Prem Sagar Gali created ARROW-11872:
---------------------------------------
Summary: Array Validation of GPU buffers fails due to incorrect
validation check
Key: ARROW-11872
URL: https://issues.apache.org/jira/browse/ARROW-11872
Project: Apache Arrow
Issue Type: Bug
Components: C++, Python
Affects Versions: 3.0.0
Reporter: Prem Sagar Gali
Whenever we try to create an array or a Table from a GPU buffer we get the
following Invalid missing values buffer error.
pyarrow/table.pxi:485: in pyarrow.lib.chunked_array
???
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ > ???
E pyarrow.lib.ArrowInvalid: In chunk 0: Invalid: Missing values buffer in
non-empty arraypyarrow/error.pxi:84: ArrowInvalid
This is because the GPU buffers aren't passing the following validation check:
[https://github.com/apache/arrow/blob/7be266be98d08214ef4b3f067059b6e6a8a6e5c0/cpp/src/arrow/array/validate.cc#L183]
But this has to be changed to something like:
static bool IsBufferValid(const ArrayData& data, int index) {
return data.buffers[index] != nullptr && (data.buffers[index]->is_cpu() ?
data.buffers[index]->data() != nullptr : true);
}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
