[concat] Move validity checks to init

Move the dimension checks for validity of the various inputs
for concat layer are moved to initialize().
They are kept back in forwarding() under DEBUG conditional.

**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped

Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>

diff --git a/nntrainer/include/tensor_dim.h b/nntrainer/include/tensor_dim.h
index 42e0429..0ff0e62 100644 (file)
--- a/nntrainer/include/tensor_dim.h
+++ b/nntrainer/include/tensor_dim.h
@@ -99,7 +99,8 @@ public:
   bool isEmpty() const { return len == 0; }
   unsigned int rank() const;
 
-  unsigned int &operator[](unsigned int index);
+  unsigned int &operator[](const unsigned int index);
+  const unsigned int &operator[](const unsigned int index) const;
 
   /**
    * @brief Calculate standard strides

diff --git a/nntrainer/src/concat_layer.cpp b/nntrainer/src/concat_layer.cpp
index 1181d2f..2f3cbad 100644 (file)
--- a/nntrainer/src/concat_layer.cpp
+++ b/nntrainer/src/concat_layer.cpp
@@ -24,12 +24,22 @@ namespace nntrainer {
 int ConcatLayer::initialize() {
   int status = ML_ERROR_NONE;
   unsigned int channel = 0;
+
   if (num_inputs == 0) {
     ml_loge("Error: number of inputs are not initialized");
     return ML_ERROR_INVALID_PARAMETER;
   }
 
-  for (unsigned int idx = 0; idx < num_inputs; ++idx) {
+  const TensorDim &d = input_dim[0];
+  channel += d.channel();
+  for (unsigned int idx = 1; idx < num_inputs; ++idx) {
+    const TensorDim &dim = input_dim[idx];
+
+    for (unsigned int i = 2; i < d.rank(); ++i) {
+      if (d[i] != dim[i])
+        throw std::runtime_error("Error: concat layer requires same "
+                                 "shape from  all input layers");
+    }
     channel += input_dim[idx].channel();
   }
 
@@ -40,35 +50,30 @@ int ConcatLayer::initialize() {
 }
 
 sharedConstTensors ConcatLayer::forwarding(sharedConstTensors in) {
+  hidden = Tensor(output_dim[0]);
 
-  TensorDim d = in[0]->getDim();
-  unsigned int concat_channel = d.channel();
-
+#ifdef DEBUG
+  const TensorDim &d = in[0]->getDim();
+  channel += d.channel();
   for (unsigned int idx = 1; idx < num_inputs; ++idx) {
-    TensorDim dim = in[idx]->getDim();
+    const TensorDim &dim = in[idx]->getDim();
 
-#ifdef DEBUG
     for (unsigned int i = 2; i < d.rank(); ++i) {
       if (d[i] != dim[i])
         throw std::runtime_error("Error: concat layer requires same "
                                  "shape from  all input layers");
     }
-#endif
-    concat_channel += dim.channel();
+    channel += input_dim[idx].channel();
   }
 
-#ifdef DEBUG
-  if (concat_channel != output_dim[0].channel()) {
+  if (channel != output_dim[0].channel())
     throw std::runtime_error(
       "Error: Sum of channel of input layers is not same with output channel");
-  }
 #endif
 
-  hidden = Tensor(output_dim[0]);
-
-  unsigned int f_size = d.width() * d.height() * (concat_channel);
+  unsigned int f_size = output_dim[0].getFeatureLen();
 
-  for (unsigned int b = 0; b < d.batch(); ++b) {
+  for (unsigned int b = 0; b < input_dim[0].batch(); ++b) {
     unsigned int position = 0;
     for (unsigned int idx = 0; idx < num_inputs; ++idx) {
       TensorDim in_dim = in[idx]->getDim();

diff --git a/nntrainer/src/tensor_dim.cpp b/nntrainer/src/tensor_dim.cpp
index 236e9b8..a815731 100644 (file)
--- a/nntrainer/src/tensor_dim.cpp
+++ b/nntrainer/src/tensor_dim.cpp
@@ -114,7 +114,14 @@ unsigned int TensorDim::rank() const {
   return rank;
 }
 
-unsigned int &TensorDim::operator[](unsigned int index) {
+unsigned int &TensorDim::operator[](const unsigned int index) {
+  if (index >= MAXDIM)
+    throw std::out_of_range(
+      "[TensorDim] Tensor Dimension index should be between 0 and 4");
+  return dim[index];
+}
+
+const unsigned int &TensorDim::operator[](const unsigned int index) const {
   if (index >= MAXDIM)
     throw std::out_of_range(
       "[TensorDim] Tensor Dimension index should be between 0 and 4");