[mobile][opencl]optimise log print , use kNOLOG to close develop time… (#2693)

* [mobile][opencl]optimise log print , ues kNOLOG to close develop time logs ,test=mobile

* [mobile][opencl]optimise log print , ues kNOLOG to close develop time logs ,test=mobile

mobile/src/common/log.h

@@ -80,7 +80,6 @@ static const char *ANDROID_LOG_TAG =
 #endif
 
 enum LogLevel {
-  kNO_LOG,
   kLOG_ERROR,
   kLOG_WARNING,
   kLOG_INFO,
@@ -89,15 +88,16 @@ enum LogLevel {
   kLOG_DEBUG1,
   kLOG_DEBUG2,
   kLOG_DEBUG3,
-  kLOG_DEBUG4
+  kLOG_DEBUG4,
+  kNO_LOG,
 };
 
 // log level
 static LogLevel log_level = kLOG_DEBUG4;
 
-static std::vector<std::string> logs{"NO     ", "ERROR  ", "WARNING", "INFO   ",
-                                     "VERBOSE", "DEBUG  ", "DEBUG1 ", "DEBUG2 ",
-                                     "DEBUG3 ", "DEBUG4 "};
+static std::vector<std::string> logs{"ERROR  ", "WARNING", "INFO   ", "VERBOSE",
+                                     "DEBUG  ", "DEBUG1 ", "DEBUG2 ", "DEBUG3 ",
+                                     "DEBUG4 ", "NO     "};
 struct ToLog;
 struct Print;
 
@@ -217,7 +217,6 @@ struct ToLog {
 #define ANDROIDLOGV(...)
 
 enum LogLevel {
-  kNO_LOG,
   kLOG_ERROR,
   kLOG_WARNING,
   kLOG_INFO,
@@ -226,7 +225,8 @@ enum LogLevel {
   kLOG_DEBUG1,
   kLOG_DEBUG2,
   kLOG_DEBUG3,
-  kLOG_DEBUG4
+  kLOG_DEBUG4,
+  kNO_LOG
 };
 
 struct ToLog;

mobile/src/framework/cl/cl_engine.h

@@ -124,9 +124,9 @@ class CLEngine {
     if (status != CL_SUCCESS || ret_size / sizeof(size_t) < 3) {
       return CLLocalWorkSizeInfo(0, 0, 0, 0);
     }
-    DLOG << max_work_item_sizes[0];
-    DLOG << max_work_item_sizes[1];
-    DLOG << max_work_item_sizes[2];
+    DLOG << " max_work_item_sizes {" << max_work_item_sizes[0] << ", "
+         << max_work_item_sizes[1] << ", " << max_work_item_sizes[2] << "}";
+
     localWorkSizeInfo_ =
         CLLocalWorkSizeInfo(max_work_group_size, max_work_item_sizes[0],
                             max_work_item_sizes[1], max_work_item_sizes[2]);
@@ -182,8 +182,8 @@ class CLEngine {
     cl_program p =
         clCreateProgramWithSource(context, 1, &source, sourceSize, &status_);
 
-    DLOG << " cl kernel from source";
-    DLOG << " source size: " << sourceSize[0];
+    LOG(kLOG_DEBUG4) << " cl kernel from source";
+    LOG(kLOG_DEBUG4) << " source size: " << sourceSize[0];
     CL_CHECK_ERRORS(status_);
 
     std::unique_ptr<_cl_program, CLProgramDeleter> program_ptr(p);

mobile/src/framework/cl/cl_helper.h

@@ -36,9 +36,9 @@ class CLHelper {
 
   void AddKernel(const std::string &kernel_name, const std::string &file_name,
                  const std::string &options = "") {
-    DLOG << " begin add kernel ";
+    LOG(kLOG_DEBUG1) << " begin add kernel ";
     auto kernel = scope_->GetKernel(kernel_name, file_name, options);
-    DLOG << " add kernel ing ";
+    LOG(kLOG_DEBUG1) << " begin add kernel ";
     kernels.emplace_back(std::move(kernel));
   }
 

mobile/src/framework/cl/cl_image.h

@@ -87,14 +87,14 @@ class CLImage {
     PADDLE_MOBILE_ENFORCE(tensor_data_ != nullptr,
                           " need call SetTensorData first");
 
