Just working in CUDA 10.1 and I’ve found that whilst in a function that has warp primitives, there are no revealed statistics. I’ve included a screenshot to make clear what I think is unusual.
Would love a hand - I’m new to this software :)
Sorry - some useful information might be that the version of ncompute is 2019.3 and I’m running on a Turing architecture (with a 2080 Ti)
For which lines metrics are available in the CUDA-C view depends on the correlated assembly instructions in the SASS view. You should check if the marked line has any correlated (highlighted) ones in the corresponding SASS. You can see that by changing to the combined CUDA-C/SASS view. If it does not, the compiler didn’t provide any correlation information for this high-level CUDA-C code.