On Intel systems, how would I measure CPU cache misses and pipeline stalls and use them in a formula to determine if hyperthreading should be on or off for my workload?
I have benchmarked and reliably get a 12% improvement with HT on but I want to gain a better theoretical framework.