Skip to content

Can't detect any instruction #1

@Andre4102

Description

@Andre4102

I wrote a Cuda implementation for the convolution in order to do fault injection on a CNN but the injector does not detect any instruction. I use Pytorch C++ as a framework.
When I open the file stdout.txt in the directory where I store the executable, I get the following output:
inspecting forward_cuda_kernel(at::GenericPackedTensorAccessor<float, 4ul, at::RestrictPtrTraits, int>, at::GenericPackedTensorAccessor<float, 4ul, at::RestrictPtrTraits, int>, at::GenericPackedTensorAccessor<float, 1ul, at::RestrictPtrTraits, int>, at::GenericPackedTensorAccessor<float, 4ul, at::RestrictPtrTraits, int>, int, int, int, int, int, int) - num instrs 456
and a very long list of instructions (more than 1300) terminating with
NVBit-igprofile; ERROR FAIL in kernel execution!!
I was wondering what could be the meaning of this and how to solve it.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions