andygrove commented on issue #547:
URL:
https://github.com/apache/datafusion-comet/issues/547#issuecomment-2168486571
@advancedxy I am working on re-implementing this now in a simpler way as an
experiement ... I will aim to have a PR up by end of day
--
This is an automated message from
andygrove commented on issue #547:
URL:
https://github.com/apache/datafusion-comet/issues/547#issuecomment-2168451782
Here are benchmark results comparing murmur3 to xxhash64. xxhash64 is 3x
slower (not sure if that is the expectation) but what is more interesting is
that there is a warnin
andygrove commented on issue #547:
URL:
https://github.com/apache/datafusion-comet/issues/547#issuecomment-2162053525
Thanks @advancedxy. Yes, it definitely needs more investigation.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to G
advancedxy commented on issue #547:
URL:
https://github.com/apache/datafusion-comet/issues/547#issuecomment-2162006398
Linking this to #517 too.
> Are we expecting a performance improvement as a result? (I doubt stack
allocations are a performance bottleneck)
+1. I think we m
andygrove opened a new issue, #547:
URL: https://github.com/apache/datafusion-comet/issues/547
### What is the problem the feature request solves?
Here is the current xxhash64 implementation:
```rust
pub(crate) fn spark_compatible_xxhash64>(data: T, seed: u64)
-> u64 {
