep 19, 2025

4.0.12

Sep 19, 2025

4.0.11

Sep 16, 2025

4.0.10

Sep 15, 2025

4.0.9

Sep 15, 2025

4.0.8

Sep 15, 2025

4.0.7

Sep 10, 2025

4.0.6

Sep 9, 2025

4.0.2

Sep 9, 2025

4.0.1

Sep 9, 2025

4.0.0

Sep 9, 2025

3.12.6

Aug 7, 2025

3.12.5

Apr 18, 2025

3.12.4

Apr 12, 2025

3.12.3

Mar 9, 2025

3.12.2

Mar 2, 2025

3.12.1

Feb 26, 2025

3.12.0

Feb 23, 2025

3.11.3

Dec 26, 2024

3.11.2

Dec 19, 2024

3.11.1

Dec 11, 2024

3.11.0

Dec 1, 2024

3.10.11

Nov 30, 2024

3.10.10

Nov 12, 2024

3.10.9

Nov 9, 2024

3.10.8

Nov 6, 2024

3.10.7

Oct 30, 2024

3.10.6

Oct 29, 2024

3.10.5

Oct 19, 2024

3.10.4

Oct 17, 2024

3.10.3

Oct 16, 2024

3.10.2

Oct 16, 2024

3.10.1

Oct 15, 2024

3.10.0

Oct 13, 2024

3.9.8

Sep 26, 2024

3.9.7

Sep 21, 2024

3.9.6

Sep 14, 2024

3.9.5

Sep 10, 2024

3.9.4

Sep 9, 2024

3.9.3

Aug 22, 2024

3.9.2

Aug 21, 2024

3.9.1

Aug 19, 2024

3.9.0

Aug 5, 2024

3.8.4

Apr 27, 2024

3.8.3

Apr 26, 2024

3.8.2

Apr 19, 2024

3.8.1

Apr 17, 2024

3.8.0

Apr 15, 2024

3.7.3

Apr 10, 2024

3.7.2

Apr 8, 2024

3.7.1

Mar 29, 2024

3.7.0

Mar 18, 2024

3.6.8

Mar 18, 2024

3.6.7

Mar 18, 2024

3.6.6

Mar 17, 2024

3.6.5

Mar 17, 2024

3.6.4

Mar 17, 2024

3.6.2

Mar 16, 2024

3.6.1

Mar 15, 2024

3.6.0

Mar 15, 2024

3.5.0

Mar 3, 2024

3.4.1

Mar 2, 2024

3.3.1

Feb 27, 2024

3.3.0

Feb 24, 2024

3.1.2

Feb 19, 2024

3.1.1

Feb 18, 2024

3.0.0

Feb 6, 2024

2.0.3

Nov 19, 2023

2.0.2

Nov 4, 2023

2.0.1

Oct 10, 2023

1.2.2

Sep 18, 2023

1.1.3

Sep 1, 2023

1.1.0

Aug 6, 2023

1.0.3

Jul 31, 2023

1.0.2

Jul 30, 2023

1.0.0

Jul 13, 2023

0.2.0

Jul 9, 2023

0.1.3

Jul 8, 2023

0.1.2

Jun 22, 2023

0.1.1

Jun 21, 2023

0.1.0

Jun 21, 2023

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

stringzilla-4.6.1.tar.gz (646.5 kB view details)

Uploaded May 5, 2026 Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

stringzilla-4.6.1-cp313-cp313-win_arm64.whl (123.3 kB view details)

Uploaded May 5, 2026 CPython 3.13Windows ARM64

stringzilla-4.6.1-cp313-cp313-win_amd64.whl (162.6 kB view details)

Uploaded May 5, 2026 CPython 3.13Windows x86-64

stringzilla-4.6.1-cp313-cp313-win32.whl (114.8 kB view details)

Uploaded May 5, 2026 CPython 3.13Windows x86

stringzilla-4.6.1-cp313-cp313-musllinux_1_2_x86_64.whl (1.9 MB view details)

Uploaded May 5, 2026 CPython 3.13musllinux: musl 1.2+ x86-64

stringzilla-4.6.1-cp313-cp313-musllinux_1_2_s390x.whl (613.5 kB view details)

Uploaded May 5, 2026 CPython 3.13musllinux: musl 1.2+ s390x

stringzilla-4.6.1-cp313-cp313-musllinux_1_2_ppc64le.whl (619.1 kB view details)

Uploaded May 5, 2026 CPython 3.13musllinux: musl 1.2+ ppc64le

stringzilla-4.6.1-cp313-cp313-musllinux_1_2_i686.whl (626.3 kB view details)

