Merge branch 'main' into check

.github/workflows/cpp_linter.yml

@@ -12,16 +12,17 @@ jobs:
           echo "Install mandatory dev packages to avoid false-positive reports from cpp-linter"
           sudo apt-get update
           sudo apt-get install libstdc++-*-dev
-      - uses: cpp-linter/cpp-linter-action@v2.8.0
+      - uses: cpp-linter/cpp-linter-action@v2.9.0
         id: linter
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:
-          style: ''
-          version: 16
+          style: file
+          version: 14
           lines-changed-only: true
           file-annotations: false
           step-summary: true
+          format-review: true
 
       - name: failing fast
         if: steps.linter.outputs.clang-format-checks-failed > 0

api/ccapi/include/layer.h

@@ -101,6 +101,7 @@ enum LayerType {
   LAYER_LOSS_CONSTANT_DERIVATIVE, /**< Synthetic loss layer to feed constant
                                      derivative */
   LAYER_RMSNORM = ML_TRAIN_LAYER_TYPE_RMSNORM, /**<RMS NORM Layer */
+  LAYER_UPSAMPLE2D,               /**< Upsample 2D Layer type */
   LAYER_UNKNOWN = ML_TRAIN_LAYER_TYPE_UNKNOWN /**< Unknown */
 };
 
@@ -557,6 +558,14 @@ Identity(const std::vector<std::string> &properties = {}) {
   return createLayer(LayerType::LAYER_IDENTITY, properties);
 }
 
+/**
+ * @brief Helper function to create Upsample2d layer
+ */
+inline std::unique_ptr<Layer>
+Upsample2D(const std::vector<std::string> &properties = {}) {
+  return createLayer(LayerType::LAYER_UPSAMPLE2D, properties);
+}
+
 /**
  * @brief Helper function to create activation layer
  */

nntrainer/app_context.cpp

@@ -73,6 +73,7 @@
 #include <rnncell.h>
 #include <split_layer.h>
 #include <time_dist.h>
+#include <upsample2d_layer.h>
 #include <zoneout_lstmcell.h>
 
 #ifdef ENABLE_TFLITE_BACKBONE
@@ -154,7 +155,7 @@ std::vector<std::string> getPluginPaths() {
    * where you would like to look for the layers, while NNTRAINER_CONF_PATH is a
    * (buildtime hardcoded @a file path) to locate configuration file *.ini file
    */
-  /*** @note for now, NNTRAINER_PATH is a SINGLE PATH rather than serise of path
+  /*** @note for now, NNTRAINER_PATH is a SINGLE PATH rather than series of path
    * like PATH environment variable. this could be improved but for now, it is
    * enough
    */
@@ -211,7 +212,7 @@ const std::string getFullPath(const std::string &path,
 std::mutex factory_mutex;
 
 /**
- * @brief finialize global context
+ * @brief finalize global context
  *
  */
 static void fini_global_context_nntrainer(void) __attribute__((destructor));
@@ -221,7 +222,7 @@ static void fini_global_context_nntrainer(void) {}
 std::once_flag global_app_context_init_flag;
 
 static void add_default_object(AppContext &ac) {
-  /// @note all layers should be added to the app_context to gaurantee that
+  /// @note all layers should be added to the app_context to guarantee that
   /// createLayer/createOptimizer class is created
   using OptType = ml::train::OptimizerType;
   ac.registerFactory(nntrainer::createOptimizer<SGD>, SGD::type, OptType::SGD);
@@ -306,6 +307,8 @@ static void add_default_object(AppContext &ac) {
                      LayerType::LAYER_POSITIONAL_ENCODING);
   ac.registerFactory(nntrainer::createLayer<IdentityLayer>, IdentityLayer::type,
                      LayerType::LAYER_IDENTITY);
+  ac.registerFactory(nntrainer::createLayer<Upsample2dLayer>,
+                     Upsample2dLayer::type, LayerType::LAYER_UPSAMPLE2D);
 
 #ifdef ENABLE_NNSTREAMER_BACKBONE
   ac.registerFactory(nntrainer::createLayer<NNStreamerLayer>,
@@ -319,7 +322,7 @@ static void add_default_object(AppContext &ac) {
   ac.registerFactory(nntrainer::createLayer<CentroidKNN>, CentroidKNN::type,
                      LayerType::LAYER_CENTROID_KNN);
 
