[OpenMP][libomptarget] Init device when printing device info

[kevinsala]

Summary

Make sure the device is ready and initialized when printing the device information

Fixes #61392

[kevinsala]

This patch is still missing a test.

[josemonsalve2]

Will this solve the problem or mitigate the error?

[jdoerfert]

In D146081#4194588, @josemonsalve2 wrote:

Will this solve the problem or mitigate the error?

I also fear this will just make the script not crash but also useless.
That said, we probably want this regardless and then a patch to initialize stuff in the omp-device-info executable.
Eventually one test will be enough, given we don't have unit tests for this stuff yet.

[kevinsala]

Explicitly initializing devices from llvm-omp-device-info tool.

[kevinsala]

The initial patch should fix the crash for the llvm-omp-device-info case, or any other case calling __tgt_print_device_info. I saw lazy initialization of devices in other API calls and tried to reproduced the same behavior. Anyway, initializing them explicitly from the tool seems more clean, so I just updated the patch with it.

Regarding the usefulness of this tool, it's true that it does not print anything relevant since nextgen plugins are not printing any information for the moment. I'm going to open a new patch to print at least the information reported by old plugins.

[kevinsala]

There is also the option to auto initialize the devices at the plugin level when GenericPluginTy::getDevice(N) is called and device N is in a uninitialized state. In my opinion, that would be the safest option and would cover all cases. But it will require implementing a similar mechanism from libomptarget level to guarantee the device creation and initialization is tread-safe and serialized (i.e., std::call_once in libomptarget).

[jdoerfert]

openmp/libomptarget/src/interface.cpp

61

Isn't this the same code as in __tgt_print_device_info? Why do we need this?

[kevinsala]

openmp/libomptarget/src/interface.cpp

61

From previous comments I understood it was better to explicitly initialize devices from the llvm-omp-device-info. Is it preferable this explicit initialization or a lazy initialization by __tgt_print_device_info?

[kevinsala]

Updating patch to initial version. We make sure in libomptarget that the device is initialized (ready) when printing its information. Together with https://reviews.llvm.org/D148178, it should fix the llvm-omp-device-info command.

openmp/libomptarget/src/interface.cpp

 EXTERN void __tgt_init_all_rtls() { PM->RTLs.initAllRTLs(); }
 
 ////////////////////////////////////////////////////////////////////////////////
 /// unloads a target shared library
 EXTERN void __tgt_unregister_lib(__tgt_bin_desc *Desc) {
   TIMESCOPE();
   PM->RTLs.unregisterLib(Desc);
   for (auto &RTL : PM->RTLs.UsedRTLs) {
     if (RTL->unregister_lib) {
       if ((*RTL->unregister_lib)(Desc) != OFFLOAD_SUCCESS) {
         DP("Could not register library with %s", RTL->RTLName.c_str());
       }
     }
   }
 }
 
 template <typename TargetAsyncInfoTy>
EXTERN void __tgt_set_info_flag(uint32_t NewInfoLevel) {
   InfoLevel.store(NewInfoLevel);
   for (auto &R : PM->RTLs.AllRTLs) {
     if (R.set_info_flag)
       R.set_info_flag(NewInfoLevel);
   }
 }
 
 EXTERN int __tgt_print_device_info(int64_t DeviceId) {
+  // Make sure the device is ready.
+  if (!deviceIsReady(DeviceId)) {
+    DP("Device %" PRId64 " is not ready\n", DeviceId);
+    return OMP_TGT_FAIL;
+  }
+
   return PM->Devices[DeviceId]->printDeviceInfo(
       PM->Devices[DeviceId]->RTLDeviceID);
 }
 
 EXTERN void __tgt_target_nowait_query(void **AsyncHandle) {
   if (!AsyncHandle || !*AsyncHandle) {
     FATAL_MESSAGE0(
         1, "Receive an invalid async handle from the current OpenMP task. Is "