[Windows] Fix compiling error on develop branch (#4383)

lite/backends/x86/math/context_project.h

@@ -161,7 +161,7 @@ class ContextProjectFunctor {
                       sequence_width});
 
         if (up_pad > 0) {  // add up pad
-          int padding_rows = std::min(
+          int padding_rows = (std::min)(
               up_pad, static_cast<int>(lod_level_0[i + 1] - lod_level_0[i]));
 
           for (int k = 0; k < padding_rows; ++k) {
@@ -180,10 +180,10 @@ class ContextProjectFunctor {
         }
         if (down_pad > 0) {  // add down pad
           int down_pad_begin_row =
-              std::max(0,
-                       (sequence_height - context_start - context_length) + 1) +
+              (std::max)(
+                  0, (sequence_height - context_start - context_length) + 1) +
               1;
-          int padding_begin = std::max(0, context_start - sequence_height);
+          int padding_begin = (std::max)(0, context_start - sequence_height);
           int padding_size =
               sequence_height - context_start >= context_length
                   ? 1

lite/backends/x86/math/pooling.cc

@@ -67,8 +67,8 @@ class Pool2dFunctor<lite::TargetType::kX86, PoolProcess, T> {
             hend = AdaptEndIndex(ph, input_height, output_height);
           } else {
             hstart = ph * stride_height - padding_height;
-            hend = std::min(hstart + ksize_height, input_height);
-            hstart = std::max(hstart, 0);
+            hend = (std::min)(hstart + ksize_height, input_height);
+            hstart = (std::max)(hstart, 0);
           }
           for (int pw = 0; pw < output_width; ++pw) {
             if (adaptive) {
@@ -76,8 +76,8 @@ class Pool2dFunctor<lite::TargetType::kX86, PoolProcess, T> {
               wend = AdaptEndIndex(pw, input_width, output_width);
             } else {
               wstart = pw * stride_width - padding_width;
-              wend = std::min(wstart + ksize_width, input_width);
-              wstart = std::max(wstart, 0);
+              wend = (std::min)(wstart + ksize_width, input_width);
+              wstart = (std::max)(wstart, 0);
             }
 
             T ele = pool_process.initial();
@@ -150,8 +150,8 @@ class Pool2dGradFunctor<lite::TargetType::kX86, PoolProcess, T> {
             hend = AdaptEndIndex(ph, input_height, output_height);
           } else {
             hstart = ph * stride_height - padding_height;
-            hend = std::min(hstart + ksize_height, input_height);
-            hstart = std::max(hstart, 0);
+            hend = (std::min)(hstart + ksize_height, input_height);
+            hstart = (std::max)(hstart, 0);
           }
           for (int pw = 0; pw < output_width; ++pw) {
             if (adaptive) {
@@ -159,8 +159,8 @@ class Pool2dGradFunctor<lite::TargetType::kX86, PoolProcess, T> {
               wend = AdaptEndIndex(pw, input_width, output_width);
             } else {
               wstart = pw * stride_width - padding_width;
-              wend = std::min(wstart + ksize_width, input_width);
-              wstart = std::max(wstart, 0);
+              wend = (std::min)(wstart + ksize_width, input_width);
+              wstart = (std::max)(wstart, 0);
             }
             int pool_size = (exclusive || adaptive)
                                 ? (hend - hstart) * (wend - wstart)
@@ -228,12 +228,12 @@ class MaxPool2dGradFunctor<lite::TargetType::kX86, T> {
       for (int c = 0; c < output_channels; ++c) {
         for (int ph = 0; ph < output_height; ++ph) {
           int hstart = ph * stride_height - padding_height;
-          int hend = std::min(hstart + ksize_height, input_height);
-          hstart = std::max(hstart, 0);
+          int hend = (std::min)(hstart + ksize_height, input_height);
+          hstart = (std::max)(hstart, 0);
           for (int pw = 0; pw < output_width; ++pw) {
             int wstart = pw * stride_width - padding_width;
-            int wend = std::min(wstart + ksize_width, input_width);
-            wstart = std::max(wstart, 0);
+            int wend = (std::min)(wstart + ksize_width, input_width);
+            wstart = (std::max)(wstart, 0);
 
