From f0afa0c1e43c43729dd26cca719d1db21844d279 Mon Sep 17 00:00:00 2001
From: Tom Deakin <tom.deakin@bristol.ac.uk>
Date: Wed, 4 May 2016 10:41:41 +0100
Subject: [PATCH] Add reference OpenMP 3.0 version

---
 OMP3Stream.cpp | 106 +++++++++++++++++++++++++++++++++++++++++++++++++
 OMP3Stream.h   |  35 ++++++++++++++++
 main.cpp       |   7 ++++
 3 files changed, 148 insertions(+)
 create mode 100644 OMP3Stream.cpp
 create mode 100644 OMP3Stream.h

diff --git a/OMP3Stream.cpp b/OMP3Stream.cpp
new file mode 100644
index 0000000..8899cff
--- /dev/null
+++ b/OMP3Stream.cpp
@@ -0,0 +1,106 @@
+
+#include "OMP3Stream.h"
+
+template <class T>
+OMP3Stream<T>::OMP3Stream(const unsigned int ARRAY_SIZE, T *a, T *b, T *c)
+{
+  array_size = ARRAY_SIZE;
+  this->a = (T*)malloc(sizeof(T)*array_size);
+  this->b = (T*)malloc(sizeof(T)*array_size);
+  this->c = (T*)malloc(sizeof(T)*array_size);
+}
+
+template <class T>
+OMP3Stream<T>::~OMP3Stream()
+{
+  free(a);
+  free(b);
+  free(c);
+}
+
+
+template <class T>
+void OMP3Stream<T>::write_arrays(const std::vector<T>& h_a, const std::vector<T>& h_b, const std::vector<T>& h_c)
+{
+  #pragma omp parallel for
+  for (int i = 0; i < array_size; i++)
+  {
+    a[i] = h_a[i];
+    b[i] = h_b[i];
+    c[i] = h_c[i];
+  }
+}
+
+template <class T>
+void OMP3Stream<T>::read_arrays(std::vector<T>& h_a, std::vector<T>& h_b, std::vector<T>& h_c)
+{
+  #pragma omp parallel for
+  for (int i = 0; i < array_size; i++)
+  {
+    h_a[i] = a[i];
+    h_b[i] = b[i];
+    h_c[i] = c[i];
+  }
+}
+
+template <class T>
+void OMP3Stream<T>::copy()
+{
+  #pragma omp parallel for
+  for (int i = 0; i < array_size; i++)
+  {
+    c[i] = a[i];
+  }
+}
+
+template <class T>
+void OMP3Stream<T>::mul()
+{
+  const T scalar = 3.0;
+  #pragma omp parallel for
+  for (int i = 0; i < array_size; i++)
+  {
+    b[i] = scalar * c[i];
+  }
+}
+
+template <class T>
+void OMP3Stream<T>::add()
+{
+  #pragma omp parallel for
+  for (int i = 0; i < array_size; i++)
+  {
+    c[i] = a[i] + b[i];
+  }
+}
+
+template <class T>
+void OMP3Stream<T>::triad()
+{
+  const T scalar = 3.0;
+  #pragma omp parallel for
+  for (int i = 0; i < array_size; i++)
+  {
+    a[i] = b[i] + scalar * c[i];
+  }
+}
+
+void listDevices(void)
+{
+  std::cout << "0: CPU" << std::endl;
+}
+
+std::string getDeviceName(const int)
+{
+  return std::string("Device name unavailable");
+}
+
+std::string getDeviceDriver(const int)
+{
+  return std::string("Device driver unavailable");
+}
+
+
+template class OMP3Stream<float>;
+template class OMP3Stream<double>;
+
diff --git a/OMP3Stream.h b/OMP3Stream.h
new file mode 100644
index 0000000..15172c3
--- /dev/null
+++ b/OMP3Stream.h
@@ -0,0 +1,35 @@
+
+#pragma once
+
+#include <iostream>
+#include <stdexcept>
+
+#include "Stream.h"
+
+#define IMPLEMENTATION_STRING "Reference OpenMP"
+
+template <class T>
+class OMP3Stream : public Stream<T>
+{
+  protected:
+    // Size of arrays
+    unsigned int array_size;
+    // Device side pointers
+    T *a;
+    T *b;
+    T *c;
+
+  public:
+    OMP3Stream(const unsigned int, T*, T*, T*);
+    ~OMP3Stream();
+
+    virtual void copy() override;
+    virtual void add() override;
+    virtual void mul() override;
+    virtual void triad() override;
+
+    virtual void write_arrays(const std::vector<T>& a, const std::vector<T>& b, const std::vector<T>& c) override;
+    virtual void read_arrays(std::vector<T>& a, std::vector<T>& b, std::vector<T>& c) override;
+
+};
+
diff --git a/main.cpp b/main.cpp
index 59411b8..18b8c72 100644
--- a/main.cpp
+++ b/main.cpp
@@ -26,6 +26,8 @@
 #include "ACCStream.h"
 #elif defined(SYCL)
 #include "SYCLStream.h"
+#elif defined(OMP3)
+#include "OMP3Stream.h"
 #endif
 
 
@@ -92,6 +94,11 @@ void run()
   // Use the SYCL implementation
   stream = new SYCLStream<T>(ARRAY_SIZE, deviceIndex);
 
+#elif defined(OMP3)
+  // Use the "reference" OpenMP 3 implementation
+  stream = new OMP3Stream<T>(ARRAY_SIZE, a.data(), b.data(), c.data());
+
+
 #endif
 
   stream->write_arrays(a, b, c);