-    DLOG << " begin init cl image ";
+    LOG(kNO_LOG) << " begin init cl image ";
     image_dims_ = converter->InitImageDimInfoWith(tensor_dims_);
 
     half_t *image_data = new half_t[product(image_dims_) * 4];
 
-    DLOG << " convert to image";
+    LOG(kNO_LOG) << " convert to image";
     converter->NCHWToImage(tensor_data_, image_data, tensor_dims_);
-    DLOG << " end convert to image";
+    LOG(kNO_LOG) << " end convert to image";
 
     InitCLImage(context, image_dims_[0], image_dims_[1], image_data);
 
@@ -105,7 +105,7 @@ class CLImage {
     tensor_data_ = nullptr;
     image_converter_ = converter;
     initialized_ = true;
-    DLOG << " end init cl image";
+    LOG(kNO_LOG) << " end init cl image";
   }
 
   void InitNImage(cl_context context, cl_command_queue command_queue) {
@@ -137,9 +137,9 @@ class CLImage {
     //    CLImageConverterFolder();
     CLImageConverterNormal *normal_converter = new CLImageConverterNormal();
     PADDLE_MOBILE_ENFORCE(!shared_mem_, "do not init mem after shared .")
-    DLOG << " to get image dims ";
+    //    LOG(kNO_LOG) << " to get image dims ";
     image_dims_ = normal_converter->InitImageDimInfoWith(dim);
-    DLOG << " end get image dims " << image_dims_;
+    //    LOG(kNO_LOG) << " end get image dims " << image_dims_;
 
     InitCLImage(context, image_dims_[0], image_dims_[1], nullptr);
 
@@ -148,7 +148,7 @@ class CLImage {
     image_converter_ = normal_converter;
     cl_event_ = CLEngine::Instance()->CreateEvent(context);
     initialized_ = true;
-    DLOG << " end init cl image";
+    //    LOG(kNO_LOG) << " end init cl image";
   }
   /**
    *  create fake size cl_mem for mem share
@@ -169,9 +169,9 @@ class CLImage {
     InitCLImage(context, real_image_dims_[0], real_image_dims_[1], nullptr);
     // cheat cl_image they got what they wanted
     image_dims_ = normal_converter->InitImageDimInfoWith(need_dims);
-    DLOG << "InitFakeSizeImage ... ";
-    DLOG << "real_image_dims:  " << real_image_dims_;
-    DLOG << "image_dims_:  " << image_dims_;
+    LOG(kNO_LOG) << "InitFakeSizeImage ... ";
+    LOG(kNO_LOG) << "real_image_dims:  " << real_image_dims_;
+    LOG(kNO_LOG) << "image_dims_:  " << image_dims_;
     PADDLE_MOBILE_ENFORCE(real_image_dims_[0] >= image_dims_[0] &&
                               real_image_dims_[1] >= image_dims_[1],
                           "real image is not enough");
@@ -182,7 +182,7 @@ class CLImage {
     initialized_ = true;
     shared_mem_ = true;
 
-    DLOG << " end init FakeSizeImage";
+    LOG(kNO_LOG) << " end init FakeSizeImage";
   }
   /**
    * init cl mem with a exist cl mem
@@ -197,15 +197,15 @@ class CLImage {
     real_image_dims_ = src.real_image_dims_;
     image_dims_ = normal_converter->InitImageDimInfoWith(need_dims);
 
-    DLOG << "InitWithExistMem ... ";
-    DLOG << "real_image_dims:  " << real_image_dims_;
-    DLOG << "image_dims_:  " << image_dims_;
+    LOG(kNO_LOG) << "InitWithExistMem ... ";
+    LOG(kNO_LOG) << "real_image_dims:  " << real_image_dims_;
+    LOG(kNO_LOG) << "image_dims_:  " << image_dims_;
 
     if (real_image_dims_[0] < image_dims_[0] ||
         real_image_dims_[1] < image_dims_[1]) {
-      DLOG << "real image is not enough!";
-      DLOG << "real_image_dims:  " << real_image_dims_;
-      DLOG << "image_dims_:  " << image_dims_;
+      LOG(kNO_LOG) << "real image is not enough!";
+      LOG(kNO_LOG) << "real_image_dims:  " << real_image_dims_;
+      LOG(kNO_LOG) << "image_dims_:  " << image_dims_;
     }
     PADDLE_MOBILE_ENFORCE(real_image_dims_[0] >= image_dims_[0] &&
                               real_image_dims_[1] >= image_dims_[1],
@@ -221,7 +221,7 @@ class CLImage {
     initialized_ = true;
     shared_mem_ = true;
 
