[jira] [Resolved] (GROOVY-12162) Harden and extend bytecode peephole optimization

["Paul King (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/GROOVY-12162?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Paul King resolved GROOVY-12162.
--------------------------------
    Fix Version/s: 6.0.0-beta-1
       Resolution: Fixed

> Harden and extend bytecode peephole optimization
> ------------------------------------------------
>
>                 Key: GROOVY-12162
>                 URL: https://issues.apache.org/jira/browse/GROOVY-12162
>             Project: Groovy
>          Issue Type: Improvement
>            Reporter: Daniel Sun
>            Assignee: Daniel Sun
>            Priority: Major
>             Fix For: 6.0.0-beta-1
>
>
> h2. Summary
> Follow-up to GROOVY-12065. Make the single-pass peephole layer apply to *every* method body, restore classgen diagnostics after visitor wrapping, and extend safe stack-local rewrites so generated bytecode stays compact without changing runtime semantics.
> h2. Background
> After GROOVY-12065, {{OperandStack}} emits many constants uniformly via {{visitLdcInsn}} and relies on {{PeepholeOptimizingMethodVisitor}} to narrow them ({{ICONST_*}}, {{BIPUSH}}, {{SIPUSH}}, etc.).
> Two practical gaps remained:
> # *Incomplete coverage* — the peephole wrapper was applied only at selected {{visitMethod}} sites. Synthetic helpers (call-site array init, large-list chunks, MOP bridges, …) often bypassed it and kept non-compact forms.
> # *Broken diagnostics* — with the peephole visitor outside {{TraceMethodVisitor}}, {{mv instanceof TraceMethodVisitor}} failed, so "Last known generated bytecode" was never printed when {{visitMaxs}} failed under classgen logging.
> h2. Goals
> * One installation point so *all* methods (user + synthetic) get the same compaction.
> * Correct classgen logging / failure dumps through the wrapper chain.
> * More rewrites that stay single-pass, stack-local, and semantics-preserving.
> * Strong unit coverage for edge cases (CHECKCAST side effects, void {{RETURN}}, signed zeros, box/unbox mismatch).
> h2. Approach
> h3. Uniform installation
> * Add {{PeepholeOptimizingClassVisitor}}, installed once from {{WriterController}}, wrapping every method visitor.
> * Remove ad-hoc {{new PeepholeOptimizingMethodVisitor(...)}} from {{AsmClassGenerator}} and {{CallSiteWriter}}.
> * {{PeepholeOptimizingMethodVisitor.wrap(null)}} propagates {{null}} unchanged (ASM may skip a method body); wrap is idempotent.
> h3. Diagnostics
> * Add {{PeepholeOptimizingMethodVisitor.printTraceBytecode}} to walk nested peephole layers and print an inner {{TraceMethodVisitor}} when present.
> * {{AsmClassGenerator}} uses it on {{visitMaxs}} failure so classgen logging still shows the last known bytecode.
> h3. Extended rewrites (single basic-block window)
> Still at most one pending load, one pending box, and one pending {{DUP}}/store:
> * *CHECKCAST attachment* — attach to pending {{ALOAD}} or reference constant so {{load}}; {{CHECKCAST}}; {{POP}} can collapse when unused; keep the cast when the value is used.
> * *Null compare* — {{ACONST_NULL}}; {{IF_ACMPEQ}}/{{IF_ACMPNE}} → {{IFNULL}}/{{IFNONNULL}}.
> * *Zero compare* — {{ICONST_0}}; {{IF_ICMP*}} → {{IF*}} via the fixed JVM opcode offset ({{IF_ICMPEQ}}…{{IF_ICMPLE}} ↔ {{IFEQ}}…{{IFLE}}).
> * *Bare DUP / DUP+store+pop* — matching {{POP}}/{{POP2}} collapses to a plain store, or eliminates the dup entirely.
> * *Void RETURN* — drop dead loads/constants (preserve paired {{IINC}}); for pending {{CHECKCAST}} (standalone or attached), emit the cast then {{POP}} so the type-check side effect is kept and the void return stays verifiable.
> * *Constants* — integers re-emitted via {{BytecodeHelper.pushConstant}} on the *delegate* (no re-buffering); {{ConstantDynamic}} is never buffered (bootstrap may have side effects); signed FP zeros still compared by raw bits (GROOVY-9797).
> * *Box/unbox cancellation* (gjit-inspired) — matching {{Wrapper.valueOf}} + {{xxxValue}}, or {{DefaultTypeTransformation.box}} + {{xxxUnbox}}, cancel to a no-op; mismatched types are left intact.
> * *Dead box* — {{valueOf}}/{{box}} + {{POP}}/{{POP2}} drops the box and pops the original primitive ({{POP2}} when wide).
> * *Boolean constant folding* — {{GETSTATIC Boolean.TRUE/FALSE}} + {{booleanValue}}/{{booleanUnbox}} → {{ICONST_1}}/{{ICONST_0}}.
> h2. Safety
> * Flush pending state before labels, frames, line numbers, invokes (except matched box/unbox), switches, maxs/end, and other non-local boundaries.
> * No second compilation pass; no full data-flow analysis.
> * Observable behaviour unchanged; only denser equivalent bytecode within a basic block.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)