From d7c17d72d56247a04bcf8c58ca4586282d1e234b Mon Sep 17 00:00:00 2001
From: Tom Deakin <tom.deakin@bristol.ac.uk>
Date: Tue, 3 May 2016 11:02:33 +0100
Subject: [PATCH] Use device index from CLI in OpenCL

---
 src/OCLStream.cpp | 40 ++++++++++++++++++++++++++--------------
 src/OCLStream.h   |  3 ++-
 src/main.cpp      |  5 ++---
 3 files changed, 30 insertions(+), 18 deletions(-)

diff --git a/src/OCLStream.cpp b/src/OCLStream.cpp
index 34b7933..422b409 100644
--- a/src/OCLStream.cpp
+++ b/src/OCLStream.cpp
@@ -1,6 +1,11 @@
 
 #include "OCLStream.h"
 
+// Cache list of devices
+bool cached = false;
+std::vector<cl::Device> devices;
+void getDeviceList(void);
+
 std::string kernels{R"CLC(
 
   constant TYPE scalar = 3.0;
@@ -43,16 +48,27 @@ std::string kernels{R"CLC(
 
 
 template <class T>
-OCLStream<T>::OCLStream(const unsigned int ARRAY_SIZE)
+OCLStream<T>::OCLStream(const unsigned int ARRAY_SIZE, const int device_index)
 {
+  if (!cached)
+    getDeviceList();
+
   array_size = ARRAY_SIZE;
 
   // Setup default OpenCL GPU
-  context = cl::Context::getDefault();
-  queue = cl::CommandQueue::getDefault();
+  if (device_index >= devices.size())
+    throw std::runtime_error("Invalid device index");
+  device = devices[device_index];
+
+  // Print out device information
+  std::cout << "Using OpenCL device " << getDeviceName(device_index) << std::endl;
+  std::cout << "Driver: " << getDeviceDriver(device_index) << std::endl;
+
+  context = cl::Context(device);
+  queue = cl::CommandQueue(context);
 
   // Create program
-  cl::Program program(kernels);
+  cl::Program program(context, kernels);
   if (sizeof(T) == sizeof(double))
     program.build("-DTYPE=double");
   else if (sizeof(T) == sizeof(float))
@@ -123,23 +139,19 @@ void OCLStream<T>::triad()
 template <class T>
 void OCLStream<T>::write_arrays(const std::vector<T>& a, const std::vector<T>& b, const std::vector<T>& c)
 {
-  cl::copy(a.begin(), a.end(), d_a);
-  cl::copy(b.begin(), b.end(), d_b);
-  cl::copy(c.begin(), c.end(), d_c);
+  cl::copy(queue, a.begin(), a.end(), d_a);
+  cl::copy(queue, b.begin(), b.end(), d_b);
+  cl::copy(queue, c.begin(), c.end(), d_c);
 }
 
 template <class T>
 void OCLStream<T>::read_arrays(std::vector<T>& a, std::vector<T>& b, std::vector<T>& c)
 {
-  cl::copy(d_a, a.begin(), a.end());
-  cl::copy(d_b, b.begin(), b.end());
-  cl::copy(d_c, c.begin(), c.end());
+  cl::copy(queue, d_a, a.begin(), a.end());
+  cl::copy(queue, d_b, b.begin(), b.end());
+  cl::copy(queue, d_c, c.begin(), c.end());
 }
 
-// Cache list of devices
-bool cached = false;
-std::vector<cl::Device> devices;
-
 void getDeviceList(void)
 {
   // Get list of platforms
diff --git a/src/OCLStream.h b/src/OCLStream.h
index 551cf28..3fd144a 100644
--- a/src/OCLStream.h
+++ b/src/OCLStream.h
@@ -27,6 +27,7 @@ class OCLStream : public Stream<T>
     cl::Buffer d_c;
 
     // OpenCL objects
+    cl::Device device;
     cl::Context context;
     cl::CommandQueue queue;
 
@@ -37,7 +38,7 @@ class OCLStream : public Stream<T>
 
   public:
 
-    OCLStream(const unsigned int);
+    OCLStream(const unsigned int, const int);
     ~OCLStream();
 
     virtual void copy() override;
diff --git a/src/main.cpp b/src/main.cpp
index 7ec7873..1171180 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -61,11 +61,11 @@ void run()
 
 #if defined(CUDA)
   // Use the CUDA implementation
-  stream = new CUDAStream<T>(ARRAY_SIZE);
+  stream = new CUDAStream<T>(ARRAY_SIZE, deviceIndex);
 
 #elif defined(OCL)
   // Use the OpenCL implementation
-  stream = new OCLStream<T>(ARRAY_SIZE);
+  stream = new OCLStream<T>(ARRAY_SIZE, deviceIndex);
 
 #endif
 
@@ -144,7 +144,6 @@ void run()
       << std::left << std::setw(12) << std::setprecision(5) << *minmax.second
       << std::left << std::setw(12) << std::setprecision(5) << average
       << std::endl;
-    
 
   }