[llvm] [AMDGPU] Set register bank for i1 register copies (PR #96155)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> > // The calling convention is to be updated such that i1 function arguments
> > // or return values are assigned to SGPRs without promoting to i32.
> 
> I was not aware of this. Why do you want to change calling convention?

Because we are not taking advantage of scalar arguments and return values, and forcing everything into VGPRs, which introduces extra code to convert from and to boolean masks in the caller/callee 

https://github.com/llvm/llvm-project/pull/96155