|
Uni20
Asynchronous tensor and runtime library.
|
Scoped read-only CUDA pointer access after a host-side synchronization. More...
#include <buffer.hpp>
Public Types | |
| using | element_type = T |
Public Member Functions | |
| BlockingReadAccess (BlockingReadAccess &&other) noexcept | |
| BlockingReadAccess (BlockingReadAccess const &)=delete | |
| T const * | data () const noexcept |
| Return the typed device pointer for a blocking runtime operation. | |
| BlockingReadAccess & | operator= (BlockingReadAccess &&other) noexcept |
| BlockingReadAccess & | operator= (BlockingReadAccess const &)=delete |
| void | release () noexcept |
| End the scoped host-synchronized read access. | |
| std::size_t | size () const noexcept |
| Return the number of typed elements in the allocation. | |
| CudaBuffer< T > const & | storage () const |
| Return the CUDA buffer retained by this access object. | |
| ~BlockingReadAccess () | |
Friends | |
| class | CudaBuffer< T > |
Scoped read-only CUDA pointer access after a host-side synchronization.
Construction waits for the buffer's prior writer completion. No CUDA completion is published on release because the caller must finish every operation using the pointer before releasing it.
| using uni20::cuda::BlockingReadAccess< T >::element_type = T |
|
delete |
|
inlinenoexcept |
|
inline |
|
inlinenoexcept |
Return the typed device pointer for a blocking runtime operation.
|
inlinenoexcept |
|
delete |
|
inlinenoexcept |
End the scoped host-synchronized read access.
|
inlinenoexcept |
Return the number of typed elements in the allocation.
|
inline |
Return the CUDA buffer retained by this access object.
|
friend |