-  /** proprocess layers */
+  /** preprocess layers */
   ac.registerFactory(nntrainer::createLayer<PreprocessFlipLayer>,
                      PreprocessFlipLayer::type,
                      LayerType::LAYER_PREPROCESS_FLIP);

nntrainer/dataset/data_iteration.h

@@ -141,7 +141,7 @@ class Sample {
    * @brief Construct a new Sample object
    * @note the batch dimension will be ignored to make a single sample
    *
-   * @param iter iteration obejcts
+   * @param iter iteration objects
    * @param batch nth batch to create the sample
    */
   Sample(const Iteration &iter, unsigned int batch);

nntrainer/dataset/data_producer.h

@@ -109,18 +109,18 @@ class DataProducer {
   }
 
   /**
-   * @brief this function helps exporting the dataproducer in a predefined
+   * @brief this function helps exporting the data producer in a predefined
    * format, while workarounding issue caused by templated function type eraser
    *
-   * @param     exporter exporter that conatins exporting logic
+   * @param     exporter exporter that contains exporting logic
    * @param     method enum value to identify how it should be exported to
    */
   virtual void exportTo(Exporter &exporter,
                         const ml::train::ExportMethods &method) const {}
 
   /**
    * @brief denote if given producer is thread safe and can be parallelized.
-   * @note if size() == SIZE_UNDEFIEND, thread safe shall be false
+   * @note if size() == SIZE_UNDEFINED, thread safe shall be false
    *
    * @return bool true if thread safe.
    */

nntrainer/dataset/databuffer.h

@@ -139,7 +139,7 @@ class DataBuffer : public ml::train::Dataset {
    * @brief this function helps exporting the dataset in a predefined format,
    * while workarounding issue caused by templated function type eraser
    *
-   * @param     exporter exporter that conatins exporting logic
+   * @param     exporter exporter that contains exporting logic
    * @param     method enum value to identify how it should be exported to
    */
   void exportTo(Exporter &exporter,

nntrainer/dataset/dir_data_producers.cpp

@@ -72,9 +72,7 @@ static void readImage(const std::string path, float *input, uint width,
 namespace nntrainer {
 
 DirDataProducer::DirDataProducer() :
-  dir_data_props(new Props()),
-  num_class(0),
-  num_data_total(0) {}
+  dir_data_props(new Props()), num_class(0), num_data_total(0) {}
 
 DirDataProducer::DirDataProducer(const std::string &dir_path) :
   dir_data_props(new Props(props::DirPath(dir_path))),
@@ -140,7 +138,7 @@ DirDataProducer::finalize(const std::vector<TensorDim> &input_dims,
   auto sz = size(input_dims, label_dims);
 
   NNTR_THROW_IF(sz == 0, std::invalid_argument)
-    << "size is zero, dataproducer does not provide anything";
+    << "size is zero, data producer does not provide anything";
 
   return [sz, input_dims, this](unsigned int idx, std::vector<Tensor> &inputs,
                                 std::vector<Tensor> &labels) {

nntrainer/dataset/random_data_producers.cpp

@@ -113,7 +113,7 @@ RandomDataOneHotProducer::finalize(const std::vector<TensorDim> &input_dims,
 
   /// @todo move this to higher order component
   NNTR_THROW_IF(size(input_dims, label_dims) == 0, std::invalid_argument)
-    << "size is zero, dataproducer does not provide anything";
+    << "size is zero, data producer does not provide anything";
 
   /** prepare states for the generator */
   std::vector<std::uniform_int_distribution<unsigned int>> label_chooser_;

nntrainer/layers/cl_layers/swiglu_cl.h

@@ -69,7 +69,7 @@ class SwiGLULayerCl final : public Layer {
   /**
    * @copydoc bool supportBackwarding() const
    */
-  bool supportBackwarding() const override { return true; };
+  bool supportBackwarding() const override { return false; };
 
   /**
    * @copydoc Layer::exportTo(Exporter &exporter, ExportMethods method)

nntrainer/layers/common_properties.h

@@ -869,10 +869,9 @@ struct ActivationTypeInfo {
     Enum::ACT_GELU,    Enum::ACT_QUICK_GELU, Enum::ACT_NONE,
     Enum::ACT_UNKNOWN};
 
