cub::TransformPrefetchPolicy

struct TransformPrefetchPolicy

The prefetch sub-policy for TransformPolicy.

Public Functions

inline friend ::std::ostream &operator<<(

::std::ostream &os,

const TransformPrefetchPolicy &policy

)

Public Members

int threads_per_block

Number of threads in a CUDA block.

int items_per_thread_no_input = 2

When there are no iterators as inputs, the kernel is just filling.

This is the number of items written per thread in this case.

int min_items_per_thread = 1

Minimum number of items per thread (inclusive)

int max_items_per_thread = 32

Maximum number of items per thread (inclusive)

int prefetch_byte_stride = 128

The stride in bytes to issue prefetch requests to memory.

Corresponds somewhat to the size of a cache line.

int unroll_factor = 0

For any value >1, the unroll factor for the transformation loop in the kernel.

The value 0 retains the compiler’s default unrolling by specifying no unroll pragma. 1 prevents unrolling.

Friends

inline friend constexpr bool operator==(

const TransformPrefetchPolicy &lhs,

const TransformPrefetchPolicy &rhs

) noexcept

inline friend constexpr bool operator!=(

const TransformPrefetchPolicy &lhs,

const TransformPrefetchPolicy &rhs

) noexcept