Fourth Draft "Unsafe fp optimizations" project description.
Geert Bosch
bosch@gnat.com
Sun Aug 12 21:34:00 GMT 2001
On Mon, 13 Aug 2001 dewar@gnat.com wrote:
I do not believe that these built in routines meet the Ada requirements
for accuracy. At least I cannot prove that from the documentation Intel
provides.
Of course the whole issue here is the -ffast-math option, which indeed
gives up on such accuracy requirements to achieve better speed, so
your comment is not relevant here. Think of this as the Ada relaxed math mode.
-Geert
PS. The accurracy difference is the value of Pi used for argument reduction.
More information about the Gcc
mailing list