Some thread Block in kernel function can't stop normally

[OOOOOmoi]

I was using GeNN 4. 9 and tried to simulate two cortical columns to observe the flow of information, but I found that the program did not finish executing properly when I ran the initializeSparse() function, I made a slight change to the generated code and at the end of the initializeSparseKernel() kernel function, I added the following Code:

 __syncthreads();
if(threadIdx.x==0){
      int finished = atomicAdd(&finishedBlockCounter, 1);
      printf("initializeSparseKernel %d-th block finished. Total finished: %d\n", blockIdx.x, finished + 1);
}

Because __syncthreads() only guarantees synchronization within each thread block, it doesn't guarantee synchronization of all threads all together, so I calculated the number of completed thread blocks within this kernel function using finishedBlockCounter, and I found that out of 2200+ thread blocks, only 1600+ completed the computation, and the rest of the threads didn't finish properly, regarding This problem, does anyone have a clue, also, my code is in the repository: https: //github. com/OOOOOOmoi/DoubleColumnOfGeNN, I made some changes to the script genn-create-user-project. sh that generates the makefile to enable the use of customized The new script, genn-create-user-project-record.sh, is also in the repository, so if you need to run the code, please use this script.

[neworderofjamie]

Hmm that is a strange one. I don't have access to that repository but all that should be happening in initializeSparse is initialization of synaptic weights and delays. If you're using the built in "NormalClipped" and "NormalClippedDelay" initialisers, there is a possibility for an infinite loop/very slow initialisation if "min", "max", "sd" and "mean" are set to weird values - might be worth printing them out to check your file parsing isn't going wrong in some way.

[neworderofjamie]

Also, always worth trying to run your model with compute sanitizer to check if there are any indexing bugs i.e.

compute-sanitizer python your_model.py

if you're using PyGeNN, otherwise:

compute-sanitizer ./simulator

[OOOOOmoi]

OK，I fix the link of the repository ,I'll try the compute sanitizer to check my program

[neworderofjamie]

#669 turned out to be the same issue, I suspect the problem is the same here