tensorrt version : v5.0.6
hardware platform: jetson xavier
problem: follow instruction, IExecutionContext::enqueue : asynchronously execute inference on a batch
howerver, but when I execute IExecutionContext::enqueue, I fonnd it consumes more than 23ms.
then, I make a funny test, replace IExecutionContext::enqueue with IExecutionContext::execute.
I fonnd it consumes more than 26ms.
In fact, enqueue is an async interface, it should return quickly when called, why it consume more than 23ms， just less 3ms, compared to sync interface。why？