6 files changed, 6 insertions, 0 deletions
diff --git a/src/routines/level3/xgemm.cpp b/src/routines/level3/xgemm.cpp
index 94392dd0..edba1f00 100644
--- a/src/routines/level3/xgemm.cpp
+++ b/src/routines/level3/xgemm.cpp
@@ -41,6 +41,7 @@ Xgemm<T>::Xgemm(Queue &queue, EventPointer event, const std::string &name):
     #include "../../kernels/level3/xgemm_part1.opencl"
     #include "../../kernels/level3/xgemm_part2.opencl"
     #include "../../kernels/level3/xgemm_part3.opencl"
+    #include "../../kernels/level3/xgemm_part4.opencl"
     }) {
 }
 
diff --git a/src/routines/level3/xher2k.cpp b/src/routines/level3/xher2k.cpp
index 2aed2781..3010dea8 100644
--- a/src/routines/level3/xher2k.cpp
+++ b/src/routines/level3/xher2k.cpp
@@ -31,6 +31,7 @@ Xher2k<T,U>::Xher2k(Queue &queue, EventPointer event, const std::string &name):
     #include "../../kernels/level3/xgemm_part1.opencl"
     #include "../../kernels/level3/xgemm_part2.opencl"
     #include "../../kernels/level3/xgemm_part3.opencl"
+    #include "../../kernels/level3/xgemm_part4.opencl"
     }) {
 }
 
diff --git a/src/routines/level3/xherk.cpp b/src/routines/level3/xherk.cpp
index d982859e..38a8420f 100644
--- a/src/routines/level3/xherk.cpp
+++ b/src/routines/level3/xherk.cpp
@@ -31,6 +31,7 @@ Xherk<T,U>::Xherk(Queue &queue, EventPointer event, const std::string &name):
     #include "../../kernels/level3/xgemm_part1.opencl"
     #include "../../kernels/level3/xgemm_part2.opencl"
     #include "../../kernels/level3/xgemm_part3.opencl"
+    #include "../../kernels/level3/xgemm_part4.opencl"
     }) {
 }
 
diff --git a/src/routines/level3/xsyr2k.cpp b/src/routines/level3/xsyr2k.cpp
index 7900eb74..5ff4769d 100644
--- a/src/routines/level3/xsyr2k.cpp
+++ b/src/routines/level3/xsyr2k.cpp
@@ -31,6 +31,7 @@ Xsyr2k<T>::Xsyr2k(Queue &queue, EventPointer event, const std::string &name):
     #include "../../kernels/level3/xgemm_part1.opencl"
     #include "../../kernels/level3/xgemm_part2.opencl"
     #include "../../kernels/level3/xgemm_part3.opencl"
+    #include "../../kernels/level3/xgemm_part4.opencl"
     }) {
 }
 
diff --git a/src/routines/level3/xsyrk.cpp b/src/routines/level3/xsyrk.cpp
index 9588c28c..79e3a4ab 100644
--- a/src/routines/level3/xsyrk.cpp
+++ b/src/routines/level3/xsyrk.cpp
@@ -31,6 +31,7 @@ Xsyrk<T>::Xsyrk(Queue &queue, EventPointer event, const std::string &name):
     #include "../../kernels/level3/xgemm_part1.opencl"
     #include "../../kernels/level3/xgemm_part2.opencl"
     #include "../../kernels/level3/xgemm_part3.opencl"
+    #include "../../kernels/level3/xgemm_part4.opencl"
     }) {
 }
 
diff --git a/src/routines/levelx/xgemmbatched.cpp b/src/routines/levelx/xgemmbatched.cpp
index 152e7194..8ce2dedc 100644
--- a/src/routines/levelx/xgemmbatched.cpp
+++ b/src/routines/levelx/xgemmbatched.cpp
@@ -38,6 +38,7 @@ XgemmBatched<T>::XgemmBatched(Queue &queue, EventPointer event, const std::strin
     #include "../../kernels/level3/xgemm_part1.opencl"
     #include "../../kernels/level3/xgemm_part2.opencl"
     #include "../../kernels/level3/xgemm_part3.opencl"
+    #include "../../kernels/level3/xgemm_part4.opencl"
     , // separated in multiple parts to prevent C1091 in MSVC 2013
     #include "../../kernels/level3/xgemm_batched.opencl"
     #include "../../kernels/level3/xgemm_direct_batched.opencl"