This is the mail archive of the 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: [PATCH] PPC64: First in the series of patches implementing POWER8 vector math.

On Mon, 25 Feb 2019, GT wrote:

> 1. svml_finite_alias.S deals only with log/logf, exp/expf and pow/powf.
> So does the discussion linked at the given url. At this time only the
> cos function has been implemented. Do we then still require
> libmvec_nonshared.a to be created?

The alternative would be to have some kind of conditional in 
math/Makefile, where it creates the linker script for, based on 
whatever logic determines whether there is libmvec_nonshared.a (i.e. 
whether there are any objects that go in it).  If you're adding new 
architecture support for libmvec that doesn't start off with any functions 
that bits/math-finite.h does anything with, you get to deal with adapting 
the generic code to handle that previously unsupported case.

> 2. I find a single reference in the entire source tree where
> libmvec_nonshared.a doesn't appear in the dependency section of
> a rule. That's in math/Makefile. Does this mean libmvec_nonshared.a
> is created with 'make install'? I rather expected it would be built

The linker scripts that are only usable in a glibc installation, not in a 
build context, are created at install time, yes.  (So is the manual, hence 
the build of the manual breaking from time to time when someone had a 
broken change to the manual and didn't run "make install" in testing.)

Joseph S. Myers

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