Uploaded May 5, 2026 CPython 3.13musllinux: musl 1.2+ i686

stringzilla-4.6.1-cp313-cp313-musllinux_1_2_armv7l.whl (587.1 kB view details)

Uploaded May 5, 2026 CPython 3.13musllinux: musl 1.2+ ARMv7l

stringzilla-4.6.1-cp313-cp313-musllinux_1_2_aarch64.whl (653.9 kB view details)

Uploaded May 5, 2026 CPython 3.13musllinux: musl 1.2+ ARM64

stringzilla-4.6.1-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl (2.1 MB view details)

Uploaded May 5, 2026 CPython 3.13manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

stringzilla-4.6.1-cp313-cp313-manylinux_2_17_s390x.manylinux2014_s390x.manylinux_2_28_s390x.whl (640.6 kB view details)

Uploaded May 5, 2026 CPython 3.13manylinux: glibc 2.17+ s390xmanylinux: glibc 2.28+ s390x

stringzilla-4.6.1-cp313-cp313-manylinux_2_17_ppc64le.manylinux2014_ppc64le.manylinux_2_28_ppc64le.whl (657.4 kB view details)

Uploaded May 5, 2026 CPython 3.13manylinux: glibc 2.17+ ppc64lemanylinux: glibc 2.28+ ppc64le

stringzilla-4.6.1-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_28_aarch64.whl (689.9 kB view details)

Uploaded May 5, 2026 CPython 3.13manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

stringzilla-4.6.1-cp313-cp313-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl (644.1 kB view details)

Uploaded May 5, 2026 CPython 3.13manylinux: glibc 2.17+ i686manylinux: glibc 2.5+ i686

stringzilla-4.6.1-cp313-cp313-macosx_11_0_arm64.whl (199.4 kB view details)

Uploaded May 5, 2026 CPython 3.13macOS 11.0+ ARM64

stringzilla-4.6.1-cp313-cp313-macosx_10_13_x86_64.whl (212.3 kB view details)

Uploaded May 5, 2026 CPython 3.13macOS 10.13+ x86-64

stringzilla-4.6.1-cp312-cp312-win_arm64.whl (123.3 kB view details)

Uploaded May 5, 2026 CPython 3.12Windows ARM64

stringzilla-4.6.1-cp312-cp312-win_amd64.whl (162.6 kB view details)

Uploaded May 5, 2026 CPython 3.12Windows x86-64

stringzilla-4.6.1-cp312-cp312-win32.whl (114.8 kB view details)

Uploaded May 5, 2026 CPython 3.12Windows x86

stringzilla-4.6.1-cp312-cp312-musllinux_1_2_x86_64.whl (1.9 MB view details)

Uploaded May 5, 2026 CPython 3.12musllinux: musl 1.2+ x86-64

stringzilla-4.6.1-cp312-cp312-musllinux_1_2_s390x.whl (613.4 kB view details)

Uploaded May 5, 2026 CPython 3.12musllinux: musl 1.2+ s390x

stringzilla-4.6.1-cp312-cp312-musllinux_1_2_ppc64le.whl (619.1 kB view details)

Uploaded May 5, 2026 CPython 3.12musllinux: musl 1.2+ ppc64le

stringzilla-4.6.1-cp312-cp312-musllinux_1_2_i686.whl (626.3 kB view details)

Uploaded May 5, 2026 CPython 3.12musllinux: musl 1.2+ i686

stringzilla-4.6.1-cp312-cp312-musllinux_1_2_armv7l.whl (587.1 kB view details)

Uploaded May 5, 2026 CPython 3.12musllinux: musl 1.2+ ARMv7l

stringzilla-4.6.1-cp312-cp312-musllinux_1_2_aarch64.whl (653.9 kB view details)

Uploaded May 5, 2026 CPython 3.12musllinux: musl 1.2+ ARM64

stringzilla-4.6.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl (2.1 MB view details)

Uploaded May 5, 2026 CPython 3.12manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

stringzilla-4.6.1-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.manylinux_2_28_s390x.whl (640.7 kB view details)

Uploaded May 5, 2026 CPython 3.12manylinux: glibc 2.17+ s390xmanylinux: glibc 2.28+ s390x

stringzilla-4.6.1-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.manylinux_2_28_ppc64le.whl (657.4 kB view details)

Uploaded May 5, 2026 CPython 3.12manylinux: glibc 2.17+ ppc64lemanylinux: glibc 2.28+ ppc64le

stringzilla-4.6.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_28_aarch64.whl (689.9 kB view details)

