cuda::experimental::stf::deallocateHostMemory#

Overloads#

deallocateHostMemory(p, sz, loc=::cuda::std::source_location::current())#

inline void cuda::experimental::stf::deallocateHostMemory(
void *p,
size_t sz,
const ::cuda::std::source_location loc = ::cuda::std::source_location::current(),
)

Deallocates memory allocated with allocateHostMemory immediately.

Parameters:

  • p – pointer to memory chunk

  • sz – size in bytes

  • loc – location of the call, defaulted

deallocateHostMemory(p, sz, stream)#

inline void cuda::experimental::stf::deallocateHostMemory(
void *p,
size_t sz,
cudaStream_t stream,
)

Deallocates memory allocated with allocateHostMemory in a stream-ordered fashion.

Will perform deallocation when all preceding operations on stream have completed.

Parameters:

  • p – pointer

  • sz – size in bytes

  • stream – the stream used for ordering

deallocateHostMemory(p, sz, graph, pDependencies, numDependencies)#

inline cudaGraphNode_t cuda::experimental::stf::deallocateHostMemory(
void *p,
size_t sz,
cudaGraph_t graph,
const cudaGraphNode_t *pDependencies,
size_t numDependencies,
)

Deallocates memory allocated with allocateHostMemory in a graph-ordered fashion.

Will perform deallocation when all dependent graph operations have completed.

Parameters:

  • p – pointer

  • sz – size in byutes

  • graph – graph used for ordering

  • pDependencies – array of dependencies

  • numDependencies – number of elements in pDependencies

Returns:

cudaGraphNode_t the newly inserted node