Matrix multiplication slower in 0.7

```
using StaticArrays
julia> A, B = rand(SMatrix{3,3}), rand(SMatrix{3,3})
julia> @btime $A*$B;
julia> @btime $A*$B;
  8.210 ns (0 allocations: 0 bytes)
```
however on Julia 0.7, with `StaticArrays v0.8.1` I get
```
julia> @btime $A*$B;
  9.442 ns (0 allocations: 0 bytes)
```
