[flang] Fix folding of EXPONENT() intrinsic function

[klausler]

Summary

The definition of the EXPONENT() intrinsic function differs by one
from the real arithmetic folding templates concept of an unbiased
exponent, and also needs special handling for zero. Fix, and add
more tests.

[PeteSteinfeld]

All builds, tests, and looks good.

flang/include/flang/Evaluate/real.h

public:
   // HYPOT(x,y)=SQRT(x**2 + y**2) computed so as to avoid spurious
   // intermediate overflows.
   ValueWithRealFlags<Real> HYPOT(
       const Real &, Rounding rounding = defaultRounding) const;
 
   template <typename INT> constexpr INT EXPONENT() const {
     if (Exponent() == maxExponent) {
       return INT::HUGE();
+    } else if (IsZero()) {
+      return {0};
     } else {
-      return {UnbiasedExponent()};
+      return {UnbiasedExponent() + 1};
     }
   }
 
   static constexpr Real EPSILON() {
     Real epsilon;
     epsilon.Normalize(
         false, exponentBias + 1 - binaryPrecision, Fraction::MASKL(1));
     return epsilon;
if constexpr (!isImplicitMSB) {
       } else {
         return result.IBCLR(significandBits);
       }
     }
   }
 
   // Extracts unbiased exponent value.
   // Corrects the exponent value of a subnormal number.
+  // Note that the result is one less than the EXPONENT intrinsic;
+  // UnbiasedExponent(1.0) is 0, not 1.
   constexpr int UnbiasedExponent() const {
     int exponent{Exponent() - exponentBias};
     if (IsSubnormal()) {
       ++exponent;
     }
     return exponent;
   }
 

flang/test/Evaluate/folding07.f90

module m
   real(16), parameter :: tiny16 = tiny(0._16), ztiny16 = real(z'00010000000000000000000000000000', kind=16)
   logical, parameter :: test_tiny2 = tiny2 == ztiny2
   logical, parameter :: test_tiny3 = tiny3 == ztiny3
   logical, parameter :: test_tiny4 = tiny4 == ztiny4
   logical, parameter :: test_tiny8 = tiny8 == ztiny8
   logical, parameter :: test_tiny10 = tiny10 == ztiny10
   logical, parameter :: test_tiny16 = tiny16 == ztiny16
 
+  logical, parameter :: test_exponent_0 = exponent(0.0) == 0
+  logical, parameter :: test_exponent_r8 = exponent(0.125) == -2
+  logical, parameter :: test_exponent_r4 = exponent(0.25) == -1
+  logical, parameter :: test_exponent_r2 = exponent(0.5) == 0
+  logical, parameter :: test_exponent_1 = exponent(1.0) == 1
+  logical, parameter :: test_exponent_4 = exponent(4.1) == 3
+  logical, parameter :: test_exponent_12 = exponent(12.9) == 4
+
   integer, parameter :: &
     max2 = maxexponent(0._2), &
     max3 = maxexponent(0._3), &
     max4 = maxexponent(0._4), &
     max8 = maxexponent(0._8), &
     max10 = maxexponent(0._10), &
     max16 = maxexponent(0._16)
   logical, parameter :: test_max2 = max2 == 16