register_cache

Applicability

Product	Supported
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	x
Atlas inference products	x
Atlas training products	x

Note: For the Atlas A2 training products/Atlas A2 inference products, only the Atlas 800I A2 inference server, Atlas 300I A2 inference card, and A200I A2 Box heterogeneous components are supported.

Function Description

In the non-PagedAttention scenario, this API is called to register a self-allocated memory.

Prototype

register_cache(cache_desc: CacheDesc, addrs: List[int], cache_keys: Union[Tuple[CacheKey], List[CacheKey]] = (), remote_accessible: Optional[bool] = None) -> Cache:

Parameters

Parameter	Data Type	Description
cache_desc	CacheDesc	Cache description.
addrs	List[int]	Cache address. The total number of addresses in register_cache and register_blocks_cache cannot exceed 240.
cache_keys	Union[Tuple[CacheKey], List[CacheKey]]	Cache index.
remote_accessible	Optional[bool]	Whether the registered memory can be used for network transmission. The default value is True for device memory and False for host memory.

Example

Click Gitcode, select the matching version, and obtain the sample from the examples/python directory.

Returns

In normal cases, the registered cache is returned.

If the input data type is incorrect (src and dst do not match), a TypeError or ValueError is thrown.

If the input parameter is None, an AttributeError is thrown.

Restrictions

The device memory must be registered before link establishment, while the host memory has no restrictions on the order of operations.

The registered memory address must be unique.

A maximum of 50 GB device memory and 20 GB host memory can be registered. Larger registered memory results in higher OS memory usage.

For D2D data transmission over HCCS, the start address must be aligned to 2 MB; otherwise, the link API may fail. HCCS does not support host memory.

For RDMA data transmission, host memory must be allocated using the aclrtMallocHost API; otherwise, the link API may fail.

Parent topic: CacheManager