Re: Support for non-default rounding modes in the softfloat layer

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Tue, Apr 30, 2013 at 07:41:33AM -0400, Greg Troxel wrote:
> I certainly don't, but pkgsrc/benchmarks/paranoia seems to check things
> like this, and it seems a regression to have that do worse.

I don't think platforms are required to implement fenv.h and the
non-default rounding at all. This is somewhat supported by GCC's general
ignorance of the whole topic until very recently.

> I am not sure I understand your global vs thread-local concern - the
> real issue is lexical source-code scope rather than dynamic thread
> context.

Eh, no. For many of the potential applications, having a single global
rounding state is impractical at best. This is completely unrelated to
the issue of source level representation and exposure of rounding modes.
The potential secondary API with explicit rounding mode as argument
allows the compiler to provide either and optimize fetching the current
state intelligently.

Joerg