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 1/3] New generic sincosf




On 12/19/2017 10:05 PM, H.J. Lu wrote:
On Sun, Dec 17, 2017 at 7:26 PM, Rajalakshmi Srinivasaraghavan
<raji@linux.vnet.ibm.com> wrote:


On 12/16/2017 08:13 PM, H.J. Lu wrote:

On Sat, Dec 16, 2017 at 12:52 AM, Rajalakshmi Srinivasaraghavan
<raji@linux.vnet.ibm.com> wrote:



On 12/16/2017 02:20 AM, Joseph Myers wrote:


On Fri, 15 Dec 2017, H.J. Lu wrote:

One comment,  should sysdeps/ieee754/flt-32/s_sincos.h be renamed
to  sysdeps/ieee754/flt-32/s_sincosf.h?



Yes, I think so.


Pushed as 984ae9967b49830173490a33ae6130880f3f70d9 with the file
name changed as s_sincosf.h.  Thanks for the review.


I noticed that sysdeps/x86_64/fpu/s_sincosf.S is still faster:

Generic:

      "max": 276.971,
      "min": 10.813,
      "mean": 29.3755

SSE2:

      "max": 138.795,
      "min": 11.686,
      "mean": 22.9463

The SSE2 is 28% faster.  Do they use the same algorithm?


Yes, they are same. One small difference is generic version calls
reduced_sin() and reduced_cos() whereas asm version handles both in the same
branch for reconstruction.


Can generic version do the same?

I don't see any difference in mean value with that change.





--
Thanks
Rajalakshmi S


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