prod.hpp

Go to the documentation of this file.

#ifndef STAN_MATH_OPENCL_PRIM_PROD_HPP
#define STAN_MATH_OPENCL_PRIM_PROD_HPP
#ifdef STAN_OPENCL
 
#include <stan/math/prim/meta.hpp>
#include <stan/math/prim/fun/prod.hpp>
#include <stan/math/opencl/matrix_cl.hpp>
#include <stan/math/opencl/kernel_generator.hpp>
 
namespace stan {
namespace math {
 
template <typename T,
          require_all_kernel_expressions_and_none_scalar_t<T>* = nullptr>
value_type_t<T> prod(const T& m) {
  if (is_matrix_cl<T>::value && m.size() < 1000) {
    // for small matrices running another kernel is not worth it
    return prod(from_matrix_cl(m));
  }
  matrix_cl<value_type_t<T>> res;
  if (m.rows() <= 8) {
    // without transpose we would use just a few threads in a work group
    res = prod_2d(transpose(m));
  } else {
    res = prod_2d(m);
  }
  return prod(from_matrix_cl(res));
}
 
}  // namespace math
}  // namespace stan
 
#endif
#endif