Merge pull request #99 from CNugteren/development

Update to version 0.9.0

1 files changed, 33 insertions, 23 deletions

diff --git a/src/routines/common.cpp b/src/routines/common.cpp
index c378df28..3969cf9f 100644
--- a/src/routines/common.cpp
+++ b/src/routines/common.cpp
@@ -12,6 +12,7 @@
 // =================================================================================================
 
 #include <vector>
+#include <chrono>
 
 #include "routines/common.hpp"
 
@@ -21,45 +22,54 @@ 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) {
+                     EventPointer event, const 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; }
+  if (!local.empty()) {
+    // 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]; }
+    // 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; }
 
+  // Prints the name of the kernel to launch in case of debugging in verbose mode
+  #ifdef VERBOSE
+    queue.Finish();
+    printf("[DEBUG] Running kernel '%s'\n", kernel.GetFunctionName().c_str());
+    const auto start_time = std::chrono::steady_clock::now();
+  #endif
+
   // Launches the kernel (and checks for launch errors)
   try {
     kernel.Launch(queue, global, local, event, waitForEvents);
   } catch (...) { return StatusCode::kKernelLaunchError; }
 
+  // Prints the elapsed execution time in case of debugging in verbose mode
+  #ifdef VERBOSE
+    queue.Finish();
+    const auto elapsed_time = std::chrono::steady_clock::now() - start_time;
+    const auto timing = std::chrono::duration<double,std::milli>(elapsed_time).count();
+    printf("[DEBUG] Completed kernel in %.2lf ms\n", timing);
+  #endif
+
   // 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