How to run attack

Terminal 1: (keep core 28 fully awake, alternatively disable C1e and C-states)
cd ~/ECE792/Final/stress_hyperthread
taskset -c 84 ./busyloop

Terminal 2:
cd build

Terminal 3:
cd amx_runner

In terminal 3, run taskset -c 28 ./main
In terminal 2, run taskset -c 28 ./infer 0 0 and taskset -c 28 ./infer 1 0
   This doesn't early exit and early exits, respectively. The second 0 is whether or not to warm up AMX, I found it doesn't make much of a diff

Terminal 3 should output roughly 6% when early exiting and 12% when running the full model, allowing us to derive intermediate model entropy which can help in leaking the class label according to Akinsanya et al. - Timing Channels in Adaptive Neural Networks