// Sets the kernel's arguments
static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
std::vector<T> &, std::vector<T> &,
std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &,
std::vector<T> &) {
auto alpha_buffer = std::vector<T>{args.alpha};
tuner.AddArgumentScalar(static_cast<int>(args.m));
tuner.AddArgumentInput(a_mat);
tuner.AddArgumentOutput(b_mat);
tuner.AddArgumentInput(alpha_buffer);
}
// Sets the kernel's arguments
static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
std::vector<T> &, std::vector<T> &,
std::vector<T> &a_mat, std::vector<T> &b_mat, std::vector<T> &c_mat,
std::vector<T> &) {
tuner.AddArgumentScalar(static_cast<int>(args.m));
tuner.AddArgumentScalar(static_cast<int>(args.n));
tuner.AddArgumentScalar(static_cast<int>(args.k));
tuner.AddArgumentScalar(GetRealArg(args.alpha));
tuner.AddArgumentScalar(GetRealArg(args.beta));
tuner.AddArgumentInput(a_mat);
tuner.AddArgumentScalar(0); // a_offset
tuner.AddArgumentScalar(static_cast<int>(args.k)); // a_ld
tuner.AddArgumentInput(b_mat);
tuner.AddArgumentScalar(0); // b_offset
tuner.AddArgumentScalar(static_cast<int>(args.n)); // b_ld
tuner.AddArgumentOutput(c_mat);
tuner.AddArgumentScalar(0); // c_offset
tuner.AddArgumentScalar(static_cast<int>(args.n)); // c_ld
tuner.AddArgumentScalar(1); // c_do_transpose
tuner.AddArgumentScalar(0); // a_conjugate
tuner.AddArgumentScalar(0); // b_conjugate
}
// Sets the kernel's arguments
static void SetArguments(cltune::Tuner &tuner, const Arguments<T> &args,
std::vector<T> &x_vec, std::vector<T> &y_vec,
std::vector<T> &a_mat, std::vector<T> &, std::vector<T> &,
std::vector<T> &) {
auto a_rotated = (V==3) ? 1 : 0;
tuner.AddArgumentScalar(static_cast<int>(args.m));
tuner.AddArgumentScalar(static_cast<int>(args.n));
tuner.AddArgumentScalar(GetRealArg(args.alpha));
tuner.AddArgumentScalar(GetRealArg(args.beta));
tuner.AddArgumentScalar(static_cast<int>(a_rotated));
tuner.AddArgumentInput(a_mat);
tuner.AddArgumentScalar(0);
tuner.AddArgumentScalar(static_cast<int>(args.m));
tuner.AddArgumentInput(x_vec);
tuner.AddArgumentScalar(0);
tuner.AddArgumentScalar(1);
tuner.AddArgumentOutput(y_vec);
tuner.AddArgumentScalar(0);
tuner.AddArgumentScalar(1);
tuner.AddArgumentScalar(0); // Conjugate transpose
tuner.AddArgumentScalar(0); // Additional parameter
tuner.AddArgumentScalar(0); // Banded 'kl'
tuner.AddArgumentScalar(0); // Banded 'ku'
}
