update trt int8 calibrator to IEntropyCalibratorV2 (#31060)

* update trt int8 calibrator to IEntropyCalibratorV2

* add delele opt_cache for trt_split_converter_test

paddle/fluid/inference/tensorrt/trt_int8_calibrator.h

@@ -34,7 +34,7 @@ namespace tensorrt {
 
 class TensorRTEngine;
 
-struct TRTInt8Calibrator : public nvinfer1::IInt8EntropyCalibrator {
+struct TRTInt8Calibrator : public nvinfer1::IInt8EntropyCalibrator2 {
  public:
   TRTInt8Calibrator(const std::unordered_map<std::string, size_t>& buffers,
                     int batch_size, std::string engine_name,

paddle/fluid/inference/tests/api/trt_dynamic_shape_ernie_serialize_deserialize_test.h

@@ -12,7 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 #pragma once
-#include <dirent.h>
 #include <glog/logging.h>
 #include <gtest/gtest.h>
 #include <unistd.h>
@@ -27,22 +26,6 @@ limitations under the License. */
 namespace paddle {
 namespace inference {
 
-static int DeleteCache(std::string path) {
-  DIR* dir = opendir(path.c_str());
-  if (dir == NULL) return 0;
-  struct dirent* ptr;
-  while ((ptr = readdir(dir)) != NULL) {
-    if (std::strcmp(ptr->d_name, ".") == 0 ||
-        std::strcmp(ptr->d_name, "..") == 0) {
-      continue;
-    } else if (ptr->d_type == 8) {
-      std::string file_rm = path + "/" + ptr->d_name;
-      return remove(file_rm.c_str());
-    }
-  }
-  return 0;
-}
-
 static void run(const AnalysisConfig& config, std::vector<float>* out_data) {
   auto predictor = CreatePaddlePredictor(config);
   auto input_names = predictor->GetInputNames();
@@ -111,7 +94,7 @@ static void trt_ernie(bool with_fp16, std::vector<float> result) {
   // Delete serialization cache to perform serialization first rather than
   // deserialization.
   std::string opt_cache_dir = FLAGS_infer_model + "/_opt_cache";
-  DeleteCache(opt_cache_dir);
+  delete_cache_files(opt_cache_dir);
 
   SetConfig(&config, model_dir, true /* use_gpu */);
 

paddle/fluid/inference/tests/api/trt_split_converter_test.cc

@@ -23,6 +23,9 @@ namespace inference {
 
 TEST(TensorRT, split_converter) {
   std::string model_dir = FLAGS_infer_model + "/split_converter";
+  std::string opt_cache_dir = model_dir + "/_opt_cache";
+  delete_cache_files(opt_cache_dir);
+
   AnalysisConfig config;
   int batch_size = 4;
   config.EnableUseGpu(100, 0);

paddle/fluid/inference/tests/api/trt_test_helper.h

@@ -12,6 +12,7 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 #pragma once
+#include <dirent.h>
 #include <string>
 #include <vector>
 
@@ -134,5 +135,20 @@ void compare_continuous_input(std::string model_dir, bool use_tensorrt) {
   }
 }
 
+void delete_cache_files(std::string path) {
+  DIR* dir = opendir(path.c_str());
+  if (dir == NULL) return;
+  struct dirent* ptr;
+  while ((ptr = readdir(dir)) != NULL) {
+    if (std::strcmp(ptr->d_name, ".") == 0 ||
+        std::strcmp(ptr->d_name, "..") == 0) {
+      continue;
+    } else if (ptr->d_type == 8) {
+      std::string file_rm = path + "/" + ptr->d_name;
+      remove(file_rm.c_str());
+    }
+  }
+}
+
 }  // namespace inference
 }  // namespace paddle