This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: RFC: Patch to avoid using _finite calls in some cases (including in libmvec names)
- From: Steve Ellcey <sellcey at marvell dot com>
- To: "Steve dot Ellcey at cavium dot com" <Steve dot Ellcey at cavium dot com>, "libc-alpha at sourceware dot org" <libc-alpha at sourceware dot org>, "Wilco dot Dijkstra at arm dot com" <Wilco dot Dijkstra at arm dot com>, "Szabolcs dot Nagy at arm dot com" <Szabolcs dot Nagy at arm dot com>, "joseph at codesourcery dot com" <joseph at codesourcery dot com>
- Cc: "nd at arm dot com" <nd at arm dot com>
- Date: Thu, 14 Mar 2019 15:21:09 +0000
- Subject: Re: RFC: Patch to avoid using _finite calls in some cases (including in libmvec names)
- References: <DB5PR08MB1030F112521F8FF5F08DB223834B0@DB5PR08MB1030.eurprd08.prod.outlook.com>
On Thu, 2019-03-14 at 13:52 +0000, Wilco Dijkstra wrote:
> Hi,
>
> Clearly the best solution is to remove all finite renaming. Over the past years
> we've removed a lot of unnecessary inlines, redirects and other nasty hacks
> from GLIBC. GCC now sees the original functions and as a result can optimize far
> better. This is just one of the last few remaining header hacks.
>
> We use no veneers in the new optimized math functions and as a result they are
> significantly faster, so there is no argument that the finite variants are useful or
> improve performance. It is better to handle special cases in the main code rather
> than do them in a veneer and thus need to check them twice...
>
> So I believe we should be bold and go ahead with disabling the header
> completely.
>
> Wilco
I agree with this approach. The header is a hack. My patch was a way
of removing (ifdefing out) functions from math-finite.h over time on
individual platforms but I would also support a rip-the-bandage off
quickly approach and just get rid of math-finite.h in one go.
Steve Ellcey
sellcey@marvell.com