Avatar for the vortex-data user
vortex-data
vortex
BlogDocsChangelog

Performance History

Latest Results

feat[gpu]: multi-stage dynamic dispatch with plan builder Evolve the dynamic dispatch kernel from a single-pipeline design (one source op + scalar ops) to a multi-stage plan. Input stages decode child arrays (dictionary values, run-end endpoints) into shared memory; the output stage references them and writes directly to global memory. Add `build_plan`, which walks a Vortex encoding tree and automatically produces a `DynamicDispatchPlan`. Supported encodings: BitPacked, FoR, ZigZag, ALP (f32), Dict, RunEnd, and Primitive. Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
ad/dyn-dispatch
5 hours ago
chore(deps): update dependency ray to v2.54.0 [security]
renovate/pypi-ray-vulnerability
1 day ago

Active Branches

feat[gpu]: multi-stage dynamic dispatch with plan builder
last run
5 hours ago
#6636
CodSpeed Performance Gauge
0%
last run
1 day ago
#6635
CodSpeed Performance Gauge
-65%
#6634
CodSpeed Performance Gauge
-27%
© 2026 CodSpeed Technology
Home Terms Privacy Docs