The default mnk seems fine. But when I tried to set m=32, n=8192, k=8192, the memory should be enough but the bin spmma_example just failed and raised Segmentation fault (core dumped). Wondering why
The n and k parameters are quite high. They can make static array overflowed: __half hB[k * n];
Can you try using dynamic allocation (using new) for the arrays?
I tried to run
spmma_example.cpp
on a A10.The default
mnk
seems fine. But when I tried to setm=32, n=8192, k=8192
, the memory should be enough but the binspmma_example
just failed and raisedSegmentation fault (core dumped)
. Wondering whyhttps://github.com/NVIDIA/CUDALibrarySamples/tree/master/cuSPARSELt/spmma