Uploaded May 5, 2026 CPython 3.12manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

stringzilla-4.6.1-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl (644.1 kB view details)

Uploaded May 5, 2026 CPython 3.12manylinux: glibc 2.17+ i686manylinux: glibc 2.5+ i686

stringzilla-4.6.1-cp312-cp312-macosx_11_0_arm64.whl (199.4 kB view details)

Uploaded May 5, 2026 CPython 3.12macOS 11.0+ ARM64

stringzilla-4.6.1-cp312-cp312-macosx_10_13_x86_64.whl (212.3 kB view details)

Uploaded May 5, 2026 CPython 3.12macOS 10.13+ x86-64

stringzilla-4.6.1-cp311-cp311-win_arm64.whl (123.1 kB view details)

Uploaded May 5, 2026 CPython 3.11Windows ARM64

stringzilla-4.6.1-cp311-cp311-win_amd64.whl (162.3 kB view details)

Uploaded May 5, 2026 CPython 3.11Windows x86-64

stringzilla-4.6.1-cp311-cp311-win32.whl (114.5 kB view details)

Uploaded May 5, 2026 CPython 3.11Windows x86

stringzilla-4.6.1-cp311-cp311-musllinux_1_2_x86_64.whl (1.9 MB view details)

Uploaded May 5, 2026 CPython 3.11musllinux: musl 1.2+ x86-64

stringzilla-4.6.1-cp311-cp311-musllinux_1_2_s390x.whl (611.0 kB view details)

Uploaded May 5, 2026 CPython 3.11musllinux: musl 1.2+ s390x

stringzilla-4.6.1-cp311-cp311-musllinux_1_2_ppc64le.whl (618.7 kB view details)

Uploaded May 5, 2026 CPython 3.11musllinux: musl 1.2+ ppc64le

stringzilla-4.6.1-cp311-cp311-musllinux_1_2_i686.whl (624.5 kB view details)

Uploaded May 5, 2026 CPython 3.11musllinux: musl 1.2+ i686

stringzilla-4.6.1-cp311-cp311-musllinux_1_2_armv7l.whl (583.8 kB view details)

Uploaded May 5, 2026 CPython 3.11musllinux: musl 1.2+ ARMv7l

stringzilla-4.6.1-cp311-cp311-musllinux_1_2_aarch64.whl (654.1 kB view details)

Uploaded May 5, 2026 CPython 3.11musllinux: musl 1.2+ ARM64

stringzilla-4.6.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl (2.1 MB view details)

Uploaded May 5, 2026 CPython 3.11manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

stringzilla-4.6.1-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.manylinux_2_28_s390x.whl (640.0 kB view details)

Uploaded May 5, 2026 CPython 3.11manylinux: glibc 2.17+ s390xmanylinux: glibc 2.28+ s390x

stringzilla-4.6.1-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.manylinux_2_28_ppc64le.whl (656.8 kB view details)

Uploaded May 5, 2026 CPython 3.11manylinux: glibc 2.17+ ppc64lemanylinux: glibc 2.28+ ppc64le

stringzilla-4.6.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_28_aarch64.whl (690.0 kB view details)

Uploaded May 5, 2026 CPython 3.11manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

stringzilla-4.6.1-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl (640.9 kB view details)

Uploaded May 5, 2026 CPython 3.11manylinux: glibc 2.17+ i686manylinux: glibc 2.5+ i686

stringzilla-4.6.1-cp311-cp311-macosx_11_0_arm64.whl (199.1 kB view details)

Uploaded May 5, 2026 CPython 3.11macOS 11.0+ ARM64

stringzilla-4.6.1-cp311-cp311-macosx_10_13_x86_64.whl (211.6 kB view details)

Uploaded May 5, 2026 CPython 3.11macOS 10.13+ x86-64

stringzilla-4.6.1-cp310-cp310-win_arm64.whl (123.1 kB view details)

Uploaded May 5, 2026 CPython 3.10Windows ARM64

stringzilla-4.6.1-cp310-cp310-win_amd64.whl (162.4 kB view details)

Uploaded May 5, 2026 CPython 3.10Windows x86-64

stringzilla-4.6.1-cp310-cp310-win32.whl (114.5 kB view details)

Uploaded May 5, 2026 CPython 3.10Windows x86

stringzilla-4.6.1-cp310-cp310-musllinux_1_2_x86_64.whl (1.9 MB view details)

