From 67fcc7cf02b07dd962e0773877c0631746393674 Mon Sep 17 00:00:00 2001
From: cyy <cyyever@outlook.com>
Date: Tue, 28 Jan 2025 04:28:18 +0000
Subject: [PATCH] [3/N] Remove unnecessary once flag usage (#145672)

Fixes #ISSUE_NUMBER

Pull Request resolved: https://github.com/pytorch/pytorch/pull/145672
Approved by: https://github.com/albanD
---
 c10/core/impl/GPUTrace.cpp        | 7 +++----
 c10/cuda/CUDACachingAllocator.cpp | 7 +++----
 c10/util/Logging.cpp              | 7 +++----
 3 files changed, 9 insertions(+), 12 deletions(-)

diff --git a/c10/core/impl/GPUTrace.cpp b/c10/core/impl/GPUTrace.cpp
index ac1f08d7e09..9a4bab6b64a 100644
--- a/c10/core/impl/GPUTrace.cpp
+++ b/c10/core/impl/GPUTrace.cpp
@@ -1,5 +1,4 @@
 #include <c10/core/impl/GPUTrace.h>
-#include <c10/util/CallOnce.h>
 
 namespace c10::impl {
 
@@ -8,11 +7,11 @@ std::atomic<const PyInterpreter*> GPUTrace::gpuTraceState{nullptr};
 bool GPUTrace::haveState{false};
 
 void GPUTrace::set_trace(const PyInterpreter* trace) {
-  static c10::once_flag flag;
-  c10::call_once(flag, [&]() {
+  static bool once_flag [[maybe_unused]] = [&]() {
     gpuTraceState.store(trace, std::memory_order_release);
     haveState = true;
-  });
+    return true;
+  }();
 }
 
 } // namespace c10::impl
diff --git a/c10/cuda/CUDACachingAllocator.cpp b/c10/cuda/CUDACachingAllocator.cpp
index 74ac1d0ff36..a84d79dd245 100644
--- a/c10/cuda/CUDACachingAllocator.cpp
+++ b/c10/cuda/CUDACachingAllocator.cpp
@@ -5,7 +5,6 @@
 #include <c10/cuda/CUDAException.h>
 #include <c10/cuda/CUDAFunctions.h>
 #include <c10/cuda/CUDAGuard.h>
-#include <c10/util/CallOnce.h>
 #include <c10/util/Gauge.h>
 #include <c10/util/ScopeExit.h>
 #include <c10/util/UniqueVoidPtr.h>
@@ -978,10 +977,10 @@ static std::string reportProcessMemoryInfo(c10::DeviceIndex device) {
   if (!nvml_handle) {
     return "";
   }
-  static c10::once_flag nvml_init;
-  c10::call_once(nvml_init, [] {
+  static bool nvml_init [[maybe_unused]] = []() {
     TORCH_INTERNAL_ASSERT(NVML_SUCCESS == DriverAPI::get()->nvmlInit_v2_());
-  });
+    return true;
+  }();
 
   cudaDeviceProp prop{};
   C10_CUDA_CHECK(cudaGetDeviceProperties(&prop, device));
diff --git a/c10/util/Logging.cpp b/c10/util/Logging.cpp
index 15d00b94422..657fea76dcd 100644
--- a/c10/util/Logging.cpp
+++ b/c10/util/Logging.cpp
@@ -1,5 +1,4 @@
 #include <c10/util/Backtrace.h>
-#include <c10/util/CallOnce.h>
 #include <c10/util/Flags.h>
 #include <c10/util/Lazy.h>
 #include <c10/util/Logging.h>
@@ -161,8 +160,7 @@ void InitEventSampledHandlers(
     std::vector<
         std::pair<std::string_view, std::unique_ptr<EventSampledHandler>>>
         handlers) {
-  static c10::once_flag flag;
-  c10::call_once(flag, [&]() {
+  static bool flag [[maybe_unused]] = [&]() {
     auto& registry = EventSampledHandlerRegistry();
     for (auto& [event, handler] : handlers) {
       auto entry = registry.find(std::string{event});
@@ -171,7 +169,8 @@ void InitEventSampledHandlers(
       }
       entry->second = std::move(handler);
     }
-  });
+    return true;
+  }();
 }
 
 const std::unique_ptr<EventSampledHandler>& GetEventSampledHandler(