Re: toolchain/50940: cc -ffast-math fails to disable denormals on amd64

[Andreas Gustafsson <gson%gson.org@localhost>]

The following reply was made to PR toolchain/50940; it has been noted by GNATS.

From: Andreas Gustafsson <gson%gson.org@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: toolchain-manager%netbsd.org@localhost,
    gnats-admin%netbsd.org@localhost
    gson%gson.org@localhost (Andreas Gustafsson)
Subject: Re: toolchain/50940: cc -ffast-math fails to disable denormals on
 amd64
Date: Fri, 11 Mar 2016 18:07:56 +0200

 Joerg Sonnenberger wrote:
 >  > When the -ffast-math option is enabled, gcc generates x86_64 floating
 >  > point code that assumes that the FTZ (Flush To Zero) and DAZ
 >  > (Denormals Are Zero) bits in the MXCSR register are set.
 >  
 >  That's not really true. The assumption with -ffast-math is that
 >  denormals and infinites don't happen at all.
 
 That may be - I haven't found any documention on what the exact
 assumptions are.  If you know of some, I would appreciate a pointer.
 
 >  Whether the CPU generates them or not is a secondary question -- at
 >  that point you are already in Undefined Behavior land for this
 >  mode.
 
 Do you at least agree that the current NetBSD behavior is incorrect?
 -- 
 Andreas Gustafsson, gson%gson.org@localhost