COMPMID-1400: Add command line option to specify the tuner's config file

Change-Id: Ib597e0dff4c8c01f7e6bd46d03824beef4bc1e9a
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/139923
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

17 files changed, 73 insertions, 18 deletions

diff --git a/arm_compute/graph/Types.h b/arm_compute/graph/Types.h
index 9cfede9cf3..c5b7fb1c51 100644
--- a/arm_compute/graph/Types.h
+++ b/arm_compute/graph/Types.h
@@ -77,10 +77,11 @@ class TensorDescriptor;
 /** Graph configuration structure */
 struct GraphConfig
 {
-    bool use_function_memory_manager{ true };   /**< Use a memory manager to manage per-funcion auxilary memory */
-    bool use_transition_memory_manager{ true }; /**< Use a memory manager to manager transition buffer memory */
-    bool use_tuner{ false };                    /**< Use a tuner in tunable backends */
-    int  num_threads{ -1 };                     /**< Number of threads to use (thread capable backends), if 0 the backend will auto-initialize, if -1 the backend will stay as it is. */
+    bool        use_function_memory_manager{ true };   /**< Use a memory manager to manage per-funcion auxilary memory */
+    bool        use_transition_memory_manager{ true }; /**< Use a memory manager to manager transition buffer memory */
+    bool        use_tuner{ false };                    /**< Use a tuner in tunable backends */
+    int         num_threads{ -1 };                     /**< Number of threads to use (thread capable backends), if 0 the backend will auto-initialize, if -1 the backend will stay as it is. */
+    std::string tuner_file{ "acl_tuner.csv" };         /**< File to load/store tuning values from */
 };
 
 /**< Device target types */
diff --git a/arm_compute/graph/backends/CL/CLDeviceBackend.h b/arm_compute/graph/backends/CL/CLDeviceBackend.h
index 0b45dfe479..c1a6a28e6c 100644
--- a/arm_compute/graph/backends/CL/CLDeviceBackend.h
+++ b/arm_compute/graph/backends/CL/CLDeviceBackend.h
@@ -66,6 +66,7 @@ private:
     bool                               _initialized; /**< Flag that specifies if the backend has been default initialized */
     CLTuner                            _tuner;       /**< CL kernel tuner */
     std::unique_ptr<CLBufferAllocator> _allocator;   /**< CL buffer affinity allocator */
+    std::string                        _tuner_file;  /** Filename to load/store the tuner's values from */
 };
 } // namespace backends
 } // namespace graph
diff --git a/examples/graph_alexnet.cpp b/examples/graph_alexnet.cpp
index bf4d131ac5..e4bc2da571 100644
--- a/examples/graph_alexnet.cpp
+++ b/examples/graph_alexnet.cpp
@@ -149,6 +149,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_googlenet.cpp b/examples/graph_googlenet.cpp
index ff7992c164..f9cb9787f1 100644
--- a/examples/graph_googlenet.cpp
+++ b/examples/graph_googlenet.cpp
@@ -121,6 +121,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_inception_v3.cpp b/examples/graph_inception_v3.cpp
index 1b7b0fd6c2..6262193b02 100644
--- a/examples/graph_inception_v3.cpp
+++ b/examples/graph_inception_v3.cpp
@@ -199,6 +199,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_inception_v4.cpp b/examples/graph_inception_v4.cpp
index 93166a41c2..def73dff7a 100644
--- a/examples/graph_inception_v4.cpp
+++ b/examples/graph_inception_v4.cpp
@@ -143,6 +143,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_lenet.cpp b/examples/graph_lenet.cpp
index 9be5bed6a1..c02e2f9ec0 100644
--- a/examples/graph_lenet.cpp
+++ b/examples/graph_lenet.cpp
@@ -105,6 +105,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_mobilenet.cpp b/examples/graph_mobilenet.cpp
index 2b662d9055..395936b8f7 100644
--- a/examples/graph_mobilenet.cpp
+++ b/examples/graph_mobilenet.cpp
@@ -109,6 +109,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_resnext50.cpp b/examples/graph_resnext50.cpp
index 0fce3a1b04..a02e406acb 100644
--- a/examples/graph_resnext50.cpp
+++ b/examples/graph_resnext50.cpp
@@ -97,6 +97,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_squeezenet.cpp b/examples/graph_squeezenet.cpp
index 09fb1c5960..9751539af6 100644
--- a/examples/graph_squeezenet.cpp
+++ b/examples/graph_squeezenet.cpp
@@ -156,6 +156,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_squeezenet_v1_1.cpp b/examples/graph_squeezenet_v1_1.cpp
index 75ea6b258b..be0c326f2a 100644
--- a/examples/graph_squeezenet_v1_1.cpp
+++ b/examples/graph_squeezenet_v1_1.cpp
@@ -158,6 +158,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_vgg16.cpp b/examples/graph_vgg16.cpp
index c2a9b82f68..8ce1221ead 100644
--- a/examples/graph_vgg16.cpp
+++ b/examples/graph_vgg16.cpp
@@ -213,6 +213,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/examples/graph_vgg19.cpp b/examples/graph_vgg19.cpp
index 8de0223a48..dcde4a2a65 100644
--- a/examples/graph_vgg19.cpp
+++ b/examples/graph_vgg19.cpp
@@ -225,6 +225,8 @@ public:
         GraphConfig config;
         config.num_threads = common_params.threads;
         config.use_tuner   = common_params.enable_tuner;
+        config.tuner_file  = common_params.tuner_file;
+
         graph.finalize(common_params.target, config);
 