-  static constexpr const char *EnumStr[] = {"tanh",    "sigmoid",    "relu",
-                                            "softmax", "leaky_relu", "swish",
-                                            "gelu",    "quick_gelu", "none",
-                                            "unknown"};
+  static constexpr const char *EnumStr[] = {
+    "tanh",  "sigmoid", "relu",       "softmax", "leaky_relu",
+    "swish", "gelu",    "quick_gelu", "none",    "unknown"};
 };
 
 /**
@@ -1068,6 +1067,34 @@ class WeightRegularizer final : public BasicRegularizer {
   static constexpr const char *key = "weight_regularizer";
 };
 
+/**
+ * @brief     Enumeration of upsample type
+ * @todo Support torch and keras supported modes like bicubic
+ */
+struct UpsampleModeInfo {
+  /**
+   * @brief   Upsampling operation type class
+   */
+  enum class Interpolation { nearest, bilinear };
+
+  using Enum = Interpolation;
+
+  static constexpr std::initializer_list<Interpolation> EnumList = {
+    Interpolation::nearest, Interpolation::bilinear};
+
+  static constexpr const char *EnumStr[] = {"nearest", "bilinear"};
+};
+
+/**
+ * @brief Upsample Type Enumeration Information
+ *
+ */
+class UpsampleMode final : public EnumProperty<UpsampleModeInfo> {
+public:
+  using prop_tag = enum_class_prop_tag;
+  static constexpr const char *key = "upsample";
+};
+
 /**
  * @brief     Enumeration of pooling type
  */

nntrainer/layers/layer_context.cpp

@@ -703,6 +703,10 @@ std::string RunLayerContext::getKernelName(LayerKernel layerKernel) {
     return "rmsnorm_cl";
   case LayerKernel::RMSNORM_FP16:
     return "rmsnorm_cl_fp16";
+  case LayerKernel::SSCAL:
+    return "sscal_cl";
+  case LayerKernel::SSCAL_FP16:
+    return "sscal_cl_fp16";
   default:
     return "";
   }

nntrainer/layers/layer_context.h

@@ -840,8 +840,10 @@ class RunLayerContext {
     ADD_FP16 = 1 << 7,   /**< placeholder for kernel name */
     SWIGLU = 1 << 8,     /**< placeholder for kernel name */
     SWIGLU_FP16 = 1 << 9, /**< placeholder for kernel name */
-    RMSNORM = 1 << 10,
-    RMSNORM_FP16 = 1 << 11
+    SSCAL = 1 << 10,      /**< placeholder for kernel name */
+    SSCAL_FP16 = 1 << 11, /**< placeholder for kernel name */
+    RMSNORM = 1 << 12,
+    RMSNORM_FP16 = 1 << 13
   };
 
   /**

nntrainer/layers/meson.build

@@ -42,7 +42,8 @@ layer_sources = [
   'reshape_layer.cpp',
   'reduce_mean_layer.cpp',
   'positional_encoding_layer.cpp',
-  'identity_layer.cpp'
+  'identity_layer.cpp',
+  'upsample2d_layer.cpp'
 ]
 
 layer_headers = [

nntrainer/layers/preprocess_l2norm_layer.h

@@ -29,7 +29,7 @@ class PreprocessL2NormLayer : public Layer {
 public:
   /**
    * @brief Construct a new L2norm Layer object
-   * that normlizes given feature with l2norm
+   * that normalizes given feature with l2norm
    */
   PreprocessL2NormLayer() : Layer() {}
 

nntrainer/layers/split_layer.cpp

@@ -44,7 +44,7 @@ void SplitLayer::finalize(InitLayerContext &context) {
 
   /**
    * The split is only done along the split_dimension dimension.
-   * (Assumes input data is continous)
+   * (Assumes input data is continuous)
    * For example, consider input dimension [b,c,h,w], split_number = n
    * 1. axis = 1, output_dim = [b,c//n,h,w], num_outputs = n
    * 2. axis = 2, output_dim = [b,c,h//n,w], num_outputs = n
@@ -75,7 +75,7 @@ void SplitLayer::finalize(InitLayerContext &context) {
    * to facilitate easier processing.
    *
    * The helper shape consolidates all the dimensions before the split_dimension
-   * together and all the dimensions after the split_dimension to faciliate
+   * together and all the dimensions after the split_dimension to facilitate
    * easier splitting of the data.
    */
   leading_helper_dim = 1;