Efficient-Scalable-Machine-Learning/EvNN

One Error of "make evnn_pytorch"

Closed this issue · 4 comments

iCGY96 commented

image

KhaleelKhan commented

Could you please provide more details such as the CUDA toolkit version and the GPU compute capability you are using?
Earlier versions of CUDA arch don't support half ops, but the directive at inline_ops.h(120) should take care of this.
#if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 600)

iCGY96 commented

image

iCGY96 commented

The above issues were resolved by upgrading to CUDA 11.

  • 👍1
KhaleelKhan commented

Thank you for raising this issue.
CUDA toolkit v 10.1 doesn't have __habs operation (See: cuda 10.1)

Can you please try using v10.2+

  • ❤️1