I’m running cusarseLt library in nvidia cuda library example in github(CUDALibrarySamples/spmma_example.cpp at master · NVIDIA/CUDALibrarySamples · GitHub). when I set m = 1024, n = 1024, k = 1024 it compile well.
But when I excute the code, my computer said it’s core dump. Why Error is made by m = 1024 n = 1024 k = 1024. How could I solve this problem?
What error are you getting? if you just change the size of the problem, the example will run in Segmentation fault as the arrays are on the stack memory and their size is limited.
but it is just 1024x1024 three matrix in my stack memory. it is just few mega byte. Am i supposed to use sppma2 code(CUDALibrarySamples/spmma2_example.cpp at master · NVIDIA/CUDALibrarySamples · GitHub) to solve this problem?
yes, even a few MBs can be larger than the stack memory. It is 8 MB on Linux in general. Anyway, the second example uses dynamic memory, so you will not have this issue