         return true;
diff --git a/src/graph/backends/CL/CLDeviceBackend.cpp b/src/graph/backends/CL/CLDeviceBackend.cpp
index b235c3aa48..6717f9f8e3 100644
--- a/src/graph/backends/CL/CLDeviceBackend.cpp
+++ b/src/graph/backends/CL/CLDeviceBackend.cpp
@@ -62,20 +62,17 @@ bool file_exists(const std::string &filename)
 /** Register CL backend */
 static detail::BackendRegistrar<CLDeviceBackend> CLDeviceBackend_registrar(Target::CL);
 
-/** Tuner export file */
-static const std::string tuner_data_filename = "acl_tuner.csv";
-
 CLDeviceBackend::CLDeviceBackend()
-    : _initialized(false), _tuner(), _allocator(nullptr)
+    : _initialized(false), _tuner(), _allocator(nullptr), _tuner_file()
 {
 }
 
 CLDeviceBackend::~CLDeviceBackend()
 {
     // TODO (geopin01) : Shouldn't call non exception safe stuff here
-    if(_tuner.tune_new_kernels() && !_tuner.lws_table().empty())
+    if(_tuner.tune_new_kernels() && !_tuner.lws_table().empty() && !_tuner_file.empty())
     {
-        _tuner.save_to_file(tuner_data_filename);
+        _tuner.save_to_file(_tuner_file);
     }
 }
 
@@ -86,12 +83,6 @@ void CLDeviceBackend::set_kernel_tuning(bool enable_tuning)
 
 void CLDeviceBackend::initialize_backend()
 {
-    // Load tuner data if available
-    if(_tuner.lws_table().empty() && file_exists(tuner_data_filename))
-    {
-        _tuner.load_from_file(tuner_data_filename);
-    }
-
     // Setup Scheduler
     CLScheduler::get().default_init(&_tuner);
 
@@ -109,6 +100,13 @@ void CLDeviceBackend::setup_backend_context(GraphContext &ctx)
     }
 
     // Setup tuner
+    _tuner_file = ctx.config().tuner_file;
+    // Load tuner data if available
+    if(file_exists(_tuner_file))
+    {
+        _tuner.load_from_file(_tuner_file);
+    }
+
     set_kernel_tuning(ctx.config().use_tuner);
 
     // Setup a management backend
@@ -193,4 +191,4 @@ std::shared_ptr<arm_compute::IMemoryManager> CLDeviceBackend::create_memory_mana
 }
 } // namespace backends
 } // namespace graph
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute
diff --git a/tests/main.cpp b/tests/main.cpp
index 2da6804c60..aaae99157a 100644
--- a/tests/main.cpp
+++ b/tests/main.cpp
@@ -61,6 +61,17 @@ std::unique_ptr<AssetsLibrary> library;
 } // namespace test
 } // namespace arm_compute
 
