One of our longest compile times is for drop_duplicates.cu measured currently at about 15 minutes. This file contains two cudf APIs: cudf::drop_duplicates and cudf::distinct_count. They share no code so it is reasonable to move the distinct_count to its own source file. Both files individually have a compile time around 7 minutes.

… HEAD

[REVIEW] Parquet writer int96 [skip ci]

…6837)

Fixes #6803

This optimizes string slice copying in the case where the string column view starts at offset 0. In that case the offset values do not need to be modified, and all the column buffers can be copied.

Addendum to #6768, to support scatter on columns of type list<struct>. The hard part was the tests.

Co-authored-by: Karthikeyan Natarajan <karthikeyann@users.noreply.github.com>
Co-authored-by: Mark Harris <mharris@nvidia.com>

closes #6465

- Add utility `cudf::detail::align_ptr_for_type`
- Add contiguous_copy_column_device_views
- reduces multiple HtoD copies in cudf::concatenate by adding create_contiguous_device_views for list of column_views

Correct the param order of writeParquetBufferBegin[skip ci]

Fixes: #6832

Fixes #521 - Replaces macros with constexpr variables and inline functions wherever applicable.

…cu (#6835)

This change reduces the compile time for hash/groupby.cu from 16 minutes to 9 minutes.
The `skip_rows_with_nulls` template parameter is set to avoid calling (and inlining) `cudf::bit_is_set()`. This function is minimal compared to the `cudf::detail::aggregate_row` function that must be inlined twice to accommodate this template parameter. Simply changing this to a member variable means we still do not incur an extra call to `cudf::bit_is_set()` when appropriate but also means we generate half as much device code for this specific function. The `cudf::detail::aggregate_row` code is quite significant.

This PR attempts to address issue #6842, which may be caused by multiple reasons. For now, one certain problem is contiguous splitting on null (fully invalid) string columns, which only contains single child column (offset column). This PR is about to fix this problem.

[REVIEW] fix uint32_t undefined errors

Closes #6857 

This function contains only C++ STL code and no CUDA code. Also, including it in the `column_device_view.cuh` required a new dependency on `cub/cub.cuh` (per the `cuda.cuh`) as found when building blazingSQL.

Authors:
  - davidwendt <dwendt@nvidia.com>

Approvers:
  - Ram (Ramakrishna Prabhu)
  - Mark Harris
  - Ray Douglass

URL: #6859

Removed unused stream parameter from type-dispatcher benchmark source

Reference #5963 

This PR adds dictionary column type support to the set of `cudf::reduce` functions.
This PR depends on utilities added in PR #6651 

Here are the reduce operations that will be included in this PR.
- [x] all
- [x] any
- [x] max
- [x] mean
- [x] median
- [x] min
- [x] nth_element
- [x] product
- [x] quantile
- [x] std
- [x] sum_of_squares
- [x] sum
- [x] unique_count
- [x] var

Authors:
  - davidwendt <dwendt@nvidia.com>

Approvers:
  - Mike Wendt
  - AJ Schmidt
  - Ram (Ramakrishna Prabhu)
  - Karthikeyan

URL: #6666