Repository navigation

[WIP][POC][BenchmarkingOnly] PFOR interleaved and fastlanes numbers - #51296

Draft

prtkgaur wants to merge 27 commits into

apache:mainfrom

prtkgaur:pgaur_interleavedPlusFastLanesDelta

Commits on Oct 3, 2026

Fold frame bias into bit unpacking
sfc-gh-pgaur
committed
c1109a7
Add the PFOR codec and page format
sfc-gh-pgaur
committed
7770ac4
Integrate PFOR with Parquet integer columns
sfc-gh-pgaur
committed
f24b140
Test and benchmark Parquet PFOR integration
sfc-gh-pgaur
committed
de76383
Export PFOR templates from the Arrow library
sfc-gh-pgaur
committed
706d549
Document PFOR encoding selection
sfc-gh-pgaur
committed
fe3d6ca
Add PFOR delta planning and searchable frames
sfc-gh-pgaur
committed
c07a099
Benchmark PFOR delta mode on correlated columns
sfc-gh-pgaur
committed
33feeff
Reject unprofitable PFOR delta plans early
sfc-gh-pgaur
committed
4c02050
Gate PFOR and delta mode with writer properties
sfc-gh-pgaur
committed
4fd7ac7
Validate PFOR delta metadata and page bounds
sfc-gh-pgaur
committed
25d786a
Let benchmarks force delta vectors
sfc-gh-pgaur
committed
6e15dc9
Add lane-interleaved packing to PFOR
sfc-gh-pgaur
committed
abbf366
Integrate and test interleaved PFOR writing
sfc-gh-pgaur
committed
3c2054c
Benchmark sequential and interleaved PFOR layouts
sfc-gh-pgaur
committed
b69d4f0
Add FastLanes-style lane-parallel delta kernels
sfc-gh-pgaur
committed
ddb3faa
Decode lane-parallel delta pages
sfc-gh-pgaur
committed
02564e2
Frame FastLanes delta pages
sfc-gh-pgaur
committed
c7f3dd0
Benchmark interleaved PFOR on the corpus
sfc-gh-pgaur
committed
954f10a
Cap bit-unpack dispatch at AVX2
sfc-gh-pgaur
committed
94a5b81
Fuse FL_ORDER transpose into unpacking
sfc-gh-pgaur
committed
e9e877f
Dispatch interleaved kernels by ISA
sfc-gh-pgaur
committed
5d0f349
Share one column corpus across PFOR benchmarks
sfc-gh-pgaur
committed
e0c5250
Transpose FL_ORDER blocks in NEON registers
sfc-gh-pgaur
committed
7f697e0

Commits on Oct 4, 2026

Generalize block kernels over element width
sfc-gh-pgaur
committed
e27369f

Commits on Oct 5, 2026

Add a bit-unpack width and footprint benchmark
sfc-gh-pgaur
committed
36f0854
Shift narrow lanes in place on AVX2
sfc-gh-pgaur
committed
37104b8