RFR: 8293972: runtime/NMT/NMTInitializationTest.java#default_long-off failed with "Suspiciously long bucket chains in lookup table." [v3]

[Gerard Ziemski]

> We improve the hash function, but using Mersenne prime (2^x - 1) i.e. 8191, which when combined with "modulo" operation, provides uniform bit distribution.
> 
> To help with testing this change to make sure we improve the hash function, I collected `NMTPreInitAllocation` data on 3 platforms: macOS, Linux and Windows, using 100 runs of `test/hotspot/jtreg/runtime/NMT/NMTInitializationTest.java` and computed `stdev` as well as average `max chain`:
> 
> 
> mac OS data:
>  BEFORE total: max chain: 676, empties: 210892, ranges: 676, std_devs: 112.771
>   AFTER total: max chain: 622, empties: 230762, ranges: 622, std_devs: 111.094
>  BEFORE avg: max chain: 5.061, empties: 18.695, ranges: 5.061, std_devs: 0.010
>   AFTER avg: max chain: 4.181, empties: 16.970, ranges: 4.161, std_devs: 0.009
> 	
> Linux data:
>  BEFORE total: max chain: 523, empties: 193174, ranges: 523, std_devs: 103.243
>   AFTER total: max chain: 432, empties: 175347, ranges: 430, std_devs: 90.670
>  BEFORE avg: max chain: 6.540, empties: 20.402, ranges: 6.540, std_devs: 0.011
>   AFTER avg: max chain: 6.017, empties: 22.324, ranges: 6.017, std_devs: 0.011
> 
> Windows data:
>  BEFORE total: max chain: 578, empties: 201224, ranges: 578, std_devs: 108.988
>   AFTER total: max chain: 491, empties: 218653, ranges: 491, std_devs: 104.879
>  BEFORE avg: max chain: 5.602, empties: 19.504, ranges: 5.602, std_devs: 0.011
>   AFTER avg: max chain: 4.759, empties: 21.193, ranges: 4.759, std_devs: 0.010
> 
> Note: the lower the number, the better.
> Note: the `stdedev` and `max chain` are the most important numbers here.
> 
> We also, mark output with PID to help identify which test process it is coming from as the attached log was very confusing to read.
> 
> Tested manually with `/Volumes/Work/tests/jtreg/jtreg/bin/jtreg -nr -va -jdk:./build/macosx-x86_64-server-fastdebug/images/jdk test/hotspot/jtreg/runtime/NMT/NMTInitializationTest.java` and `Mach5 hs-tier1,2,3,4,5
> `

Gerard Ziemski has updated the pull request incrementally with one additional commit since the last revision:

  do not clip bits to 32, we need full 64 bits

-------------

Changes:
  - all: https://git.openjdk.org/jdk/pull/14607/files
  - new: https://git.openjdk.org/jdk/pull/14607/files/c346ffa2..683fb0e6

Webrevs:
 - full: https://webrevs.openjdk.org/?repo=jdk&pr=14607&range=02
 - incr: https://webrevs.openjdk.org/?repo=jdk&pr=14607&range=01-02

  Stats: 3 lines in 1 file changed: 0 ins; 0 del; 3 mod
  Patch: https://git.openjdk.org/jdk/pull/14607.diff
  Fetch: git fetch https://git.openjdk.org/jdk.git pull/14607/head:pull/14607

PR: https://git.openjdk.org/jdk/pull/14607