Enum CudaHostAllocFlags

Namespace

DotCompute.Backends.CUDA.Types.Native

Assembly

DotCompute.Backends.CUDA.dll

Flags that control the allocation behavior of CUDA host memory.

[Flags]
public enum CudaHostAllocFlags : uint

Fields

Default = 0

Default host allocation (page-locked / pinned, no special properties). Corresponds to cudaHostAllocDefault (0x00).

Mapped = 2

Map the allocation into the device address space. The returned host pointer can be used directly by CUDA kernels. Corresponds to cudaHostAllocMapped (0x02).

Pinned = 0

Alias for Default; all cudaHostAlloc allocations are page-locked (pinned).

Portable = 1

Make the allocation portable across CUDA contexts. This allows the memory to be used by multiple CUDA contexts. Corresponds to cudaHostAllocPortable (0x01).

WriteCombined = 4

Allocate write-combined memory. This can improve transfer performance in some cases but may reduce CPU read performance. Corresponds to cudaHostAllocWriteCombined (0x04).

Remarks

These flags are used with cudaHostAlloc and related functions to specify special properties for host memory allocations. Multiple flags can be combined using bitwise OR operations.