vortex-data/vortex - CodSpeed

vortex

Blog Docs Changelog

Performance History

Latest Results

fix Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>

myrrc/duckdb-bench-timer

5 minutes ago

fix(cuda): stage the OnPair dictionary from device-resident buffers OnPairExecutor::stage_dict builds the split dictionary layout on the host and copies it to the device, but it reached its two inputs through host-only accessors. Arrays read through the CUDA flat layout arrive with their buffers already device-resident, so dict_view panicked in BufferHandle::as_host with "expected host buffer", and its collect_widened call panicked in the CPU cast kernel with "as_slice must be called on host buffer". Copy the dictionary blob back with BufferHandle::try_into_host, and decode the dict_offsets child on the GPU like the other three children before widening it on the host. Both are bounded by the dictionary size rather than the array length. Building the dictionary from those parts needs a constructor that does not go through the array, so add OnPairDictionary to the onpair crate. This does not address the other GPU OnPair gap: the Delta arrays OnPair emits internally still have no CUDA kernel. Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

claude/gpu-compress-benchmarks-x788xc

6 minutes ago

Fold the drifting-values helper into its only caller Workspace-wide clippy unifies features, so vortex-btrblocks builds with pco enabled but not unstable_encodings. In that combination both users of the ArrayRef import were compiled out and the import was left unused, failing the clippy-default lint. Inlining the helper into the test that uses it leaves ArrayRef with a single user, gated the same way, so no feature combination strands it. Verified with clippy over default, pco, zstd, all-features, and no-default-features. Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

claude/array-compression-sampling-qqbhm7

7 minutes ago

feat: support SpatialBench in DataFusion benchmark Signed-off-by: Baris Palaska <barispalaska@gmail.com>

bp/datafusion-spatialbench

13 minutes ago

Merge remote-tracking branch 'origin/develop' into claude/array-compression-sampling-qqbhm7

claude/array-compression-sampling-qqbhm7

18 minutes ago

Make chunked decompression an explicit capability, not a silent fallback The default VTable::decompress_chunks previously fell back to executing the array to canonical and streaming the materialized result, which silently defeated the purpose of the API: decompressing and transforming blocks while they are L1-resident. Streaming is now advertised via VTable::supports_decompress_chunks (default false); wrapper encodings propagate the check through the children they stream from, so support of the whole tree is decided before any chunk is emitted. ArrayRef::decompress_chunks errors on unsupported trees without emitting anything, and the materializing two-pass fallback moves to the explicitly named ArrayRef::decompress_chunks_or_materialize. Primitive gains a real streaming implementation (it is already decompressed; chunks are copied through one reusable L1 scratch buffer), Constant and BitPacked advertise support directly, FoR advertises support when its fused BitPacked path applies or its child supports streaming, and Patched advertises support when its inner array does. Signed-off-by: "Joe Isaacs" <joe.isaacs@live.co.uk> Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

claude/decompression-chunked-iteration-u3kcns

20 minutes ago

fix: omit baseline-only formats from the file-size report (#9589) Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

develop

23 minutes ago

fix: update the delta snapshot (#9608) ## Summary Fixes: https://github.com/vortex-data/vortex/actions/runs/32849513434/job/97806878852 Updates the snapshot now that both delta cost estimation issues are fixed. Now that the delta estimation is accurate, pco is correctly picked by compressor and results in better compression ratio. Signed-off-by: rapour <reza.apour@gmail.com>

develop

26 minutes ago

Latest Branches

-1%

Add BlockedFoR: a frame of reference per 1024 values#9609

15 minutes ago

ee07a98

claude/array-compression-sampling-qqbhm7

+17%

Support SpatialBench in DataFusion benchmark#9615

15 minutes ago

ffe767d

bp/datafusion-spatialbench

0%

bench: run the full compress suite under `--gpu-decompress`#9399

5 days ago

6c85266

claude/gpu-compress-benchmarks-x788xc

© 2026 CodSpeed Technology

Home Terms Privacy Docs