update io file to surport int8 tensor feed and fetch fixed#1602 (#1603)

* update concat and split kernel and related files in FPGA v2(v3) track

* update

* update

* update kernel and related files in FPGA v2 track

* update

* update

* update kernel and related files for static quantization in FPGA v2 track

* update

* update feed and fetch kernel in FPGA v2 track

* update io file

src/fpga/V2/api.cpp

@@ -32,7 +32,7 @@ void format_image(framework::Tensor *image_tensor) {
   int8_t *p_data = external_ptr == nullptr ? data_ptr : external_ptr;
 
   image::format_image<int8_t>(&p_data, channel, height, width);
-  if (p_data != data_ptr && external_ptr == nullptr) {
+  if (p_data != data_ptr) {
     image_tensor->reset_data_ptr(p_data);
   }
 }
@@ -43,7 +43,6 @@ void format_ofm(framework::Tensor *ofm_tensor) {
   } else {
     format_int8_ofm(ofm_tensor);
   }
-  format_int8_ofm(ofm_tensor);
 }
 
 void format_int8_ofm(framework::Tensor *ofm_tensor) {

src/io/api_paddle_mobile.cc

@@ -131,9 +131,12 @@ void ConvertTensors(const framework::Tensor &src, PaddleTensor *des) {
   if (src.type() == type_id<float>()) {
     des->data.Reset(const_cast<float *>(src.data<float>()),
                     num * sizeof(float));
-  } else {
+  } else if (src.type() == type_id<half>()) {
     des->data.Reset(const_cast<int16_t *>(src.data<int16_t>()),
                     num * sizeof(int16_t));
+  } else {
+    des->data.Reset(const_cast<int8_t *>(src.data<int8_t>()),
+                    num * sizeof(int8_t));
   }
 }
 
@@ -143,7 +146,11 @@ void PaddleMobilePredictor<Device, T>::FeedPaddleTensors(
   auto num = inputs.size();
   std::vector<framework::Tensor> tensors(num, framework::Tensor());
   for (int i = 0; i < num; i++) {
-    tensors[i].init(type_id<float>().hash_code());
+    if (inputs[i].dtypeid == type_id<int8_t>().hash_code()) {
+      tensors[i].init(type_id<int8_t>().hash_code());
+    } else {
+      tensors[i].init(type_id<float>().hash_code());
+    }
     ConvertPaddleTensors(inputs[i], &tensors[i]);
   }
   paddle_mobile_->FeedTensorData(tensors);

src/operators/kernel/fpga/V2/feed_kernel.cpp

@@ -44,6 +44,7 @@ void FeedKernel<FPGA, float>::Compute(const FeedParam<FPGA> &param) {
   }
   fpga::format_image(input);
   output->ShareDataWith(*input);
+  input->external_data = nullptr;
 }
 template class FeedKernel<FPGA, float>;