[llvm] [APFloat] Fix literals with long significands. (PR #102051)

[Jakub Kuderski via llvm-commits]

================
@@ -2950,18 +2956,21 @@ IEEEFloat::opStatus
 IEEEFloat::roundSignificandWithExponent(const integerPart *decSigParts,
                                         unsigned sigPartCount, int exp,
                                         roundingMode rounding_mode) {
-  unsigned int parts, pow5PartCount;
-  fltSemantics calcSemantics = { 32767, -32767, 0, 0 };
-  integerPart pow5Parts[maxPowerOfFiveParts];
+  unsigned int parts;
+  fltSemantics calcSemantics = {std::numeric_limits<ExponentType>::max(),
+                                std::numeric_limits<ExponentType>::min(), 0, 0};
   bool isNearest;
 
   isNearest = (rounding_mode == rmNearestTiesToEven ||
                rounding_mode == rmNearestTiesToAway);
 
   parts = partCountForBits(semantics->precision + 11);
 
+  // Make sure that abs(exp) is representable.
+  assert(exp > INT_MIN);
----------------
kuhar wrote:

Also prefer `std::numeric_limits`

https://github.com/llvm/llvm-project/pull/102051