Re: PATH Optimized with SSE2 sinf and cof for x86_32

[Dmitrieva Liubov <liubov dot dmitrieva at gmail dot 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

Attachment: new_test_cases.patch
Description: Binary data