CopyDeviceMemory function

Άρθρο
01/26/2024

The CopyDeviceMemory function copies memory from one location to another without interference from compiler optimizations in situations where the developer needs to additionally be sure that alignment faults will not be generated when accessing device memory.

Important

Some information relates to a prerelease product which may be substantially modified before it's commercially released. Microsoft makes no warranties, express or implied, with respect to the information provided here.

Parameters

Param Destination [out]

A pointer to the starting address of the copied block's destination.

Param Source [in]

A pointer to the starting address of the block of memory to copy.

Param Length [in]

The size of the block of memory to copy, in bytes.

Syntax

volatile void*
  __cdecl
  CopyDeviceMemory (
    _Out_writes_bytes_all_(Length) volatile void* Destination,
    _In_reads_bytes_(Length) volatile const void* Source,
    SIZE_T Length
  );

Remarks

This API exists to provide CopyVolatileMemory behavior (i.e., copying memory from one location to another without interference from compiler optimizations) in situations where the developer needs to additionally be sure that alignment faults will not be generated when accessing device memory. The API has the following properties:

The API is not recognized as a compiler intrinsic so the compiler will never optimize away the call (either entirely, or replace the call with an "equivalent" sequence of instructions). This differs from CopyMemory which is subject to a variety of compiler optimizations.
When the call returns, the data has been copied from Source to Destination. This functions memory accesses to the Source and Destination will only be performed within the function (i.e. the compiler cannot move memory accesses out of this function).
The API may perform unaligned memory accesses only if the CPU supports unaligned memory accesses on device memory. If the CPU does not support unaligned device memory accesses, only aligned accesses will be performed.
The API may access memory locations more than once as part of its copy operation.
Does not support copy operations when Source and Destination overlap each other. If overlapping buffers are provided, fast-fails with the error code FAST_FAIL_INVALID_ARG.

Note

This function only guarantees that the CPU’s requirements for accessing memory mapped as device memory are respected. If a specific device has its own specific requirements for being accessed, this function should not be used (and instead, the developer must implement their own accessor functions). For example, this function makes no guarantee about the size of memory accesses generated (unless the CPU itself enforces these requirements).

Note

This function works on all versions of Windows, not just the latest. You need to consume the latest SDK to get the function declaration from the winbase.h header. You also need the library (volatileaccessu.lib) from the latest SDK. However, the resulting binary will run fine on older versions of Windows.

Example

UCHAR* CopyBuffer;

// In this scenario we are copying data from memory mapped
// as "device memory" (i.e. memory not backed by RAM). On
// some platforms like ARM64, device memory cannot tolerate
// memory accesses that are not naturally aligned (i.e. a 4-byte
// load must be 4-byte aligned). Functions like mempcy, CopyMemory,
// and even CopyVolatileMemory may perform unaligned memory accesses
// because it is typically faster to do this.
// To ensure only naturally aligned accesses happen, use CopyDeviceMemory.

CopyDeviceMemory(CopyBuffer, DeviceMemoryBuffer, 100);

Requirements

Minimum supported client: Windows 11 Insider Preview Build TBD

Header: winbase.h (include Winbase.h)

Kernel-mode library: volatileaccessk.lib

User-mode library: volatileaccessu.lib

Κοινή χρήση μέσω