Re: [Familiar] xscale & soft-float vs hard-float

on Tue, Nov 25, 2003 at 07:51:52PM -0500, David Tauriainen wrote:
> On a PXA_250 processor (H3950), which option in gcc will produce faster
> floating point operations: soft-float or hard-float? Also, if
> soft-float, is there any way to reduce the precision to make things a
> little faster?

Soft-float is faster and more flexible, though the price you pay for
that is incompatibility with all current binaries and libraries. You
can select a library implementation that gives the appropriate balance
between precision, functionality and speed for your application, for
example by flushing denormals to zero or omitting rounding modes other
than to-nearest.

p.
Received on Wed Nov 26 12:19:00 2003