Optimization to decoding of parquet level streams#13203

Merged

rapids-bot[bot] merged 16 commits intorapidsai:branch-23.06from nvdbaranec:parquet_level_optimization

May 15, 2023

+725-203

Commits on Apr 23, 2023

Rework of level decoding to be considerably more parallel. Previously, it was only 1 warp wide. Now it is block-wide.
nvdbaranec
committed
Merge branch 'branch-23.06' into parquet_level_optimization
nvdbaranec
committed

Commits on Apr 24, 2023

Commits on May 1, 2023

Remove definition and repetition levels from page_data_s struct to deal with a performance issue introduced in gpuDecodePageData by
nvdbaranec
committed

Commits on May 3, 2023

Commits on May 4, 2023

Commits on May 8, 2023

Use a dynamically sized type for level/repetition data. In almost all cases, we only need 1 byte to store level
nvdbaranec
committed

Commits on May 9, 2023

Commits on May 10, 2023

Commits on May 15, 2023