This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: Re: [PATCH 4/4] Do not wrap expf and exp2f

From: Arjan van de Ven <arjan at linux dot intel dot com>
To: Joseph Myers <joseph at codesourcery dot com>, Szabolcs Nagy <szabolcs dot nagy at arm dot com>
Cc: GNU C Library <libc-alpha at sourceware dot org>, nd at arm dot com
Date: Wed, 13 Sep 2017 17:31:18 -0700
Subject: Re: Re: [PATCH 4/4] Do not wrap expf and exp2f
Authentication-results: sourceware.org; auth=none
References: <alpine.DEB.2.20.1709132017370.28319@digraph.polyomino.org.uk>

On 9/13/2017 1:19 PM, Joseph Myers wrote:

On Wed, 13 Sep 2017, Szabolcs Nagy wrote:

	* sysdeps/x86_64/fpu/w_expf.c: New file.


I don't think having w_expf.c alongside w_expf.S in the same directory
makes sense.  (An IFUNC w_expf.c in the multiarch subdirectory, used to
build both fma and non-fma versions and replacing the existing
e_expf-fma.S, may well make sense if benchmarks show improvements on the
fma and non-fma .S files.)


fwiw I am assuming we'll retire the x86-64 .S files and use the C version
(for both FMA and non-FMA with some IFUNC selector) once this lands.
Optimizing the .S file was fun for FMA, but the C version has a smaller
polynomial and is faster ... data speaks and this wins.

References:
- Re: [PATCH 4/4] Do not wrap expf and exp2f
  - From: Joseph Myers

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]