Spread bits in computing bucket index

[liufengyun]

Spread bits in computing bucket index

Summary

This is a common optimization in power-of-two hash tables (%16, %32, etc.), to distribute hash codes that have similar lower bits but different high bits more evenly.

This is a performance improvement. The following test is an extreme case, but it does show the optimization pays off:

Backend	Before	After	Speedup
Ruby	11.33 s	0.349 s	32.5×
Python	14.09 s	0.347 s	40.6×
JavaScript (Node.js)	0.280 s	0.131 s	2.1×
JavaScript (Node.js, --jitless)	14.68 s	0.493 s	29.8×

4,000 distinct Long values (i << 32). Median of five runs after one warm-up.

Checklist

• Added / updated tests under tests/pos/ or tests/warn/
• Docs updated if the change affects user-visible behavior
• A JIP is required for language change
• All commits are signed off (why?)

How to sign off commits

Use git commit -s to add the Signed-off-by line automatically:

To add a sign-off to the last commit retroactively:

git commit --amend -s --no-edit

To add sign-off to the last 3 commits:

git rebase --signoff HEAD~3

Security impact

No

Compatibility impact

No

• Source compatibility
  • syntax parsing
  • type checking
• SAST compatibility
  • forward compatibility: new libraries can be used by old compiler
  • backward comopatibility: old libraries can be used by new compiler
• Standard Library compatibility:
  • forward compatibility: new code can work with old stdlib
  • backward compatibility: old code work with the new stdlib
• Runtime Library compatibility:
  • forward compatibility: new code can work with old runtime
  • backward compatibility: old code work with the new runtime
• Build tool
  • Build spec compatibility: old projects continue to build
  • Joy package compatibility: old .joy can be consumed by new build tool

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.

[everdance]

we should ensure cap is prime number

[everdance]

this is more important than the hash code value

[liufengyun]

Using prime number is a design choice, but not the only one. Here we are following the common design that use power-of-two capacity and spread higher-bits.

It's used e.g. in JDK: https://github.com/openjdk/jdk/blob/8610e806260f4eec6d98a5b418e4a1bc11a9a3cb/src/java.base/share/classes/java/util/HashMap.java#L320-L339