[Bug math/14473] New: Inaccurate CPOWF function on x86_32 and x86_64

[liubov.dmitrieva at gmail dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14473

             Bug #: 14473
           Summary: Inaccurate CPOWF  function on x86_32 and x86_64
           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 <math.h>
#include <complex.h>
#include <fenv.h>
#include <stdlib.h>  // abs function for int argument

#define ireal(a)    (((int*)&(a))[0])
#define iimag(a)    (((int*)&(a))[1])
#define  real(a)    (((float*)&(a))[0])
#define  imag(a)    (((float*)&(a))[1])

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

    ireal(a)=0xbf0e38e4;    iimag(a)=0xbde38e3e;    //argument
    ireal(b)=0x3f0e38e3;    iimag(b)=0xbde38e3e;    //argument

    fesetround(FE_TONEAREST);

    r1 = cpowf(a,b);  //actual result
    r2 = cpow(a,b);   //actual result 2

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

    return 0;
}

Result x86_32 (core2):

inputs: a.real = (0xbf0e38e4)   -5.555556e-01
        a.imag = (0xbde38e3e)   -1.111111e-01

inputs: b.real = (0x3f0e38e3)   5.555555e-01
        b.imag = (0xbde38e3e)   -1.111111e-01

actual.real    = (0xba6f8923)   -9.1375614283606e-04
expected.real  = (0xba6f8d77)   -9.1382063692436e-04
error.real     = 1108 ulp

actual.imag    = (0xbf069c6d)   -5.2582436800003e-01
expected.imag  = (0xbf069c6d)   -5.2582436800003e-01
error.imag     = 0 ulp


Result x86_64 (core2):

inputs: a.real = (0xbf0e38e4)   -5.555556e-01
        a.imag = (0xbde38e3e)   -1.111111e-01

inputs: b.real = (0x3f0e38e3)   5.555555e-01
        b.imag = (0xbde38e3e)   -1.111111e-01

actual.real    = (0xba6f8923)   -9.1375614283606e-04
expected.real  = (0xba6f8d77)   -9.1382063692436e-04
error.real     = 1108 ulp

actual.imag    = (0xbf069c6d)   -5.2582436800003e-01
expected.imag  = (0xbf069c6d)   -5.2582436800003e-01
error.imag     = 0 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.