finish execute

Change-Id: I76a7a42deaab3f748d958d03c26d88e808f6c6d8

BCLOUD

@@ -84,6 +84,8 @@ def UT_FILE(filename):
     UT_DIR = 'paddle/fluid/train/custom_trainer/feed/unit_test'
     return os.path.join(UT_DIR, filename)
 
+CPPFLAGS('-DPYBIND_AVX_MKLML') # for paddle with avx and mklml
+
 custom_trainer_src = GLOB('paddle/fluid/train/custom_trainer/feed/*/*.cc', Exclude(UT_FILE('*')))
 Application('feed_trainer', Sources('paddle/fluid/train/custom_trainer/feed/main.cc', custom_trainer_src), CppFlags('-DHPPL_STUB_FUNC -DLAPACK_FOUND -DPADDLE_DISABLE_PROFILER -DPADDLE_NO_PYTHON -DCUSTOM_TRAINER -DPADDLE_ON_INFERENCE -DPADDLE_USE_DSO -DPADDLE_USE_PTHREAD_BARRIER -DPADDLE_USE_PTHREAD_SPINLOCK -DPADDLE_VERSION=0.0.0 -DPADDLE_WITH_AVX -DPADDLE_WITH_MKLML -DPADDLE_WITH_XBYAK -DXBYAK64 -DXBYAK_NO_OP_NAMES -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -DPYBIND_AVX_MKLML' + r" -DPADDLE_REVISION=\"%s@%s@%s\"" % (REPO_URL(), REPO_BRANCH(), REPO_REVISION())), CFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -fopenmp -mavx -O3 -DNDEBUG '), CxxFlags('-std=c++11 -m64 -fPIC -fno-omit-frame-pointer -Werror -Wall -Wextra -Wnon-virtual-dtor -Wdelete-non-virtual-dtor -Wno-unused-parameter -Wno-unused-function -Wno-error=literal-suffix -Wno-error=sign-compare -Wno-error=unused-local-typedefs -Wno-error=maybe-uninitialized -fopenmp -mavx -O3 -DNDEBUG '), Libraries('$OUT/lib/libpaddle_fluid_avx_mklml.a'))
 

paddle/fluid/train/custom_trainer/feed/executor/executor.cc

@@ -5,6 +5,7 @@
 #include "paddle/fluid/platform/init.h"
 #include "paddle/fluid/platform/cpu_helper.h"
 #include "paddle/fluid/inference/api/details/reset_tensor_array.h"
+#include "paddle/fluid/platform/enforce.h"
 
 
 namespace paddle {
@@ -16,7 +17,7 @@ namespace {
 int ReadBinaryFile(const std::string& filename, std::string* contents) {
     std::ifstream fin(filename, std::ios::in | std::ios::binary);
     if (!fin) {
-        VLOG(4) << "Cannot open file " << filename;
+        VLOG(2) << "Cannot open file " << filename;
         return -1;
     }
     fin.seekg(0, std::ios::end);
@@ -73,37 +74,53 @@ int SimpleExecute::initialize(YAML::Node& exe_config,
 
     if (!exe_config["startup_program"] || 
         !exe_config["main_program"]) {
-        VLOG(4) << "fail to load config";
+        VLOG(2) << "fail to load config";
         return -1;
     }
 
-    _context.reset(new SimpleExecute::Context(context_ptr->cpu_place));
-	auto startup_program = Load(&_context->executor, exe_config["startup_program"].as<std::string>());
-    if (startup_program == nullptr) {
-        VLOG(4) << "fail to load startup_program: " << exe_config["startup_program"].as<std::string>();
+    try {
+        _context.reset(new SimpleExecute::Context(context_ptr->cpu_place));
+        auto startup_program = Load(&_context->executor, exe_config["startup_program"].as<std::string>());
+        if (startup_program == nullptr) {
+            VLOG(2) << "fail to load startup_program: " << exe_config["startup_program"].as<std::string>();
+            return -1;
+        }
+        
+        _context->executor.Run(*startup_program, this->scope(), 0, false, true);
+
+        _context->main_program = Load(&_context->executor, exe_config["main_program"].as<std::string>());
+        if (_context->main_program == nullptr) {
+            VLOG(2) << "fail to load main_program: " << exe_config["main_program"].as<std::string>();
+            return -1;
+        }
+        _context->prepare_context = _context->executor.Prepare(*_context->main_program, 0);
+        _context->executor.CreateVariables(*_context->main_program, this->scope(), 0);
+    } catch (::paddle::platform::EnforceNotMet& err) {
+        VLOG(2) << err.what();
+        _context.reset(nullptr);
         return -1;
     }
-    
-    _context->executor.Run(*startup_program, this->scope(), 0, false, true);
 
