From a0195132e8739fdea9cbfcef5d04ab1afcfb17c8 Mon Sep 17 00:00:00 2001
From: moneta <lorenzo.moneta@cern.ch>
Date: Tue, 21 Aug 2018 10:10:43 +0200
Subject: [PATCH] Use fast zero GPU initialisation for Cuda matrices
---
.../src/DNN/Architectures/Cuda/Initialization.cu | 13 ++-----------
1 file changed, 2 insertions(+), 11 deletions(-)
diff --git a/tmva/tmva/src/DNN/Architectures/Cuda/Initialization.cu b/tmva/tmva/src/DNN/Architectures/Cuda/Initialization.cu
index 6efd88ddcd2..3acb8afebfd 100644
--- a/tmva/tmva/src/DNN/Architectures/Cuda/Initialization.cu
+++ b/tmva/tmva/src/DNN/Architectures/Cuda/Initialization.cu
@@ -158,17 +158,8 @@ void TCuda<AFloat>::InitializeIdentity(TCudaMatrix<AFloat> & A)
template<typename AFloat>
void TCuda<AFloat>::InitializeZero(TCudaMatrix<AFloat> & A)
{
- size_t m,n;
- m = A.GetNrows();
- n = A.GetNcols();
- TMatrixT<AFloat> B(m, n);
-
- for (size_t i = 0; i < m; i++) {
- for (size_t j = 0; j < n ; j++) {
- B(i,j) = 0.0;
- }
- }
- A = B;
+ // use fast zero initialization on the device
+ A.Zero();
}
} // namespace DNN
--
GitLab