[code-reflection] RFR: Bytecode locals compaction [v3]

[Paul Sandoz]

On Tue, 10 Sep 2024 15:16:35 GMT, Adam Sotona <asotona at openjdk.org> wrote:

>> `LocalsCompactor` is a bytecode transformation reducing generated method `maxLocals` by re-using local slots.
>> Roundtrip of lift and bytecode generation doubles the `maxLocals` value, while `LocalsCompactor` reduces the regression by 50%.
>> 
>> Lifting of compacted locals revealed bugs in `LocalsTypeMapper` variables graph construction. This patch includes significantly  refactored `LocalsTypeMapper` to support lifting of compacted locals.
>> 
>> `BytecodeGenerator` has slightly improved performance (so `TestSmallCorpus` finishes in a reasonable time). However performance has not been main focus of this PR.
>> 
>> `TestSmallCorpus` has been slightly refactored to provide better debugging feedback and compaction statistics. Overall stability is still >99%.
>> 
>> Please review.
>> 
>> Thanks,
>> Adam
>
> Adam Sotona has updated the pull request incrementally with one additional commit since the last revision:
> 
>   added comments

This is getting to a level of complexity that i cannot provide a useful review on the algorithmic details as I don't fully comprehend them. Once you reach a stable point it will be very useful to describe the approach.

src/java.base/share/classes/java/lang/reflect/code/bytecode/BytecodeGenerator.java line 355:

> 353:     // Some of the operations can be deferred
> 354:     private boolean canDefer(Op op) {
> 355:         Boolean can = deferCache.get(op);

Use `computeIfAbsent`?

src/java.base/share/classes/java/lang/reflect/code/bytecode/BytecodeLift.java line 275:

> 273:     }
> 274: 
> 275:     public static CoreOp.FuncOp lift(MethodModel methodModel, MethodHandles.Lookup lookup) {

Place the lookup as the first parameter, like for the bytecode generator and the interpreter

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

Marked as reviewed by psandoz (Lead).

PR Review: https://git.openjdk.org/babylon/pull/224#pullrequestreview-2294105665
PR Review Comment: https://git.openjdk.org/babylon/pull/224#discussion_r1752899280
PR Review Comment: https://git.openjdk.org/babylon/pull/224#discussion_r1752905979