add error handling for xpu profiler (#989)

Summary: Pull Request resolved: #989 Reviewed By: sraikund16 Differential Revision: D63052607 Pulled By: briancoutinho fbshipit-source-id: f8e27e1c8a44f1bf5dd850835f82b5136008a48b
pytorch · Sep 20, 2024 · 45cc65a · 45cc65a
1 parent 3d355d1
commit 45cc65a
Show file tree

Hide file tree

Showing 2 changed files with 23 additions and 3 deletions.
diff --git a/libkineto/src/init.cpp b/libkineto/src/init.cpp
@@ -203,10 +203,15 @@ void libkineto_init(bool cpuOnly, bool logOnError) {
   libkineto::api().registerProfilerFactory([]() -> std::unique_ptr<IActivityProfiler> {
     auto returnCode = ptiViewGPULocalAvailable();
     if (returnCode != PTI_SUCCESS) {
-      std::string errCode = std::to_string(returnCode);
-      std::string errMsg(
+      std::string errPrefixMsg(
           "Fail to enable Kineto Profiler on XPU due to error code: ");
-      throw std::runtime_error(errMsg + errCode);
+      errPrefixMsg = errPrefixMsg + std::to_string(returnCode);
+#if PTI_VERSION_MAJOR > 0 || PTI_VERSION_MINOR > 9
+      std::string errMsg(ptiResultTypeToString(returnCode));
+      throw std::runtime_error(errPrefixMsg + std::string(". The detailed error message is: ") + errMsg);
+#else
+      throw std::runtime_error(errPrefixMsg);
+#endif
     }
     return std::make_unique<XPUActivityProfiler>();
   });

diff --git a/libkineto/src/plugin/xpupti/XpuptiProfilerMacros.h b/libkineto/src/plugin/xpupti/XpuptiProfilerMacros.h
@@ -11,6 +11,20 @@ namespace KINETO_NAMESPACE {
 
 using namespace libkineto;
 
+#if PTI_VERSION_MAJOR > 0 || PTI_VERSION_MINOR > 9
+#define XPUPTI_CALL(returnCode)                                                 \
+  {                                                                             \
+    if (returnCode != PTI_SUCCESS) {                                            \
+      std::string funcMsg(__func__);                                            \
+      std::string codeMsg = std::to_string(returnCode);                         \
+      std::string HeadMsg("Kineto Profiler on XPU got error from function ");   \
+      std::string Msg(". The error code is ");                                  \
+      std::string detailMsg(". The detailed error message is ");                \
+      detailMsg = detailMsg + std::string(ptiResultTypeToString(returnCode));   \
+      throw std::runtime_error(HeadMsg + funcMsg + Msg + codeMsg + detailMsg);  \
+    }                                                                           \
+  }
+#else
 #define XPUPTI_CALL(returnCode)                                               \
   {                                                                           \
     if (returnCode != PTI_SUCCESS) {                                          \
@@ -21,6 +35,7 @@ using namespace libkineto;
       throw std::runtime_error(HeadMsg + funcMsg + Msg + codeMsg);            \
     }                                                                         \
   }
+#endif
 
 class XpuptiActivityApi;
 using DeviceIndex_t = int8_t;