More comments on the allocators.

This commit is contained in:
Christoffer Lerno
2025-05-07 12:52:19 +02:00
parent 54e70cae0f
commit 164c901ae6
6 changed files with 68 additions and 11 deletions

View File

@@ -1,12 +1,15 @@
module std::core::mem::allocator;
import std::io, std::math;
struct AllocChunk @local
{
usz size;
char[*] data;
}
<*
The backed arena allocator provides an allocator that will allocate from a pre-allocated chunk of memory
provided by it's backing allocator. The allocator supports mark / reset operations, so it can be used
as a stack (push-pop) allocator. If the initial memory is used up, it will fall back to regular allocations,
that will be safely freed on `reset`.
While this allocator is similar to the dynamic arena, it supports multiple "save points", which the dynamic arena
doesn't.
*>
struct BackedArenaAllocator (Allocator)
{
Allocator backing_allocator;
@@ -16,6 +19,12 @@ struct BackedArenaAllocator (Allocator)
char[*] data;
}
struct AllocChunk @local
{
usz size;
char[*] data;
}
const usz PAGE_IS_ALIGNED @local = (usz)isz.max + 1u;
struct ExtraPage @local

View File

@@ -4,6 +4,17 @@
module std::core::mem::allocator;
import std::math;
<*
The dynamic arena allocator is an arena allocator that can grow by adding additional arena "pages".
It only supports reset, at which point all pages except the first one is released to the backing
allocator.
If you want multiple save points, use the BackedArenaAllocator instead.
The advantage over the BackedArenaAllocator, is that when allocating beyond the first "page", it will
retain the characteristics of an arena allocator (allocating a large piece of memory then handing off
memory from that memory), wheras the BackedArenaAllocator will have heap allocator characteristics.
*>
struct DynamicArenaAllocator (Allocator)
{
Allocator backing_allocator;

View File

@@ -5,6 +5,13 @@
module std::core::mem::allocator;
import std::math;
<*
The SimpleHeapAllocator implements a simple heap allocator on top of an allocator function.
It uses the given allocator function to allocate memory from some source, but never frees it.
This allocator is intended to be used in environments where there isn't any native libc malloc,
and it has to be emulated from a memory region, or wrapping linear memory as is the case for plain WASM.
*>
struct SimpleHeapAllocator (Allocator)
{
MemoryAllocFn alloc_fn;

View File

@@ -1,13 +1,15 @@
// Copyright (c) 2021-2024 Christoffer Lerno. All rights reserved.
// Copyright (c) 2021-2025 Christoffer Lerno. All rights reserved.
// Use of this source code is governed by the MIT license
// a copy of which can be found in the LICENSE_STDLIB file.
module std::core::mem::allocator @if(env::LIBC);
import std::io;
import libc;
const LibcAllocator LIBC_ALLOCATOR = {};
<*
The LibcAllocator is a wrapper around malloc to conform to the Allocator interface.
*>
typedef LibcAllocator (Allocator, Printable) = uptr;
const LibcAllocator LIBC_ALLOCATOR = {};
fn String LibcAllocator.to_string(&self, Allocator allocator) @dynamic => "Libc allocator".copy(allocator);
fn usz? LibcAllocator.to_format(&self, Formatter *format) @dynamic => format.print("Libc allocator");

View File

@@ -1,5 +1,14 @@
module std::core::mem::allocator;
<*
The OnStackAllocator is similar to the ArenaAllocator: it allocates from a chunk of memory
given to it.
The difference is that when it runs out of memory it will go directly to its backing allocator
rather than failing.
It is utilized by the @stack_mem macro as an alternative to the temp allocator.
*>
struct OnStackAllocator (Allocator)
{
Allocator backing_allocator;
@@ -8,7 +17,6 @@ struct OnStackAllocator (Allocator)
OnStackAllocatorExtraChunk* chunk;
}
struct OnStackAllocatorExtraChunk @local
{
bool is_aligned;