Fix Windows Build Warnings

This PR resolves Windows build warnings.

- Type Conversion Warnings (C4244, C4267, C4146, C4293)
- MSVC Compiler Flag Warnings (D9024, D9025, D9027)
- Hexadecimal Floating-Point Warnings (C4305, C4309)

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

Signed-off-by: Donghyeon Jeong <[email protected]>

Author: djeong20

meson.build

@@ -131,7 +131,7 @@ elif arch == 'x86' or arch == 'x86_64'
   message('Build for X86 architecture')
   if cxx_compiler_id == 'msvc'
     add_project_arguments(['/arch:AVX2'], language: ['c','cpp'])
-   add_project_arguments(['cpp_args=/MT'], language: ['c','cpp'] )
+    # Note: C runtime library (/MT or /MD) is controlled by b_vscrt option
   else
     add_project_arguments(['-march=native'], language: ['c','cpp'])
     add_project_arguments(['-mavx2', '-mfma'], language: ['c','cpp'])

nntrainer/dataset/databuffer.cpp

@@ -197,7 +197,7 @@ void DataBuffer::displayProgress(const int count, float loss) {
     else
       progress = (((float)(count * batch_size)) / (float)samples_per_epoch);
 
-    int pos = barWidth * progress;
+    int pos = (int)(barWidth * progress);
     std::cout << " [ ";
     for (int l = 0; l < barWidth; ++l) {
       if (l <= pos)

nntrainer/dataset/dir_data_producers.cpp

@@ -118,7 +118,7 @@ DirDataProducer::finalize(const std::vector<TensorDim> &input_dims,
       std::string p = std::filesystem::absolute(entry.path()).string();
       if (p.compare(".") && p.compare("..")) {
         num_data++;
-        data_list.push_back(std::make_pair(id, p));
+        data_list.push_back(std::make_pair((unsigned int)id, p));
       }
       itr++;
     }
@@ -152,7 +152,7 @@ DirDataProducer::finalize(const std::vector<TensorDim> &input_dims,
 
     unsigned int c_id = data_list[idx].first;
 
-    std::memset(labels[0].getData(), 0.0, num_class * sizeof(float));
+    std::memset(labels[0].getData(), 0, num_class * sizeof(float));
 
     labels[0].getData()[c_id] = 1.0;
 

nntrainer/dataset/raw_file_data_producer.cpp

@@ -50,12 +50,12 @@ RawFileDataProducer::finalize(const std::vector<TensorDim> &input_dims,
   auto sz = size(input_dims, label_dims);
   auto path_prop = std::get<props::FilePath>(*raw_file_props);
 
-  auto size_accumulator = [](const unsigned int &a, const TensorDim &b) {
+  auto size_accumulator = [](const size_t &a, const TensorDim &b) {
     return a + b.getFeatureLen();
   };
 
-  auto sample_size =
-    std::accumulate(input_dims.begin(), input_dims.end(), 0u, size_accumulator);
+  auto sample_size = std::accumulate(input_dims.begin(), input_dims.end(),
+                                     (size_t)0, size_accumulator);
   sample_size = std::accumulate(label_dims.begin(), label_dims.end(),
                                 sample_size, size_accumulator);
 
@@ -85,12 +85,12 @@ RawFileDataProducer::finalize(const std::vector<TensorDim> &input_dims,
 unsigned int
 RawFileDataProducer::size(const std::vector<TensorDim> &input_dims,
                           const std::vector<TensorDim> &label_dims) const {
-  auto size_accumulator = [](const unsigned int &a, const TensorDim &b) {
+  auto size_accumulator = [](const size_t &a, const TensorDim &b) {
     return a + b.getFeatureLen();
   };
 
-  auto sample_size =
-    std::accumulate(input_dims.begin(), input_dims.end(), 0u, size_accumulator);
+  auto sample_size = std::accumulate(input_dims.begin(), input_dims.end(),
+                                     (size_t)0, size_accumulator);
   sample_size = std::accumulate(label_dims.begin(), label_dims.end(),
                                 sample_size, size_accumulator);
   NNTR_THROW_IF(sample_size == 0, std::invalid_argument)

nntrainer/layers/conv2d_layer.cpp

@@ -95,8 +95,8 @@ static void col2im(const Tensor &col_matrix, const TensorDim &kdim,
    */
   auto apply_data = [&]<typename T>(T *val) {
     unsigned col_w = 0;
-    for (int hs = -pt; hs <= h_stride_end; hs += hstride) {
-      for (int ws = -pl; ws <= w_stride_end; ws += wstride) {
+    for (int hs = -(int)pt; hs <= h_stride_end; hs += hstride) {
+      for (int ws = -(int)pl; ws <= w_stride_end; ws += wstride) {
         unsigned col_h = 0;
         int patch_height_end = hs + eff_k_height;
         int patch_width_end = ws + eff_k_width;
@@ -238,7 +238,7 @@ static void im2col(const Tensor &in, const TensorDim &kdim,
     /// hs is height_strided, ws is width_strided
     unsigned int owidth = out.width();
     unsigned int base_im_w = 0;
-    for (int hs = -pt; hs <= h_stride_end; hs += mstride[0]) {
+    for (int hs = -(int)pt; hs <= h_stride_end; hs += mstride[0]) {
       unsigned int base_im_h = 0;
       int patch_height_end = eff_k_height + hs;
       /// map the patch to a single line looping through channel
@@ -252,7 +252,7 @@ static void im2col(const Tensor &in, const TensorDim &kdim,
           }
 
           unsigned int im_w = base_im_w;
-          for (int ws = -pl; ws <= w_stride_end; ws += mstride[1]) {
+          for (int ws = -(int)pl; ws <= w_stride_end; ws += mstride[1]) {
             unsigned int im_h = base_im_h;
             int patch_width_end = eff_k_width + ws;
 
