[REVIEW] Optimize cudf.concat for axis=0

[galipremsagar]

This PR optimizes cudf.concat when axis=0 by not materializing RangeIndex objects present as index to the Dataframe objects.

Partially addresses #9200, This is 1/2 of full optimizations. A follow-up PR to optimize axis=1 will be opened as there are multiple large changes.

Here is a benchmark:
On branch-21.10:

IPython 7.27.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: import cudf

In [2]: df = cudf.DataFrame({'a':[1, 2, 3]*100})

In [3]: df2 = cudf.DataFrame({'a':[1, 2, 3]*100}, index=cudf.RangeIndex(300, 600))

In [4]: %timeit cudf.concat([df, df2])
806 µs ± 8.02 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

This PR:

IPython 7.27.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: import cudf

In [2]: df = cudf.DataFrame({'a':[1, 2, 3]*100})

In [3]: df2 = cudf.DataFrame({'a':[1, 2, 3]*100}, index=cudf.RangeIndex(300, 600))

In [4]: %timeit cudf.concat([df, df2])
434 µs ± 4.35 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.10@c6ddd46). Click here to learn what that means.
The diff coverage is n/a.

❗ Current head dd6338c differs from pull request most recent head f3f134d. Consider uploading reports for the commit f3f134d to get more accurate results

@@               Coverage Diff               @@
##             branch-21.10    #9222   +/-   ##
===============================================
  Coverage                ?   10.81%           
===============================================
  Files                   ?      115           
  Lines                   ?    19170           
  Branches                ?        0           
===============================================
  Hits                    ?     2074           
  Misses                  ?    17096           
  Partials                ?        0           

---

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 c6ddd46...f3f134d. Read the comment docs.

[shwina]

Is this case not included in line 1218?

[galipremsagar]

Nope, that line's expectation is to have the index columns materialized. Whereas we don't want to materialize and hit the specialized concat rangeindex logic already present in index.py:

https://github.com/rapidsai/cudf/blob/branch-21.10/python/cudf/cudf/core/index.py#L688-L702

[galipremsagar]

Especially we want to hit _concat_range_index in this case:

cudf/python/cudf/cudf/core/index.py

Line 2311 in eab2486

def _concat_range_index(indexes: List[RangeIndex]) -> BaseIndex:

[shwina]

@gpucibot merge