When I used the ltrace command to trace the PyTorch training process of ResNet50, I encountered 3 API-calls that were not listed in the NVIDIA documentation. So what these API-calls are for?
NVIDIA-DOC link:CUDA cuBLAS
ltrace command: ltrace -f -x ‘cu*’ -x ‘__cu*’ -n 4 -o outputfile.txt -tt python resnet50.py
3 API-calls:
cublasLtCtxInit
cublasLtSSSMatmul
cublasLtSSSMatmulAlgoGetHeuristic