Remove tacit double usage in ldbl-128
Commit Message
There is quiet truncation to double arithmetic in several
files. I noticed them when building ldbl-128 in a
soft-fp context. This did not change any test results.
* sysdeps/ieee754/ldbl-128/k_tanl.c (__kernal_tanl):
Use fabsl instead of fabs.
* sysdeps/ieee754/ldbl-128/s_logbl.c (__logbl):
Likewise.
* sysdeps/ieee754/ldbl-128/s_atanl.c (__kernel_tanl):
Add long double suffix to constant to ensure proper
promotion to long double.
---
sysdeps/ieee754/ldbl-128/k_tanl.c | 2 +-
sysdeps/ieee754/ldbl-128/s_atanl.c | 2 +-
sysdeps/ieee754/ldbl-128/s_logbl.c | 2 +-
3 files changed, 3 insertions(+), 3 deletions(-)
Comments
On Tue, 2 Aug 2016, Paul E. Murphy wrote:
> There is quiet truncation to double arithmetic in several
> files. I noticed them when building ldbl-128 in a
> soft-fp context. This did not change any test results.
>
> * sysdeps/ieee754/ldbl-128/k_tanl.c (__kernal_tanl):
> Use fabsl instead of fabs.
> * sysdeps/ieee754/ldbl-128/s_logbl.c (__logbl):
> Likewise.
> * sysdeps/ieee754/ldbl-128/s_atanl.c (__kernel_tanl):
> Add long double suffix to constant to ensure proper
> promotion to long double.
OK.
On 08/02/2016 01:57 PM, Paul E. Murphy wrote:
> There is quiet truncation to double arithmetic in several
> files. I noticed them when building ldbl-128 in a
> soft-fp context. This did not change any test results.
>
> * sysdeps/ieee754/ldbl-128/k_tanl.c (__kernal_tanl):
> Use fabsl instead of fabs.
> * sysdeps/ieee754/ldbl-128/s_logbl.c (__logbl):
> Likewise.
> * sysdeps/ieee754/ldbl-128/s_atanl.c (__kernel_tanl):
> Add long double suffix to constant to ensure proper
> promotion to long double.
Committed as cad1d60.
@@ -98,7 +98,7 @@ __kernel_tanl (long double x, long double y, int iy)
{ /* generate inexact */
if ((ix | u.parts32.w1 | u.parts32.w2 | u.parts32.w3
| (iy + 1)) == 0)
- return one / fabs (x);
+ return one / fabsl (x);
else if (iy == 1)
{
math_check_force_underflow (x);
@@ -231,7 +231,7 @@ __atanl (long double x)
Roundoff to integer is asymmetrical to avoid cancellation when t < 0
(cf. fdlibm). */
k = 8.0 * x + 0.25;
- u = 0.125 * k;
+ u = 0.125L * k;
/* Small arctan argument. */
t = (x - u) / (1.0 + x * u);
}
@@ -34,7 +34,7 @@ __logbl (long double x)
GET_LDOUBLE_WORDS64 (hx, lx, x);
hx &= 0x7fffffffffffffffLL; /* high |x| */
if ((hx | lx) == 0)
- return -1.0 / fabs (x);
+ return -1.0 / fabsl (x);
if (hx >= 0x7fff000000000000LL)
return x * x;
if ((ex = hx >> 48) == 0) /* IEEE 754 logb */