[llvm] [APFloat] Extend fltSemantics with field for explicit integer bit (PR #204860)

[via llvm-commits]

================
@@ -3813,6 +3805,21 @@ void IEEEFloat::initFromIEEEAPInt(const APInt &api) {
     is_nan = is_zero && sign;
   }
 
+  if constexpr (S.hasExplicitIntegerBit) {
+    // This is only used and tested for x87DoubleExtended
+    static_assert(S.precision == 64);
+    const integerPart myintegerbit = mysignificand[0] >> 63;
+    constexpr integerPart significand_mask_no_int_bit =
+        (uint64_t{1} << (trailing_significand_bits - 1)) - 1;
+    if (myexponent - bias != ::exponentNaN(S) && myexponent != 0 &&
+        myintegerbit == 0) {
+      is_nan = true;
----------------
janr-bay wrote:

I've changed this now so the special case handling is part of the other NaN detection logic. Now hasExplicitIntegerBit and nanEncoding interact but only by turning handling of those special cases on and this should be clear from the code

https://github.com/llvm/llvm-project/pull/204860