concat+fetch bug for FPGA V1 track (#1678)

* marker1-anchor update

* concat-fetch bug

src/fpga/V1/api.cpp

@@ -273,6 +273,7 @@ void format_concat_output(framework::Tensor *out, int height, int width,
   auto ddim = framework::make_ddim({1, sum_channel, height, width});
   out->Resize(ddim);
   out->reset_data_ptr(data_ptr);
+  out->fpga_data_num = sum_cw * height;
   out->set_type(type_id<half>().hash_code());
 }
 void format_conv_data(framework::Tensor *filter_tensor,

src/framework/executor.cpp

@@ -561,9 +561,9 @@ template <typename Device, typename T>
 void Executor<Device, T>::FeedData(const std::vector<void *> &v) {
   auto input_size = v.size();
   int index = 0;
-  auto vars = program_.scope->VarContain("feed", &index);
-  PADDLE_MOBILE_ENFORCE(input_size == vars.size(),
-                        "input data number not correct");
+  // auto vars = program_.scope->VarContain("feed", &index);
+  // PADDLE_MOBILE_ENFORCE(input_size == vars.size(),
+  //                    "input data number not correct");
   for (int i = 0; i < input_size; i++) {
     auto var = program_.scope->Var("feed", i + index);
     auto feed_tensor = var->template GetMutable<LoDTensor>();

src/io/api_paddle_mobile.cc

@@ -19,6 +19,9 @@
 #include <vector>
 #include "common/enforce.h"
 #include "framework/tensor.h"
+#ifdef PADDLE_MOBILE_FPGA
+#include <fpga/common/fpga_common.h>
+#endif
 
 namespace paddle_mobile {
 
@@ -179,6 +182,19 @@ void PaddleMobilePredictor<Device, T>::FetchPaddleTensors(PaddleTensor *output,
                                                           int id) {
   std::shared_ptr<framework::Tensor> tensor_ptr =
       paddle_mobile_->FetchResult(id);
+  void *data_addr = nullptr;
+  int data_sizeof = 1;
+  if (tensor_ptr.get()->type() == type_id<half>().hash_code()) {
+    data_addr = tensor_ptr.get()->data<half>();
+    data_sizeof = sizeof(half);
+  } else if (tensor_ptr.get()->type() == type_id<float>().hash_code()) {
+    data_addr = tensor_ptr.get()->data<float>();
+    data_sizeof = sizeof(float);
+  } else {
+    PADDLE_MOBILE_ENFORCE(0, "output typeid is not supported");
+  }
+  size_t size = tensor_ptr.get()->numel() * data_sizeof;
+  fpga::fpga_invalidate(data_addr, size);
   ConvertTensors(*(tensor_ptr.get()), output);
   return;
 }

src/operators/kernel/fpga/V1/relu_kernel.cpp

@@ -26,7 +26,9 @@ bool ReluKernel<FPGA, float>::Init(ReluParam<FPGA> *param) {
 }
 
 template <>
-void ReluKernel<FPGA, float>::Compute(const ReluParam<FPGA> &param) {}
+void ReluKernel<FPGA, float>::Compute(const ReluParam<FPGA> &param) {
+  PADDLE_MOBILE_ENFORCE(0, "relu as a single op is wrong");
+}
 }  // namespace operators
 }  // namespace paddle_mobile