This is the mail archive of the
mailing list for the GCC project.
Re: Draft "Unsafe fp optimizations" project description.
- To: dewar at gnat dot com, toon at moene dot indiv dot nluug dot nl
- Subject: Re: Draft "Unsafe fp optimizations" project description.
- From: dewar at gnat dot com
- Date: Mon, 6 Aug 2001 20:45:04 -0400 (EDT)
- Cc: aj at suse dot de, gcc at gcc dot gnu dot org, lucier at math dot purdue dot edu,tprince at computer dot org
<<Unfortunately, on the system he works on, the kernel traps now start to
form a noticable performance problem.
Absolutely, this is a well known problem. If you look back at the thread,
you will see that very early on I suggested that sudden underflow was a
perfectly reasonable approximation for many purposes. Furthermore, my
viewpoint is that on machines where denormals are very expensive, it
is reasonable to do flush-to-zero in default mode, and reserve the
expensive denormal mode for -mieee (so in fact I go *further* than
agreeing this should be in -ffast-math :-)
But what I was saying was that casually, many programmers will not understand
why denormals are so valuable. Yes, if you are on a machine which cannot
provide this facility with reasonable efficiency, then it is reasonable
to abandon it, but many people are unaware of why this is a real loss!