[llvm] [WIP] v_lshrrev uses vector register operand in wave space after scaling offset materialization in v_mad (PR #101649)

[Christudasan Devadasan via llvm-commits]

cdevadas wrote:

Avoid such a long name for the PR. Change it to something like: 
[AMDGPU] Fixed the wrong register used in the shift operation.

Indicate the PR in the description where this bug was introduced (https://github.com/llvm/llvm-project/pull/99556).
Why is this PR marked WIP?


https://github.com/llvm/llvm-project/pull/101649