[llvm] AMDGPU: fix isSafeToSink expecting exactly one predecessor (PR #89224)

[Jay Foad via llvm-commits]

================
@@ -217,10 +217,20 @@ bool SIInstrInfo::isSafeToSink(MachineInstr &MI,
         SmallVector<MachineBasicBlock *, 1> ExitBlocks;
         FromCycle->getExitBlocks(ExitBlocks);
         assert(ExitBlocks.size() == 1);
-        assert(ExitBlocks[0]->getSinglePredecessor());
-
+        // After structurize-cfg, cycle exit block should have exactly one
----------------
jayfoad wrote:

Isn't it simpler to write this code so that it makes no assumptions about structurized CFGs? E.g.
```c++
  FromCycle->getExitBlocks(ExitBlocks);
  for (ExitBlock : ExitBlocks)
    for (Pred : ExitBlock->predecessors())
      if (FromCycle->contains(Pred) && hasDivergentBranch(Pred))
        return false;
```

https://github.com/llvm/llvm-project/pull/89224