IntelPython / dpnp / 42434092273

84%

push

github

Fix the duplicate-indices example in the `dpnp.tensor.put` docstring (#3165)

The example of handling duplicate indices in the `dpnp.tensor.put`
docstring could not be run as written:
- it imported `dpctl.tensor` instead of `dpnp.tensor`;
- `assert vec.ndim, ind.ndim, vals.ndim == 1, 1, 1` is a `SyntaxError`,
so the snippet did not parse; it is now `assert vec.ndim == ind.ndim ==
vals.ndim == 1`;
- the positions of the last occurrences, found in the flipped `ind`,
were mapped back with `vec.size - 1` instead of `ind.size - 1`. The
example itself passed only because there `x.size == ind.size`; with a
longer array writes were lost:

```python
import dpnp.tensor as dpt

vec = dpt.zeros(10, dtype="int32")
put_vec_duplicates(vec, dpt.asarray([1, 1, 3]), dpt.asarray([10, 20, 30], dtype="int32"))
# before: [ 0  0  0 30  0  0  0  0  0  0]
# after:  [ 0 20  0 30  0  0  0  0  0  0]
```

The example from the updated docstring, extracted and run as is, passes
on a Level Zero GPU and an OpenCL CPU device, together with the case
above.

1792 of 2998 branches covered (59.77%)

Branch coverage included in aggregate %.

29619 of 34220 relevant lines covered (86.55%)

39090.05 hits per line