SIMD-accelerated XML parser with full XPath 1.0 support.

Parses XML into flat arrays instead of a DOM tree using structural indexing adapted from simdjson. Evaluates XPath expressions against those arrays with cache-friendly access patterns. Includes sxq, a fast command-line tool for XML/XPath queries.

Benchmarked on Apple M4 Max (NEON) and AMD Ryzen 9 3950X (AVX2).

End-to-end XPath queries (M4 Max, NEON):

Attribute-heavy XML (Ryzen 9 3950X, AVX2):

Library parse throughput vs Rust parsers (M4 Max, criterion, parse only):

Wins are largest on attribute-dense XML (where SIMD quote masking shines) and multi-gigabyte files (where flat-array memory efficiency dominates). pugixml wins on scalar aggregation queries (count()) and text predicates (contains()) where its single-pass DOM build has lower overhead. On small files (<1 MB), all native tools are within startup noise. Unlike quick-xml (streaming events) and roxmltree (read-only DOM), simdxml produces a queryable structural index with XPath support.

cargo install simdxml-cli

sxq '//title' book.xml
sxq '//claim[@type="independent"]' patents/*.xml
sxq -c '//record' huge.xml
sxq -r '//path' drawing.svg
sxq -j '//author' papers/*.xml
sxq 'count(//article)' dblp.xml
cat feed.xml | sxq '//item/title'
sxq info large.xml

Files are mmap'd. Large files are parsed in parallel. 684 KB binary, zero dependencies.

cargo add simdxml

use simdxml::{parse, CompiledXPath};

let xml = b"<library><book><title>Rust</title></book></library>";
let index = parse(xml).unwrap();

// One-shot query
let titles = index.xpath_text("//title").unwrap();
assert_eq!(titles, vec!["Rust"]);

// Compiled query (reusable across documents)
let query = CompiledXPath::compile("//title").unwrap();
let titles = query.eval_text(&index).unwrap();

// Scalar expressions
let mut index = parse(xml).unwrap();
let count = index.eval("count(//book)").unwrap(); // Number(1)

Full support for all 13 axes, predicates, functions, operators, and unions. 327/327 libxml2 conformance tests, 1015/1023 pugixml conformance tests.

All axes: child, descendant, descendant-or-self, parent, ancestor, ancestor-or-self, following-sibling, preceding-sibling, following, preceding, self, attribute, namespace

Functions: string(), count(), sum(), boolean(), not(), contains(), starts-with(), substring(), concat(), normalize-space(), translate(), string-length(), number(), floor(), ceiling(), round(), id(), name(), local-name(), namespace-uri(), position(), last(), true(), false(), lang()

Two-pass structural indexing (no DOM tree):

1. Parse — Scan XML bytes with SIMD-accelerated character classification, build flat arrays (tag offsets, types, depths, parents, text ranges). ~16 bytes/tag vs ~35 for a DOM node.
2. Query — Evaluate XPath against the arrays using CSR indices, pre/post-order numbering for O(1) ancestor checks, and inverted name posting lists.

Additional features: parallel parsing across cores, lazy index building, bloom filter prescan for batch skip, persistent .sxi index files.

crates/simdxml/      Rust library (crates.io: simdxml)
crates/simdxml-cli/  CLI tool (crates.io: simdxml-cli, binary: sxq)
bench/               Benchmark scripts and pugixml comparison shim
testdata/            Conformance test suites (libxml2, pugixml)

MIT OR Apache-2.0