Uploaded May 5, 2026 CPython 3.10musllinux: musl 1.2+ x86-64

stringzilla-4.6.1-cp310-cp310-musllinux_1_2_s390x.whl (604.6 kB view details)

Uploaded May 5, 2026 CPython 3.10musllinux: musl 1.2+ s390x

stringzilla-4.6.1-cp310-cp310-musllinux_1_2_ppc64le.whl (612.3 kB view details)

Uploaded May 5, 2026 CPython 3.10musllinux: musl 1.2+ ppc64le

stringzilla-4.6.1-cp310-cp310-musllinux_1_2_i686.whl (620.5 kB view details)

Uploaded May 5, 2026 CPython 3.10musllinux: musl 1.2+ i686

stringzilla-4.6.1-cp310-cp310-musllinux_1_2_armv7l.whl (578.8 kB view details)

Uploaded May 5, 2026 CPython 3.10musllinux: musl 1.2+ ARMv7l

stringzilla-4.6.1-cp310-cp310-musllinux_1_2_aarch64.whl (647.5 kB view details)

Uploaded May 5, 2026 CPython 3.10musllinux: musl 1.2+ ARM64

stringzilla-4.6.1-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl (2.1 MB view details)

Uploaded May 5, 2026 CPython 3.10manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

stringzilla-4.6.1-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.manylinux_2_28_s390x.whl (633.8 kB view details)

Uploaded May 5, 2026 CPython 3.10manylinux: glibc 2.17+ s390xmanylinux: glibc 2.28+ s390x

stringzilla-4.6.1-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.manylinux_2_28_ppc64le.whl (649.8 kB view details)

Uploaded May 5, 2026 CPython 3.10manylinux: glibc 2.17+ ppc64lemanylinux: glibc 2.28+ ppc64le

stringzilla-4.6.1-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_28_aarch64.whl (684.2 kB view details)

Uploaded May 5, 2026 CPython 3.10manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

stringzilla-4.6.1-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl (633.8 kB view details)

Uploaded May 5, 2026 CPython 3.10manylinux: glibc 2.17+ i686manylinux: glibc 2.5+ i686

stringzilla-4.6.1-cp310-cp310-macosx_11_0_arm64.whl (199.1 kB view details)

Uploaded May 5, 2026 CPython 3.10macOS 11.0+ ARM64

stringzilla-4.6.1-cp310-cp310-macosx_10_13_x86_64.whl (211.6 kB view details)

Uploaded May 5, 2026 CPython 3.10macOS 10.13+ x86-64

stringzilla-4.6.1-cp39-cp39-win_arm64.whl (123.1 kB view details)

Uploaded May 5, 2026 CPython 3.9Windows ARM64

stringzilla-4.6.1-cp39-cp39-win_amd64.whl (162.5 kB view details)

Uploaded May 5, 2026 CPython 3.9Windows x86-64

stringzilla-4.6.1-cp39-cp39-win32.whl (114.6 kB view details)

Uploaded May 5, 2026 CPython 3.9Windows x86

stringzilla-4.6.1-cp39-cp39-musllinux_1_2_x86_64.whl (1.9 MB view details)

Uploaded May 5, 2026 CPython 3.9musllinux: musl 1.2+ x86-64

stringzilla-4.6.1-cp39-cp39-musllinux_1_2_s390x.whl (604.5 kB view details)

Uploaded May 5, 2026 CPython 3.9musllinux: musl 1.2+ s390x

stringzilla-4.6.1-cp39-cp39-musllinux_1_2_ppc64le.whl (611.3 kB view details)

Uploaded May 5, 2026 CPython 3.9musllinux: musl 1.2+ ppc64le

stringzilla-4.6.1-cp39-cp39-musllinux_1_2_i686.whl (620.8 kB view details)

Uploaded May 5, 2026 CPython 3.9musllinux: musl 1.2+ i686

stringzilla-4.6.1-cp39-cp39-musllinux_1_2_armv7l.whl (578.4 kB view details)

Uploaded May 5, 2026 CPython 3.9musllinux: musl 1.2+ ARMv7l

stringzilla-4.6.1-cp39-cp39-musllinux_1_2_aarch64.whl (647.3 kB view details)

Uploaded May 5, 2026 CPython 3.9musllinux: musl 1.2+ ARM64

stringzilla-4.6.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl (2.1 MB view details)

Uploaded May 5, 2026 CPython 3.9manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

stringzilla-4.6.1-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.manylinux_2_28_s390x.whl (633.4 kB view details)

