use auto in intermediate expr

Author: zrr1999

paddle/phi/kernels/cpu/add_position_encoding_kernel.cc

@@ -76,8 +76,9 @@ void AddPositionEncodingKernel(const Context& dev_ctx,
 
   const int half_size = enc_size / 2;
   for (int i = 0; i < batch_size; ++i) {
-    const int max_length =
-        x_lod.empty() ? max_seq_len : x_lod[0][i + 1] - x_lod[0][i];
+    const auto max_length(x_lod.empty() ? max_seq_len
+                                        : x_lod[0][i + 1] - x_lod[0][i]);
+
     for (int j = 0; j < max_length; ++j) {
       for (int k = 0; k < half_size; ++k) {
         const double val =

paddle/phi/kernels/cpu/batch_norm_grad_kernel.cc

@@ -181,7 +181,7 @@ void BatchNormGradFunctor(const Context& dev_ctx,
     bias_arr.setZero();
   }
 
-  int scale_coeff = use_global_stats ? 1 : N * sample_size;
+  auto scale_coeff = use_global_stats ? 1 : N * sample_size;
   const auto scale_inv_var_nhw = scale_arr * inv_var_arr / scale_coeff;
 
   DenseTensor dy_sum;

paddle/phi/kernels/cpu/box_coder_kernel.cc

@@ -120,7 +120,7 @@ void DecodeCenterSize(const DenseTensor *target_box,
       std::array<T, 4> var_data{1., 1., 1., 1.};
       T *var_ptr = var_data.data();
       size_t offset = i * col * len + j * len;
-      int prior_box_offset = axis == 0 ? j * len : i * len;
+      auto prior_box_offset = axis == 0 ? j * len : i * len;
 
       T prior_box_width = prior_box_data[prior_box_offset + 2] -
                           prior_box_data[prior_box_offset] +
@@ -135,7 +135,7 @@ void DecodeCenterSize(const DenseTensor *target_box,
 
       T target_box_center_x = 0, target_box_center_y = 0;
       T target_box_width = 0, target_box_height = 0;
-      int prior_var_offset = axis == 0 ? j * len : i * len;
+      auto prior_var_offset = axis == 0 ? j * len : i * len;
       if (var_size == 2) {
         std::memcpy(var_ptr,
                     prior_box_var->data<T>() + prior_var_offset,

paddle/phi/kernels/cpu/broadcast_tensors_grad_kernel.cc

@@ -112,8 +112,8 @@ void BroadcastTensorsGradKernel(const Context& dev_ctx,
     std::vector<int> reduce_dims_vec;
     std::vector<int> reshape_dims_vec;
     for (int j = 0; j < in_rank; j++) {
-      int out_axis = out_rank - j - 1;
-      int in_axis = in_rank - j - 1;
+      auto out_axis = out_rank - j - 1;
+      auto in_axis = in_rank - j - 1;
 
       reshape_dims_vec.push_back(static_cast<int>(input_dims[j]));
       if (out_axis < 0 || output_dims[out_axis] != input_dims[in_axis]) {

paddle/phi/kernels/cpu/conv_util.h

@@ -77,8 +77,9 @@ inline int ConvOutSize(int input_size,
                        int pad_left,
                        int pad_right,
                        int stride) {
-  const int dkernel = dilation * (filter_size - 1) + 1;
-  int output_size =
+  const auto dkernel(dilation * (filter_size - 1) + 1);
+
+  auto output_size =
       (input_size + (pad_left + pad_right) - dkernel) / stride + 1;
 
   PADDLE_ENFORCE_GT(

paddle/phi/kernels/cpu/cross_entropy_grad_kernel.cc

@@ -95,8 +95,8 @@ void CrossEntropyWithSoftmaxGradCPUKernel(const CPUContext& dev_ctx,
       const int remain = d / axis_dim;
       for (int i = 0; i < n; ++i) {         // for each sample_1_dim
         for (int j = 0; j < remain; j++) {  // for each sample_other_dims
-          int idx = i * remain + j;  // this sample's label_idx. for 1d case,
-                                     // remain=1 and j=0, so, idx = i
+          auto idx = i * remain + j;  // this sample's label_idx. for 1d case,
+                                      // remain=1 and j=0, so, idx = i
           auto lbl = static_cast<int64_t>(label_data[idx]);  // NOLINT
           if (lbl == ignore_index) {
             for (int k = 0; k < axis_dim; ++k) {  // for each class id's label
@@ -147,8 +147,8 @@ void CrossEntropyWithSoftmaxGradCPUKernel(const CPUContext& dev_ctx,
     const int remain = d / axis_dim;
     for (int i = 0; i < n; ++i) {         // for each sample_1_dim
       for (int j = 0; j < remain; j++) {  // for each sample_other_dims
-        int idx = i * remain + j;  // this sample's label_idx. for 1d case,
-                                   // remain=1 and j=0, so, idx = i
+        auto idx = i * remain + j;  // this sample's label_idx. for 1d case,
+                                    // remain=1 and j=0, so, idx = i
         auto lbl = static_cast<int64_t>(label_data[idx]);  // NOLINT
         if (lbl == ignore_index) {
           for (int k = 0; k < axis_dim; ++k) {  // for each class id's label

paddle/phi/kernels/cpu/distribute_fpn_proposals_kernel.cc

@@ -33,7 +33,7 @@ void DistributeFpnProposalsKernel(
     std::vector<DenseTensor*> multi_fpn_rois,
     std::vector<DenseTensor*> multi_level_rois_num,
     DenseTensor* restore_index) {
-  const int num_level = max_level - min_level + 1;
+  const auto num_level(max_level - min_level + 1);
 
   // check that the fpn_rois is not empty
   if (!rois_num.get_ptr()) {

paddle/phi/kernels/cpu/lookup_table_dequant_kernel.cc

@@ -82,7 +82,7 @@ void LookupTableDequantKernel(const Context &dev_ctx,
               ids[i]));
       float min = *(table + ids[i] * quant_number);
       float max = *(table + ids[i] * quant_number + 1);
-      int offset = ids[i] * quant_number + 2;
+      auto offset = ids[i] * quant_number + 2;
       const unsigned char *tensor_buf =
           reinterpret_cast<const unsigned char *>(table + offset);
       dequant(

paddle/phi/kernels/cpu/lrn_kernel.cc

@@ -91,7 +91,7 @@ struct LRNFunctor<phi::CPUContext, T> {
       }
       for (int c = 1; c < C; ++c) {
         // copy previous scale
-        int mid_offset = i * fea_size + c * img_size;
+        auto mid_offset = i * fea_size + c * img_size;
         std::memcpy(mdata + mid_offset,
                     mdata + mid_offset - img_size,
                     img_size * sizeof(T));

paddle/phi/kernels/cpu/matrix_rank_tol_kernel.cc

@@ -42,7 +42,7 @@ void LapackSVD(const T* x_data,
   int mn = std::min(rows, cols);
   T* a = const_cast<T*>(x_data);  // NOLINT
   int lda = rows;
-  int lwork = 3 * mn + std::max(mx, 7 * mn);
+  auto lwork = 3 * mn + std::max(mx, 7 * mn);
   std::vector<phi::dtype::Real<T>> rwork(
       std::max(5 * mn * mn + 5 * mn, 2 * mx * mn + 2 * mn * mn + mn));
   std::vector<T> work(lwork);