This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: PATCH: SSE sin and cos for x86 and x86_64
On Thursday, August 23, 2012 17:28:26 Dmitrieva Liubov wrote:
> Ping.
>
> SSE2 sin and cos patches (up to 10 times faster) are waiting for
> review and my patch with new test cases as well.
> I've attached all 3 again (the same as last time)
The patches are all fine with one minor nit: Please reformat them so that
we do not have overlong lines (> 78 chars). A couple of comments are
going over.
Note: do not wrap libm-test.inc, those long lines are fine.
> #1
> 2012-08-16 Liubov Dmitrieva <liubov.dmitrieva@gmail.com>
>
> * math/libm-test.inc: Update
> Add new test cases in large arguments path.
This should name the functions you change:
* math/libm-test.inc (cos_test): Add more test cases.
(sin_test): Likewise.
(sincos_test): Likewise.
> #2
> 2012-08-13 Liubov Dmitrieva <liubov.dmitrieva@gmail.com>
>
> * sysdeps/i386/i686/fpu/multiarch/Makefile: Update
> (sysdep_routines): Add s_sinf-sse2, s_conf-sse2
Just name the variable you update, no need to say "Update" for the file:
* sysdeps/i386/i686/fpu/multiarch/Makefile (sysdep_routines):
Add s_sinf-sse2, s_conf-sse2
>
> * sysdeps/i386/i686/fpu/multiarch/s_sinf-sse2.S: New file
> * sysdeps/i386/i686/fpu/multiarch/s_cosf-sse2.S: New file
> * sysdeps/i386/i686/fpu/multiarch/s_sinf.c: New file
> * sysdeps/i386/i686/fpu/multiarch/s_cosf.c: New file
> * sysdeps/ieee754/flt-32/s_sinf.c: Update
> (SINF): Add macro for using routine as __sinf_ia32
see below
> * sysdeps/ieee754/flt-32/s_cosf.c: Update
> (COSF): Add macro for using routine as __cosf_ia32
Don't update the Sun copyrights.
And write the changelog as:
* sysdeps/ieee754/flt-32/s_cosf.c (COSF, COFS_FUNC): Add macros
for using routine as __cosf_ia32.
Use macro for function declaration and weak_alias.
> * sysdeps/i386/i686/fpu/multiarch/e_expf-sse2.S: Fix
> Copyright * sysdeps/i386/i686/fpu/multiarch/e_expf.c: Fix Copyright
>
> #3
> 2012-08-15 Liubov Dmitrieva <liubov.dmitrieva@gmail.com>
>
> * sysdeps/x86_64/fpu/s_sinf.S: New file.
> * sysdeps/x86_64/fpu/s_cosf.S: New file.
> * sysdeps/x86_64/fpu/libm-test-ulps: Update.
Could you resend everything with the above changes, please? I'll commit
then on your behalf...
Thanks a lot!
Andreas
> --
> Liubov Dmitrieva
> Intel Corporation
>
> 2012/8/16 Dmitrieva Liubov <liubov.dmitrieva@gmail.com>:
> > False alarm. Our functions work correctly.
> >
> > But anyway I have a separate patch for adding that test cases to
> > "make check" (attached).
> > It does not reveal new fails in current GLIBC, but several 1-ulp new
> > errors on IA.
> > And it does not reveal new fails in our new sinf/cosf functions
> > (and
> > no 1-ulp new errors)
> >
> >
> > 2012-08-16 Liubov Dmitrieva <liubov.dmitrieva@gmail.com>
> >
> > * math/libm-test.inc: Update
> > Add new test cases in large arguments path.
> >
> > So, no need to fix here, both patches are ok.
> >
> > The latest version were attached to:
> > http://sourceware.org/ml/libc-alpha/2012-08/msg00267.html
> > http://sourceware.org/ml/libc-alpha/2012-08/msg00265.html
> >
> > --
> > Liubov Dmitrieva
> >
> > 2012/8/15 Joseph S. Myers <joseph@codesourcery.com>:
> >> On Wed, 15 Aug 2012, Dmitrieva Liubov wrote:
> >>> > This code is wrong. You cannot perform argument reduction for
> >>> > large
> >>> > arguments by using a single, double, or even extended-precision
> >>> > approximation for pi.
> >>>
> >>> Yes, that's wrong in x86_32 version and will be fixed but 64 bit
> >>> version looks ok.
> >>
> >> If that didn't get detected by the testsuite, I suppose we should
> >> add
> >> 0x1p+120 (or some such value that detects the problem) to the tests
> >> for cos and sin in libm-test.inc. (The larest float value for cos
> >> in the testsuite is 0x1p65; sin also tests 0x1.7f4134p+103.)
> >>
> >> --
> >> Joseph S. Myers
> >> joseph@codesourcery.com
--
Andreas Jaeger aj@{suse.com,opensuse.org} Twitter/Identica: jaegerandi
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn,Jennifer Guild,Felix Imendörffer,HRB16746 (AG Nürnberg)
GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126