Pretty interesting work, liked the companion paper, too! (https://eecs.oregonstate.edu/aidarc/paper/CNNBranchPrediction.pdf …) Offline training could also enable revisiting a wider range of ideas from the adaptive runtime systems (Self, Smalltalk, http://bibliography.selflanguage.org/ ; C++ context back in '94, http://www.cs.cornell.edu/courses/cs612/2000SP/papers/calder-overhead-in-C++.pdf …).
-
-
-
Considering the training dataset (global history data mentioned in the companion paper), there's also a potential for cross-pollination of ideas across disciplines (e.g., temporal nature of branch pred. data makes me think of time series competitions, https://robjhyndman.com/publications/forecasting-competitions/ …).
Kraj razgovora
Novi razgovor -
-
-
Is there any information about the kind of branch predication that is used in SKL or SNC?
-
Everything points to TAGE-SC-L but with either smaller stores, or sloppier implementation, than Apple. Apple prob also uses TAGE for indirect branches, less clear for Intel.
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.