Fix destructor Seg Faults for SYCL

[vikaskurapati]

On SuperMUC Phase 2, there were two double-free or Invalid Pointer Access errors, which caused the program to exit with a Segmentation fault (SIGSEGV). The issue was in freeMem method. It was crashing when we tried to destruct the map this->currentMemoryToSizeMap if it had already been destructed in another destructor. I modified the checks to return when we encounter empty objects.

[davschneller]

Generally LGTM (IMO almost merge-ready); but might it be possible to get a more exact picture of what's going on?
I.e. the probably statistics get deleted—and there's still memory to be de-allocated.
Maybe we could move the deletion of the statistics to the destructor?

[davschneller]

Add brackets

[davschneller]

use the current device ID

[vikaskurapati]

Generally LGTM (IMO almost merge-ready); but might it be possible to get a more exact picture of what's going on? I.e. the probably statistics get deleted—and there's still memory to be de-allocated. Maybe we could move the deletion of the statistics to the destructor?

On SuperMUC NG phase 2, I was running a few cases to study scaling, but it always ended with a Segfault. The backtrace looked like this

Thread 1 "SeisSol_Release" received signal SIGSEGV, Segmentation fault.

0x00001555491512be in device::ConcreteAPI::freeMem(void*) () from /dss/dsshome1/00/ge26cet3/seissol/build-pvc/submodules/Device/libdevice_Release_oneapi.so

#0  0x00001555491512be in device::ConcreteAPI::freeMem(void*) () from /dss/dsshome1/00/ge26cet3/seissol/build-pvc/submodules/Device/libdevice_Release_oneapi.so

#1  0x00000000004aea54 in seissol::writer::EnergyOutput::~EnergyOutput() ()

#2  0x000000000044b505 in seissol::SeisSol::~SeisSol() ()

#3  0x00000000004341aa in main ()

I thought that the currentMemoryToSizeMap was being destroyed and then being used again in freeMem method. I tried adding a check for that too in the if(devPtr != nullptr) condition, but then CI tests involving acpp complained that it wasn't a pointer, and I had to modify it to what it is now to get it fixed.

Probably, you are right, the API is still new to me, if you think that is a better solution, we could try that.