             bool stop = false;
             for (int h = hstart; h < hend && !stop; ++h) {
@@ -337,8 +337,8 @@ class Pool3dFunctor<lite::TargetType::kX86, PoolProcess, T> {
             dend = AdaptEndIndex(pd, input_depth, output_depth);
           } else {
             dstart = pd * stride_depth - padding_depth;
-            dend = std::min(dstart + ksize_depth, input_depth);
-            dstart = std::max(dstart, 0);
+            dend = (std::min)(dstart + ksize_depth, input_depth);
+            dstart = (std::max)(dstart, 0);
           }
           for (int ph = 0; ph < output_height; ++ph) {
             if (adaptive) {
@@ -346,8 +346,8 @@ class Pool3dFunctor<lite::TargetType::kX86, PoolProcess, T> {
               hend = AdaptEndIndex(ph, input_height, output_height);
             } else {
               hstart = ph * stride_height - padding_height;
-              hend = std::min(hstart + ksize_height, input_height);
-              hstart = std::max(hstart, 0);
+              hend = (std::min)(hstart + ksize_height, input_height);
+              hstart = (std::max)(hstart, 0);
             }
             for (int pw = 0; pw < output_width; ++pw) {
               if (adaptive) {
@@ -355,8 +355,8 @@ class Pool3dFunctor<lite::TargetType::kX86, PoolProcess, T> {
                 wend = AdaptEndIndex(pw, input_width, output_width);
               } else {
                 wstart = pw * stride_width - padding_width;
-                wend = std::min(wstart + ksize_width, input_width);
-                wstart = std::max(wstart, 0);
+                wend = (std::min)(wstart + ksize_width, input_width);
+                wstart = (std::max)(wstart, 0);
               }
               int output_idx = (pd * output_height + ph) * output_width + pw;
               T ele = pool_process.initial();
@@ -441,8 +441,8 @@ class Pool3dGradFunctor<lite::TargetType::kX86, PoolProcess, T> {
             dend = AdaptEndIndex(pd, input_depth, output_depth);
           } else {
             dstart = pd * stride_depth - padding_depth;
-            dend = std::min(dstart + ksize_depth, input_depth);
-            dstart = std::max(dstart, 0);
+            dend = (std::min)(dstart + ksize_depth, input_depth);
+            dstart = (std::max)(dstart, 0);
           }
           for (int ph = 0; ph < output_height; ++ph) {
             if (adaptive) {
@@ -450,8 +450,8 @@ class Pool3dGradFunctor<lite::TargetType::kX86, PoolProcess, T> {
               hend = AdaptEndIndex(ph, input_height, output_height);
             } else {
               hstart = ph * stride_height - padding_height;
-              hend = std::min(hstart + ksize_height, input_height);
-              hstart = std::max(hstart, 0);
+              hend = (std::min)(hstart + ksize_height, input_height);
+              hstart = (std::max)(hstart, 0);
             }
             for (int pw = 0; pw < output_width; ++pw) {
               if (adaptive) {
@@ -459,8 +459,8 @@ class Pool3dGradFunctor<lite::TargetType::kX86, PoolProcess, T> {
                 wend = AdaptEndIndex(pw, input_width, output_width);
               } else {
                 wstart = pw * stride_width - padding_width;
-                wend = std::min(wstart + ksize_width, input_width);
-                wstart = std::max(wstart, 0);
+                wend = (std::min)(wstart + ksize_width, input_width);
+                wstart = (std::max)(wstart, 0);
               }
 
               int pool_size =
@@ -540,16 +540,16 @@ class MaxPool3dGradFunctor<lite::TargetType::kX86, T> {
       for (int c = 0; c < output_channels; ++c) {
         for (int pd = 0; pd < output_depth; ++pd) {
           int dstart = pd * stride_depth - padding_depth;
-          int dend = std::min(dstart + ksize_depth, input_depth);
-          dstart = std::max(dstart, 0);
+          int dend = (std::min)(dstart + ksize_depth, input_depth);
+          dstart = (std::max)(dstart, 0);
           for (int ph = 0; ph < output_height; ++ph) {
             int hstart = ph * stride_height - padding_height;
-            int hend = std::min(hstart + ksize_height, input_height);
-            hstart = std::max(hstart, 0);
+            int hend = (std::min)(hstart + ksize_height, input_height);
+            hstart = (std::max)(hstart, 0);
             for (int pw = 0; pw < output_width; ++pw) {
               int wstart = pw * stride_width - padding_width;
-              int wend = std::min(wstart + ksize_width, input_width);
-              wstart = std::max(wstart, 0);
+              int wend = (std::min)(wstart + ksize_width, input_width);
+              wstart = (std::max)(wstart, 0);
               bool stop = false;
               for (int d = dstart; d < dend && !stop; ++d) {
                 for (int h = hstart; h < hend && !stop; ++h) {
@@ -651,8 +651,8 @@ class MaxPool2dWithIndexFunctor<lite::TargetType::kX86, T1, T2> {
             hend = AdaptEndIndex(ph, input_height, output_height);
           } else {
             hstart = ph * stride_height - padding_height;
-            hend = std::min(hstart + ksize_height, input_height);
-            hstart = std::max(hstart, 0);
+            hend = (std::min)(hstart + ksize_height, input_height);
+            hstart = (std::max)(hstart, 0);
           }
           for (int pw = 0; pw < output_width; ++pw) {
             if (adaptive) {
@@ -660,8 +660,8 @@ class MaxPool2dWithIndexFunctor<lite::TargetType::kX86, T1, T2> {
               wend = AdaptEndIndex(pw, input_width, output_width);
             } else {
               wstart = pw * stride_width - padding_width;
-              wend = std::min(wstart + ksize_width, input_width);
-              wstart = std::max(wstart, 0);
+              wend = (std::min)(wstart + ksize_width, input_width);
+              wstart = (std::max)(wstart, 0);
             }
 
