[BUG] RangeIndex shouldn't be materialized in cudf.concat

[galipremsagar]

Describe the bug
There is a performance overhead in cudf.concat when the type of index in dataframes is RangeIndex because the RangeIndex objects are being materialized. This is also a behavioural deviation from what pandas does, i.e., don't materialize RangeIndex if they can be homogeneously concatenated.

Steps/Code to reproduce bug

>>> import cudf
>>> df = cudf.DataFrame({'a':[1, 2, 3]})
>>> df2 = cudf.DataFrame({'a':[1, 2, 3]})
>>> df2 = cudf.DataFrame({'a':[1, 2, 3]}, index=cudf.RangeIndex(3, 6))
>>> cudf.concat([df, df2])
   a
0  1
1  2
2  3
3  1
4  2
5  3
>>> cudf.concat([df, df2]).index
Int64Index([0, 1, 2, 3, 4, 5], dtype='int64')
>>> import pandas as pd
>>> pd.concat([df.to_pandas(), df2.to_pandas()]).index
RangeIndex(start=0, stop=6, step=1)

Expected behavior
Returned dataframe should have a RangeIndex.

Environment overview (please complete the following information)

• Environment location: [Bare-metal]
• Method of cuDF install: [from source]

[vyasr]

Did I break this in my recent refactoring somewhere? I distinctly recall you bringing up and fixing this issue at some point.

[vyasr]

Ah no I see you fixed it for concatenation of standalone RangeIndex in #6872 but not for when concatenating DataFrames that have a RangeIndex index.

[galipremsagar]

Ah no I see you fixed it for concatenation of standalone RangeIndex in #6872 but not for when concatenating DataFrames that have a RangeIndex index.

Yes, that's correct. Not a regression due to refactoring.

[galipremsagar]

Part of this issue has been resolved in #9222 and addressing other part of the issue for axis=1 cases need non-trivial amount of changes, hence pushing to 21.12.

[galipremsagar]

Completely Fixed by #9333