This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: soft-fp: add macro FP_NO_EXCEPTIONS
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: Richard Henderson <rth at redhat dot com>
- Cc: <libc-alpha at sourceware dot org>, "David S. Miller" <davem at davemloft dot net>
- Date: Tue, 15 Oct 2013 19:19:16 +0000
- Subject: Re: soft-fp: add macro FP_NO_EXCEPTIONS
- Authentication-results: sourceware.org; auth=none
- References: <Pine dot LNX dot 4 dot 64 dot 1310092324560 dot 12619 at digraph dot polyomino dot org dot uk> <525D8AFE dot 4080401 at redhat dot com>
On Tue, 15 Oct 2013, Richard Henderson wrote:
> On 10/09/2013 04:26 PM, Joseph S. Myers wrote:
> > David, Richard: I think the SPARC and Alpha uses of soft-fp could
> > benefit from this. Alpha already does something similar of its own
> > regarding redefining FP_ROUNDMODE, while if the relevant SPARC files
>
> I'm not sure this is applicable to Alpha.
>
> The thing Alpha does wrt redefining FP_ROUNDMODE is to handle
> a couple of routines that are defined to take the rounding mode
> as an argument rather than taking it from the fp control register.
I'm thinking of ots_cvtqux.c, ots_cvtqx.c and ots_cvttx.c defining
FP_ROUNDMODE to FP_RND_ZERO - those being equivalent to floatunditf.c,
floatditf.c and extenddftf2.c, the first two of which define
FP_NO_EXCEPTIONS (and so don't need FP_DECL_EX or anything else related to
exceptions). I think ots_cvtqux.c and ots_cvtqx.c should be able to
define FP_NO_EXCEPTIONS (and then not need the local FP_ROUNDMODE
definitions for anything), while the FP_ROUNDMODE definition in
ots_cvttx.c shouldn't actually be doing anything (none of the macros there
does any rounding; I haven't checked, but likely the definition dates back
to when there was a single macro for all floating-point conversions rather
than separate ones for extensions and truncations).
--
Joseph S. Myers
joseph@codesourcery.com