Iterator interface

* @brief Returns stream associated with executor */

* @brief Get underlying shape object * * @return Shape object */

* Get the rank of the tensor * * @returns Rank of the tensor * */

* Constructor to assign an operator to a view * * @param out * Output destination view * * @param op * Input operator */

* Get the total number of elements in the tensor * * * @returns Total number of elements across all dimensions * */

* @brief Allocate memory * * Can be used for managed, pinned, malloced, device, and async device allocations * * @param ptr Pointer to store all

Core interface

* @brief Get total size of descriptor * * @return Product of all sizes */

Scalar sqrt overload so unary operator dispatch can handle fltflt expressions

Provide host/device wrappers for CUDA round-to-nearest intrinsics. On host, we fall back to standard operations. These helpers allow fltflt arithmetic

fltflt_fast_two_sum is the Fast-Two-Sum algorithm given by Thall, which he attributes to Dekker. This corresponds to function quickTwoSum() from Thall

fltflt_add is the df64_add() function given by Thall. This function uses two_sum() for the hi and lo components followed by addition of the cross term

fltflt_two_prod_fma is the Two-Product-FMA algorithm given by Thall, which he attributes to Hida. This corresponds to function FMA-twoProd() from Thal

* Compute the ceiling of every element in the tensor * @param t * Tensor or operator input */

Swaps two tensor implementations * * Swaps members of two tensor implementations * * @param lhs * Left argument * @par

Scalar abs overload so unary operator dispatch can handle fltflt expressions

fltflt_sub() subtracts b from a. It delegates to fltflt_add() with a negated b.

fltflt_two_sum is the Two-Sum algorithm given by Thall, which he attributes to Knuth. This corresponds to function twoSum() from Thall's paper, which

* @brief Rank of IF/ELSE operator * * @return Rank */