tiny.gpalloc

Overview · API · Code relationships · Verification · Audit

Overview

GpAllocator provides a general-purpose heap over a caller-supplied std.mem.Allocator stored at init and asked for pages and large blocks, the backing allocator. A request up to 32 KiB takes the first class at least as large as itself from a fixed list of 40 block sizes, its size class, with blocks carved from a 64 KiB span, a page. Each thread keeps its own array of free blocks per size class in thread-local storage, the thread cache. With the thread cache on, an allocation pops a block from the running thread's bin for that class and a free pushes one back, so the common path takes no lock.

Per-size-class state guarded by that class's mutex, the bin, holds that class's pages. The bounded list inside the bin that collects blocks drained from thread caches and refills them in batches is the transfer list. An empty thread bin refills from it, drawing a batch of blocks from the transfer list under the class mutex. A full thread bin drains a batch back to that transfer list under the same mutex, and a block freed by a thread other than the one that allocated it goes back the same way. Batching keeps the class mutex out of the common path, because one lock covers a whole batch of blocks.

A request above the small range goes to the backing allocator through a cache of retained blocks per class, the large cache, bounded by a byte limit.

Small pages come from the page provider, which is the backing allocator by default, and .page_provider = .os maps them from the operating system.

Counters are off by default, and turning them on with .collect_stats = true also turns off the thread cache and the large cache, so a run that collects counters measures a different path from the one that runs without them.

A thread that exits hands its cache back to the heap, and a thread the runtime cannot hook does the same by calling flushThreadCacheForCurrentThread.

Definitions

Types and contracts

Public types and contracts.

• Config
• PageProvider
• RetainedEmptyPagePolicy
• Stats
• GpAllocator

Namespaces

Public namespaces.

• class
• stats
• cache
• large
• page
• config

Values and defaults

Public values and defaults.

• class_sizes
• max_small_size
• min_alignment
• page_size

Code relationships

Direct static dependencies extracted from parsed source by semantic graph analysis.

Uses: tiny.bench, tiny.bumpalo, tiny.deadalloc, tiny.hypothesis, tiny.sql, tiny.sys, tiny.tldr
Used by: tiny.alloc

Verification

No verification records are cataloged for this module in this build.

Audit