diff options
Diffstat (limited to 'src/routines/level3')
-rw-r--r-- | src/routines/level3/xgemm.cpp | 10 |
1 files changed, 6 insertions, 4 deletions
diff --git a/src/routines/level3/xgemm.cpp b/src/routines/level3/xgemm.cpp index f4611aba..136eec43 100644 --- a/src/routines/level3/xgemm.cpp +++ b/src/routines/level3/xgemm.cpp @@ -280,11 +280,13 @@ void Xgemm<T>::GemmDirect(const size_t m, const size_t n, const size_t k, kernel.SetArgument(16, static_cast<int>(b_conjugate)); // Computes the global and local thread sizes - //const auto m_ceiled = Ceil(m, db_["WGD"]); - //const auto n_ceiled = Ceil(n, db_["WGD"]); + const auto m_ceiled = Ceil(m, db_["WGD"]); + const auto n_ceiled = Ceil(n, db_["WGD"]); const auto global = std::vector<size_t>{ - CeilDiv(m * db_["MDIMCD"], db_["WGD"]), - CeilDiv(n * db_["NDIMCD"], db_["WGD"]) + // CeilDiv(m * db_["MDIMCD"], db_["WGD"]), + // CeilDiv(n * db_["NDIMCD"], db_["WGD"]) + (m_ceiled * db_["MDIMCD"]) / db_["WGD"], + (n_ceiled * db_["NDIMCD"]) / db_["WGD"] }; const auto local = std::vector<size_t>{db_["MDIMCD"], db_["NDIMCD"]}; |