The Trouble with long double

[GregL]

It has always disappointed me that Microsoft dropped support for extended-precision 80-bit floating-point variables from Visual C++. Other compilers like Intel C++, Borland C++, GCC, PowerBASIC, Intel Fortran, and a few others do support them. And of course MASM supports them. Eric Fleegal from Microsoft talks about it here.

He mentions a 128-bit floating-point data type. If Intel and AMD would support it in hardware then we would really have something.

[xanatose]

The point is now rapidly becoming moot.  80-bit doubles seem so “last century”.  We are approaching the day when most new Windows systems will have FPUs that easily support the 128bit long double format.

:eek Err, What FPU supports 128bit long double format? AFAIK there is only 32,64 and 80 bit on x86 machines. Unless he is taking about emulation.

[GregL]

I'm not sure if he means emulation or that it could easily be done in hardware. It would fit nicely in an SSE register.

Here's a thought, extend the x87 instructions to work with the new format. That would breathe some fire into x87 assembly code. ... Not likely.

[hutch--]

Sooner or later the hardware will have to support higher precision than 64 bit as FP x87 is being phased out over time. A 128 bit precision SSE? would be a good substitute and probably a lot faster than the old x87.