ieee real(16) support

[dave_frank]

Now that ieee has defined what real(16) format is, will Intel provide future support?

[Steven_L_Intel1]

Dave,

Intel Fortran has provided IEEE-style REAL(16) support for at least a couple of years now.

Steve

[dave_frank]

Amazing since the format is still in draft form.
Sooo you are saying you guys guessed right a couple years
ago and are using the draft format's 15-bits exponent and 112-bits fraction?

[Steven_L_Intel1]

Dave,

I can't readily find a description of the IEEE effort here (even on the IEEE web site), so can't verify the details. But I can tell you that vendor support for an "IEEE-style" 128-bit floating format, with 15-bit exponent and 112-bit fraction, has been available in most high-end platforms for more than a decade. Digital supported this format on Alpha processors in 1993 (and had a similar VAX format dating back to 1980.)

In Windows Fortran compilers, Intel Fortran is one of several that currently support this format - CVF is not among them, though the future Intel Visual Fortran will be.

Steve

[TimP]

As Steve points out, DEC and HP supported this format 10 years ago. It is IEEE-854 compliant, and those Windows compilers (not only Intel's), which support real*16, use the same format. It could have been difficult for IEEE to persuade people to change horses. No, I didn't know about that meeting.

Other major vendors who support real*16 on their own architectures (e.g. IBM and SGI) have a non-IEEE compliant format with 11-bit exponent and 107 bit precision. When I last compared them, neither SGI's nor HP's transcendental functions took advantage of the full precision of their floating point format, apparently favoring speed over accuracy. Speed also favors those non-standard IBM and SGI implementations over the software floating point real*16 employed on many systems.