Ilya Enkovich

Don't return nullptr for empty results and fix empty record batch to make it suitable for Arrow Table construction.

Shared execution context draft

2

A shared execution context is used to reduce initialization and reduction costs. The feature is in a draft state, works only for a keyless perfect hash with COUNT and SUM...

Use tbb::parallel_sort instead of std::sort.

5

Trivial change to get parallel sort. Get 4x speed-up for TCP-H Q11 on scale=500.

Hi, On my system I have failing test `StringTransform.toString`. The output is: ``` ../Tests/StringTransformTest.cpp:247: Failure Expected equality of these values: ::toString(v) Which is: "[&A1, &A2]" "[A1, A2]" ``` Here is...

This patch introduces optimized buffer initializers for join hash tables and groupby buffers. For join hash table we split work between threads in a more CPU-friendly manner, giving each thread...

Currently we can get huge perfect hash tables with just a few valid elements. We found it especially hurtful for window functions where each empty entry means a partition requiring...

There is unnested arrays related logic in QueryMemoryDescriptor::pick_target_compact_width which seems to be broken. The loop [here](https://github.com/omnisci/omniscidb/blob/master/QueryEngine/Descriptors/QueryMemoryDescriptor.cpp#L670) iterates through target exprs and input columns at the same time. The logic for...

Trying to run H2O Groupby Query 10 I encountered the following error: ``` F 62751 StringDictionary.cpp:547 Check failed: string_id < static_cast(str_count_) (1119526925 < 100) ``` I run the server with...