Function: reserve_mem_address

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product/Atlas A3 inference product	√
Atlas A2 training product/Atlas A2 inference product	√
Atlas training product	√
Atlas inference product	√
Atlas 200I/500 A2 inference product	√

Reserves virtual memory.

This API needs to work with other APIs to allocate virtual memory with consecutive addresses and maximize the use of physical memory.

C Prototype

aclError aclrtReserveMemAddress(void **virPtr, size_t size, size_t alignment, void *expectPtr, uint64_t flags)

Python Function

vir_ptr, ret = acl.rt.reserve_mem_address(size, alignment, expect_ptr, flags)

Parameter	Description
size	Int, size of the virtual address space, in bytes. Must not be 0.
alignment	Int, alignment value of the virtual address. This parameter is reserved and can only be set to 0.
expectPtr	Int, start address of the virtual address space to be returned. This parameter is reserved and can only be set to 0.
flags	Int, huge page/common page flag. This parameter is reserved. The recommended value is 0.

Return Value	Description
virPtr	Int, pointing to the address of the allocated virtual address space.
ret	Int, error code. 0 on success; else, failure.

Ascend RC does not support this API.
Currently, the following models support the Ascend RC form:
- Atlas 200I/500 A2 inference product
To use the virtual memory reserved by this API, only memcpy_async can be called to copy data between two devices in the single-process scenario.

Parent topic: Memory Management