diff options
| author | Joseph Myers <joseph@codesourcery.com> | 2015-09-26 00:27:06 +0000 |
|---|---|---|
| committer | Joseph Myers <joseph@codesourcery.com> | 2015-09-26 00:27:06 +0000 |
| commit | fa752c698146ca3e9f7747d33059fbef9bb02b0e (patch) | |
| tree | a41295f8a99770d272a53c714effd0f5749b2735 /sysdeps | |
| parent | 6ace39382120dd0668b2168591611f71b04ac8f3 (diff) | |
| download | glibc-fa752c698146ca3e9f7747d33059fbef9bb02b0e.zip glibc-fa752c698146ca3e9f7747d33059fbef9bb02b0e.tar.gz glibc-fa752c698146ca3e9f7747d33059fbef9bb02b0e.tar.bz2 | |
Fix powf inaccuracy (bug 18956).
The flt-32 version of powf can be inaccurate because of bugs in the
extra-precision calculation of (x-1)/(x+1) or (x-1.5)/(x+1.5) as part
of calculating log(x) with extra precision: a constant used (as part
of adding 1 or 1.5 through integer arithmetic) is incorrect, and then
the code fails to mask a computed high part before using it in
arithmetic that relies on s_h*t_h being exactly representable. This
patch fixes these bugs.
Tested for x86_64 and x86. x86_64 ulps for powf removed and
regenerated to reflect reduced ulps from the increased accuracy for
existing tests.
[BZ #18956]
* sysdeps/ieee754/flt-32/e_powf.c (__ieee754_powf): Add 0x00400000
not 0x0040000 for high bit of mantissa. Mask with 0xfffff000 when
extracting high part.
* math/auto-libm-test-in: Add another test of pow.
* math/auto-libm-test-out: Regenerated.
* sysdeps/x86_64/fpu/libm-test-ulps: Update.
Diffstat (limited to 'sysdeps')
| -rw-r--r-- | sysdeps/ieee754/flt-32/e_powf.c | 4 | ||||
| -rw-r--r-- | sysdeps/x86_64/fpu/libm-test-ulps | 16 |
2 files changed, 11 insertions, 9 deletions
diff --git a/sysdeps/ieee754/flt-32/e_powf.c b/sysdeps/ieee754/flt-32/e_powf.c index 1804296..c72fe37 100644 --- a/sysdeps/ieee754/flt-32/e_powf.c +++ b/sysdeps/ieee754/flt-32/e_powf.c @@ -166,7 +166,9 @@ __ieee754_powf(float x, float y) GET_FLOAT_WORD(is,s_h); SET_FLOAT_WORD(s_h,is&0xfffff000); /* t_h=ax+bp[k] High */ - SET_FLOAT_WORD(t_h,((ix>>1)|0x20000000)+0x0040000+(k<<21)); + SET_FLOAT_WORD (t_h, + ((((ix>>1)|0x20000000)+0x00400000+(k<<21)) + & 0xfffff000)); t_l = ax - (t_h-bp[k]); s_l = v*((u-s_h*t_h)-s_h*t_l); /* compute log(ax) */ diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps index 127a8e1..12c3dd1 100644 --- a/sysdeps/x86_64/fpu/libm-test-ulps +++ b/sysdeps/x86_64/fpu/libm-test-ulps @@ -1966,8 +1966,8 @@ Function: "log_vlen8_avx2": float: 2 Function: "pow": -float: 3 -ifloat: 3 +float: 1 +ifloat: 1 ildouble: 1 ldouble: 1 @@ -2003,25 +2003,25 @@ ldouble: 2 Function: "pow_downward": double: 1 -float: 4 +float: 1 idouble: 1 -ifloat: 4 +ifloat: 1 ildouble: 4 ldouble: 4 Function: "pow_towardzero": double: 1 -float: 8 +float: 1 idouble: 1 -ifloat: 8 +ifloat: 1 ildouble: 1 ldouble: 1 Function: "pow_upward": double: 1 -float: 8 +float: 1 idouble: 1 -ifloat: 8 +ifloat: 1 ildouble: 2 ldouble: 2 |