[ADT] APFloatBase: Prevent collapsing semPPCDoubleDouble and semBogus

[mgorny]

Summary

Provide a distinct contents for semBogus and semPPCDoubleDouble in order
to prevent compilers from collapsing them to a single memory address,
while we heavily rely on every semantic having distinct address.

This happens if insecure optimization collapsing identical values is
enabled. As a result, APFloats of semBogus are indistinguishable from
semPPCDoubleDouble -- and whenever the move constructor is used, the old
value beings being incorrectly recognized as a semPPCDoubleDouble.

Since the values in semPPCDoubleDouble are not used anywhere,
we can easily solve this issue via altering the value of one of the
fields and therefore ensuring that the collapse can not occur.

llvm/trunk/lib/Support/APFloat.cpp

/* The PowerPC format consists of two doubles.  It does not map cleanly
      mantissa bits.  Note that for exponents near the double minimum,
      we no longer can represent the full 106 mantissa bits, so those
      will be treated as denormal numbers.
 
      FIXME: While this approximation is equivalent to what GCC uses for
      compile-time arithmetic on PPC double-double numbers, it is not able
      to represent all possible values held by a PPC double-double number,
      for example: (long double) 1.0 + (long double) 0x1p-106
-     Should this be replaced by a full emulation of PPC double-double?  */
-  static const fltSemantics semPPCDoubleDouble = {0, 0, 0, 0};
+     Should this be replaced by a full emulation of PPC double-double?
+
+     Note: we need to make the value different from semBogus as otherwise
+     an unsafe optimization may collapse both values to a single address,
+     and we heavily rely on them having distinct addresses.             */
+  static const fltSemantics semPPCDoubleDouble = {-1, 0, 0, 0};
 
   /* There are temporary semantics for the real PPCDoubleDouble implementation.
      Currently, APFloat of PPCDoubleDouble holds one PPCDoubleDoubleImpl as the
      high part of double double, and one IEEEdouble as the low part, so that
      the old operations operate on PPCDoubleDoubleImpl, while the newly added
      operations also populate the IEEEdouble.
 
      TODO: Once all functions support DoubleAPFloat mode, we'll change all