[llvm] [AMDGPU][DAG] Enable ganging up of memcpy loads/stores for AMDGPU (PR #96185)

[Fabian Ritter via llvm-commits]

================

----------------
ritter-x2a wrote:

The test functions are now normal functions instead of kernels and I added another test with `inreg` arguments.
The differences in the generated code between those are not too fundamental, so I skipped most address space combinations and sizes for the sreg test to avoid making it too big.

https://github.com/llvm/llvm-project/pull/96185