[llvm] [IndVarSimplify] Ensure fp values can be represented as consecutive integers (PR #166649)

[Yingwei Zheng via llvm-commits]

================
@@ -196,6 +196,24 @@ static bool ConvertToSInt(const APFloat &APF, int64_t &IntVal) {
   return true;
 }
 
+// Ensure we stay within the bounds of fp values that can be represented as
+// integers without gaps, namely 2^24 and 2^53 for IEEE-754 single and double
+// precision respectively (both on negative and positive side).
+static bool isRepresentableAsExactInteger(ConstantFP *FPVal, int64_t IntVal) {
+  const auto &InitValueFltSema = FPVal->getValueAPF().getSemantics();
+  if (!APFloat::isIEEELikeFP(InitValueFltSema))
+    return false;
+
+  uint64_t InitValuePrecision = APFloat::semanticsPrecision(InitValueFltSema);
+  if (InitValuePrecision >= 64)
+    return false;
+
+  uint64_t InitValueIntegerLimit = 1LL << InitValuePrecision;
+  if (AbsoluteValue(IntVal) >= InitValueIntegerLimit)
+    return false;
+  return true;
----------------
dtcxzyw wrote:

```suggestion
  return isUIntN(InitValuePrecision, AbsoluteValue(IntVal));
```


https://github.com/llvm/llvm-project/pull/166649