Optimize RLE decoding using a warp-balanced chunking approach - #23271

Merged

rapids-bot[bot] merged 41 commits into

NVIDIA:mainfrom

vyasr:opt/rle-chunked-expand

Aug 12, 2026

Commits on Jul 15, 2026

perf(parquet): add chunked-parse expand path to rle_stream (opt-in)
vyasr
committed
5fec7c1
perf(parquet): use chunked-parse expand in preprocess_levels_kernel
vyasr
committed
0a9f45a
docs(parquet): document decode_fixed SMEM constraint on chunked RLE
vyasr
committed
3f3e4fc
fix(parquet): fix signed UB in rle_stream chunked meta flag; add boundary tests
vyasr
committed
910da6e
refactor(parquet): use cuda::std::span and upper_bound in rle_stream chunked expand
vyasr
committed
401cdfc
refactor(parquet): remove unused chunked_smem_t struct
vyasr
committed
bb226c8
Some minor cleanup
vyasr
committed
c60de71
More comments
vyasr
committed
c3f6c05
Hoist level_bits==0 fast path into decode_next
vyasr
committed
1e5d130
test(parquet): pass cooperative group to rle_stream::init in equivalence test
vyasr
committed
8f2b7f4
perf(parquet): raise kGenRuns to 1024 on sm_80+ for chunked RLE expand
vyasr
committed
042d8c5
docs(parquet): note the 2 GiB payload-offset assumption in gen_meta
vyasr
committed
be9a068
refactor(parquet): hoist level_mask out of chunked-expand literal loop
vyasr
committed
c4facac
perf(parquet): use fixed kGenRuns=512 across all archs
vyasr
committed
fca7320
style(parquet): use snake_case for new chunked-expand identifiers
vyasr
committed
3388901
refactor(parquet): inline warp_fill into its sole caller
vyasr
committed
e2c899a
perf(parquet): use max_runs_per_chunk=1024 on sm_80+
vyasr
committed
5396008

Commits on Jul 17, 2026

Commits on Jul 23, 2026

Merge branch 'main' into opt/rle-chunked-expand
mhaseeb123
authored
4317bd2

Commits on Jul 25, 2026

Commits on Jul 26, 2026

Per-lane linear-walk phase-2 expand in decode_next_chunked
vyasr
committed
798d545

Commits on Jul 27, 2026

Use stable owners for RLE stream state

vyasr
and
sisyphus-dev-ai
committed
9e9d8bf

Commits on Jul 29, 2026

Merge branch 'main' into opt/rle-chunked-expand
mhaseeb123
authored
b120823

Commits on Jul 30, 2026

Merge remote-tracking branch 'upstream/main' into opt/rle-chunked-expand
vyasr
committed
b3515c7

Commits on Jul 31, 2026

Address safety and correctness in edge cases raised during PR review
vyasr
committed
64c6af2
Fix signedness issues with run_desc
vyasr
committed
e3738f3
Various style and code quality improvements from PR feedback
vyasr
committed
aebcc26
Single-source rle_stream type alias
vyasr
committed
44d0571
Don't waste runs shmem on chunked expansion rle_streams
vyasr
committed
ceb81d6
Remove untested cross-call partial-run resume from chunked rle_stream
vyasr
committed
789b511
docs(parquet): explain level_bits==0 fast path cur_values offset
vyasr
committed
9128028

Commits on Aug 2, 2026

Commits on Aug 10, 2026

Commits on Aug 11, 2026

fix(parquet): clamp chunked RLE run length to output window
vyasr
committed
947bb58