[llvm] AMDGPU/GlobalISel: Switch to extended LLTs (PR #196522)

[Petar Avramovic via llvm-commits]

petar-avramovic wrote:

ping
I commited aarch64 changes. We are left with amdgpu changes here.
For common code changes, in most cases there is nothing to test with since we need amdgpu with extended LLTs.
But I assume that there good possibility that some of the common code changes would not be required if we decide to not use LLT::scalar mixed with LLT::integer/float. But that is too big of a change.

Any suggestions how to move this forward?
I can remove common code changes and look at them later but will need to disable ~300 tests

https://github.com/llvm/llvm-project/pull/196522