Open dcoutts opened 1 month ago
This is here for benchmarking comparison purposes, and so see what we think of the generality vs performance trade-off.
This allocator approach is a direct style, not using arenas. It's more like a slab allocator, but only for 4k page allocations.
This allows the slab cache to be manipulated only using atomic IORef operations.
@dcoutts is this a PR you would still like to get merged at some point now that the existing page allocator was optimised using SPECIALISE pragmas in #330?
SPECIALISE
This is here for benchmarking comparison purposes, and so see what we think of the generality vs performance trade-off.
This allocator approach is a direct style, not using arenas. It's more like a slab allocator, but only for 4k page allocations.
This allows the slab cache to be manipulated only using atomic IORef operations.