tisonkun opened a new pull request, #312: URL: https://github.com/apache/datasketches-rust/pull/312
Scalar TDigest rank queries can regress when shared batch-query code changes its layout and inlining. Specialize scalar, sparse-batch, and dense-scan paths at compile time, inline query wrappers, and reuse known endpoint bounds. Dense scans accumulate weights as they advance, and PMF carries the previous CDF value to avoid rereading it. Adds CDF/scalar-rank equivalence coverage and benchmarks for mixed queries, small digests, mutable rank, and sparse/dense PMF batches. Measured on Apple M4 Max with Rust 1.99.1 release, relative to the TDigest implementation before #305: | Workload | Latency change | | --- | ---: | | Uniform-data point rank | -4.0% | | Mutable rank, already compressed | -3.8% | | Repeated minimum-value rank | -55.1% | | CDF, 100 split points | -75.7% | | PMF, 100 split points | -76.4% | Single-split PMF retains the approximately 2.4% microbenchmark slowdown already present in #305; these measurements do not establish a universal zero-regression guarantee. Validation: `cargo x check`, `cargo x test`, and `cargo x lint`. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
