Use cuda::std::is_arithmetic in cudf::is_numeric trait.

[bdice]

The current implementation of cudf::is_numeric is equivalent to the implementation of cuda::std::is_arithmetic. This PR simplifies the implementation in cuDF and aligns it with libcudacxx.

Notes:

• bool returns true from both (cuda::)std::is_integral and (cuda::)std::is_arithmetic, so bool is considered a "numeric" type. This behavior is unchanged.
• We must use cuda::std::is_arithmetic rather than std::is_arithmetic to support 128-bit integer types (as we do with cuda::std::is_integral in the current implementation)

See also: https://en.cppreference.com/w/cpp/types/is_arithmetic

[davidwendt]

I think we should just use cuda::std::is_arithmetic directly and get rid of cudf::is_numeric. No need to have a cudf wrapper around a single std trait.

[bdice]

I think we should just use cuda::std::is_arithmetic directly and get rid of cudf::is_numeric. No need to have a cudf wrapper around a single std trait.

That's possible. @jrhemstad mentioned in conversation with me that it is nice that the "numeric" name aligns with make_numeric_column, among other "numeric" names (like numeric scalars). I'm open to whatever reviewers think is most appropriate here. Personally I lean towards keeping our own wrappers for ensuring internal consistency, since we define and use a lot of type traits that are not easily expressible in terms of those in <cuda/std/type_traits>.

[davidwendt]

That's fine. At the very least we could remove the cudf::is_numeric() that is only needed for the dispatch impl:

template <typename T>
constexpr inline bool is_numeric()
{
  return cuda::std::is_arithmetic<T>();
}

struct is_numeric_impl {
  template <typename T>
  constexpr bool operator()()
  {
    return cuda::std::is_arithmetic<T>();
  }
};

[bdice]

That's fine. At the very least we could remove the cudf::is_numeric() that is only needed for the dispatch impl:

I'm not sure if that's possible here -- lots of code appears to use the template form of this trait as well as the form accepting a data_type that does a type dispatch through is_numeric_impl: https://github.com/rapidsai/cudf/search?q=is_numeric (This search lists both cases, since GitHub code search ignores < and >.) Examples:

• cudf/cpp/src/unary/cast_ops.cu

  Line 38 in 024003c

  typename std::enable_if_t<(cudf::is_numeric<SourceT>() &&

• cudf/cpp/src/filling/sequence.cu

  Line 60 in 31f92d7

  typename std::enable_if_t<cudf::is_numeric<T>() and not cudf::is_boolean<T>()>* = nullptr>

• cudf/cpp/tests/rolling/rolling_test.hpp

  Line 33 in 31f92d7

  } else if (cudf::is_numeric<ColumnType>() or cudf::is_duration<ColumnType>()) {

[codecov]

Codecov Report

Merging #9996 (17be02a) into branch-22.02 (de8c0b8) will increase coverage by 0.03%.
The diff coverage is n/a.

@@               Coverage Diff                @@
##           branch-22.02    #9996      +/-   ##
================================================
+ Coverage         10.41%   10.45%   +0.03%     
================================================
  Files               119      119              
  Lines             20023    20417     +394     
================================================
+ Hits               2086     2134      +48     
- Misses            17937    18283     +346     

Impacted Files	Coverage Δ
python/cudf/cudf/io/hdf.py	0.00% <0.00%> (ø)
python/cudf/cudf/io/orc.py	0.00% <0.00%> (ø)
python/cudf/cudf/_version.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/abc.py	0.00% <0.00%> (ø)
python/cudf/cudf/api/types.py	0.00% <0.00%> (ø)
python/cudf/cudf/io/dlpack.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/frame.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/index.py	0.00% <0.00%> (ø)
python/cudf/cudf/io/feather.py	0.00% <0.00%> (ø)
python/cudf/cudf/io/parquet.py	0.00% <0.00%> (ø)
... and 48 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 de8c0b8...17be02a. Read the comment docs.

[bdice]

@gpucibot merge