Date: Thu, 8 Oct 2015 09:38:30 +0300 From: Konstantin Belousov <kostikbel@gmail.com> To: Peter Jeremy <peter@rulingia.com> Cc: FreeBSD-arm@freebsd.org Subject: Re: svn commit: r288983 - in head/sys/arm: arm include Message-ID: <20151008063830.GN2257@kib.kiev.ua> In-Reply-To: <20151008053507.GD85040@server.rulingia.com> References: <201510070912.t979Co21054138@repo.freebsd.org> <20151008053507.GD85040@server.rulingia.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Oct 08, 2015 at 04:35:07PM +1100, Peter Jeremy wrote: > [moved from src-* lists] > > On 2015-Oct-07 09:12:50 +0000, Konstantin Belousov <kib@FreeBSD.org> wrote: > >Log: > > A follow-up to r288492. In fact, revert the mentioned commit for > > pre-VFPv3 processors, since they do require software support code to > > handle denormals. For VFPv3 and later, enable flush-to-zero if > > hardware does not claim full denormals arithmetic support by VMVFR1_FZ > > field in mvfr1 register. > > Is there any reason we can't re-enable denormals on ARM11 and use the > existing softfloat code to handle the underflow exceptions? Do you mean a technical reason, or the current state of efforts ? Sure, the referenced support code was developed by ARM exactly to fullfill the goal you stated, having the working denormals on FPU which does not handle them. The ARM Application Note 98 'The VFP support code' seems to contain detailed explanation how to integrate the code with libc. Somebody should do the work. First step actually is to make an effort to obtain the said support code and to read the license before looking at the single line of the code. The App Note 98 stated that the code is 'provided with the note', but I failed to make any useful consequences of the statement. > > This is the approach SPARC uses to handle non-normalised numbers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20151008063830.GN2257>