Use memory pool for macro-op fusion arrays

[jserv]

This replaces malloc/free with mpool in macro-op fusion, eliminating per-allocation overhead and reduces memory fragmentation.

The fixed-size pool approach trades some memory (unused slot space for short sequences) for allocation speed and cache locality. Sequences exceeding 16 instructions gracefully degrade to non-fused execution.

---

Summary by cubic

Switch macro-op fusion arrays to a fixed-size memory pool to cut allocation overhead and fragmentation. Fused sequences are capped at 16 ops; longer sequences fall back to non-fused execution.

• Refactors
  • Replace malloc/free with mpool_alloc/mpool_free for ir->fuse arrays.
  • Add fuse_mp pool and FUSE_MAX_ENTRIES/FUSE_SLOT_SIZE constants.
  • Update cleanup paths: block_map_clear, block_map_destroy, rv_create failure handling, and rv_delete.

^{Written for commit 7dad667. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 3 files

[jserv]

Benchmarks

Details

Benchmark suite	Current: 7dad667	Previous: f4aff45	Ratio
Dhrystone	1657 DMIPS	1659 DMIPS	1.00
CoreMark	1017.224 iterations/sec	1022.858 iterations/sec	1.01

This comment was automatically generated by workflow using github-action-benchmark.