128 bit floating point real

[Neil Roberts]

Hi,

I've had a user come to me asking about whether Alpha has support for 128 bit floating point reals (H-floating on VAX).

Any help on this would be appreciated.

thanks

Neil

[Ian Miller. · Accepted solution]

According to the migration manual
http://h71000.www7.hp.com/doc/73final/documentation/pdf/OVMS_MIGR_APPL.pdf

H float is not supported. Consider using X float which is IEEE extended double precision and has a similar range. There are compiler qualifers (depending on the language used) to select this and conversion routines to convert to various formats.

VAX code translated to Alpha using DECmigration can use H Float.

[Kris Clippeleyr]

Neil,
The FORTRAN compiler supports REAL*16.
On VAX processors, H_floating format is used to implement REAL*16. On Alpha processors, IEEE X_floating format is used to implement REAL*16.
Regards,
Kris (aka Qkcl)

[Joseph Huber_1]

I don't know off-hand what floating type it is (depending on compile /FLOAT switches ?), but this quick test shows:

create f128.for
program f128
* test if we have real*16
real*16 test
test=1.1234567899999999
print *, ' ',test
print *, ' ',test/test
end Exit
fortran f128.for
link f128
run f128
1.12345683574676513671875000000000
1.00000000000000000000000000000000

[Joseph Huber_1]

In addition on Alphas, the DOUBLE PRECISION (F90 REAL(KIND=16)) can be made by default 128 bit (X-floating), with the /DOUBLE_SIZE=128 Fortran compiler switch.

[Neil Roberts]

Thanks for all the responses.

Unfortunately we're using ADA rather than Fortran, which doesn't have the X or H switches available. Does anyone have any thoughts on the ADA side?

Neil

[Kris Clippeleyr]

Neil,
AFAIK, on Alpha, the FORTRAN, the HP C, the HP Pascal compilers support 128-bit floating points (IEEE X_float); the ADA compiler does not.
Regards,
Kris (aka Qkcl)

[Ian Miller.]

The manual I referenced has info on ADA. Search it for 128.

[Neil Roberts]

Thank you all for your assistance. Looks like a Fortran licence may be in the offing!!

Neil

[Neil Roberts]

No solution for problem. Time to look at Fortran.