A 64-byte-aligned bump-pointer allocator over a single contiguous block of host memory. Every chunk handed out via AllocateFloats() is itself 64-byte aligned, satisfying this codebase's strict alignment requirement for AVX2/AVX-512 loads.
More...
#include <MemoryArena.h>
|
| | MemoryArena (size_t total_floats, bool use_huge_pages=false) |
| | Allocates a single 64-byte-aligned buffer large enough to hold total_floats floats, rounded up to the nearest 64-byte boundary.
|
| | ~MemoryArena () |
| | Frees the underlying allocation, via whichever deallocator matches the allocator actually used (see used_huge_pages).
|
| | MemoryArena (const MemoryArena &)=delete |
| | Deleted: MemoryArena owns a single aligned allocation, so copying would risk a double-free.
|
| MemoryArena & | operator= (const MemoryArena &)=delete |
| | Deleted: MemoryArena owns a single aligned allocation, so copy-assignment would risk a double-free.
|
| float * | AllocateFloats (size_t num_floats) |
| | Allocates a 64-byte aligned chunk of floats from the arena.
|
| size_t | GetUsedBytes () const |
| | Returns how many bytes have been allocated from the arena so far.
|
| size_t | GetCapacityBytes () const |
| | Returns the arena's total capacity.
|
| bool | UsedHugePages () const |
| | Returns whether this arena's memory is actually huge-page-backed. Since huge pages are requested on a best-effort basis, this can be false even if use_huge_pages was passed to the constructor, if the request wasn't honored (system lacks configuration/privilege, or the allocation call failed) and the arena silently fell back to the standard aligned allocator instead.
|
A 64-byte-aligned bump-pointer allocator over a single contiguous block of host memory. Every chunk handed out via AllocateFloats() is itself 64-byte aligned, satisfying this codebase's strict alignment requirement for AVX2/AVX-512 loads.
◆ MemoryArena() [1/2]
| Deep::MemoryArena::MemoryArena |
( |
size_t | total_floats, |
|
|
bool | use_huge_pages = false ) |
|
inlineexplicit |
Allocates a single 64-byte-aligned buffer large enough to hold total_floats floats, rounded up to the nearest 64-byte boundary.
- Parameters
-
| total_floats | Total number of floats this arena can hand out across all future AllocateFloats() calls combined. |
| use_huge_pages | Opt-in, best-effort request for huge-page-backed memory (see HugePageAllocPortable() for the platform-specific caveats). Defaults to false, preserving existing behavior exactly for anyone not explicitly opting in. Falls back safely to the standard aligned allocator if the request isn't honored – this never causes a hard failure on its own. |
- Exceptions
-
| std::bad_alloc | if the underlying allocation fails (including the fallback path, if huge pages were requested but unavailable). |
◆ ~MemoryArena()
| Deep::MemoryArena::~MemoryArena |
( |
| ) |
|
|
inline |
Frees the underlying allocation, via whichever deallocator matches the allocator actually used (see used_huge_pages).
◆ MemoryArena() [2/2]
| Deep::MemoryArena::MemoryArena |
( |
const MemoryArena & | | ) |
|
|
delete |
Deleted: MemoryArena owns a single aligned allocation, so copying would risk a double-free.
◆ AllocateFloats()
| float * Deep::MemoryArena::AllocateFloats |
( |
size_t | num_floats | ) |
|
|
inline |
Allocates a 64-byte aligned chunk of floats from the arena.
- Parameters
-
| num_floats | Number of floats to allocate from the arena. The actual reservation is rounded up to the nearest 64-byte boundary, so the returned chunk may have more usable capacity than requested. |
- Returns
- Pointer to the start of the allocated chunk, valid for the lifetime of this MemoryArena. 64-byte aligned.
- Exceptions
-
| std::runtime_error | if the requested allocation would exceed the arena's total capacity. |
- Warning
- Individual chunks are never freed independently; the entire arena is released at once in the destructor.
◆ GetCapacityBytes()
| size_t Deep::MemoryArena::GetCapacityBytes |
( |
| ) |
const |
|
inline |
Returns the arena's total capacity.
- Returns
- The capacity, in bytes, rounded up to a multiple of 64.
◆ GetUsedBytes()
| size_t Deep::MemoryArena::GetUsedBytes |
( |
| ) |
const |
|
inline |
Returns how many bytes have been allocated from the arena so far.
- Returns
- The current offset, in bytes.
◆ operator=()
Deleted: MemoryArena owns a single aligned allocation, so copy-assignment would risk a double-free.
◆ UsedHugePages()
| bool Deep::MemoryArena::UsedHugePages |
( |
| ) |
const |
|
inline |
Returns whether this arena's memory is actually huge-page-backed. Since huge pages are requested on a best-effort basis, this can be false even if use_huge_pages was passed to the constructor, if the request wasn't honored (system lacks configuration/privilege, or the allocation call failed) and the arena silently fell back to the standard aligned allocator instead.
The documentation for this class was generated from the following file:
1.16.1