Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
This is essential to reduce microperformance overhead as has been discussed and investigated many other places. This first stage of work is to remove the use of Datum and ExecBatch from the input side of only scalar kernels, so that we can work toward using span/view data structures as the inputs (and eventually outputs) of all kernels.
Attachments
Issue Links
- is a child of
-
ARROW-16755 [C++] Improve array expression and kernel evaluation performance on small inputs
- Open
- is related to
-
ARROW-16819 [C++] arrow::compute::CallFunction needs a batch length for nullary functions
- Resolved
-
ARROW-16824 [C++] Migrate VectorKernel implementations to use ExecSpan, ArraySpan
- Resolved
- links to