CODEWITHSUNDEEP
performancecudagpgpugpu
Amir
Amir

Reputation: 11

CUDA Warp Execution Efficiency

I can obtain warp_execution_efficiency for each kernel, but how can I obtain total warp_execution_efficiency of my program?

Upvotes: 0

Views: 423

Answers (1)

wirefox
wirefox

Reputation: 454

Note: Calculating a single number across all kernels is unlikely to be useful for optimization purposes.

NVIDIA's profiler is kernel-centric, so directly querying that information is impossible. You can, however, run nvprof --csv --metrics warp_execution_efficiency, then put the output into Excel, R, or your favorite statistical software and mess with it there.

One problem with this is that warp_execution_efficiency involves an average, so you can't just take the average. You might be able to get some handy visualizations that the Visual Profiler doesn't offer.

Upvotes: 3

Related Questions