Add detail interface for split and slice(table_view), refactors both function with host_span

[isVoid]

cudf::detail::slice performs a segmented_count_unset_bits that requires stream ordering. The depending split interface does not have an internal version that accepts a stream argument. Similarly for slice(table_view). This PR fixes that.

Besides, slice/split interface is refactored to accept host_span to specify indices/splits, and is overloaded with std::initializer_list. This allows specifying the argument with both a container and a brace-init-list.

[codecov]

Codecov Report

Merging #9226 (aa67638) into branch-21.12 (ab4bfaa) will decrease coverage by 0.03%.
The diff coverage is 0.00%.

❗ Current head aa67638 differs from pull request most recent head d21b646. Consider uploading reports for the commit d21b646 to get more accurate results

@@               Coverage Diff                @@
##           branch-21.12    #9226      +/-   ##
================================================
- Coverage         10.79%   10.75%   -0.04%     
================================================
  Files               116      116              
  Lines             18869    19481     +612     
================================================
+ Hits               2036     2096      +60     
- Misses            16833    17385     +552     

Impacted Files	Coverage Δ
python/cudf/cudf/__init__.py	0.00% <0.00%> (ø)
python/cudf/cudf/_lib/__init__.py	0.00% <ø> (ø)
python/cudf/cudf/core/_base_index.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/categorical.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/column.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/datetime.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/lists.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/numerical.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/string.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/column/timedelta.py	0.00% <0.00%> (ø)
... and 76 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4d7e69a...d21b646. Read the comment docs.

[ttnghia]

I think slice is used in more places than those currently modified in this PR. Did you try to search all?

[thomcom]

If the tests passed it would lgtm.

[vyasr]

rerun tests

[isVoid]

I think slice is used in more places than those currently modified in this PR. Did you try to search all?

Changed a couple more places with use of slice. What confused me at one point is that there are two slice interfaces:

This does not count nulls, so no need to use stream.

cudf/cpp/include/cudf/detail/copy.hpp

Line 45 in 015f15c

ColumnView slice(ColumnView const& input, cudf::size_type begin, cudf::size_type end)

This one does.

cudf/cpp/include/cudf/detail/copy.hpp

Lines 74 to 76 in 015f15c

	std::vector<column_view> slice(column_view const& input,
	std::vector<size_type> const& indices,
	rmm::cuda_stream_view stream = rmm::cuda_stream_default);

And I only refactor usages of the latter.

[jrhemstad]

Since you're already touching it, can you change split/slice input parameters from std::vector to host_span?

[isVoid]

The PR became a bit larger because of refactoring the indices specification with host_span. But most are just trivially replacing the brace initialization for vector with an explicit std::vector construction.

Vscode regex find and replace helped.

[isVoid]

Added an overload of split/slice interface with initializer_list for the indices/splits argument. This allows no change in existing usage of the API.

[isVoid]

@gpucibot merge