aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/ieee754/ldbl-128ibm
diff options
context:
space:
mode:
authorPaul E. Murphy <murphyp@linux.vnet.ibm.com>2016-08-08 15:58:28 -0500
committerPaul E. Murphy <murphyp@linux.vnet.ibm.com>2016-08-19 11:29:43 -0500
commit4482ff226e4b286ab171f3c5841ae1f7e61780cd (patch)
tree69bf06744f290b3656ad238c0b5dd700c850f9e9 /sysdeps/ieee754/ldbl-128ibm
parent01ee387015a2075c45a4e1ad45d39e50b5a6d40b (diff)
downloadglibc-4482ff226e4b286ab171f3c5841ae1f7e61780cd.zip
glibc-4482ff226e4b286ab171f3c5841ae1f7e61780cd.tar.gz
glibc-4482ff226e4b286ab171f3c5841ae1f7e61780cd.tar.bz2
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-128ibm')
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/s_fmal.c27
-rw-r--r--sysdeps/ieee754/ldbl-128ibm/x2y2m1l.c32
2 files changed, 2 insertions, 57 deletions
diff --git a/sysdeps/ieee754/ldbl-128ibm/s_fmal.c b/sysdeps/ieee754/ldbl-128ibm/s_fmal.c
index 177a048..1405763 100644
--- a/sysdeps/ieee754/ldbl-128ibm/s_fmal.c
+++ b/sysdeps/ieee754/ldbl-128ibm/s_fmal.c
@@ -22,6 +22,7 @@
#include <math.h>
#include <math_private.h>
#include <math_ldbl_opt.h>
+#include <mul_split.h>
#include <stdlib.h>
/* Calculate X + Y exactly and store the result in *HI + *LO. It is
@@ -36,32 +37,6 @@ add_split (double *hi, double *lo, double x, 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 void
-mul_split (double *hi, double *lo, double x, double y)
-{
-#ifdef __FP_FAST_FMA
- /* Fast built-in fused multiply-add. */
- *hi = x * y;
- *lo = __builtin_fma (x, y, -*hi);
-#else
- /* Apply Dekker's algorithm. */
- *hi = x * y;
-# define C ((1 << (DBL_MANT_DIG + 1) / 2) + 1)
- double x1 = x * C;
- double y1 = y * C;
-# undef C
- x1 = (x - x1) + x1;
- y1 = (y - y1) + y1;
- double x2 = x - x1;
- double y2 = y - y1;
- *lo = (((x1 * y1 - *hi) + x1 * y2) + x2 * y1) + x2 * y2;
-#endif
-}
-
/* Value with extended range, used in intermediate computations. */
typedef struct
{
diff --git a/sysdeps/ieee754/ldbl-128ibm/x2y2m1l.c b/sysdeps/ieee754/ldbl-128ibm/x2y2m1l.c
index da2e929..a8b4895 100644
--- a/sysdeps/ieee754/ldbl-128ibm/x2y2m1l.c
+++ b/sysdeps/ieee754/ldbl-128ibm/x2y2m1l.c
@@ -18,7 +18,7 @@
#include <math.h>
#include <math_private.h>
-#include <float.h>
+#include <mul_split.h>
#include <stdlib.h>
/* Calculate X + Y exactly and store the result in *HI + *LO. It is
@@ -33,36 +33,6 @@ add_split (double *hi, double *lo, double x, 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 (double *hi, double *lo, double x, double y)
-{
-#ifdef __FP_FAST_FMA
- /* Fast built-in fused multiply-add. */
- *hi = x * y;
- *lo = __builtin_fma (x, y, -*hi);
-#elif defined FP_FAST_FMA
- /* Fast library fused multiply-add, compiler before GCC 4.6. */
- *hi = x * y;
- *lo = __fma (x, y, -*hi);
-#else
- /* Apply Dekker's algorithm. */
- *hi = x * y;
-# define C ((1 << (DBL_MANT_DIG + 1) / 2) + 1)
- double x1 = x * C;
- double y1 = y * C;
-# undef C
- x1 = (x - x1) + x1;
- y1 = (y - y1) + y1;
- double x2 = x - x1;
- 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. */