-    DLOG << " end init WithExistMem";
+    LOG(kNO_LOG) << " end init WithExistMem";
   }
 
   void InitConv2dTransposeFilterCLImage(cl_context context,

mobile/src/framework/cl/cl_scope.h

@@ -47,14 +47,14 @@ class CLScope {
   std::unique_ptr<_cl_kernel, CLKernelDeleter> GetKernel(
       const std::string &kernel_name, const std::string &file_name,
       const std::string &options) {
-    DLOG << " to get program " << file_name;
+    LOG(kLOG_DEBUG2) << " to get program " << file_name;
     auto program = Program(file_name, kernel_name, options);
-    DLOG << " end get program ~ ";
-    DLOG << " to create kernel: " << kernel_name;
+    LOG(kLOG_DEBUG2) << " end get program ~ ";
+    LOG(kLOG_DEBUG2) << " to create kernel: " << kernel_name;
     std::unique_ptr<_cl_kernel, CLKernelDeleter> kernel(
         clCreateKernel(program, kernel_name.c_str(), &status_));
     CL_CHECK_ERRORS(status_);
-    DLOG << " end create kernel ~ ";
+    LOG(kLOG_DEBUG2) << " end create kernel ~ ";
     return std::move(kernel);
   }
 
@@ -81,9 +81,11 @@ class CLScope {
       auto program = CLEngine::Instance()->CreateProgramWithSource(
           context_, source.c_str());
 
-      DLOG << " --- begin build program -> " << program_key << " --- ";
+      LOG(kLOG_DEBUG3) << " --- begin build program -> " << program_key
+                       << " --- ";
       CLEngine::Instance()->BuildProgram(program.get(), options);
-      DLOG << " --- end build program -> " << program_key << " --- ";
+      LOG(kLOG_DEBUG3) << " --- end build program -> " << program_key
+                       << " --- ";
 
       programs_[program_key] = std::move(program);
       return programs_[program_key].get();
@@ -100,9 +102,11 @@ class CLScope {
           context_,
           CLEngine::Instance()->GetCLPath() + "/cl_kernel/" + file_name);
 
-      DLOG << " --- begin build program -> " << program_key << " --- ";
+      LOG(kLOG_DEBUG3) << " --- begin build program ele-> " << program_key
+                       << " --- ";
       CLEngine::Instance()->BuildProgram(program.get(), options);
-      DLOG << " --- end build program -> " << program_key << " --- ";
+      LOG(kLOG_DEBUG3) << " --- end build program ele-> " << program_key
+                       << " --- ";
 
       programs_[program_key] = std::move(program);
       return programs_[program_key].get();

mobile/src/framework/executor.cpp

@@ -80,7 +80,7 @@ Executor<Device, T>::Executor(const Program<Device> &program,
   std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
   for (int j = 0; j < ops.size(); ++j) {
     std::shared_ptr<OpDesc> op_desc = ops[j];
-    DLOG << "create op: " << op_desc->Type();
+    LOG(kLOG_INFO) << "create op[" << j << "]: " << op_desc->Type();
 
