Execution Methods

Execution methods are used to run the Solver function as defined by user with cuSolverDx operators.

Note

Right now, cuSolverDx supports only execution on CUDA thread block level (block execution).

Block Execute Methods

The block execution methods are available if the descriptor has been constructed using the Block Operator and is_complete_solver_execution Trait is true. All threads within a block must collectively participate in a block execution method; otherwise, the behavior is undefined.

execute(...) runs the calculations defined by the Solver descriptor, and its arguments vary depending on the descriptor’s function.

using Solver = decltype(cusolverdx::Size<...>() + ...);

// function::potrf and getrf_no_pivot
//
// with compile-time lda
void Solver::execute(typename Solver::a_data_type* A, typename Solver::status_type* status);
// with the runtime lda
void Solver::execute(typename Solver::a_data_type* A, const unsigned int runtime_lda,
                     typename Solver::status_type* status);

// function::potrs and getrs_no_pivot
//
// with compile-time lda and ldb
void Solver::execute(typename Solver::a_data_type* A, typename Solver::b_data_type* B);
// with runtime lda and ldb
void Solver::execute(typename Solver::a_data_type* A, const unsigned int runtime_lda,
                     typename Solver::b_data_type* B, const unsigned int runtime_ldb);

// function::posv and gesv_no_pivot
//
// with compile-time lda and ldb
void Solver::execute(typename Solver::a_data_type* A, typename Solver::b_data_type* B,
                     typename Solver::status_type* status);
// with runtime lda and ldb
void Solver::execute(typename Solver::a_data_type* A, const unsigned int runtime_lda,
                     typename Solver::b_data_type* B, const unsigned int runtime_ldb,
                     typename Solver::status_type* status);