From 2411ed42862481ac07d3191e9b537698b35cd857 Mon Sep 17 00:00:00 2001
From: Tao Luo <luotao02@baidu.com>
Date: Mon, 14 Jan 2019 15:00:44 +0800
Subject: [PATCH] fix multi-threads in ZeroCopyProfile

test=develop
---
 .../fluid/inference/tests/api/analyzer_rnn1_tester.cc | 11 ++++-------
 .../inference/tests/api/analyzer_seq_pool1_tester.cc  | 11 ++++-------
 2 files changed, 8 insertions(+), 14 deletions(-)

diff --git a/paddle/fluid/inference/tests/api/analyzer_rnn1_tester.cc b/paddle/fluid/inference/tests/api/analyzer_rnn1_tester.cc
index 315b49533..22e6366fb 100644
--- a/paddle/fluid/inference/tests/api/analyzer_rnn1_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_rnn1_tester.cc
@@ -370,15 +370,12 @@ TEST(Analyzer_rnn1, ZeroCopyMultiThread) {
   auto base_predictor = CreatePaddlePredictor<AnalysisConfig>(config);
   double total_time_of_threads{0};
   std::vector<std::thread> threads;
-  std::vector<std::unique_ptr<PaddlePredictor>> predictors;
-  for (int tid = 0; tid < FLAGS_num_threads; tid++) {
-    predictors.emplace_back(CreatePaddlePredictor<AnalysisConfig>(config));
-  }
 
   for (int tid = 0; tid < FLAGS_num_threads; tid++) {
-    threads.emplace_back([config, &total_time_of_threads, &predictors, tid] {
-      // auto predictor = base_predictor->Clone();
-      auto &predictor = predictors[tid];
+    threads.emplace_back([&, tid] {
+      // To ensure the thread binding correctly,
+      // please clone inside the threadpool.
+      auto predictor = base_predictor->Clone();
       NEW_TENSOR(data_lod_attention);
       NEW_TENSOR(cell_init);
       NEW_TENSOR(data);
diff --git a/paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester.cc b/paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester.cc
index fb4c5c0a0..c13709087 100644
--- a/paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_seq_pool1_tester.cc
@@ -263,15 +263,12 @@ TEST(Analyzer_seq_pool1, zerocopy_profile_threads) {
   auto base_predictor = CreatePaddlePredictor<AnalysisConfig>(config);
   double total_time_of_threads{0};
   std::vector<std::thread> threads;
-  std::vector<std::unique_ptr<PaddlePredictor>> predictors;
-  for (int tid = 0; tid < FLAGS_num_threads; tid++) {
-    predictors.emplace_back(base_predictor->Clone());
-    // predictors.emplace_back(CreatePaddlePredictor<AnalysisConfig>(config));
-  }
 
   for (int tid = 0; tid < FLAGS_num_threads; tid++) {
-    threads.emplace_back([config, &total_time_of_threads, &predictors, tid] {
-      auto &predictor = predictors[tid];
+    threads.emplace_back([&, tid] {
+      // To ensure the thread binding correctly,
+      // please clone inside the threadpool.
+      auto predictor = base_predictor->Clone();
       std::vector<std::unique_ptr<ZeroCopyTensor>> inputs;
       PrepareZeroCopyInputs(predictor, &inputs);
       auto output_tensor = predictor->GetOutputTensor(out_var_name);
-- 
GitLab