cuda::experimental::places::cute_partition#
-
template<size_t rank, size_t num_place_leaves, size_t num_local_leaves>
class cute_partition# Structured tensor partition with compile-time rank and leaf counts.
Extents and strides remain runtime values, while exact array extents make the execution descriptor compact and give device code static loop bounds.
Public Functions
- inline explicit cute_partition(
- const cute_partition_descriptor &descriptor
-
inline const dim4 &true_dims() const#
-
inline const dim4 &padded_dims() const#
-
inline const dim4 &grid_dims() const#
-
inline ::cuda::std::span<const layout_leaf> place_leaves() const#
-
inline ::cuda::std::span<const int> place_axes() const#
-
inline ::cuda::std::span<const layout_leaf> local_leaves() const#
-
inline size_t num_places() const#
-
inline size_t tiles_per_place() const#
-
inline pos4 owner(pos4 data_coords) const#
-
template<size_t dims>
inline auto apply( - const box<dims> &b,
- pos4 place_position,
- dim4 grid_dims
-
inline size_t place_offset(size_t place_index) const#
-
inline cute_partition_descriptor descriptor() const#
-
inline int cmp(const cute_partition &other) const#
-
inline bool operator==(const cute_partition &other) const#
-
inline bool operator!=(const cute_partition &other) const#
Public Static Attributes
-
static constexpr size_t place_leaf_count_v = num_place_leaves#
-
static constexpr size_t local_leaf_count_v = num_local_leaves#