Added an option to the performance clients to do a warm-up run before timing

4 files changed, 20 insertions, 2 deletions

diff --git a/CHANGELOG b/CHANGELOG
index b0d70c80..15948bbd 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -4,6 +4,7 @@ Development version (next release)
 - Fixed proper MSVC dllimport and dllexport declarations
 - Fixed memory leaks related to events not being released
 - Fixed a bug with a size_t and cl_ulong mismatch on 32-bit systems
+- Added an option (-warm_up) to do a warm-up run before timing in the performance clients
 - Added tuned parameters for various devices (see README)
 
 Version 0.8.0
diff --git a/src/utilities.hpp b/src/utilities.hpp
index 5a4eef0f..d5efab9f 100644
--- a/src/utilities.hpp
+++ b/src/utilities.hpp
@@ -80,8 +80,9 @@ constexpr auto kArgComparecblas = "cblas";
 constexpr auto kArgStepSize = "step";
 constexpr auto kArgNumSteps = "num_steps";
 constexpr auto kArgNumRuns = "runs";
+constexpr auto kArgWarmUp = "warm_up";
 
-// The client-specific arguments in string form
+// The test-specific arguments in string form
 constexpr auto kArgFullTest = "full_test";
 constexpr auto kArgVerbose = "verbose";
 
diff --git a/test/performance/client.cpp b/test/performance/client.cpp
index d0068f8b..aaaab22e 100644
--- a/test/performance/client.cpp
+++ b/test/performance/client.cpp
@@ -113,6 +113,7 @@ Arguments<U> Client<T,U>::ParseArguments(int argc, char *argv[], const size_t le
   args.print_help     = CheckArgument(argc, argv, help, kArgHelp);
   args.silent         = CheckArgument(argc, argv, help, kArgQuiet);
   args.no_abbrv       = CheckArgument(argc, argv, help, kArgNoAbbreviations);
+  warm_up_            = CheckArgument(argc, argv, help, kArgWarmUp);
 
   // Prints the chosen (or defaulted) arguments to screen. This also serves as the help message,
   // which is thus always displayed (unless silence is specified).
@@ -244,12 +245,24 @@ template <typename T, typename U>
 double Client<T,U>::TimedExecution(const size_t num_runs, const Arguments<U> &args,
                                    Buffers<T> &buffers, Queue &queue,
                                    Routine run_blas, const std::string &library_name) {
+  auto status = StatusCode::kSuccess;
+
+  // Do an optional warm-up to omit compilation times and initialisations from the measurements
+  if (warm_up_) {
+    try {
+      status = run_blas(args, buffers, queue);
+    } catch (...) { status = static_cast<StatusCode>(kUnknownError); }
+    if (status != StatusCode::kSuccess) {
+      throw std::runtime_error(library_name+" error: "+ToString(static_cast<int>(status)));
+    }
+  }
+
+  // Start the timed part
   auto timings = std::vector<double>(num_runs);
   for (auto &timing: timings) {
     auto start_time = std::chrono::steady_clock::now();
 
     // Executes the main computation
-    auto status = StatusCode::kSuccess;
     try {
       status = run_blas(args, buffers, queue);
     } catch (...) { status = static_cast<StatusCode>(kUnknownError); }
diff --git a/test/performance/client.hpp b/test/performance/client.hpp
index 5ff2aec7..6d35fced 100644
--- a/test/performance/client.hpp
+++ b/test/performance/client.hpp
@@ -82,6 +82,9 @@ class Client {
   const std::vector<std::string> options_;
   const GetMetric get_flops_;
   const GetMetric get_bytes_;
+
+  // Extra arguments
+  bool warm_up_; // if enabled, do a warm-up run first before measuring execution time
 };
 
 // =================================================================================================