Guo Feng

Thanks ! I've fixed these places and will do a benchmark based on the newest code tonight.

Thanks @uschindler . > I don't fully understand the variant that supports negative values. Do we still need it as the default one again supports negative values. The code looks...

Here is the report (30 JVM * 200 repeat): ``` TaskQPS baseline StdDevQPS my_modified_version StdDev Pct diff p-value HighTermTitleBDVSort 34.07 (21.0%) 32.13 (12.4%) -5.7% ( -32% - 35%) 0.201 HighTermMonthSort...

FYI here are some of my thoughts based on all these benchmark reports: * Rolling up loops for vint/vlong seems not bring a significant speed up (neither significant regression), IMO...

Here is the bechmark result on Java 11: ``` TaskQPS baseline StdDevQPS my_modified_version StdDev Pct diff p-value BrowseMonthSSDVFacets 6.28 (17.6%) 6.02 (16.9%) -4.2% ( -32% - 36%) 0.441 IntNRQ 30.10...

Thanks. > Did you check for other ReadVint variants? There may be more. Maybe use the type hierarchy functions to investigate all child classes of DataInput. I'm using IntelliJ IDEA...

Thanks a lot for so quick feedback! Using FOR/PFOR sounds promising, i'll give it a try.

Thank you very much for your guidance! This is of great help to me. I will fork a new `DocvalueFormat` for this experiment. Actually, what I thought at first was...

@jpountz Thanks a lot for the suggestion! > Let's update this pull request to limit the impact to the API to DirectMonotonicReader, since it's the only LongValues impl where we...

Thanks @jpountz ! I did not realize that reading ummapped bytebuffer will make JVM crach, sorry! > we should keep calling ensureValid before reading any byte from the ByteBuffer while...