Uploaded May 5, 2026 CPython 3.9manylinux: glibc 2.17+ s390xmanylinux: glibc 2.28+ s390x

stringzilla-4.6.1-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.manylinux_2_28_ppc64le.whl (648.6 kB view details)

Uploaded May 5, 2026 CPython 3.9manylinux: glibc 2.17+ ppc64lemanylinux: glibc 2.28+ ppc64le

stringzilla-4.6.1-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_28_aarch64.whl (683.6 kB view details)

Uploaded May 5, 2026 CPython 3.9manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

stringzilla-4.6.1-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl (633.4 kB view details)

Uploaded May 5, 2026 CPython 3.9manylinux: glibc 2.17+ i686manylinux: glibc 2.5+ i686

stringzilla-4.6.1-cp39-cp39-macosx_11_0_arm64.whl (199.0 kB view details)

Uploaded May 5, 2026 CPython 3.9macOS 11.0+ ARM64

stringzilla-4.6.1-cp39-cp39-macosx_10_13_x86_64.whl (211.5 kB view details)

Uploaded May 5, 2026 CPython 3.9macOS 10.13+ x86-64

stringzilla-4.6.1-cp38-cp38-win_amd64.whl (162.4 kB view details)

Uploaded May 5, 2026 CPython 3.8Windows x86-64

stringzilla-4.6.1-cp38-cp38-win32.whl (114.5 kB view details)

Uploaded May 5, 2026 CPython 3.8Windows x86

stringzilla-4.6.1-cp38-cp38-musllinux_1_2_x86_64.whl (1.9 MB view details)

Uploaded May 5, 2026 CPython 3.8musllinux: musl 1.2+ x86-64

stringzilla-4.6.1-cp38-cp38-musllinux_1_2_s390x.whl (601.6 kB view details)

Uploaded May 5, 2026 CPython 3.8musllinux: musl 1.2+ s390x

stringzilla-4.6.1-cp38-cp38-musllinux_1_2_ppc64le.whl (608.5 kB view details)

Uploaded May 5, 2026 CPython 3.8musllinux: musl 1.2+ ppc64le

stringzilla-4.6.1-cp38-cp38-musllinux_1_2_i686.whl (618.6 kB view details)

Uploaded May 5, 2026 CPython 3.8musllinux: musl 1.2+ i686

stringzilla-4.6.1-cp38-cp38-musllinux_1_2_armv7l.whl (576.1 kB view details)

Uploaded May 5, 2026 CPython 3.8musllinux: musl 1.2+ ARMv7l

stringzilla-4.6.1-cp38-cp38-musllinux_1_2_aarch64.whl (644.6 kB view details)

Uploaded May 5, 2026 CPython 3.8musllinux: musl 1.2+ ARM64

stringzilla-4.6.1-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl (2.1 MB view details)

Uploaded May 5, 2026 CPython 3.8manylinux: glibc 2.17+ x86-64manylinux: glibc 2.28+ x86-64

stringzilla-4.6.1-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.manylinux_2_28_s390x.whl (630.4 kB view details)

Uploaded May 5, 2026 CPython 3.8manylinux: glibc 2.17+ s390xmanylinux: glibc 2.28+ s390x

stringzilla-4.6.1-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.manylinux_2_28_ppc64le.whl (645.2 kB view details)

Uploaded May 5, 2026 CPython 3.8manylinux: glibc 2.17+ ppc64lemanylinux: glibc 2.28+ ppc64le

stringzilla-4.6.1-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.manylinux_2_28_aarch64.whl (679.9 kB view details)

Uploaded May 5, 2026 CPython 3.8manylinux: glibc 2.17+ ARM64manylinux: glibc 2.28+ ARM64

stringzilla-4.6.1-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl (630.5 kB view details)

Uploaded May 5, 2026 CPython 3.8manylinux: glibc 2.17+ i686manylinux: glibc 2.5+ i686

stringzilla-4.6.1-cp38-cp38-macosx_11_0_arm64.whl (198.8 kB view details)

Uploaded May 5, 2026 CPython 3.8macOS 11.0+ ARM64

stringzilla-4.6.1-cp38-cp38-macosx_10_13_x86_64.whl (211.2 kB view details)

Uploaded May 5, 2026 CPython 3.8macOS 10.13+ x86-64

File details

Details for the file stringzilla-4.6.1.tar.gz.

