While I was in undergrad I toyed around with abusing the branch predictor on a few different machines, compiling something like the following with optimizations

While I was in undergrad I toyed around with abusing the branch predictor on a few different machines, compiling something like the following with optimizations off - it performs an identical computation regardless of branch outcome: void branchLoop(unsigned int condition, unsigned int &sum) { // put something suitably large here unsigned int loopCount = 0x0fffffff; unsigned int i; // compile with -O0 or this gets optimized away for (i = 0; i < loopCount; i++) if ((i & condition) == 0) sum++; else sum++; } The Core Duo on my Thinkpad T60 had some very distinct slowdowns on certain bit patterns, which were not repeatable on the handful of other CPUs I had access to at the time. I haven't tried this with more modern CPUs, however.