Trace level env (#39926)

* add align for WorkQueue

* add spinlock

* merge develop

* merge

* Add EventsWaiter

* Revert "Add EventsWaiter"

This reverts commit e206173aa9be7401b83a53581627bfaf557c8fb2.

* Add host_trace_level env variable

* Revert "Optimize perf of softmax_with_cross_entropy (#39553)"

This reverts commit bbe5228c.
Co-authored-by: Nliutiexing <liutiexing@google.com>
Co-authored-by: NZzSean <18818272991@163.com>

paddle/fluid/framework/new_executor/interpretercore.cc

@@ -389,7 +389,7 @@ void InterpreterCore::RunInstruction(const Instruction& instr_node) {
   auto op_with_kernel = dynamic_cast<const framework::OperatorWithKernel*>(op);
   {
     platform::RecordEvent infershape_event(
-        "InferShape", platform::TracerEventType::OperatorInner, 1,
+        "infer_shape", platform::TracerEventType::OperatorInner, 1,
         platform::EventRole::kInnerOp);
     // If it is OperatorBase, InferShape do nothing.
     if (op_with_kernel != nullptr)
@@ -411,7 +411,7 @@ void InterpreterCore::RunInstruction(const Instruction& instr_node) {
   }
   {
     platform::RecordEvent compute_event(
-        "Compute", platform::TracerEventType::OperatorInner, 1,
+        "compute", platform::TracerEventType::OperatorInner, 1,
         platform::EventRole::kInnerOp);
     if (op_with_kernel == nullptr) {
       instr_node.OpBase()->Run(*local_scope, place_);
@@ -561,7 +561,8 @@ void InterpreterCore::RunInstructionAsync(size_t instr_id) {
             << " runs on " << platform::GetCurrentThreadName();
 
     auto* op = instr_node.OpBase();
-    platform::RecordEvent instruction_event(op->Type().c_str());
+    platform::RecordEvent instruction_event(
+        op->Type(), platform::TracerEventType::Operator, 1);
     interpreter::WaitEvent(instr_node, place_);
 
     try {

paddle/fluid/imperative/tracer.cc

@@ -175,7 +175,7 @@ void Tracer::TraceOp(const std::string& type, const NameVarMap<VarType>& ins,
                      paddle::framework::AttributeMap* passed_default_attrs_,
                      bool use_default_attr_map) {
   platform::RecordEvent op_type_record_event(
-      type, platform::TracerEventType::Operator, 2);
+      type, platform::TracerEventType::Operator, 1);
   platform::ScopedFlushDenormal flush;
   VLOG(1) << "Trace Op: " << type;
   if (FLAGS_use_mkldnn) {

paddle/fluid/platform/profiler/host_tracer.cc

@@ -14,9 +14,16 @@
 
 #include "paddle/fluid/platform/profiler/host_tracer.h"
 #include "glog/logging.h"
+#include "paddle/fluid/platform/flags.h"
 #include "paddle/fluid/platform/profiler/common_event.h"
 #include "paddle/fluid/platform/profiler/host_event_recorder.h"
 
+// Used to filter events, works like glog VLOG(level).
+// RecordEvent will works if host_trace_level >= level.
+PADDLE_DEFINE_EXPORTED_int64(host_trace_level, 2,
+                             "RecordEvent will works "
+                             "if host_trace_level >= level.");
+
 namespace paddle {
 namespace platform {
 

paddle/fluid/platform/profiler/profiler.h

@@ -23,11 +23,13 @@
 #include "paddle/fluid/platform/profiler/event_node.h"
 #include "paddle/fluid/platform/profiler/tracer_base.h"
 
+DECLARE_int64(host_trace_level);
+
 namespace paddle {
 namespace platform {
 
 struct ProfilerOptions {
-  uint32_t trace_level = 0;
+  uint32_t trace_level = FLAGS_host_trace_level;
 };
 
 class Profiler {