File metadata

Download URL: stringzilla-4.6.1.tar.gz
Upload date: May 5, 2026
Size: 646.5 kB
Tags: Source
Uploaded using Trusted Publishing? No
Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for stringzilla-4.6.1.tar.gz
Algorithm	Hash digest
SHA256	`47e9b0b95337857146e0dd4309998af7acab6bdca1f85c069adc1c32bbd57cb0`
MD5	`d049fd65889ada9c4539a7f1fe794066`
BLAKE2b-256	`4dc008c222e1c89871970121f8ead9764f63df925e32879f68bb479c61f7ad89`

Algorithm	Hash digest
SHA256	`c2cc745ce598af8cdd5067b923ba7631bbe5e77b9d552b7d5a29135566ea3bec`
MD5	`05b9d311111cb2148eb609bfa41e8f3c`
BLAKE2b-256	`e97c350482e963262ff6cfb7d094fe5254c6a432c21ce971d83e98a23268a2ed`

Algorithm	Hash digest
SHA256	`a9da1feb660085a00f8c8e781be85a4984e6a4504f7804af1b13fdf5563a41b7`
MD5	`427a74c9f7f694c4d63b246c39f57a2d`
BLAKE2b-256	`97dbc917b8c9428f647bcd04433e66110cbcd63cee1075f735bb3c7ae6b2ba20`

Algorithm	Hash digest
SHA256	`e8b2ab69ee28eb79f54754e9b85cad827580f83aab8d5929ac1eb140fe6e7ade`
MD5	`9fc07f496d2caca95a4cb4545925efe3`
BLAKE2b-256	`16834b694084746a7e58604a46dc5a6911341b5ae332360492262a84a265d110`

Algorithm	Hash digest
SHA256	`0e162700650d25f6e9040fb9d196fec19337c072cad5b1df11fc4f612e7d86c7`
MD5	`ca2a1e3709c65db96450a5d31c0833fd`
BLAKE2b-256	`a53bc49a971f9bbec98deb3b388dd51a1ef458472e871223f86608bfdad9cc22`

Algorithm	Hash digest
SHA256	`19d91a5a4e6a3c7827d89a3b8bfe5e691f90987adaff9418c63bd17380fe1c38`
MD5	`40d97b14837ffbd3f1d0eadba4aeb6e4`
BLAKE2b-256	`2ea3ccf774223db71021d0af56d385a17d7f78245d8d34ba5cadc6517fd0e58c`

Algorithm	Hash digest
SHA256	`f40b8973efa7bc127d07a26c5bbd753197cd9d7b539810896bf8b98a9542f51b`
MD5	`b438c5257982ffefe0bee3f93ec94239`
BLAKE2b-256	`3765333f702b366a9595fea03de805379bb672fa3cafe52178697a09f430fc1a`

Algorithm	Hash digest
SHA256	`1360b78634f495272a85704cacb2197ecce7f00b9b8be1c0d2d6e7c05ffc4efc`
MD5	`72180915eb085fe6e0393e2fec3b66a4`
BLAKE2b-256	`dafb33702a7365aba99b0bb7e6866fe35860142a8e9752d6de3de66250bc9f90`

Algorithm	Hash digest
SHA256	`b23d3172830e5136691a9a13bb2f6bc391fff7d3a3399e02a690bcbfd0f5bbda`
MD5	`989d0eed2440edb1d235df3f3279848d`
BLAKE2b-256	`dde8281d32b15adda19346edc0a722e6433e6fc3e9356cfcf9bfef512d9f2a78`

Algorithm	Hash digest
SHA256	`3c4248555165e1dc5c743932941070c2f523cdaf17b3938ed04dd9a74cc543e8`
MD5	`5af7eb65e2ce2a9a3a8cf56f2b3eb195`
BLAKE2b-256	`f0b07e71463c0545f13c9e893391a1c7ba61cd4344240b1c972095bcef06f489`

Algorithm	Hash digest
SHA256	`e288fd253872c6a098d4478b2653a05ec0c6be51b1618a45f25dd78527950a96`
MD5	`6b6c3c996c36d51a24c04a21a55438fb`
BLAKE2b-256	`e30486ee8bde71c1898ce2bcf1cbc3f2b80ec749852b9031ffbf3eb7e8802aa1`

Algorithm	Hash digest
SHA256	`4a924be9e4ca165da0c88017ac33caa888d7b33c2ec9c5cf3cc17e9bf090faff`
MD5	`a4e0ac6db8bb03da6344f402172d2d5e`
BLAKE2b-256	`f8d406a86053caa38834ec91c90d1823928a03a9f9080a37dbf087a5cad98c77`

