add threads test

paddle/fluid/inference/tests/book/test_inference_nlp.cc

@@ -14,6 +14,7 @@ limitations under the License. */
 
 #include <sys/time.h>
 #include <time.h>
+#include <thread>  // NOLINT
 #include "gflags/gflags.h"
 #include "gtest/gtest.h"
 #include "paddle/fluid/inference/tests/test_helper.h"
@@ -31,7 +32,14 @@ TEST(inference, understand_sentiment) {
 
   LOG(INFO) << "FLAGS_dirname: " << FLAGS_dirname << std::endl;
   std::string dirname = FLAGS_dirname;
-
+  const bool model_combined = false;
+  int total_work = 100;
+  int num_threads = 10;
+  int work_per_thread = total_work / num_threads;
+  std::vector<std::unique_ptr<std::thread>> infer_threads;
+  for (int i = 0; i < num_threads; ++i) {
+    infer_threads.emplace_back(new std::thread([&, i]() {
+      for (int j = 0; j < work_per_thread; ++j) {
         // 0. Call `paddle::framework::InitDevices()` initialize all the devices
         // In unittests, this is done in paddle/testing/paddle_gtest_main.cc
         paddle::framework::LoDTensor words;
@@ -42,19 +50,21 @@ TEST(inference, understand_sentiment) {
                          static_cast<int64_t>(word_dict_len - 1));
          */
         std::vector<int64_t> srcdata{
-      784,    784,   1550,   6463,   56,     75693, 6189,  784,    784,  1550,
-      198391, 6463,  42468,  4376,   10251,  10760, 6189,  297,    396,  6463,
-      6463,   1550,  198391, 6463,   22564,  1612,  291,   68,     164,  784,
-      784,    1550,  198391, 6463,   13659,  3362,  42468, 6189,   2209, 198391,
-      6463,   2209,  2209,   198391, 6463,   2209,  1062,  3029,   1831, 3029,
-      1065,   2281,  100,    11216,  1110,   56,    10869, 9811,   100,  198391,
-      6463,   100,   9280,   100,    288,    40031, 1680,  1335,   100,  1550,
-      9280,   7265,  244,    1550,   198391, 6463,  1550,  198391, 6463, 42468,
-      4376,   10251, 10760};
+            784,   784,    1550,   6463,   56,     75693,  6189,  784,    784,
+            1550,  198391, 6463,   42468,  4376,   10251,  10760, 6189,   297,
+            396,   6463,   6463,   1550,   198391, 6463,   22564, 1612,   291,
+            68,    164,    784,    784,    1550,   198391, 6463,  13659,  3362,
+            42468, 6189,   2209,   198391, 6463,   2209,   2209,  198391, 6463,
+            2209,  1062,   3029,   1831,   3029,   1065,   2281,  100,    11216,
+            1110,  56,     10869,  9811,   100,    198391, 6463,  100,    9280,
+            100,   288,    40031,  1680,   1335,   100,    1550,  9280,   7265,
+            244,   1550,   198391, 6463,   1550,   198391, 6463,  42468,  4376,
+            10251, 10760};
         paddle::framework::LoD lod{{0, srcdata.size()}};
         words.set_lod(lod);
         int64_t* pdata = words.mutable_data<int64_t>(
-      {static_cast<int64_t>(srcdata.size()), 1}, paddle::platform::CPUPlace());
+            {static_cast<int64_t>(srcdata.size()), 1},
+            paddle::platform::CPUPlace());
         memcpy(pdata, srcdata.data(), words.numel() * sizeof(int64_t));
 
         LOG(INFO) << "number of input size:" << words.numel();
@@ -66,7 +76,6 @@ TEST(inference, understand_sentiment) {
         cpu_fetchs1.push_back(&output1);
 
         // Run inference on CPU
-  const bool model_combined = false;
         if (FLAGS_prepare_vars) {
           if (FLAGS_prepare_context) {
             TestInference<paddle::platform::CPUPlace, false, true>(
@@ -90,17 +99,7 @@ TEST(inference, understand_sentiment) {
         }
         LOG(INFO) << output1.lod();
         LOG(INFO) << output1.dims();
-
-#ifdef PADDLE_WITH_CUDA
-  paddle::framework::LoDTensor output2;
-  std::vector<paddle::framework::LoDTensor*> cpu_fetchs2;
-  cpu_fetchs2.push_back(&output2);
-
-  // Run inference on CUDA GPU
-  TestInference<paddle::platform::CUDAPlace>(dirname, cpu_feeds, cpu_fetchs2);
-  LOG(INFO) << output2.lod();
-  LOG(INFO) << output2.dims();
-
-  CheckError<float>(output1, output2);
-#endif
+      }
+    }));
+  }
 }