[submodule] Pin fmtlib/fmt to 10.1.0 (pytorch#106672)

fmt10.1.0 fixes a bug of format_string_checker initialisation order which is important to our improved clang-tidy checks pytorch#103058. This PR upgrades third_party fmt to 10.1.0, in the meanwhile, kineto is also upgraded to avoid fmt errors.

Pull Request resolved: pytorch#106672
Approved by: https://github.com/Skylion007

aten/src/ATen/mps/MPSProfiler.mm

@@ -504,7 +504,7 @@
                opInfo->runCount,
                fmt::format("{:.3f}", opInfo->totalSchedulingTime / double(opInfo->runCount)),
                fmt::format("{:.3f}", opInfo->totalGpuTime / double(opInfo->runCount)),
-               fmt::format("{:.3f}", opInfo->totalGpuTime),
+               fmt::format("{:.3f}", opInfo->totalGpuTime.load()),
                opInfo->strKey);
   }
 }
@@ -556,7 +556,7 @@
                cpuFbInfo->profileId,
                cpuFbInfo->runCount,
                fmt::format("{:.3f}", cpuFbInfo->totalSchedulingTime / double(cpuFbInfo->runCount)),
-               fmt::format("{:.3f}", cpuFbInfo->totalSchedulingTime),
+               fmt::format("{:.3f}", cpuFbInfo->totalSchedulingTime.load()),
                getIMPSAllocator()->formatSize(cpuFbInfo->totalCopyOverhead),
                cpuFbInfo->opName);
   }
@@ -607,8 +607,8 @@
           copyStat.kindStr,
           copyStat.totalCount,
           getIMPSAllocator()->formatSize(copyStat.length),
-          fmt::format("{:.3f}", copyStat.totalSchedulingTime),
-          fmt::format("{:.3f}", copyStat.totalGpuTime),
+          fmt::format("{:.3f}", copyStat.totalSchedulingTime.load()),
+          fmt::format("{:.3f}", copyStat.totalGpuTime.load()),
           copyStat.scalarsCount,
           fmt::format("{:.2f} %",
                       copyStat.totalGpuTime > 0.0

torch/csrc/distributed/rpc/agent_utils.cpp

@@ -156,13 +156,17 @@ static std::atomic<int> barrierId(0);
 
 static std::tuple<std::string, std::string, std::string> getNextKeyIds() {
   barrierId++;
-  std::string processCountKey =
-      fmt::format("{}{}{}", storeKeyProcessCount, storeKeyBarrierId, barrierId);
+  auto newBarrierId = barrierId.load();
+  std::string processCountKey = fmt::format(
+      "{}{}{}", storeKeyProcessCount, storeKeyBarrierId, newBarrierId);
   std::string activeCallCountKey = fmt::format(
-      "{}{}{}", storeKeyActiveCallCount, storeKeyBarrierId, barrierId);
+      "{}{}{}", storeKeyActiveCallCount, storeKeyBarrierId, newBarrierId);
   std::string barrierKey =
-      fmt::format("{}{}{}", storeKeyReady, storeKeyBarrierId, barrierId);
-  return std::make_tuple(processCountKey, activeCallCountKey, barrierKey);
+      fmt::format("{}{}{}", storeKeyReady, storeKeyBarrierId, newBarrierId);
+  return std::make_tuple(
+      std::move(processCountKey),
+      std::move(activeCallCountKey),
+      std::move(barrierKey));
 }
 
 // Synchronize process with all other agent processes strictly using store