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: [PATCH v0] PPC64: Attach SIMD attribute to cosf, sin, sinf function declarations.


‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Thursday, February 13, 2020 2:26 PM, Tulio Magno Quites Machado Filho <tuliom@ascii.art.br> wrote:

> GT tnggil@protonmail.com writes:
>
> > This is the first of 2 patches required to enable GCC auto-vectorization
> > of calls to libmvec.
> > The changes here should have been part of earlier patches to libmvec
> > which added SIMD versions of the functions.
> > There are a few other small patches in tuliom/libmvec that should
> > similarly have been part of earlier patches. Can we re-organize the
> > patches in tuliom/libmvec using git rebase before submitting them for
> > inclusion in master?
>
> Could you be more specific, please?

The commits whose abbreviated ids are c1782a410f, 1bd6ae3a5a, and 29d3adfe09.
Each of them introduces a test to ensure that feature(s) used in immediately preceding
patches are available on the system.

> What kind of reorganization are you suggesting?

My understanding is that by using git rebase, each of those 3 commits can be combined
with the earlier patch that should have had the feature tests initially. I am assuming that
it is ok to re-write history in branch tuliom/libmvec in this manner. If not so, then you
may dismiss the suggestion.

> > Subject: [PATCH v0] PPC64: Attach SIMD attribute to cosf, sin, sinf function
> > declarations.
> > These changes were mistakenly left out of the patches that added SIMD
> > versions of these functions to libmvec.
>
> This patch is not just adding missing data, it's also moving lines around.
> Some of them seem to be re-sorted, but not all of them.
>
..
..
..
>
> Why are these 2 lines moving to...
>
> > +# undef __DECL_SIMD_log
> > +# define __DECL_SIMD_log __DECL_SIMD_PPC64
> > +# undef __DECL_SIMD_logf
> > +# define __DECL_SIMD_logf __DECL_SIMD_PPC64
> >
> > undef __DECL_SIMD_exp
> >
> > ======================
> >
> > define __DECL_SIMD_exp __DECL_SIMD_PPC64
> >
> > =========================================
> >
> > -# undef __DECL_SIMD_powf
> > -# define __DECL_SIMD_powf __DECL_SIMD_PPC64
> > +# undef __DECL_SIMD_expf
> > +# define __DECL_SIMD_expf __DECL_SIMD_PPC64
>
> ... here?
>

I wanted to use the same order of defines/undefines as the math-vector.h for x86-64 at:
sysdeps/x86/fpu/bits/

Thanks.
Bert.


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