Wire AMO, HESS, VMPC into core allocation path - #3

Open

shift wants to merge 18 commits into

mainfrom

feature/wire-advanced-features

Commits on Apr 1, 2026

feat: wire AMO, HESS, VMPC into core allocation path
shift
committed
7fe34f5
ci: run benchmarks on feature branches, add realloc/fragmentation benchmarks
shift
committed
7fdd63e
ci: full benchmark matrix with 5 runs across all feature configs
shift
committed
d92f5b6
fix: support_core_test uses real allocations for FreeBlock commands
shift
committed
0cca429
fix: clippy derivable_impls and missing_safety_doc
shift
committed
c348ebd
ci: fix benchmark matrix workflow syntax and simplify
shift
committed
c3f5ef7
ci: fix summarize job to handle benchmark failures gracefully
shift
committed
e4ffa77
ci: handle empty benchmark results in summarize job
shift
committed
10b1ddd
perf: use mremap with MAYMOVE for large realloc
shift
committed
fe77776
perf: aggressive sleep in support core, gate VMPC behind feature flag
shift
committed
15fd7bd
perf: optimize realloc hot path - reorder get_alloc_size checks, inline small memcpy
shift
committed
1189243
perf: make AMO ring buffer opt-in via feature flag
shift
committed
726f2fe
perf: double magazine capacity to 128 to reduce global pool contention
shift
committed
d040013

Commits on Apr 2, 2026

perf: add 64-entry LUT for size class classification
shift
committed
8d07f90
perf: allocate large allocations with 2x padding for in-place realloc
shift
committed
8e2a74f
perf: restore 2x padding for large allocations
shift
committed
b78de07
perf: gate get_alloc_size call behind AMO feature flag
shift
committed
ff0e8a4
perf: change size_to_class to inline(always) for hot path
shift
committed
b6f985c