[oneDNN] Added Unit Test for Multiple instances prediction (#29501)

* - Added infrastructre for new test

 - Added UT for Multiple models prediction

- cosmetic fixes

- lint

- lint fixes

* - Removed timeout for MMP test

paddle/fluid/inference/tests/api/CMakeLists.txt

@@ -54,6 +54,12 @@ function(inference_analysis_api_test target install_dir filename)
         ARGS --infer_model=${install_dir}/model --infer_data=${install_dir}/data.txt --refer_result=${install_dir}/result.txt)
 endfunction()
 
+function(inference_multiple_models_analysis_api_test target install_dir filename)
+    inference_analysis_test(${target} SRCS ${filename}
+        EXTRA_DEPS ${INFERENCE_EXTRA_DEPS}
+        ARGS --infer_model=${install_dir}/mobilenet_v2_models/1 --infer_model2=${install_dir}/mobilenet_v2_models/xx --infer_model3=${install_dir}/mobilenet_v2_models/3)
+endfunction()
+
 function(inference_analysis_api_test_build TARGET_NAME filename)
 	inference_analysis_test_build(${TARGET_NAME} SRCS ${filename}
         EXTRA_DEPS ${INFERENCE_EXTRA_DEPS})
@@ -462,6 +468,11 @@ set(BERT_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/bert_emb128")
 download_model_and_data(${BERT_INSTALL_DIR} "bert_emb128_model.tar.gz" "bert_data_len20.txt.tar.gz")
 inference_analysis_api_test(test_analyzer_bert ${BERT_INSTALL_DIR} analyzer_bert_tester.cc)
 
+# multiple models prediction
+set(MMP_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/multi_model_prediction")
+download_data(${MMP_INSTALL_DIR} PaddleInference/mobilenet_v2_models.tar.gz) 
+inference_multiple_models_analysis_api_test(test_analyzer_multi_model_prediction ${MMP_INSTALL_DIR} analyzer_mmp_tester.cc)
+
 if(WITH_GPU AND TENSORRT_FOUND)
     set(TRT_MODEL_INSTALL_DIR "${INFERENCE_DEMO_INSTALL_DIR}/trt_models")
     if (NOT EXISTS ${TRT_MODEL_INSTALL_DIR}/trt_inference_test_models.tar.gz)

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

+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// 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.
+
+#include "paddle/fluid/framework/transfer_scope_cache.h"
+#include "paddle/fluid/inference/tests/api/tester_helper.h"
+
+#include <random>
+
+// Here add missing commands
+DEFINE_string(infer_model2, "", "model path");
+DEFINE_string(infer_model3, "", "model path");
+
+namespace paddle {
+namespace inference {
+
+// Shape of Input to models
+const int N = 1, C = 3, H = 224, W = 224;
+
+void SetConfig(AnalysisConfig* config, const std::string& infer_model) {
+  config->SetModel(infer_model + "/__model__", infer_model + "/__params__");
+  config->DisableFCPadding();
+  config->SwitchUseFeedFetchOps(false);
+  config->SwitchSpecifyInputNames(true);
+}
+
+std::unique_ptr<PaddlePredictor> InitializePredictor(
+    const std::string& infer_model, std::vector<float>& data, bool use_mkldnn) {
+  AnalysisConfig cfg;
+  SetConfig(&cfg, infer_model);
+  if (use_mkldnn) {
+    cfg.EnableMKLDNN();
+  }
+
+  auto predictor = ::paddle::CreatePaddlePredictor<AnalysisConfig>(cfg);
+  auto input_name = predictor->GetInputNames()[0];
+  auto input = predictor->GetInputTensor(input_name);
+  std::vector<int> shape{N, C, H, W};
+  input->Reshape(std::move(shape));
+  input->copy_from_cpu(data.data());
+
+  return predictor;
+}
+
+// Compare result of NativeConfig and AnalysisConfig
+void compare(bool use_mkldnn = false) {
+  // Create Input to models
+  std::vector<float> data(N * C * H * W);
+  std::default_random_engine re{1234};
+  std::uniform_real_distribution<float> sampler{0.0, 1.0};
+  for (auto& v : data) {
+    v = sampler(re);
+  }
+
+  // Initialize Models predictors
+  auto predictor_1 = InitializePredictor(FLAGS_infer_model, data, use_mkldnn);
+  auto predictor_xx = InitializePredictor(FLAGS_infer_model2, data, use_mkldnn);
+  auto predictor_3 = InitializePredictor(FLAGS_infer_model3, data, use_mkldnn);
+
+  // Run single xx model
+  predictor_xx->ZeroCopyRun();
+  auto output =
+      predictor_xx->GetOutputTensor(predictor_xx->GetOutputNames()[0]);
+  auto output_shape = output->shape();
+  int numel = std::accumulate(output_shape.begin(), output_shape.end(), 1,
+                              std::multiplies<int>());
+  std::vector<float> xx_output(numel);
+  output->copy_to_cpu(xx_output.data());
+
+  // Initialize xx model's predictor to trigger oneDNN cache clearing
+  predictor_xx =
+      std::move(InitializePredictor(FLAGS_infer_model2, data, use_mkldnn));
+
+  // Run sequence of models
+  predictor_1->ZeroCopyRun();
+  predictor_xx->ZeroCopyRun();
+  predictor_3->ZeroCopyRun();
+
+  // Get again output of xx model , but when all three models were executed
+  std::vector<float> xx2_output(numel);
+  output = predictor_xx->GetOutputTensor(predictor_xx->GetOutputNames()[0]);
+  output->copy_to_cpu(xx2_output.data());
+
+  // compare results
+  auto result = std::equal(
+      xx_output.begin(), xx_output.end(), xx2_output.begin(),
+      [](const float& l, const float& r) { return fabs(l - r) < 1e-4; });
+
+  PADDLE_ENFORCE_EQ(result, true, paddle::platform::errors::Fatal(
+                                      "Results of model run independently "
+                                      "differs from results of the same model "
+                                      "run as a sequence of models"));
+}
+
+TEST(Analyzer_mmp, compare) { compare(); }
+#ifdef PADDLE_WITH_MKLDNN
+TEST(Analyzer_mmp, compare_mkldnn) { compare(true /* use_mkldnn */); }
+#endif
+
+}  // namespace inference
+}  // namespace paddle