[llvm] [AMDGPU] Fix machine verification failure from INIT_EXEC lowering (PR #98333)

[Jay Foad via llvm-commits]

================
@@ -1676,6 +1676,8 @@ bool SIWholeQuadMode::runOnMachineFunction(MachineFunction &MF) {
   if (!(GlobalFlags & (StateWQM | StateStrict)) && LowerToCopyInstrs.empty() &&
----------------
jayfoad wrote:

I am now even more convinced that we should not bother with this fast path. We should just make sure that the general path is fast when there is no work to do.

https://github.com/llvm/llvm-project/pull/98333