Originally posted by artivision
View Post
(maybe it might even become a thing to prebuffer both paths out of a branch, if they happen to fall within the same simultaneously-fetchable chunk of memory, which would A/ need, and make full use of a longer prefetch queue, so long as it could be tagged and handled properly, or if the "outer" stages of it were twice as wide as the inner ones; and B/ not suffer anything like the same massive performance hit as the P4 did from an incorrect branch prediction and subsequent queue flush...)
Comment