(MAXDECEXP, MINDECEXP): New decimal exponent limits
that vary with definition of LONG_DOUBLE_TYPE_SIZE.
(endian, ereal_atof, real_value_truncate, einfin, emdnorm, asctoeg):
Add cases for TFmode.
(etartdouble): New function converts REAL_VALUE_TYPE to TFmode
for use by ASM_OUTPUT_LONG_DOUBLE.
(edivm, emulm): Ifdef out, replace by faster multiply and divide.
(etoe113, toe113, e113toe): New type conversions for TFmode.
(asctoe113, e113toasc): New TFmode binary <-> decimal conversions.
(at top level): Define constants ezero, eone, emtens, etens, ...
in a new 20-byte format when LONG_DOUBLE_TYPE_SIZE = 128 and
set NE to 10.  Otherwise, the internal format remains 12 bytes wide.
(etoudi, etodi, ditoe, uditoe): New functions, signed and unsigned
DImode float and fix, for real.c used in a libgcc-of-last-resort.
(esqrt): New function, correctly rounded square root for libgcc.
(etodec): Delete ifdef'd version.
(eroundi, eroundui): Rename to efixi, efixui and always
round towards zero.
From frank@atom.ansto.gov.au (Frank Crawford):
(etoibm, toibm, ibmtoe): New conversions for IBM 370 float format.
(e53toe, e64toe, toe64, etoe53, toe53, etoe24, toe24, asctoe53,
asctoeg, make_nan): Ifdef for IBM.

From-SVN: r5153