I have a question about the performance of Hexagon-sim. I am running a benchmark in hexagon-sim provided in hexagon SDK v3.5.2, and the benchmark seems to be running 20000 - 35000 times slower than on my host machine with --timing enabled. Is this normal or am I doing something very wrong?
This is for an academic research project studying the performance of software pipelining in LLVM (hexagon is the only platform where the feature is enabled).
Running the simulation on a laptop with Linux and an i7-7700HQ. The benchmark runs on my host machine in 0.173s, simulation takes an hour and one minute. Second benchmark runs on my host machine in 0.136s, simulation takes an hour and 23 minutes.
Since I am trying to count cycles, I do need the --timing feature enabled.