[BUG] DataFrame setitem causing data corruption for cupy array inputs

[esnvidia]

Describe the bug
Setting the values obtained from multiple columns of a StandardScaler to the GDF is not working (although this might be user error)

Steps/Code to reproduce bug

Code is here
Untitled.zip

Expected behavior
Set the values to the GDF. All the columns have the same values in the GDF despite the output not showing it should be the case.
Looks like values are

Environment overview (please complete the following information)
conda install rapids 21.06, bare-metal.

[beckernick]

The following is a minimal, reproducible example that demonstrates the underlying setitem issue causing data corruption in 21.08. In this example, the first 10 values of the contiguous array are being set into the 10 rows for each column.

import cudf
import cupy as cp
​
df = cudf.DataFrame({
    "a": range(10),
    "b": range(10),
})
arr = cp.arange(20,40).reshape(-1, 2)
df.loc[:, ["a","b"]] = arr
​
print(arr)
print(df)
[[20 21]
 [22 23]
 [24 25]
 [26 27]
 [28 29]
 [30 31]
 [32 33]
 [34 35]
 [36 37]
 [38 39]]
    a   b
0  20  20
1  21  21
2  22  22
3  23  23
4  24  24
5  25  25
6  26  26
7  27  27
8  28  28
9  29  29