Perhaps if DEC had carried over full support for D_FLOAT and G_FLOAT
from VAX to ALPHA (both are 8-bit floating-point representations, but
the ratio of mantissa to exponent bits is different (and neither is
IEEE)) (there is no DEC (now Compaq) Fortran90 for VAX), people wouldn't
have gotten into the kind-number=star-number=relative precision trap.
Actually, I'm somewhat surprised that choices like this are not more
widespread. My experience is that people usually use more bytes because
they want more precision, not more range. Thus, a DOUBLE PRECISION type
variable with more mantissa would in some cases be an attractive
alternative to quadruple precision, especially where large amounts of
data are concerned and thus storage space is at a premium. (I'm
thinking that any roll-your-own solutions here, involving derived types,
TRANSFER, whatever would not be as efficient, especially with regard to
intrinsic functions etc.)
--
Phillip Helbig Email ......... [log in to unmask]
University of Manchester Tel. ... +44 1477 571 321 (ext. 2635)
Jodrell Bank Observatory Fax ................ +44 1477 571 618
Macclesfield Telex ................ 36149 JODREL G
UK-Cheshire SK11 9DL Web ... http://www.jb.man.ac.uk/~pjh/
************************ currently working at *******************************
Kapteyn Instituut Email (above preferred) [log in to unmask]
Rijksuniversiteit Groningen Tel. ...................... +31 50 363 4067
Postbus 800 Fax ....................... +31 50 363 6100
NL-9700 AV Groningen
The Netherlands Web ... http://gladia.astro.rug.nl/~helbig/
My opinions are not necessarily those of either of the above institutes.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|