             T1 ele = static_cast<T1>(-FLT_MAX);
@@ -794,8 +794,8 @@ class MaxPool3dWithIndexFunctor<lite::TargetType::kX86, T1, T2> {
             dend = AdaptEndIndex(pd, input_depth, output_depth);
           } else {
             dstart = pd * stride_depth - padding_depth;
-            dend = std::min(dstart + ksize_depth, input_depth);
-            dstart = std::max(dstart, 0);
+            dend = (std::min)(dstart + ksize_depth, input_depth);
+            dstart = (std::max)(dstart, 0);
           }
           for (int ph = 0; ph < output_height; ++ph) {
             if (adaptive) {
@@ -803,8 +803,8 @@ class MaxPool3dWithIndexFunctor<lite::TargetType::kX86, T1, T2> {
               hend = AdaptEndIndex(ph, input_height, output_height);
             } else {
               hstart = ph * stride_height - padding_height;
-              hend = std::min(hstart + ksize_height, input_height);
-              hstart = std::max(hstart, 0);
+              hend = (std::min)(hstart + ksize_height, input_height);
+              hstart = (std::max)(hstart, 0);
             }
             for (int pw = 0; pw < output_width; ++pw) {
               if (adaptive) {
@@ -812,8 +812,8 @@ class MaxPool3dWithIndexFunctor<lite::TargetType::kX86, T1, T2> {
                 wend = AdaptEndIndex(pw, input_width, output_width);
               } else {
                 wstart = pw * stride_width - padding_width;
-                wend = std::min(wstart + ksize_width, input_width);
-                wstart = std::max(wstart, 0);
+                wend = (std::min)(wstart + ksize_width, input_width);
+                wstart = (std::max)(wstart, 0);
               }
 
               int output_idx = (pd * output_height + ph) * output_width + pw;

lite/backends/x86/math/sequence_padding.h

@@ -35,7 +35,7 @@ inline static uint64_t MaximumSequenceLength(
   uint64_t seq_num = seq_offset.size() - 1;
   uint64_t max_seq_len = 0;
   for (size_t i = 0; i < seq_num; ++i) {
-    max_seq_len = std::max(max_seq_len, seq_offset[i + 1] - seq_offset[i]);
+    max_seq_len = (std::max)(max_seq_len, seq_offset[i + 1] - seq_offset[i]);
   }
   return max_seq_len;
 }

lite/backends/x86/parallel.h

@@ -26,7 +26,7 @@ namespace x86 {
 
 static void SetNumThreads(int num_threads) {
 #ifdef PADDLE_WITH_MKLML
-  int real_num_threads = std::max(num_threads, 1);
+  int real_num_threads = (std::max)(num_threads, 1);
   x86::MKL_Set_Num_Threads(real_num_threads);
   omp_set_num_threads(real_num_threads);
 #endif
@@ -52,14 +52,14 @@ static inline void RunParallelFor(const int64_t begin,
   }
 
