add compiler hints to reduce number of slice bounds checks

[NyaaaWhatsUpDoc]

Just some quick changes that hint to the compiler expected lengths in places to reduce the number of slice bounds checks it needs to do. You can confirm before and after results by using -gcflags="-d=ssa/check_bce/debug=1", or more usefully i was specifically doing go test -gcflags="-d=ssa/check_bce/debug=1" -v 2>&1 1>/dev/null | grep $fileNameToCheck

[NyaaaWhatsUpDoc]

in particular a good example of another use of the underscored slice value accessing is in binary.{LittleEndian,BigEndian}

[ncruces]

This looks like a great idea, and you covered some of the hot points.

Was it profile guided, should I look elsewhere?

Thanks!

[NyaaaWhatsUpDoc]

This looks like a great idea, and you covered some of the hot points.

Was it profile guided, should I look elsewhere?

Thanks!

i wish i could say it was anything as intelligent as having been directed by profile guided optimizations... it was more just i was bored, fiddling around, and noticed an opportunity for improvement :')

[ncruces]

The func.go ones can be really hot points, maybe not on “normal” usage (where callbacks are mostly used to read data from disk), but certainly if used in (e.g.) an aggregate function or collation sequence (where a callback may be called millions of times to sort a row set, or calculate statistics).

Columns is not that hot, but it's still almost single handedly responsible for this driver being close to as fast in queries as mattn.

[ncruces]

This lead to a 7% improvement on this benchmark.
Still lots of overhead, but every bit helps, so thanks!