Update to Kokkos r2.04.04 and add workaround for performance regression

lib/kokkos/bin/nvcc_wrapper

@@ -78,6 +78,9 @@ temp_dir=${TMPDIR:-/tmp}
 # Check if we have an optimization argument already
 optimization_applied=0
 
+# Check if we have -std=c++X  or --std=c++X already
+stdcxx_applied=0
+
 #echo "Arguments: $# $@"
 
 while [ $# -gt 0 ]
@@ -130,10 +133,16 @@ do
     cuda_args="$cuda_args $1 $2"
     shift
     ;;
-  #Handle c++11 setting
-  --std=c++11|-std=c++11)
-    shared_args="$shared_args $1"
+  #Handle c++11
+  --std=c++11|-std=c++11|--std=c++14|-std=c++14|--std=c++1z|-std=c++1z)
+    if [ $stdcxx_applied -eq 1 ]; then
+       echo "nvcc_wrapper - *warning* you have set multiple optimization flags (-std=c++1* or --std=c++1*), only the first is used because nvcc can only accept a single std setting"
+    else
+       shared_args="$shared_args $1"
+       stdcxx_applied=1
+    fi
     ;;
+
   #strip of -std=c++98 due to nvcc warnings and Tribits will place both -std=c++11 and -std=c++98
   -std=c++98|--std=c++98)
     ;;