 #ifdef PADDLE_WITH_MKLML
-  int64_t num_threads = std::min(GetMaxThreads(), end - begin);
+  int64_t num_threads = (std::min)(GetMaxThreads(), end - begin);
   if (num_threads > 1) {
 #pragma omp parallel num_threads(num_threads)
     {
       int64_t tid = omp_get_thread_num();
       int64_t chunk_size = (end - begin + num_threads - 1) / num_threads;
       int64_t begin_tid = begin + tid * chunk_size;
-      f(begin_tid, std::min(end, chunk_size + begin_tid));
+      f(begin_tid, (std::min)(end, chunk_size + begin_tid));
     }
     return;
   }

lite/core/mir/memory_optimize_pass.cc

@@ -148,7 +148,7 @@ void MemoryOptimizePass::CollectLifeCycleByDevice(
           int cur_life =
               (*lifecycles)[TargetToStr(target_type)][var_name].second;
           (*lifecycles)[TargetToStr(target_type)][var_name].second =
-              std::max(max_lifecycle_, cur_life);
+              (std::max)(max_lifecycle_, cur_life);
         }
       }
       ++max_lifecycle_;

lite/core/mir/static_kernel_pick_pass.h

@@ -61,7 +61,7 @@ class StaticKernelPickPass : public mir::StmtPass {
     float final_score{-1.};
     Place winner_place{places[0]};
     const int kMax =
-        std::numeric_limits<core::KernelPickFactor::value_type>::max();
+        (std::numeric_limits<core::KernelPickFactor::value_type>::max)();
     size_t place_size = places.size();
 
     // NOTE: We compare kernel's place with place in valid_places to select the

lite/kernels/host/crf_decoding_compute.h

@@ -52,7 +52,7 @@ void Decode(const Tensor& emission_weights,
 
   for (int k = 1; k < seq_len; ++k) {
     for (int i = 0; i < tag_num; ++i) {
-      T max_score = -std::numeric_limits<T>::max();
+      T max_score = -(std::numeric_limits<T>::max)();
       int max_j = 0;
       for (size_t j = 0; j < tag_num; ++j) {
         T score = alpha_value[(k - 1) * tag_num + j] +
@@ -67,7 +67,7 @@ void Decode(const Tensor& emission_weights,
     }
   }
 
-  T max_score = -std::numeric_limits<T>::max();
+  T max_score = -(std::numeric_limits<T>::max)();
   int max_i = 0;
   for (size_t i = 0; i < tag_num; ++i) {
     T score = alpha_value[(seq_len - 1) * tag_num + i] + w[tag_num + i];

lite/kernels/host/multiclass_nms_compute.cc

@@ -72,10 +72,10 @@ static T JaccardOverlap(const T* box1, const T* box2, const bool normalized) {
       box2[3] < box1[1]) {
     return static_cast<T>(0.);
   } else {
-    const T inter_xmin = std::max(box1[0], box2[0]);
-    const T inter_ymin = std::max(box1[1], box2[1]);
-    const T inter_xmax = std::min(box1[2], box2[2]);
-    const T inter_ymax = std::min(box1[3], box2[3]);
+    const T inter_xmin = (std::max)(box1[0], box2[0]);
+    const T inter_ymin = (std::max)(box1[1], box2[1]);
+    const T inter_xmax = (std::min)(box1[2], box2[2]);
+    const T inter_ymax = (std::min)(box1[3], box2[3]);
     T norm = normalized ? static_cast<T>(0.) : static_cast<T>(1.);
     T inter_w = inter_xmax - inter_xmin + norm;
     T inter_h = inter_ymax - inter_ymin + norm;

lite/kernels/host/print_compute.cc

@@ -128,7 +128,7 @@ class TensorFormatter {
   void FormatData(const Tensor& print_tensor, std::stringstream& log_stream) {
     int64_t print_size = summarize_ == -1
                              ? print_tensor.numel()
-                             : std::min(summarize_, print_tensor.numel());
+                             : (std::min)(summarize_, print_tensor.numel());
     const T* data = print_tensor.data<T>();  // Always kHost, so unnessary to
                                              // copy the data from device
     log_stream << "  - data: [";

lite/kernels/host/retinanet_detection_output_compute.cc

@@ -83,10 +83,10 @@ static inline T JaccardOverlap(const std::vector<T>& box1,
       box2[3] < box1[1]) {
     return static_cast<T>(0.);
   } else {
-    const T inter_xmin = std::max(box1[0], box2[0]);
-    const T inter_ymin = std::max(box1[1], box2[1]);
-    const T inter_xmax = std::min(box1[2], box2[2]);
-    const T inter_ymax = std::min(box1[3], box2[3]);
+    const T inter_xmin = (std::max)(box1[0], box2[0]);
+    const T inter_ymin = (std::max)(box1[1], box2[1]);
+    const T inter_xmax = (std::min)(box1[2], box2[2]);
+    const T inter_ymax = (std::min)(box1[3], box2[3]);
     T norm = normalized ? static_cast<T>(0.) : static_cast<T>(1.);
     T inter_w = inter_xmax - inter_xmin + norm;
     T inter_h = inter_ymax - inter_ymin + norm;
@@ -183,10 +183,10 @@ void DeltaScoreToPrediction(
     pred_box_xmax = pred_box_xmax / im_scale;
     pred_box_ymax = pred_box_ymax / im_scale;
 
