etseidl commented on PR #9945:
URL: https://github.com/apache/arrow-rs/pull/9945#issuecomment-4432401208
> I ran the benchmarks locally on my machine (Intel Tigerlake), and this
branch seems to be around 5% faster on all metadata benchmarks, compared to
`main`.
Same, but all on intel as well. I've been playing around on godbolt and
honestly do not understand why there's any difference at all. However I write
`read_field_begin` it always boils down to:
```rust
let field_type = self.read_byte()?;
if field_type == 0 {
// exit early
} else if field_type < 16 { // unlikely
// read full i16 field_id and return
} else {
// compute field_id from delta and return
}
```
Lacking an aarch64 machine to profile on, it's hard to see why the
benchmarks here are so different. It's possible that the difference comes from
either the new `bool_val()` function or some quirk in the error handling.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
