Pete's Log: doh!
Entry #738, (Coding, Hacking, & CS stuff)(posted when I was 22 years old.)
hmmm... what do you do when you write a branch predictor that performs horribly? I mean, it does worse then simply assuming all branches are not taken or randomly guessing taken or not. Well, one option, I suppose, is to do the opposite of what your predictor tells you to. Reverse psychology in computer architecture. But seriously, branch prediction accuracy of sub 15% is pretty bad. Especially seeing how as you give the predictor more storage, accuracy goes down!
but let's see ... other fun predictor ideas... opcode based. different ways of storing histories... target based? hmmm... wish these benchmarks ran faster...