From 61203453aaca4e47c05c598a673150522160ca87 Mon Sep 17 00:00:00 2001
From: Cedric Nugteren <web@cedricnugteren.nl>
Date: Sun, 19 Jun 2016 13:55:49 +0200
Subject: Renamed all C++ source files to .cpp to match the .hpp extension
 better

---
 src/routines/common.cpp | 65 +++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 65 insertions(+)
 create mode 100644 src/routines/common.cpp

(limited to 'src/routines/common.cpp')

diff --git a/src/routines/common.cpp b/src/routines/common.cpp
new file mode 100644
index 00000000..c378df28
--- /dev/null
+++ b/src/routines/common.cpp
@@ -0,0 +1,65 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
+// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
+// width of 100 characters per line.
+//
+// Author(s):
+//   Cedric Nugteren <www.cedricnugteren.nl>
+//
+// This file implements the common routine functions (see the header for more information).
+//
+// =================================================================================================
+
+#include <vector>
+
+#include "routines/common.hpp"
+
+namespace clblast {
+// =================================================================================================
+
+// Enqueues a kernel, waits for completion, and checks for errors
+StatusCode RunKernel(Kernel &kernel, Queue &queue, const Device &device,
+                     std::vector<size_t> global, const std::vector<size_t> &local,
+                     EventPointer event, std::vector<Event>& waitForEvents) {
+
+  // Tests for validity of the local thread sizes
+  if (local.size() > device.MaxWorkItemDimensions()) {
+    return StatusCode::kInvalidLocalNumDimensions; 
+  }
+  const auto max_work_item_sizes = device.MaxWorkItemSizes();
+  for (auto i=size_t{0}; i<local.size(); ++i) {
+    if (local[i] > max_work_item_sizes[i]) { return StatusCode::kInvalidLocalThreadsDim; }
+  }
+  auto local_size = size_t{1};
+  for (auto &item: local) { local_size *= item; }
+  if (local_size > device.MaxWorkGroupSize()) { return StatusCode::kInvalidLocalThreadsTotal; }
+
+  // Make sure the global thread sizes are at least equal to the local sizes
+  for (auto i=size_t{0}; i<global.size(); ++i) {
+    if (global[i] < local[i]) { global[i] = local[i]; }
+  }
+
+  // Tests for local memory usage
+  const auto local_mem_usage = kernel.LocalMemUsage(device);
+  if (!device.IsLocalMemoryValid(local_mem_usage)) { return StatusCode::kInvalidLocalMemUsage; }
+
+  // Launches the kernel (and checks for launch errors)
+  try {
+    kernel.Launch(queue, global, local, event, waitForEvents);
+  } catch (...) { return StatusCode::kKernelLaunchError; }
+
+  // No errors, normal termination of this function
+  return StatusCode::kSuccess;
+}
+
+// As above, but without an event waiting list
+StatusCode RunKernel(Kernel &kernel, Queue &queue, const Device &device,
+                     std::vector<size_t> global, const std::vector<size_t> &local,
+                     EventPointer event) {
+  auto emptyWaitingList = std::vector<Event>();
+  return RunKernel(kernel, queue, device, global, local, event, emptyWaitingList);
+}
+
+// =================================================================================================
+} // namespace clblast
-- 
cgit v1.2.3