Thanks. I think I’ll have to change how VectorizationBase builds and defines flags, so that it can be 16 on 1.4.2 and 32 with Julia 1.6.
That said, 16 on both should work and at least shouldn’t hurt performance of 1.6 relative to 1.4.
So I believe the issue is something else.
Mind providing example code I can run to test the solve regression and see if I can reproduce it locally?