-	_context->main_program = Load(&_context->executor, exe_config["main_program"].as<std::string>());
-    if (_context->main_program == nullptr) {
-        VLOG(4) << "fail to load main_program: " << exe_config["main_program"].as<std::string>();
-        return -1;
-    }
-    _context->prepare_context = _context->executor.Prepare(*_context->main_program, 0);
-    _context->executor.CreateVariables(*_context->main_program, this->scope(), 0);
     return 0;
 }
 
 int SimpleExecute::run() {
-    _context->executor.RunPreparedContext(_context->prepare_context.get(), this->scope(),
-                            false, /* don't create local scope each time*/
-                            false /* don't create variable each time */);
-
-    // For some other vector like containers not cleaned after each batch.
-    _context->tensor_array_batch_cleaner.CollectNoTensorVars(this->scope());
-    _context->tensor_array_batch_cleaner.ResetNoTensorVars();
+    if (_context == nullptr) {
+        VLOG(2) << "need initialize before run";
+        return -1;
+    }
+    try {
+        _context->executor.RunPreparedContext(_context->prepare_context.get(), this->scope(),
+                                false, /* don't create local scope each time*/
+                                false /* don't create variable each time */);
+
+        // For some other vector like containers not cleaned after each batch.
+        _context->tensor_array_batch_cleaner.CollectNoTensorVars(this->scope());
+        _context->tensor_array_batch_cleaner.ResetNoTensorVars();
+    } catch (::paddle::platform::EnforceNotMet& err) {
+        VLOG(2) << err.what();
+        return -1;
+    }
     return 0;
 }
     

paddle/fluid/train/custom_trainer/feed/main.cc

@@ -5,6 +5,7 @@
 #include "paddle/fluid/train/custom_trainer/feed/trainer_context.h"
 #include "paddle/fluid/train/custom_trainer/feed/process/process.h"
 #include "paddle/fluid/train/custom_trainer/feed/process/init_env_process.h"
+#include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/pybind/pybind.h"
 
 using namespace paddle::custom_trainer::feed;

paddle/fluid/train/custom_trainer/feed/unit_test/main.cc

 #include <gtest/gtest.h>
 #include <gflags/gflags.h>
 #include <glog/logging.h>
+#include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/pybind/pybind.h"
 
 int32_t main(int32_t argc, char** argv) {

paddle/fluid/train/custom_trainer/feed/unit_test/test_executor.cc

@@ -12,9 +12,11 @@ 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 <iostream>
 #include <gtest/gtest.h>
 
 #include "paddle/fluid/train/custom_trainer/feed/executor/executor.h"
+#include "paddle/fluid/framework/tensor_util.h"
 
 namespace paddle {
 namespace custom_trainer {
@@ -31,12 +33,46 @@ TEST(testSimpleExecute, initialize) {
     ASSERT_EQ(0, execute.initialize(config, context_ptr));
 }
 
+float uniform(float min, float max) {
+    float result = (float)rand() / RAND_MAX;
+    return min + result * (max - min);
+}
+
+void next_batch(int batch_size, const paddle::platform::Place& place, paddle::framework::LoDTensor* x_tensor, paddle::framework::LoDTensor* y_tensor) {
+	
+	x_tensor->Resize({batch_size, 2});
+	auto x_data = x_tensor->mutable_data<float>(place);
+
+	y_tensor->Resize({batch_size, 1});
+	auto y_data = y_tensor->mutable_data<float>(place);
+
+    for (int i = 0; i < batch_size; ++i) {
+        x_data[i * 2] = uniform(-2, 2);
+        x_data[i * 2 + 1] = uniform(-2, 2);
+        float dis = x_data[i * 2] * x_data[i * 2] + x_data[i * 2 + 1] * x_data[i * 2 + 1];
+        y_data[i] = dis < 1.0 ? 1.0 : 0.0;
+    }
+}
+
 TEST(testSimpleExecute, run) {
     SimpleExecute execute;
     auto context_ptr = std::make_shared<TrainerContext>();
     auto config = YAML::Load("{thread_num: 2, startup_program: ./data/startup_program, main_program: ./data/main_program}");
     ASSERT_EQ(0, execute.initialize(config, context_ptr));
+
+    
+	auto x_var = execute.mutable_var<::paddle::framework::LoDTensor>("x");
+	auto y_var = execute.mutable_var<::paddle::framework::LoDTensor>("y");
+    ASSERT_NE(nullptr, x_var);
+    ASSERT_NE(nullptr, y_var);
+
+    next_batch(1024, context_ptr->cpu_place, x_var, y_var);
+
     ASSERT_EQ(0, execute.run());
+
+	auto loss_var = execute.var<::paddle::framework::LoDTensor>("loss");
+    auto loss = loss_var.data<float>()[0];
+    std::cout << "loss: " << loss << std::endl;
 }
 
 }  // namespace feed