-    pred_box_xmin = std::max(std::min(pred_box_xmin, im_width - 1), zero);
-    pred_box_ymin = std::max(std::min(pred_box_ymin, im_height - 1), zero);
-    pred_box_xmax = std::max(std::min(pred_box_xmax, im_width - 1), zero);
-    pred_box_ymax = std::max(std::min(pred_box_ymax, im_height - 1), zero);
+    pred_box_xmin = (std::max)((std::min)(pred_box_xmin, im_width - 1), zero);
+    pred_box_ymin = (std::max)((std::min)(pred_box_ymin, im_height - 1), zero);
+    pred_box_xmax = (std::max)((std::min)(pred_box_xmax, im_width - 1), zero);
+    pred_box_ymax = (std::max)((std::min)(pred_box_ymax, im_height - 1), zero);
 
     std::vector<T> one_pred;
     one_pred.push_back(pred_box_xmin);

lite/kernels/x86/elementwise_op_function.h

@@ -71,7 +71,7 @@ inline void get_mid_dims(const lite::DDim &x_dims,
         for (size_t j = 0; j < i; ++j) {
           (*pre) *= y_dims[j];
         }
-        *n = std::max(x_dims[i + axis], y_dims[i]);
+        *n = (std::max)(x_dims[i + axis], y_dims[i]);
         *mid_flag = 1;
         mid = i;
         break;

lite/kernels/x86/sequence_arithmetic_compute.h

@@ -55,7 +55,7 @@ class SequenceArithmeticCompute
         auto input_x = x_data + x_seq_offset[i] * inner_size;
         auto input_y = y_data + y_seq_offset[i] * inner_size;
         auto t_out = out_data + x_seq_offset[i] * inner_size;
-        int len = std::min(len_x, len_y);
+        int len = (std::min)(len_x, len_y);
         for (int j = 0; j < len; j++) {
           t_out[j] = input_x[j] + input_y[j];
         }
@@ -73,7 +73,7 @@ class SequenceArithmeticCompute
         auto input_x = x_data + x_seq_offset[i] * inner_size;
         auto input_y = y_data + y_seq_offset[i] * inner_size;
         auto t_out = out_data + x_seq_offset[i] * inner_size;
-        int len = std::min(len_x, len_y);
+        int len = (std::min)(len_x, len_y);
         for (int j = 0; j < len; j++) {
           t_out[j] = input_x[j] - input_y[j];
         }
@@ -91,7 +91,7 @@ class SequenceArithmeticCompute
         auto input_x = x_data + x_seq_offset[i] * inner_size;
         auto input_y = y_data + y_seq_offset[i] * inner_size;
         auto t_out = out_data + x_seq_offset[i] * inner_size;
-        int len = std::min(len_x, len_y);
+        int len = (std::min)(len_x, len_y);
         for (int j = 0; j < len; j++) {
           t_out[j] = input_x[j] * input_y[j];
         }

lite/kernels/x86/sequence_conv_compute.h

@@ -49,8 +49,8 @@ class SequenceConvCompute : public KernelLite<TARGET(kX86), PRECISION(kFloat)> {
     bool padding_trainable = false;
     const Tensor* padding_data = nullptr;
 
-    int up_pad = std::max(0, -context_start);
-    int down_pad = std::max(0, context_start + context_length - 1);
+    int up_pad = (std::max)(0, -context_start);
+    int down_pad = (std::max)(0, context_start + context_length - 1);
     auto sequence_width = static_cast<int64_t>(in->dims()[1]);
 
