/sci/ - Science & Math

>>11586260
In regards to computer architecture and branch prediction, was there any thought or research into executing both branches and discarding the incorrect results in a processor? I'm sure someone put some thought into it but I can't find anything about it. My immediate thoughts are that at 100% usage, the CPU performance would degrade due to essentially lengthening your program, but at lower usages, maybe 30%, you would received faster performance due to no mispredicted branch stalls. There would be increased hardware associated with computing both branches, but you would get rid of the prediction hardware.