Can we generate assembly code that tensor dimensions are not fixed?

[gfvvz]

Seems not? Any expert can help confirm this?