I am trying the new CUDA 9 release candidate and am getting some new errors that I did not encounter with previous versions. I am trying to do a grid synchronization using cooperative groups on a Pascal TITAN Xp GPU.
I am implementing the grid sync like so
cooperative_groups::grid_group grid = cooperative_groups::this_grid(); cooperative_groups::sync(grid);
Compiling gives me the error
ptxas fatal : Unresolved extern function 'cudaCGGetIntrinsicHandle'
I tried adding the -dc flag to my compile line which instead gives me the error
lib/myLib.so: undefined reference to `__cudaRegisterLinkedBinary_51_tmpxft_00006f11_00000000_6_myKernel_cpp1_ii_6f9e48ae'
If I comment out the sync lines, and don’t use the -dc flag everything compiles fine. Has anyone seen this issue or know how to go about fixing it?