From 73d135c2cef9763b47d410b125eb8bb89ece8432 Mon Sep 17 00:00:00 2001 From: Cedric Nugteren Date: Sun, 25 Sep 2016 14:48:34 +0200 Subject: Added a first version of a tuner for the GEMM direct kernel; collapsed MWGD, NWGD and KWGD into one WGD parameter --- src/routines/level3/xgemm.cpp | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'src/routines') diff --git a/src/routines/level3/xgemm.cpp b/src/routines/level3/xgemm.cpp index 2fb9f1fd..e050e844 100644 --- a/src/routines/level3/xgemm.cpp +++ b/src/routines/level3/xgemm.cpp @@ -300,11 +300,11 @@ StatusCode Xgemm::GemmDirect(const size_t m, const size_t n, const size_t k, kernel.SetArgument(18, static_cast(b_conjugate)); // Computes the global and local thread sizes - const auto m_ceiled = Ceil(m, db_["MWGD"]); - const auto n_ceiled = Ceil(n, db_["NWGD"]); + const auto m_ceiled = Ceil(m, db_["WGD"]); + const auto n_ceiled = Ceil(n, db_["WGD"]); const auto global = std::vector{ - (m_ceiled * db_["MDIMCD"]) / db_["MWGD"], - (n_ceiled * db_["NDIMCD"]) / db_["NWGD"] + (m_ceiled * db_["MDIMCD"]) / db_["WGD"], + (n_ceiled * db_["NDIMCD"]) / db_["WGD"] }; const auto local = std::vector{db_["MDIMCD"], db_["NDIMCD"]}; -- cgit v1.2.3