     std::vector<int64_t> col_shape{in->dims()[0],

lite/kernels/x86/slice_compute.h

@@ -102,9 +102,9 @@ void slice_compute(const lite::Tensor* in,
 
         start = starts[i] < 0 ? (starts[i] + dim_value) : starts[i];
         end = ends[i] < 0 ? (ends[i] + dim_value) : ends[i];
-        start = std::max(start, 0);
-        end = std::max(end, 0);
-        end = std::min(end, dim_value);
+        start = (std::max)(start, 0);
+        end = (std::max)(end, 0);
+        end = (std::min)(end, dim_value);
         CHECK_GT(end, start) << "end should greater than start";
         out_dims[axes[i]] = end - start;
       }
@@ -172,7 +172,7 @@ void slice_compute(const lite::Tensor* in,
     if (start < 0) {
       start = (start + in_dims[axes[i]]);
     }
-    start = std::max(start, 0);
+    start = (std::max)(start, 0);
     offsets[axes[i]] = start;
   }
   auto in_t =

lite/model_parser/model_parser.cc

@@ -391,7 +391,7 @@ void TensorToStream(std::ostream &os, const lite::Tensor &tensor) {
   }
   {  // the 3rd field, tensor data
     uint64_t size = tensor.memory_size();
-    CHECK_LT(size, std::numeric_limits<std::streamsize>::max())
+    CHECK_LT(size, (std::numeric_limits<std::streamsize>::max)())
         << "Index overflow when writing tensor";
 
 #ifdef LITE_WITH_CUDA
@@ -461,7 +461,7 @@ void SetParamInfoNaive(naive_buffer::ParamDesc *param_desc,
   }
   desc.SetDim(tensor.dims().Vectorize());
   uint64_t size = tensor.memory_size();
-  CHECK_LT(size, std::numeric_limits<std::streamsize>::max())
+  CHECK_LT(size, (std::numeric_limits<std::streamsize>::max)())
       << "Index overflow when writing tensor";
 
 #ifdef LITE_WITH_CUDA

lite/operators/conv_op.cc

@@ -62,7 +62,7 @@ void UpdatePaddingAndDilation(std::vector<int>* paddings,
   if (padding_algorithm == "SAME") {
     for (size_t i = 0; i < strides.size(); ++i) {
       int out_size = (data_dims[i + 2] + strides[i] - 1) / strides[i];
-      int pad_sum = std::max(
+      int pad_sum = (std::max)(
           (out_size - 1) * strides[i] + ksize[i + 2] - data_dims[i + 2],
           (int64_t)0);
       int pad_0 = pad_sum / 2;

lite/operators/elementwise_ops.cc

@@ -75,7 +75,7 @@ bool ElementwiseOp::InferShapeImpl() const {
       if (x_dims_array[i] == -1 || y_dims_array[i] == -1) {
         out_dims_array[i] = -1;
       } else {
-        out_dims_array[i] = std::max(x_dims_array[i], y_dims_array[i]);
+        out_dims_array[i] = (std::max)(x_dims_array[i], y_dims_array[i]);
       }
     }
     param_.Out->Resize(DDim(out_dims_array));

lite/operators/pool_op.h

@@ -128,8 +128,8 @@ inline void UpdatePadding(std::vector<int> *paddings,
     for (size_t i = 0; i < strides.size(); ++i) {
       int out_size = (data_dims[i + 2] + strides[i] - 1) / strides[i];
       int pad_sum =
-          std::max((out_size - 1) * strides[i] + ksize[i] - data_dims[i + 2],
-                   (int64_t)0);
+          (std::max)((out_size - 1) * strides[i] + ksize[i] - data_dims[i + 2],
+                     (int64_t)0);
       int pad_0 = pad_sum / 2;
       int pad_1 = pad_sum - pad_0;
       *(paddings->begin() + i * 2) = pad_0;

lite/operators/slice_op.cc

@@ -51,9 +51,9 @@ bool SliceOp::InferShapeImpl() const {
       if (dim_value > 0) {
         start = starts[i] < 0 ? (starts[i] + dim_value) : starts[i];
         end = ends[i] < 0 ? (ends[i] + dim_value) : ends[i];
-        start = std::max(start, 0);
-        end = std::max(end, 0);
-        end = std::min(end, dim_value);
+        start = (std::max)(start, 0);
+        end = (std::max)(end, 0);
+        end = (std::min)(end, dim_value);
         out_dims[axes[i]] = end - start;
       }
     }

lite/tools/build_windows.bat

@@ -100,7 +100,6 @@ cd "%build_directory%"
             -DPYTHON_EXECUTABLE="%python_path%"
 
 call "%vcvarsall_dir%" amd64
-cd "%build_directory%"
 
 if "%BUILD_FOR_CI%"=="ON" (
     msbuild /m /p:Configuration=Release lite\lite_compile_deps.vcxproj