diff options
author | Paul E. Murphy <murphyp@linux.vnet.ibm.com> | 2016-08-08 15:58:28 -0500 |
---|---|---|
committer | Paul E. Murphy <murphyp@linux.vnet.ibm.com> | 2016-08-19 11:29:43 -0500 |
commit | 4482ff226e4b286ab171f3c5841ae1f7e61780cd (patch) | |
tree | 69bf06744f290b3656ad238c0b5dd700c850f9e9 /sysdeps/ieee754/ldbl-128/x2y2m1l.c | |
parent | 01ee387015a2075c45a4e1ad45d39e50b5a6d40b (diff) | |
download | glibc-4482ff226e4b286ab171f3c5841ae1f7e61780cd.tar.gz glibc-4482ff226e4b286ab171f3c5841ae1f7e61780cd.tar.xz glibc-4482ff226e4b286ab171f3c5841ae1f7e61780cd.zip |
Merge common usage of mul_split function
A number of files share identical code for the mul_split function. This moves the duplicated function mul_split into its own header, and refactors the fma usage into a single selection macro. Likewise, mul_split when used by a long double implementation is renamed mul_splitl for clarity.
Diffstat (limited to 'sysdeps/ieee754/ldbl-128/x2y2m1l.c')
-rw-r--r-- | sysdeps/ieee754/ldbl-128/x2y2m1l.c | 37 |
1 files changed, 4 insertions, 33 deletions
diff --git a/sysdeps/ieee754/ldbl-128/x2y2m1l.c b/sysdeps/ieee754/ldbl-128/x2y2m1l.c index 733742da04..4e97885d59 100644 --- a/sysdeps/ieee754/ldbl-128/x2y2m1l.c +++ b/sysdeps/ieee754/ldbl-128/x2y2m1l.c @@ -18,9 +18,10 @@ #include <math.h> #include <math_private.h> -#include <float.h> +#include <mul_splitl.h> #include <stdlib.h> + /* Calculate X + Y exactly and store the result in *HI + *LO. It is given that |X| >= |Y| and the values are small enough that no overflow occurs. */ @@ -33,36 +34,6 @@ add_split (long double *hi, long double *lo, long double x, long double y) *lo = (x - *hi) + y; } -/* Calculate X * Y exactly and store the result in *HI + *LO. It is - given that the values are small enough that no overflow occurs and - large enough (or zero) that no underflow occurs. */ - -static inline void -mul_split (long double *hi, long double *lo, long double x, long double y) -{ -#ifdef __FP_FAST_FMAL - /* Fast built-in fused multiply-add. */ - *hi = x * y; - *lo = __builtin_fmal (x, y, -*hi); -#elif defined FP_FAST_FMAL - /* Fast library fused multiply-add, compiler before GCC 4.6. */ - *hi = x * y; - *lo = __fmal (x, y, -*hi); -#else - /* Apply Dekker's algorithm. */ - *hi = x * y; -# define C ((1LL << (LDBL_MANT_DIG + 1) / 2) + 1) - long double x1 = x * C; - long double y1 = y * C; -# undef C - x1 = (x - x1) + x1; - y1 = (y - y1) + y1; - long double x2 = x - x1; - long double y2 = y - y1; - *lo = (((x1 * y1 - *hi) + x1 * y2) + x2 * y1) + x2 * y2; -#endif -} - /* Compare absolute values of floating-point values pointed to by P and Q for qsort. */ @@ -88,8 +59,8 @@ __x2y2m1l (long double x, long double y) { long double vals[5]; SET_RESTORE_ROUNDL (FE_TONEAREST); - mul_split (&vals[1], &vals[0], x, x); - mul_split (&vals[3], &vals[2], y, y); + mul_splitl (&vals[1], &vals[0], x, x); + mul_splitl (&vals[3], &vals[2], y, y); vals[4] = -1.0L; qsort (vals, 5, sizeof (long double), compare); /* Add up the values so that each element of VALS has absolute value |