You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* @brief Given the provided `execution_space` enum value, tries to get the `std::variant` alternative for the corresponding Kokkos::ExecutionSpace type.
94
96
* @tparam space the `execution_space` enum value
97
+
* @tparam use_usm_allocations if `true` use USM allocations
95
98
* @return the Kokkos::View instance (`[[nodiscard]]`)
/// `true` if USM allocations and, therefore, other Kokkos::MemorySpaces, are used.
181
+
bool use_usm_allocations_;
167
182
};
168
183
169
184
/**
170
185
* @brief Given a execution @p space and the number of elements @p size, creates a Kokkos::View in the respective memory space.
171
186
* @tparam T the value type of the underlying Kokkos::View
172
187
* @param[in] device the device for which this view should be allocated
173
188
* @param[in] size the size of the Kokkos::View (number of elements **not** byte!)
174
-
* @return a Kokkos::View wrapper where the active member of the internal `std::variant` corresponds to the Kokkos::View in the Kokkos::ExecutionSpace specified by @p space (`[[nodiscard]]`)
189
+
* @param[in] use_usm_allocations decide whether a USM memory space should be used or not
190
+
* @return a Kokkos::View wrapper where the active member of the internal `std::variant` corresponds to the Kokkos::View in the Kokkos::MemorySpace based on the requested Kokkos::ExecutionSpace and @p use_usm_allocations (`[[nodiscard]]`)
* @brief Perform an explicit BLAS SYMM operation: `C = alpha * A * B + beta * C` where @p A is a `m x k` symmetric matrix (memory optimized), @p B is a `k x n` matrix, @p C is a `m x n` matrix, and @p alpha and @p beta are scalars.
27
28
* @tparam ExecutionSpace the Kokkos::ExecutionSpace used to execute the kernel
29
+
* @tparam USMEnabledMemorySpace the Kokkos::MemorySpace that may use USM allocations
* @brief Perform an explicit BLAS SYMM operation: `C = alpha * A * B + beta * C` where @p A is a `m x k` symmetric matrix (memory optimized), @p B is a `k x n` matrix, @p C is a `m x n` matrix, and @p alpha and @p beta are scalars.
188
195
* @details In a multi-GPU setting, this function is responsible for mirroring down the columns this device is responsible for!
189
196
* @tparam ExecutionSpace the Kokkos::ExecutionSpace used to execute the kernel
197
+
* @tparam USMEnabledMemorySpace the Kokkos::MemorySpace that may use USM allocations
0 commit comments