     auto op_handler = OpRegistry<Device>::CreateOp(
         op_desc->Type(), op_desc->GetInputs(), op_desc->GetOutputs(),
@@ -111,7 +111,8 @@ Executor<Device, T>::Executor(const Program<Device> &program,
     clock_gettime(CLOCK_MONOTONIC, &ts);
     profile[op_index].runBegin = (uint64_t)ts.tv_sec * 1e9 + ts.tv_nsec;
 #endif
-    DLOG << "Initialize op[" << count++ << "]: " << op_handler->Type();
+    LOG(kLOG_INFO) << "Initialize op[" << count++
+                   << "]: " << op_handler->Type();
     if (op_handler->Type() == "feed" || op_handler->Type() == "fetch") {
       op_handler->setPrePostType(config_.pre_post_type);
     }
@@ -1015,7 +1016,7 @@ void Executor<GPU_CL, float>::InitMemory() {
           const TensorDesc &desc = var_desc->Tensor_desc();
           //          DDim ddim = make_ddim(desc.Dims());
           DDim ddim = cl_image->dims();
-          DLOG << var_desc->Name();
+          LOG(kLOG_DEBUG1) << "init image of " << var_desc->Name();
           cl_image->InitEmptyImage(context, command_queue, ddim);
         }
       }

mobile/src/framework/loader.cpp

@@ -13,6 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "framework/loader.h"
+#include <memory>
 
 #include "framework/lod_tensor.h"
 #include "framework/program/program-optimize/program_optimize.h"
@@ -173,7 +174,7 @@ static size_t ReadBuffer(const char *file_name, uint8_t **out) {
   rewind(fp);
 
   DLOG << "model size: " << size;
-
+  PADDLE_MOBILE_ENFORCE(size > 0, "model size should > 0")
   *out = reinterpret_cast<uint8_t *>(malloc(size));
 
   size_t cur_len = 0;

mobile/src/pass/memory_optimize_cl.cpp

@@ -14,6 +14,7 @@ limitations under the License. */
 #ifdef PADDLE_MOBILE_CL
 #include "pass/memory_optimize_cl.h"
 #include <algorithm>
+#include <utility>
 #include "framework/cl/cl_image.h"
 #include "framework/lod_tensor.h"
 namespace paddle_mobile {
@@ -79,7 +80,7 @@ void MemoryOptPassCl::operator()(
 
     std::vector<ClVarNode *> fetch_var_nodes;
     for (const auto &op : block->Ops()) {
-      DLOG << "op_desc->Type(): " << op->Type();
+      LOG(kNO_LOG) << "op_desc->Type(): " << op->Type();
       for (const auto &outputs : op->GetOutputs()) {
         for (const auto &output : outputs.second) {
           // not a persistable and not a exclude one ,then add it to
@@ -87,7 +88,7 @@ void MemoryOptPassCl::operator()(
           if (!IsPersistable(output) &&
               std::find(exclude_var_names.begin(), exclude_var_names.end(),
                         output) == exclude_var_names.end()) {
-            DLOG << "output: " << output;
+            LOG(kNO_LOG) << "output: " << output;
             ClVarNode *node = CreateNode(output);
             analysis_nodes_.push(node);
           }
@@ -100,7 +101,7 @@ void MemoryOptPassCl::operator()(
           if (!IsPersistable(input) &&
               std::find(exclude_var_names.begin(), exclude_var_names.end(),
                         input) == exclude_var_names.end()) {
-            DLOG << "input: " << input;
+            LOG(kNO_LOG) << "input: " << input;
             ClVarNode *node = CreateNode(input);
             analysis_nodes_.push(node);
             if (op->Type() == "fetch") {
@@ -114,7 +115,7 @@ void MemoryOptPassCl::operator()(
           if (!IsPersistable(output) &&
               std::find(exclude_var_names.begin(), exclude_var_names.end(),
                         output) == exclude_var_names.end()) {
-            DLOG << "output: " << output;
+            LOG(kNO_LOG) << "output: " << output;
             ClVarNode *node = CreateNode(output);
             analysis_nodes_.push(node);
           }
@@ -164,8 +165,8 @@ void MemoryOptPassCl::ShareData(
   cl_command_queue command_queue = scope->GetCLScpoe()->CommandQueue();
 
   for (const auto &list : reused_nodes_) {
-    DLOG << "\n";
-    DLOG << "gpu . share memory within these variables";
+    LOG(kNO_LOG) << "\n";
+    LOG(kNO_LOG) << "gpu . share memory within these variables";
     int64_t x_based_max_numl = -1;
     int64_t y_based_max_numl = -1;
     int64_t x_based_max_x = -1;