Fixed number of threads launched for AXPY

This commit is contained in:
CNugteren 2015-06-14 10:08:23 +02:00
parent 4b3e3dcfe0
commit 9aa2989447

View file

@ -88,8 +88,8 @@ StatusCode Xaxpy<T>::DoAxpy(const size_t n, const T alpha,
status = RunKernel(kernel, global, local); status = RunKernel(kernel, global, local);
} }
else { else {
auto n_ceiled = Ceil(n, db_["WGS"]); auto n_ceiled = Ceil(n, db_["WGS"]*db_["WPT"]);
auto global = std::vector<size_t>{CeilDiv(n_ceiled, db_["WPT"])}; auto global = std::vector<size_t>{n_ceiled/db_["WPT"]};
auto local = std::vector<size_t>{db_["WGS"]}; auto local = std::vector<size_t>{db_["WGS"]};
status = RunKernel(kernel, global, local); status = RunKernel(kernel, global, local);
} }