How to debug cudaSignalExternalSemaphoresAsync having invalid value?

[Gatchan]

-		externalSemaphoreSignalParams	{params={fence={value=6 } nvSciSync={fence=0x0000000000000000 {…} reserved=0 } keyedMutex={key=0 } …} …}	cudaExternalSemaphoreSignalParams
-		params	{fence={value=6 } nvSciSync={fence=0x0000000000000000 reserved=0 } keyedMutex={key=0 } …}	cudaExternalSemaphoreSignalParams::
+		fence	{value=6 }	cudaExternalSemaphoreSignalParams::::
-		nvSciSync	{fence=0x0000000000000000 reserved=0 }	cudaExternalSemaphoreSignalParams::::
		fence	0x0000000000000000	void *
		reserved	0	unsigned __int64
-		keyedMutex	{key=0 }	cudaExternalSemaphoreSignalParams::::
		key	0	unsigned __int64
+		reserved	0x000000aac79fca58 {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}	unsigned int[12]
		flags	0	unsigned int
+		reserved	0x000000aac79fca8c {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}	unsigned int[16]

This is the loc: CudaD3D12Update/simpleD3D12update.cpp at cd59b7d6eb635a564df88c2528f42f5a8a9a5ba0 · mprevot/CudaD3D12Update · GitHub

(I actually have a different kernel, which is triggered by a c# WPF app, but the rest is the same.)

Is there a usual way to debug such problem ? In my case, I most of the time do not have invalid value, so I am surprised to get this sometimes.

Driver is 566.26, cuda is 12.6.3, OS is W10 19045.5247

[veraj]

Hi, @Gatchan
Sorry for the issue you met. Can you provide the step to repro this ?
Also please clarify which CUDA/GPU/Driver do you use ?

[Gatchan]

Thanks. I updated the OP with details. Aren’t the values provided sufficient ? why not ?

[pvakeely]

Hello Gatchan and thank you for bringing this issue to our attention. I’m trying to understand if your question is regarding the API or the debugger. In other words:

• If you are wondering what the correct values for the members of externalSemaphoreSignalParams should be, then that is an API question and I can help route your question to the relevant team.
• But if you believe you have values in the member that the debugger is showing incorrectly, then that’s a question for us, the debugger team.
  From what I can tell, it seems like the first case. Can you please confirm so I can pull in the right time for further help?

Thank you
EP

[Gatchan]

Many thanks for those opportunities.
It appears that the wrong values come from a previous kernel.
I am now in familiar grounds.

Many thanks again EP.

[pvakeely]

Our pleasure! 🙂 As always, we look forward to hearing from you how to further improve our products!

Kind regards from NVIDIA!