[Bug math/14472] New: Inaccurate y1f function on x86_32

[Bug math/14472] New: Inaccurate y1f function on x86_32

[liubov.dmitrieva at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14472

             Bug #: 14472
           Summary: Inaccurate y1f function on x86_32
           Product: glibc
           Version: 2.17
            Status: NEW
          Severity: normal
          Priority: P2
         Component: math
        AssignedTo: unassigned@sourceware.org
        ReportedBy: liubov.dmitrieva@gmail.com
    Classification: Unclassified


// gcc main.c -std=c99 -lm -m32

#include <stdio.h>
#include <fenv.h>
#include <stdlib.h>  // abs function for int argument

extern float y1f(float);
extern double y1(double);

#define i32(a)    (((int*)&(a))[0])

int main() {
    float a,r1,r2;

    i32(a)=0x422cc0bc;    //argument

    fesetround(FE_TONEAREST); // FE_TONEAREST

    r1 = y1f(a);  //actual result
    r2 = y1(a);   //actual result 2

    printf("inputs: a = (0x%08x)\t%e\n", i32(a), a);
    printf("actual    = (0x%08x)\t%.13e\n", i32(r1), r1);
    printf("expected  = (0x%08x)\t%.13e\n", i32(r2), r2);
    printf("error     = %d ulp\n", abs(i32(r1) - i32(r2)));

    return 0;
}

Result:

inputs: a = (0x422cc0bc)        4.318822e+01
actual    = (0x33f3b02b)        1.1347614048418e-07
expected  = (0x33f3934c)        1.1342362427058e-07
error     = 7391 ulp

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug math/14472] Inaccurate y1f function on x86_32

[liubov.dmitrieva at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14472

--- Comment #1 from Liubov Dmitrieva <liubov.dmitrieva at gmail dot com> 2012-08-15 14:31:17 UTC ---
The same test case shows bad accuracy for x86_64 as well:
(core2 machine)

-bash-4.2$ gcc main_y1f.c -lm -odo
-bash-4.2$ ./do
inputs: a = (0x422cc0bc)        4.318822e+01
actual    = (0x33f3a4bd)        1.1345535000373e-07
expected  = (0x33f3934c)        1.1342362427058e-07
error     = 4465 ulp

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug math/14472] Inaccurate y1f function on x86_32 and x86_64

[liubov.dmitrieva at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14472

Liubov Dmitrieva <liubov.dmitrieva at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Inaccurate y1f function on  |Inaccurate y1f function on
                   |x86_32                      |x86_32 and x86_64

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug math/14472] Inaccurate y1f function on x86_32 and x86_64

[liubov.dmitrieva at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14472

--- Comment #2 from Liubov Dmitrieva <liubov.dmitrieva at gmail dot com> 2012-08-16 08:23:12 UTC ---
>> represent the test argument using a hex float value

Updated test:

#include <stdio.h>
#include <math.h>
#include <fenv.h>

extern float y1f(float);
extern double y1(double);

int main() {
    float a,r1;
    double r2;

    a=0x1.598178p5;    //argument

    fesetround(FE_TONEAREST);

    r1 = y1f(a);  //actual result
    r2 = y1(a);   //actual result 2

    printf("inputs: a = %.6a\n", a);
    printf("actual    = %.6a\n", r1);
    printf("expected  = %.6a\n", r2);
    printf("error     = %f ulp\n", fabs(r1 - r2) * exp2(23.0-logbf(r1)));
    return 0;
}

Result x86:
inputs: a = 0x1.598178p+5
actual    = 0x1.e76056p-24
expected  = 0x1.e72698p-24
error     = 7390.926985 ulp

Result x86_64:
inputs: a = 0x1.598178p+5
actual    = 0x1.e7497ap-24
expected  = 0x1.e72698p-24
error     = 4464.927009 ulp

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug math/14472] Inaccurate y1f function

[jsm28 at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=14472

Joseph Myers <jsm28 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Inaccurate y1f function on  |Inaccurate y1f function
                   |x86_32 and x86_64           |

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/14472] Inaccurate y1f function

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=14472

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|                            |security-

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/14472] Inaccurate y1f function

[zimmerma+gcc at loria dot fr]

https://sourceware.org/bugzilla/show_bug.cgi?id=14472

--- Comment #4 from Paul Zimmermann <zimmerma+gcc at loria dot fr> ---
with glibc-2.32 the maximum error on x86_64 is now 6177902 ulps (for
x=3.153646966e+38). This is the same value where the maximal error is obtained
for j0. This is not surprising since y1(x) ~ -j0(x) for large x, probably the
same code is used (I did not check).

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug math/14472] Inaccurate y1f function

[zimmerma+gcc at loria dot fr]

https://sourceware.org/bugzilla/show_bug.cgi?id=14472

Paul Zimmermann <zimmerma+gcc at loria dot fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
   Target Milestone|---                         |2.34
         Resolution|---                         |FIXED

--- Comment #5 from Paul Zimmermann <zimmerma+gcc at loria dot fr> ---
fixed by commit 9acda61

-- 
You are receiving this mail because:
You are on the CC list for the bug.