@@ -610,7 +610,7 @@ void Conv2DLayer::calcGradient(RunLayerContext &context) {
        * for the whole batch. try this while benchmarking.
        */
       im2col(in_sub, filter_dim, padding, stride, dilation, result);
-      deriv_sub.dot(result, delK, false, false, b == 0 ? 0 : 1);
+      deriv_sub.dot(result, delK, false, false, b == 0 ? 0.0f : 1.0f);
     }
     result.deallocate();
   }

nntrainer/layers/conv2d_transpose_layer.cpp

@@ -527,7 +527,7 @@ void Conv2DTransposeLayer::calcGradient(RunLayerContext &context) {
        * for the whole batch. try this while benchmarking.
        */
       im2col_transpose(in_sub, filter_dim, padding, stride, dilation, result);
-      deriv_sub.dot(result, delK, false, false, b == 0 ? 0 : 1);
+      deriv_sub.dot(result, delK, false, false, b == 0 ? 0.0f : 1.0f);
     }
     result.deallocate();
   }

nntrainer/layers/embedding.cpp

@@ -187,7 +187,7 @@ void EmbeddingLayer::calcGradient(RunLayerContext &context) {
 
     if (djdw.getDataType() == TensorDim::DataType::FP32) {
       for (unsigned int i = 0; i < input_.width(); ++i) {
-        unsigned int embed_idx = ((float *)(in_data))[i];
+        unsigned int embed_idx = (unsigned int)((float *)(in_data))[i];
         // Assume padding is 0 and index always start from 1.
         // If in_data[i] - 1 < 0, then it skips.
         // if (embed_idx == 0)
@@ -203,7 +203,7 @@ void EmbeddingLayer::calcGradient(RunLayerContext &context) {
     } else if (djdw.getDataType() == TensorDim::DataType::FP16) {
 #ifdef ENABLE_FP16
       for (unsigned int i = 0; i < input_.width(); ++i) {
-        unsigned int embed_idx = ((float *)(in_data))[i];
+        unsigned int embed_idx = (unsigned int)((float *)(in_data))[i];
         // Assume padding is 0 and index always start from 1.
         // If in_data[i] - 1 < 0, then it skips.
         // if (embed_idx == 0)

nntrainer/layers/loss/cross_entropy_sigmoid_loss_layer.cpp

@@ -38,9 +38,9 @@ void CrossEntropySigmoidLossLayer::forwarding(RunLayerContext &context,
     // @note: the output should be logit before applying sigmoid
     // log(1 + exp(-abs(y))) + max(y, 0)
     Tensor mid_term = y.apply<float>(static_cast<float (*)(float)>(&std::fabs))
-                        .multiply(-1.0)
+                        .multiply(-1.0f)
                         .apply<float>(static_cast<float (*)(float)>(&std::exp))
-                        .add(1.0)
+                        .add(1.0f)
                         .apply<float>(logFloat<float>);
     mid_term = mid_term.add(y.apply<float>(ActiFunc::relu<float>));
 
@@ -62,7 +62,8 @@ void CrossEntropySigmoidLossLayer::calcDerivative(RunLayerContext &context) {
 
   y.apply<float>(ActiFunc::sigmoid<float>, ret_derivative);
   ret_derivative.subtract_i(y2);
-  if (ret_derivative.divide_i(ret_derivative.size()) != ML_ERROR_NONE) {
+  if (ret_derivative.divide_i(static_cast<float>(ret_derivative.size())) !=
+      ML_ERROR_NONE) {
     throw std::runtime_error("[CrossEntropySigmoidLossLayer::calcDerivative] "
                              "Error when calculating loss");
   }

nntrainer/layers/loss/cross_entropy_softmax_loss_layer.cpp

@@ -38,7 +38,7 @@ void CrossEntropySoftmaxLossLayer::forwarding(RunLayerContext &context,
       Tensor &y2 = context.getLabel(SINGLE_INOUT_IDX);
       l = y2.multiply(hidden_.apply<float>(logFloat<float>))
             .sum_by_batch()
-            .multiply(-1);
+            .multiply(-1.0f);
 
       // update the loss value
       LossLayer::updateLoss(context, l);
@@ -84,7 +84,8 @@ void CrossEntropySoftmaxLossLayer::calcDerivative(RunLayerContext &context) {
   // TODO: verify y and ret_derivative must not be same as loss layer is not
   // working in-place
   ret.subtract(y2, ret_derivative);
-  if (ret_derivative.divide_i(ret.batch()) != ML_ERROR_NONE) {
+  if (ret_derivative.divide_i(static_cast<float>(ret.batch())) !=
+      ML_ERROR_NONE) {
     throw std::runtime_error("[CrossEntropySoftmaxLossLayer::calcDerivative] "
                              "Error when calculating loss");
   }

nntrainer/layers/mol_attention_layer.cpp

@@ -220,12 +220,12 @@ void MoLAttentionLayer::forwarding(RunLayerContext &context, bool training) {
   for (unsigned int b = 0; b < batch; b++) {
     for (unsigned int h = 0; h < u_base.height(); h++) {
       float *u_data = u_base.getAddress<float>(b, 0, h, 0);
-      std::fill(u_data, u_data + u_base.width(), h + 1);
+      std::fill(u_data, u_data + u_base.width(), static_cast<float>(h + 1));
     }
   }
 
-  Tensor u_pos = u_base.add(0.5);
-  u_base.add_i(-0.5);
+  Tensor u_pos = u_base.add(0.5f);
+  u_base.add_i(-0.5f);
   Tensor u_neg = u_base;
 
   Tensor beta_eps = beta.add(1e-8f);