feat: single-pass value_at_percentiles / values_at_quantiles batch API (7.2x)#138
Open
fcostaoliveira wants to merge 1 commit into
Open
Conversation
fcostaoliveira
pushed a commit
to redis-performance/hdr-agent-workspace
that referenced
this pull request
Jul 2, 2026
…+7.2x (PR #138) Race-driven optimizations of the Go and Rust ports (submodules kept at official tips; wins on fork branches + PRs): - GO-EXP-001: flat counts[] scan in ValueAtPercentile — +133% (0.0457->0.1066 Mq/s), gnr1 same-session A/B, go test green, results byte-identical. PR HdrHistogram/hdrhistogram-go#57. - RUST-EXP-001: single-pass value_at_percentiles/values_at_quantiles batch API — +616% (7.2x) over 7x singular (24.9K->178.3K calls/sec). v1 naive was slower; v2 hoisted next-target -> tight loop. bsink byte-identical. PR HdrHistogram/HdrHistogram_rust#138. Logged EXPERIMENTS.md + RACE.md (progress + post-opt standings) + memory.
Add batch queries that resolve N quantiles/percentiles in ONE scan over counts[], instead
of N separate value_at_quantile calls. Requested quantiles are resolved in ascending-target
order; the next target is hoisted into a local so the hot loop stays a tight
'total += counts[i]; if total >= next_target' (same shape as the singular scan).
Naming matches the existing value_at_quantile/value_at_percentile pair; both new methods are
#[must_use]. Results are byte-identical to per-quantile calls including unsorted/duplicate
inputs, q==0.0/1.0, q>1.0 clamp, and empty histograms (verified by tests). NaN quantiles do
not panic (the sort is over the derived u64 targets, not the f64 inputs).
Batch throughput for all 7 of {50,75,90,95,99,99.9,99.99} (Granite Rapids, single core):
178,326 calls/sec vs 24,896 for 7x value_at_percentile = +616% (7.2x).
fcostaoliveira
force-pushed
the
perf/value-at-percentiles-batch
branch
from
96fa8ab to
26e3d39
Compare
Author
|
Updated to address review: renamed |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Add single-pass batch queries:
values_at_quantiles(&[f64]) -> Vec<u64>value_at_percentiles(&[f64]) -> Vec<u64>They resolve N quantiles/percentiles in one scan over
counts[]instead of N separatevalue_at_quantilecalls. Requested quantiles are resolved in ascending-target order; the nexttarget is hoisted into a local so the hot loop stays a tight
total += counts[i]; if total >= next_target(the same shape as the singular scan) — per-crossingbookkeeping runs only when a threshold is actually reached. Results are returned in input order.
Benchmark
Getting all 7 of {50,75,90,95,99,99.9,99.99}, single core (Intel Granite Rapids):
value_at_percentilevalue_at_percentiles(this PR)+616% (7.2×).
Correctness
value_at_quantile/value_at_percentile, verified forboth sorted and unsorted inputs including the
0.0/100.0edges.recordand the singular query paths are unchanged.Happy to add doc examples and a criterion/
#[bench]entry if you'd like.