Topic: REAL*10 in 64-bit

DanRRight

Posts: 2877 South Pole, Antarctica

Back to Top

11 Mar 2017 8:30 #19056

'Error 1210: Extended precision floating point (REAL*10) is not supported in 64-bit mode'

Any suggestions to get more then real*8 ?

PaulLaidler

Posts: 7974 Salford, UK

Back to Top

11 Mar 2017 9:30 #19059

I don't have any immediate suggestions relating to FTN95.

JohnCampbell

Posts: 2526 Sydney

Back to Top

11 Mar 2017 10:22 #19063

Dan,

You need a real16 higher precision software emulation library; they must exist. (google quad precision ?) I have no experience with using these but expect it would be messy and slow !! I presume you would require a derived type to store the 'real16' variables and have limited access to the calculations. Would complex(2) provide the storage ?

For simple multiplication and accumulation, see https://en.wikipedia.org/wiki/Kahan_summation_algorithm
This may help.

John

simon

Posts: 314

Back to Top

11 Mar 2017 1:09 #19069

I don't have access to the book right now to check whether it is what you need, but there is a section on 'Arbitrary precision' in 'Numerical Recipes'

mecej4

Posts: 1911

Back to Top

11 Mar 2017 3:28 #19071

See http://crd-legacy.lbl.gov/~dhbailey/mpdist/ for one package that enables multiple precision.

Other compilers have support for REAL*16 using compiler switches.

Multiple precision is not to be chosen without due deliberation and a well-reasoned conclusion that it is necessary.