diff --git a/paddle/fluid/platform/profiler.cc b/paddle/fluid/platform/profiler.cc index 50bc0aba6aa0f056dc0b2d49f6b3b745433e0756..2fb5c6dc6b8ad25fa1ad5fcf7c2acfedd5be4a83 100644 --- a/paddle/fluid/platform/profiler.cc +++ b/paddle/fluid/platform/profiler.cc @@ -173,8 +173,9 @@ void PopEvent(const std::string& name, const DeviceContext* dev_ctx) { } RecordEvent::RecordEvent(const std::string& name, const DeviceContext* dev_ctx) - : start_ns_(PosixInNsec()) { + : is_enabled_(false), start_ns_(PosixInNsec()) { if (g_state == ProfilerState::kDisabled) return; + is_enabled_ = true; dev_ctx_ = dev_ctx; name_ = name; PushEvent(name_, dev_ctx_); @@ -183,7 +184,7 @@ RecordEvent::RecordEvent(const std::string& name, const DeviceContext* dev_ctx) } RecordEvent::~RecordEvent() { - if (g_state == ProfilerState::kDisabled) return; + if (g_state == ProfilerState::kDisabled || !is_enabled_) return; DeviceTracer* tracer = GetDeviceTracer(); if (tracer) { tracer->AddCPURecords(CurAnnotation(), start_ns_, PosixInNsec(), @@ -193,14 +194,16 @@ RecordEvent::~RecordEvent() { PopEvent(name_, dev_ctx_); } -RecordBlock::RecordBlock(int block_id) : start_ns_(PosixInNsec()) { +RecordBlock::RecordBlock(int block_id) + : is_enabled_(false), start_ns_(PosixInNsec()) { if (g_state == ProfilerState::kDisabled) return; + is_enabled_ = true; SetCurBlock(block_id); name_ = string::Sprintf("block_%d", block_id); } RecordBlock::~RecordBlock() { - if (g_state == ProfilerState::kDisabled) return; + if (g_state == ProfilerState::kDisabled || !is_enabled_) return; DeviceTracer* tracer = GetDeviceTracer(); if (tracer) { // We try to put all blocks at the same nested depth in the diff --git a/paddle/fluid/platform/profiler.h b/paddle/fluid/platform/profiler.h index 61b98143e41abb9e47d2c717c7876f1bab7f5077..643bb6183d144ec11a4890d9ea1ca970acb08b4c 100644 --- a/paddle/fluid/platform/profiler.h +++ b/paddle/fluid/platform/profiler.h @@ -74,6 +74,7 @@ struct RecordEvent { ~RecordEvent(); + bool is_enabled_; uint64_t start_ns_; // The device context is used by Event to get the current cuda stream. const DeviceContext* dev_ctx_; @@ -89,6 +90,7 @@ struct RecordBlock { ~RecordBlock(); private: + bool is_enabled_; std::string name_; uint64_t start_ns_; };