[llvm] [AMDGPU] Gate runtime unroll of LDS loops instead of overriding it (PR #222853)

[Matt Arsenault via llvm-commits]

================
@@ -225,9 +225,9 @@ void AMDGPUTTIImpl::getUnrollingPreferences(
             (!isa<GlobalVariable>(GEP->getPointerOperand()) &&
              !isa<Argument>(GEP->getPointerOperand())))
           continue;
-        LLVM_DEBUG(dbgs() << "Allow unroll runtime for loop:\n"
+        LLVM_DEBUG(dbgs() << "Gating unroll runtime by local knob for loop:\n"
                           << *L << " due to LDS use.\n");
-        UP.Runtime = UnrollRuntimeLocal;
+        UP.Runtime &= UnrollRuntimeLocal;
----------------
arsenm wrote:

```suggestion
        if (!UP.Runtime && UnrollRuntimeLocal) {
        LLVM_DEBUG(dbgs() << "Gating unroll runtime by local knob for loop:\n"
<< *L << " due to LDS use.\n");
        UP.Runtime = true;
```

Try to make the debug printing more accurate?

https://github.com/llvm/llvm-project/pull/222853