You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The current feature difference between the reference interpreter, brili, and its Rust counter-part brilirs is that brilirs does not implement the speculative execution extension. The implementation semantics can probably be borrowed from brili and converted to rust. However, brilir uses iterators along basic block sequences instead of random access of the entire function which is more performant but may be an issue when trying to commit, and then return to an instruction part of the way through a code block. Use of the speculation extension is rare, so it would be nice to do this with a limited performance penalty on the longer running benchmarks.
The text was updated successfully, but these errors were encountered:
Cool. FWIW, that might be OK… the speculation extension is very squirrelly and wasn't really designed to admit an efficient implementation. 😃 The mere fact that one needs to snapshot the entire state of the world stops it from being all that amenable to fast execution. So it wouldn't be terrible to leave this in the domain of the silly/slow TS interpreter.
The current feature difference between the reference interpreter,
brili
, and its Rust counter-partbrilirs
is thatbrilirs
does not implement the speculative execution extension. The implementation semantics can probably be borrowed frombrili
and converted to rust. However, brilir uses iterators along basic block sequences instead of random access of the entire function which is more performant but may be an issue when trying to commit, and then return to an instruction part of the way through a code block. Use of the speculation extension is rare, so it would be nice to do this with a limited performance penalty on the longer running benchmarks.The text was updated successfully, but these errors were encountered: