[llvm] [AMDGPU][LibCallSimplify] Use target type's float-semantics in `ConstantFP::get` (PR #213721)

[Juan Manuel Martinez Caamaño via llvm-commits]

================
@@ -2017,16 +2023,10 @@ bool AMDGPULibCalls::evaluateCall(CallInst *aCI, const FuncInfo &FInfo) {
     }
   }
 
-  Constant *nval0 = nullptr, *nval1 = nullptr;
-  if (FuncVecSize == 1) {
-    nval0 = ConstantFP::get(aCI->getType(), Val0[0]);
-    if (hasTwoResults)
-      nval1 = ConstantFP::get(aCI->getType(), Val1[0]);
-  } else {
-    nval0 = getConstantFloatVector(Val0, aCI->getType());
-    if (hasTwoResults)
-      nval1 = getConstantFloatVector(Val1, aCI->getType());
-  }
+  Constant *nval0 = getConstantFloat(Val0, aCI->getType());
+  Constant *nval1 = nullptr;
+  if (hasTwoResults)
+    nval1 = getConstantFloat(Val1, aCI->getType());
----------------
jmmartinez wrote:

In the end, I've ended up sinking `nval1` closer to its user. Then we have just an assignment and no need for the ternary (it user is guarded by `hasTwoResults`).

I didn't add the `const` because the code gets slightly awkward. Since the pointer is `const`, but not the pointee.

```cpp
Constant * const nval1 = ... ;
```

https://github.com/llvm/llvm-project/pull/213721