fix nan problem, the static int approach #18
Conversation
| Octavian.matmul!(o, a, b, α, β) | ||
| end | ||
| end | ||
| end | ||
| # NOTE: benchmark shows, the type instability here can be optimized by the compiler | ||
| # so you do not need to worry about the overheads. |
There was a problem hiding this comment.
It still makes me nervous, but it is good at optimizing small unions. Most obviously, for loops all have unions.
But is this really what you want?
That alpha isn't allowed to be anything other than 0 or 1?
There was a problem hiding this comment.
I compare this version with the explicit version with if ... else .... Both cases does not introduce new allocations. So I think it is safe to do so.
Yes, I am sure this is what I want. Actually, here alpha can be a tropical number. A Tropical number is a semiring that can not be converted to/from a regular number. When the input is a regular number, I just want to to treat the zero element and one element properly. There is a correspondence of zero-element and one element in different algebras, but not for other numbers.
No description provided.