Algorithm	Hash digest
SHA256	`cf49c4f06e309a2e0b99b01037ad6595a546cdcc32206f15c8b796dc666adb03`
MD5	`8c89ef45f55507b441ec0559cd324eee`
BLAKE2b-256	`4c531b017ffc46cde8f64055ac91b18a0b37b708f13e79ba8cd51da337e030f8`

Algorithm	Hash digest
SHA256	`810d86b24865832ee61641080d59f6de148f95b300dbf5a94cf45145b33c897b`
MD5	`ab6b5ad0c6038c2ab1f91d547d141fd4`
BLAKE2b-256	`90d84f1154c70e1da8ff14d10df55e5c0e64602567051e0e9d9f1e0172a158bb`

Algorithm	Hash digest
SHA256	`58327cc6e53951f2f3df02f697ba2a429ab7a1a123cce58db708398ad24fadd2`
MD5	`3eea142e422f723d9dbb8c061707405a`
BLAKE2b-256	`b190cedbc0615d5d8a8b448b954c9d4818768ac5c45e114afee11e5918982a35`

Algorithm	Hash digest
SHA256	`9f5d8f0475d85fa5999b3c474f5c8f7fb1eaea8b38599b7c119cf4fc3ef42757`
MD5	`abd5b20dd7481226bc52d8a5af21e603`
BLAKE2b-256	`0f66a641718b57adc22ba9dffa9c3eedb8bc7719726edcb09b385d2d79b24f75`

Algorithm	Hash digest
SHA256	`30468b0c5953947bda2bba2234f0c03e724c93c89e28f816ab0f694f63fd15d0`
MD5	`de9720f8058a1dc50f5ef9f2b5f044c9`
BLAKE2b-256	`643713e3d79ebe2425769617b95510b5d2a4e1128bc6f2392ceba452b4bf8e45`

Algorithm	Hash digest
SHA256	`f928181e55f49ff14a4c47c651ff853e5dbaf21d1e5ede1f98fe53a94261346f`
MD5	`99b4f4113d5b698761543df620ef97c0`
BLAKE2b-256	`af35548fc3cb3131b5c6ebba6ede0c40951f358c6f2df71453a9d9843b358f49`

Algorithm	Hash digest
SHA256	`d9beb6cbd5d2de22816d7e3e5503dea2bb47723c0868c9cd4144b0304112b2d8`
MD5	`f68c46691ef742241c98c50699f53afd`
BLAKE2b-256	`cd1e54dd3db9c48bff0de3bd085be57c6efeef0489a8e47e84fdc00f311f4cf5`

Algorithm	Hash digest
SHA256	`4c31cdfcfc4ce304ffb970a353f563d8e668d9819d6c73f0c39c162e7175f83e`
MD5	`5ffab3a3b31e8820b3fee4e6611ad9cb`
BLAKE2b-256	`b6458897eff5a8c3e155a2c26639a1c3ba49d5d571b18e5c46abd60642a2ef4f`

Algorithm	Hash digest
SHA256	`ed0ddb8e3ab71812a37f34ee15d45688021ae0e971cd337a55a8ebf04a70b883`
MD5	`018d337cdc0308613d7b45132959fcef`
BLAKE2b-256	`b88d057d939d09ef261bdcb5273c00e6a9380d7fe2e6937eef68a2be14aad189`

stringzilla 4.6.1

Navigation

Verified details

Maintainers

Unverified details

Project links

Meta

Classifiers

Project description

StringZilla 🦖

Performance

Functionality

Quick Start: Python

Basic Usage

Basic Operations

Advanced Operations

Character Set Operations

Hash

SHA-256 Checksums

Unicode Case-Folding and Case-Insensitive Search

Collection-Level Operations

Iterators and Memory Efficiency

Low-Level Python API

Similarity Scores

Rolling Fingerprints

Serialization

Filesystem

PyArrow

Quick Start: C/C++

Basic Usage with C 99 and Newer

Basic Usage with C++ 11 and Newer

Unicode Case-Folding and Case-Insensitive Search

Similarity Scores

Rolling Fingerprints

CUDA

Memory Ownership and Small String Optimization

What's Wrong with the C Standard Library?

What's Wrong with the C++ Standard Library?

Beyond the C++ Standard Library - Learning from Python

Splits and Ranges

Concatenating Strings without Allocations

Random Generation

Bulk Replacements

Sorting in C and C++

Standard C++ Containers with String Keys

Compilation Settings and Debugging

Quick Start: Rust

Hash

SHA-256 Checksums

Unicode Case-Folding and Case-Insensitive Search

Similarity Scores

Rolling Fingerprints

Quick Start: JavaScript

Unicode Case-Folding and Case-Insensitive Search

Hash

SHA-256 Checksums

Quick Start: Swift

Unicode Case-Folding and Case-Insensitive Search

Hash

SHA-256 Checksums

Quick Start: GoLang

Unicode Case-Folding and Case-Insensitive Search

Hash

SHA-256 Checksums

Algorithms & Design Decisions

Exact Substring Search

Exact Multiple Substring Search

Levenshtein Edit Distance

Needleman-Wunsch and Smith-Waterman Scores for Bioinformatics

Memory Copying, Fills, and Moves

Hashing

SHA-256 Checksums

Random Generation

Sorting

Unicode 17, UTF-8, and Wide Characters

Case-Folding and Case-Insensitive Search

Dynamic Dispatch

Contributing 👾

License 📜

Project details

Algorithm	Hash digest
SHA256	`e351b69d77d45a96293ee065c62376531d58163dd2b74c8db2f53cafbb970896`
MD5	`fd3be62e952171f0196971b999aff96d`
BLAKE2b-256	`2ccbfc4648dcaee966bc7f7ee9b6298fe8c758ac8672f76aa58f145ad153c537`

Algorithm	Hash digest
SHA256	`a1218590cb4d644549e51c17e020cc5b95bd7129f31025c2dd1e234adb2b5817`
MD5	`b15d653a9f7ab82447ff9c4acc9d3e1f`
BLAKE2b-256	`76ea667dc6cc4f77514c6286a8ba302c371ad00fc1d36e7b39f41431f4cb564e`

Algorithm	Hash digest
SHA256	`fcb8d9e8d5ca26d56c25f04d9d55a864814a8e29ef48987a28e778c5144a3605`
MD5	`006b45073bc5ea93d2f4f1c1a184e2aa`
BLAKE2b-256	`7e9d82d3b696268bc07f3f71d2bb152dd60888047f7ea85cc6da92ca5b14d050`

Algorithm	Hash digest
SHA256	`0a37390c949fbd38a4cfb32c1482faab51322a7af34d2ad9afe71f34f2614c3c`
MD5	`3302cfbe8e2e6a189c2a60457f8fd2f7`
BLAKE2b-256	`3afff07d0896cfee42bae82a98068ff34254f354da872db58ec28a78b596e1e8`

Algorithm	Hash digest
SHA256	`39018a9b33995650fb0e5d144608024629c4e90cc9d47c90eaa4cc8cfdcbdc21`
MD5	`1145fa19cec25f95c0553cb96559ecae`
BLAKE2b-256	`0ae4763a653ca1bdef70323c8f064ba09d4437ff52cd85e3ceab128d69b6aeba`

Algorithm	Hash digest
SHA256	`dd75a62340b09cfedfddf9359bbfc20465c25f24e704331027b994199a34766e`
MD5	`a19c860f8410dae45564c6ac63f33bb2`
BLAKE2b-256	`f03c86c930bf4ef7ad4f89ac8b6768bd4afd7010c4d3b45db60a1711fc72e89c`

Algorithm	Hash digest
SHA256	`4f7d8bdbe20a61acfffbecf3cab9a741c382f8805cb68e513b78234a614548fb`
MD5	`2cf73dfdb764e3802d1bc427c9c780d8`
BLAKE2b-256	`8c1a2298271b5b8f35a9fca57841b4f37e60c91e095110dc37691ba32a0d02e0`

Algorithm	Hash digest
SHA256	`393ca09165194b08664ecd615205d26f2d40666ef971e5bdabb205e713ae2269`
MD5	`41c10050c236694971b714ff1c57484b`
BLAKE2b-256	`6adbbe3a3d58fc34512de73bc6b84ba0c423e1b5210f07e9c2874ecdd1b5bcaf`

Algorithm	Hash digest
SHA256	`25454262cd73db2d9718498ea27d61f2cebed89c10c5e8e404049d36502934c5`
MD5	`655a9a270af0643a25d608b838cf45c5`
BLAKE2b-256	`87219aa322a1618f2e63e26211f8c96d91b2eb850bd57faef4fff7dbd44e4778`

Algorithm	Hash digest
SHA256	`1d50158354618ed1b9bf18172be8ee491af30259e0d156394da9dd41d7400264`
MD5	`31c7c2d5f253be030660c89f56dea96c`
BLAKE2b-256	`6af941ebe73ab07ab56096c5f76839c98074394f8d1266ad85b850444734cb6d`