"unify flags" (#7973)

* "unify flags"

* "fix init"

paddle/framework/executor.cc

@@ -25,7 +25,7 @@ limitations under the License. */
 #include "paddle/platform/place.h"
 #include "paddle/platform/profiler.h"
 
-DECLARE_bool(do_memory_benchmark);
+DECLARE_bool(benchmark);
 DEFINE_bool(check_nan_inf, false,
             "Checking whether operator produce NAN/INF or not. It will be "
             "extremely slow so please use this flag wisely.");
@@ -125,7 +125,7 @@ void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
 
     op->Run(*local_scope, place_);
     VLOG(3) << op->DebugStringEx(local_scope);
-    if (FLAGS_do_memory_benchmark) {
+    if (FLAGS_benchmark) {
       VLOG(2) << "Memory used after operator " + op->Type() + " running: "
               << memory::memory_usage(place_);
     }
@@ -142,7 +142,7 @@ void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id,
   if (create_vars && create_local_scope) {
     scope->DeleteScope(local_scope);
   }
-  if (FLAGS_do_memory_benchmark) {
+  if (FLAGS_benchmark) {
     VLOG(2) << "-------------------------------------------------------";
     VLOG(2) << "Memory used after deleting local scope: "
             << memory::memory_usage(place_);

paddle/framework/operator.cc

@@ -22,9 +22,7 @@ limitations under the License. */
 #include "paddle/framework/shape_inference.h"
 #include "paddle/framework/var_type.h"
 
-DEFINE_bool(op_sync, false,
-            "Default cuda is asynchronous device, set to True will"
-            "force op run in synchronous mode.");
+DECLARE_bool(benchmark);
 
 namespace paddle {
 namespace framework {
@@ -531,7 +529,7 @@ void OperatorWithKernel::Run(const Scope& scope,
       ExecutionContext(*this, new_scope, *new_dev_ctx));
 
   /*For profiling/benchmark only*/
-  if (FLAGS_op_sync) {
+  if (FLAGS_benchmark) {
     new_dev_ctx->Wait();
   }
 }

paddle/framework/scope.cc

@@ -20,9 +20,11 @@ limitations under the License. */
 #include "paddle/framework/threadpool.h"
 #include "paddle/string/printf.h"
 
-DEFINE_bool(do_memory_benchmark, false,
+DEFINE_bool(benchmark, false,
             "Doing memory benchmark. It will make deleting scope synchronized, "
-            "and add some memory usage logs");
+            "and add some memory usage logs."
+            "Default cuda is asynchronous device, set to True will"
+            "force op run in synchronous mode.");
 
 namespace paddle {
 namespace framework {
@@ -93,7 +95,7 @@ void Scope::DeleteScope(Scope* scope) {
   PADDLE_ENFORCE(it != this->kids_.end(), "Cannot find %p as kid scope", scope);
   this->kids_.erase(it);
   // When making memory benchmark on Fluid, we have to delete scope sync.
-  if (FLAGS_do_memory_benchmark) {
+  if (FLAGS_benchmark) {
     delete scope;
   } else {
     Async([scope] { delete scope; });

python/paddle/v2/fluid/__init__.py

@@ -86,11 +86,9 @@ def __bootstrap__():
 
     os.environ['OMP_NUM_THREADS'] = str(num_threads)
 
-    read_env_flags = [
-        'use_pinned_memory', 'check_nan_inf', 'do_memory_benchmark'
-    ]
+    read_env_flags = ['use_pinned_memory', 'check_nan_inf', 'benchmark']
     if core.is_compiled_with_cuda():
-        read_env_flags += ['fraction_of_gpu_memory_to_use', 'op_sync']
+        read_env_flags += ['fraction_of_gpu_memory_to_use']
     core.init_gflags([sys.argv[0]] +
                      ["--tryfromenv=" + ",".join(read_env_flags)])
     core.init_glog(sys.argv[0])