+namespace
+{
+#ifdef ARM_COMPUTE_CL
+bool file_exists(const std::string &filename)
+{
+    std::ifstream file(filename);
+    return file.good();
+}
+#endif /* ARM_COMPUTE_CL */
+} //namespace
+
 int main(int argc, char **argv)
 {
 #ifdef ARM_COMPUTE_CL
@@ -106,6 +117,8 @@ int main(int argc, char **argv)
 #ifdef ARM_COMPUTE_CL
     auto enable_tuner = parser.add_option<utils::ToggleOption>("enable-tuner");
     enable_tuner->set_help("Enable OpenCL dynamic tuner");
+    auto tuner_file = parser.add_option<utils::SimpleOption<std::string>>("tuner-file", "");
+    tuner_file->set_help("File to load/save CLTuner values");
 #endif /* ARM_COMPUTE_CL */
     auto threads = parser.add_option<utils::SimpleOption<int>>("threads", 1);
     threads->set_help("Number of threads to use");
@@ -127,6 +140,16 @@ int main(int argc, char **argv)
         if(enable_tuner->is_set())
         {
             cl_tuner.set_tune_new_kernels(enable_tuner->value());
+            // If that's the first run then the file won't exist yet
+            if(file_exists(tuner_file->value()))
+            {
+                cl_tuner.load_from_file(tuner_file->value());
+            }
+        }
+        else if(!tuner_file->value().empty())
+        {
+            //If we're not tuning and the file doesn't exist then we should raise an error:
+            cl_tuner.load_from_file(tuner_file->value());
         }
 #endif /* ARM_COMPUTE_CL */
         if(options.log_level->value() > framework::LogLevel::NONE)
@@ -216,6 +239,10 @@ int main(int argc, char **argv)
 
 #ifdef ARM_COMPUTE_CL
         CLScheduler::get().sync();
+        if(enable_tuner->is_set() && enable_tuner->value() && tuner_file->is_set())
+        {
+            cl_tuner.save_to_file(tuner_file->value());
+        }
 #endif /* ARM_COMPUTE_CL */
 
         return (success ? 0 : 1);
diff --git a/utils/CommonGraphOptions.cpp b/utils/CommonGraphOptions.cpp
index bf882f5164..7ac7bbce2b 100644
--- a/utils/CommonGraphOptions.cpp
+++ b/utils/CommonGraphOptions.cpp
@@ -83,6 +83,7 @@ namespace utils
     os << "Data type : " << common_params.data_type << std::endl;
     os << "Data layout : " << common_params.data_layout << std::endl;
     os << "Tuner enabled? : " << (common_params.enable_tuner ? true_str : false_str) << std::endl;
+    os << "Tuner file : " << common_params.tuner_file << std::endl;
     os << "Fast math enabled? : " << (common_params.fast_math_hint == FastMathHint::ENABLED ? true_str : false_str) << std::endl;
     if(!common_params.data_path.empty())
     {
@@ -122,7 +123,8 @@ CommonGraphOptions::CommonGraphOptions(CommandLineParser &parser)
       labels(parser.add_option<SimpleOption<std::string>>("labels")),
       validation_file(parser.add_option<SimpleOption<std::string>>("validation-file")),
       validation_path(parser.add_option<SimpleOption<std::string>>("validation-path")),
-      validation_range(parser.add_option<SimpleOption<std::string>>("validation-range"))
+      validation_range(parser.add_option<SimpleOption<std::string>>("validation-range")),
+      tuner_file(parser.add_option<SimpleOption<std::string>>("tuner-file"))
 {
     std::set<arm_compute::graph::Target> supported_targets
     {
@@ -161,6 +163,7 @@ CommonGraphOptions::CommonGraphOptions(CommandLineParser &parser)
     validation_file->set_help("File used to validate the graph");
     validation_path->set_help("Path to the validation data");
     validation_range->set_help("Range of the images to validate for (Format : start,end)");
+    tuner_file->set_help("File to load/save CLTuner values");
 }
 
 CommonGraphParams consume_common_graph_parameters(CommonGraphOptions &options)
@@ -183,6 +186,7 @@ CommonGraphParams consume_common_graph_parameters(CommonGraphOptions &options)
     common_params.validation_path        = options.validation_path->value();
     common_params.validation_range_start = validation_range.first;
     common_params.validation_range_end   = validation_range.second;
+    common_params.tuner_file             = options.tuner_file->value();
 
     return common_params;
 }
diff --git a/utils/CommonGraphOptions.h b/utils/CommonGraphOptions.h
index b429a18069..24c8dbb67c 100644
--- a/utils/CommonGraphOptions.h
+++ b/utils/CommonGraphOptions.h
@@ -49,6 +49,7 @@ struct CommonGraphParams
     std::string                      labels{};
     std::string                      validation_file{};
     std::string                      validation_path{};
+    std::string                      tuner_file{};
     unsigned int                     validation_range_start{ 0 };
     unsigned int                     validation_range_end{ std::numeric_limits<unsigned int>::max() };
 };
@@ -103,6 +104,7 @@ public:
     SimpleOption<std::string>              *validation_file;  /**< Validation file */
     SimpleOption<std::string>              *validation_path;  /**< Validation data path */
     SimpleOption<std::string>              *validation_range; /**< Validation range */
+    SimpleOption<std::string>              *tuner_file;       /**< File to load/store the tuner's values from */
 };
 
 /** Consumes the common graph options and creates a structure containing any information