Merge pull request #1214 from Eclipsess/develop

fix #1213 add test-eng

src/operators/kernel/central-arm-func/fusion_fc_arm_func.h

@@ -29,10 +29,9 @@ void FusionFcCompute(const FusionFcParam<CPU> &param) {
   auto *input_z_data = input_z->data<float>();
   int axis = param.Axis();
   Tensor *out = param.Out();
-  auto *out_data = out->mutable_data<float>();
   //  int m = out->dims()[0];
   //  int n = out->dims()[1];
-
+  auto *out_data = out->mutable_data<float>();
   const Tensor x_matrix =
       input_x->dims().size() > 2
           ? framework::ReshapeToMatrix(*input_x, param.XNumColDims())

src/operators/kernel/central-arm-func/pool_arm_func.h

@@ -83,6 +83,7 @@ void PoolCompute(const PoolParam<CPU> &param) {
 #if __aarch64__
     PoolBasic(pooling_type, ksize, strides, paddings, in_x, out);
 #else
+    /// todo: fix bug in Pool2x2
     if (pooling_type == "max") {
       math::Pool2x2Maxs2p0(strides, paddings, in_x, out);
     } else if (pooling_type == "avg") {

src/operators/kernel/central-arm-func/softmax_arm_func.h

@@ -24,6 +24,7 @@ void SoftmaxCompute(const SoftmaxParam<CPU> &param) {
   Tensor *out = param.Out();
   auto x_dims = in_x->dims();
   out->Resize(x_dims);
+  out->mutable_data<float>();
   math::SoftmaxFuntor<CPU, float>()(in_x, out);
 }
 }  // namespace operators

src/operators/op_param.h

@@ -2147,9 +2147,9 @@ class Im2SequenceParam : public OpParam {
     paddings_ = GetAttr<vector<int>>("paddings", attrs);
   }
 
-  const RType *Input() const { return input_x_; }
+  const GType *Input() const { return input_x_; }
 
-  RType *Output() const { return out_; }
+  GType *Output() const { return out_; }
 
   const vector<int> &Kernels() const { return kernels_; }
 
@@ -2158,8 +2158,8 @@ class Im2SequenceParam : public OpParam {
   const vector<int> &Paddings() const { return paddings_; }
 
  private:
-  RType *input_x_;
-  RType *out_;
+  GType *input_x_;
+  GType *out_;
   vector<int> kernels_;
   vector<int> strides_;
   vector<int> paddings_;

test/net/test_eng.cpp

@@ -23,13 +23,13 @@ int main() {
   //    paddle_mobile.SetThreadNum(4);
   auto time1 = time();
   if (paddle_mobile.Load(std::string(g_eng) + "/model",
-                         std::string(g_eng) + "/params", false, false, 1,
+                         std::string(g_eng) + "/params", true, false, 1,
                          true)) {
     auto time2 = time();
     std::cout << "load cost :" << time_diff(time1, time1) << "ms" << std::endl;
-    std::vector<int64_t> dims{1, 1, 48, 512};
+    std::vector<int64_t> dims{1, 1, 48, 400};
     LoDTensor input_tensor;
-    SetupTensor<float>(&input_tensor, {1, 1, 48, 512}, static_cast<float>(0),
+    SetupTensor<float>(&input_tensor, {1, 1, 48, 400}, static_cast<float>(0),
                        static_cast<float>(1));
 
     std::vector<float> input(input_tensor.data<float>(),