Better cap for sat_benchs' IPC
The IPC of sat_benchs
is fixed, therefor the IPC of S^(4*IPC(S)/IPC(I) I
can be capped more accurately. This would avoid non-feasibly aux LP (e.g. when no edge is present, and the measured IPC is too high compared to the maximum reachable IPC).
Edited by DERUMIGNY Nicolas