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
First step in improving the shuffling situation.
Right now we have to all of the time go and recheck all of the intrinsics for every new shuffle we add.
No guarantees we didn't miss anything, no way to prioritise intrinsics.
The first step to improve our situation is a perfect_shuffle function.
A perfect_shuffle is a shuffle in 0 (like a cast) or 1 instruction that does not need any extra masks.
If the shuffle is not supported, perfect shuffle should fail in a sfinae friendly way.
I also think it would be helpful to introduce a byte_pattern class:
First step in improving the shuffling situation.
Right now we have to all of the time go and recheck all of the intrinsics for every new shuffle we add.
No guarantees we didn't miss anything, no way to prioritise intrinsics.
The first step to improve our situation is a
perfect_shuffle
function.A
perfect_shuffle
is ashuffle
in 0 (like a cast) or 1 instruction that does not need any extra masks.If the shuffle is not supported, perfect shuffle should fail in a sfinae friendly way.
I also think it would be helpful to introduce a
byte_pattern
class:-1 could indicate a 0.
-2 could indicate - we don't care.
This will result in less template instantiations I think.
The text was updated successfully, but these errors were encountered: