From dead6d0d44a4b7a24f99a1ad8296721c81ed7de3 Mon Sep 17 00:00:00 2001
From: Tom Deakin <tom.deakin@bristol.ac.uk>
Date: Thu, 15 Feb 2018 03:32:27 +0000
Subject: [PATCH] [Kokkos] Use tempate type throughout instead of double

Fixes #44. Also requires the typedef keyword in a few places.
---
 CHANGELOG.md     |  1 +
 KokkosStream.cpp | 50 ++++++++++++++++++++++++------------------------
 KokkosStream.hpp | 12 ++++++------
 main.cpp         |  3 ---
 4 files changed, 32 insertions(+), 34 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9765632..bc1d14d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -19,6 +19,7 @@ All notable changes to this project will be documented in this file.
 
 ### Fixed
 - Kokkos now compiles and links separately to fix complication with Kokkos 2.05.00.
+- Kokkos can now instantiate single and double precision.
 - OpenMP 4.5 map and reduction clause order to ensure reduction result copied back.
 
 
diff --git a/KokkosStream.cpp b/KokkosStream.cpp
index 7054fba..19d4b24 100644
--- a/KokkosStream.cpp
+++ b/KokkosStream.cpp
@@ -14,12 +14,12 @@ KokkosStream<T>::KokkosStream(
 {
   Kokkos::initialize();
 
-  d_a = new Kokkos::View<double*>("d_a", ARRAY_SIZE);
-  d_b = new Kokkos::View<double*>("d_b", ARRAY_SIZE);
-  d_c = new Kokkos::View<double*>("d_c", ARRAY_SIZE);
-  hm_a = new Kokkos::View<double*>::HostMirror();
-  hm_b = new Kokkos::View<double*>::HostMirror();
-  hm_c = new Kokkos::View<double*>::HostMirror();
+  d_a = new Kokkos::View<T*>("d_a", ARRAY_SIZE);
+  d_b = new Kokkos::View<T*>("d_b", ARRAY_SIZE);
+  d_c = new Kokkos::View<T*>("d_c", ARRAY_SIZE);
+  hm_a = new typename Kokkos::View<T*>::HostMirror();
+  hm_b = new typename Kokkos::View<T*>::HostMirror();
+  hm_c = new typename Kokkos::View<T*>::HostMirror();
   *hm_a = create_mirror_view(*d_a);
   *hm_b = create_mirror_view(*d_b);
   *hm_c = create_mirror_view(*d_c);
@@ -34,9 +34,9 @@ KokkosStream<T>::~KokkosStream()
 template <class T>
 void KokkosStream<T>::init_arrays(T initA, T initB, T initC)
 {
-  Kokkos::View<double*> a(*d_a);
-  Kokkos::View<double*> b(*d_b);
-  Kokkos::View<double*> c(*d_c);
+  Kokkos::View<T*> a(*d_a);
+  Kokkos::View<T*> b(*d_b);
+  Kokkos::View<T*> c(*d_c);
   Kokkos::parallel_for(array_size, KOKKOS_LAMBDA (const long index)
   {
     a[index] = initA;
@@ -64,9 +64,9 @@ void KokkosStream<T>::read_arrays(
 template <class T>
 void KokkosStream<T>::copy()
 {
-  Kokkos::View<double*> a(*d_a);
-  Kokkos::View<double*> b(*d_b);
-  Kokkos::View<double*> c(*d_c);
+  Kokkos::View<T*> a(*d_a);
+  Kokkos::View<T*> b(*d_b);
+  Kokkos::View<T*> c(*d_c);
 
   Kokkos::parallel_for(array_size, KOKKOS_LAMBDA (const long index)
   {
@@ -78,9 +78,9 @@ void KokkosStream<T>::copy()
 template <class T>
 void KokkosStream<T>::mul()
 {
-  Kokkos::View<double*> a(*d_a);
-  Kokkos::View<double*> b(*d_b);
-  Kokkos::View<double*> c(*d_c);
+  Kokkos::View<T*> a(*d_a);
+  Kokkos::View<T*> b(*d_b);
+  Kokkos::View<T*> c(*d_c);
 
   const T scalar = startScalar;
   Kokkos::parallel_for(array_size, KOKKOS_LAMBDA (const long index)
@@ -93,9 +93,9 @@ void KokkosStream<T>::mul()
 template <class T>
 void KokkosStream<T>::add()
 {
-  Kokkos::View<double*> a(*d_a);
-  Kokkos::View<double*> b(*d_b);
-  Kokkos::View<double*> c(*d_c);
+  Kokkos::View<T*> a(*d_a);
+  Kokkos::View<T*> b(*d_b);
+  Kokkos::View<T*> c(*d_c);
 
   Kokkos::parallel_for(array_size, KOKKOS_LAMBDA (const long index)
   {
@@ -107,9 +107,9 @@ void KokkosStream<T>::add()
 template <class T>
 void KokkosStream<T>::triad()
 {
-  Kokkos::View<double*> a(*d_a);
-  Kokkos::View<double*> b(*d_b);
-  Kokkos::View<double*> c(*d_c);
+  Kokkos::View<T*> a(*d_a);
+  Kokkos::View<T*> b(*d_b);
+  Kokkos::View<T*> c(*d_c);
 
   const T scalar = startScalar;
   Kokkos::parallel_for(array_size, KOKKOS_LAMBDA (const long index)
@@ -122,12 +122,12 @@ void KokkosStream<T>::triad()
 template <class T>
 T KokkosStream<T>::dot()
 {
-  Kokkos::View<double*> a(*d_a);
-  Kokkos::View<double*> b(*d_b);
+  Kokkos::View<T*> a(*d_a);
+  Kokkos::View<T*> b(*d_b);
 
   T sum = 0.0;
 
-  Kokkos::parallel_reduce(array_size, KOKKOS_LAMBDA (const long index, double &tmp)
+  Kokkos::parallel_reduce(array_size, KOKKOS_LAMBDA (const long index, T &tmp)
   {
     tmp += a[index] * b[index];
   }, sum);
@@ -153,5 +153,5 @@ std::string getDeviceDriver(const int device)
   return "Kokkos";
 }
 
-//template class KokkosStream<float>;
+template class KokkosStream<float>;
 template class KokkosStream<double>;
diff --git a/KokkosStream.hpp b/KokkosStream.hpp
index 9ba6fea..1470cf4 100644
--- a/KokkosStream.hpp
+++ b/KokkosStream.hpp
@@ -25,12 +25,12 @@ class KokkosStream : public Stream<T>
     unsigned int array_size;
 
     // Device side pointers to arrays
-    Kokkos::View<double*>* d_a;
-    Kokkos::View<double*>* d_b;
-    Kokkos::View<double*>* d_c;
-    Kokkos::View<double*>::HostMirror* hm_a;
-    Kokkos::View<double*>::HostMirror* hm_b;
-    Kokkos::View<double*>::HostMirror* hm_c;
+     typename Kokkos::View<T*>* d_a;
+     typename Kokkos::View<T*>* d_b;
+     typename Kokkos::View<T*>* d_c;
+     typename Kokkos::View<T*>::HostMirror* hm_a;
+     typename Kokkos::View<T*>::HostMirror* hm_b;
+     typename Kokkos::View<T*>::HostMirror* hm_c;
 
   public:
 
diff --git a/main.cpp b/main.cpp
index 97443d4..d35cc76 100644
--- a/main.cpp
+++ b/main.cpp
@@ -85,12 +85,9 @@ int main(int argc, char *argv[])
   }
   else
   {
-    // TODO: Fix Kokkos to allow multiple template specializations
-#ifndef KOKKOS
     if (use_float)
       run<float>();
     else
-#endif
       run<double>();
   }