pub fn profile_gpu(
runnable: &Arc<dyn Runnable>,
bench_limits: &BenchLimits,
sub_matches: &ArgMatches,
dg: &mut Annotations,
inputs: &RunTensors,
before_node: &dyn Fn(usize),
after_iteration: &dyn Fn(&mut Annotations, &[(usize, String)]) -> TractResult<()>,
) -> TractResult<()>