cuda::is_valid_kernel_argument< T > Struct Template Reference
A trait characterizing those types which can be used as kernel parameters. More...
#include <kernel_launch.hpp>
Inheritance diagram for cuda::is_valid_kernel_argument< T >:
Collaboration diagram for cuda::is_valid_kernel_argument< T >:
Detailed Description
template<typename T>
struct cuda::is_valid_kernel_argument< T >
A trait characterizing those types which can be used as kernel parameters.
When passing arguments to a kernel function, their representation is copied from the host to the GPU device; so their type must allow for a construction by means of such copying; this is essentially the std::is_trivially_copy_constructible trait. which serves to identify types that can be safely passed to CUDA kernels.
- Template Parameters
-
T The prospective kernel parameter type
- Note
- This trait is used internally in the CUDA API wrappers to enforce the constraints on kernel arguments at compile time.
- The trait is based on the CUDA C++ Programming Guide's requirements from kernel parameter types, mostly (see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#global-function-argument-processing ); There is also a requirement for supporting destructor calls before the kernel execution is actually concluded - which is all but impossible to check via a trait; so it is up to users of the library to either exercise care, or otherwise specialize the trait for known-to-be-problematic types, precluding their use.
The documentation for this struct was generated from the following file:
- src/cuda/api/kernel_launch.hpp
