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/dbl-64 | |
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/dbl-64')
-rw-r--r-- | sysdeps/ieee754/dbl-64/gamma_product.c | 32 | ||||
-rw-r--r-- | sysdeps/ieee754/dbl-64/lgamma_product.c | 32 | ||||
-rw-r--r-- | sysdeps/ieee754/dbl-64/x2y2m1.c | 32 |
3 files changed, 3 insertions, 93 deletions
diff --git a/sysdeps/ieee754/dbl-64/gamma_product.c b/sysdeps/ieee754/dbl-64/gamma_product.c index 7ae144aeb3..51407b196c 100644 --- a/sysdeps/ieee754/dbl-64/gamma_product.c +++ b/sysdeps/ieee754/dbl-64/gamma_product.c @@ -18,37 +18,7 @@ #include <math.h> #include <math_private.h> -#include <float.h> - -/* 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); -#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 -} +#include <mul_split.h> /* Compute the product of X + X_EPS, X + X_EPS + 1, ..., X + X_EPS + N - 1, in the form R * (1 + *EPS) where the return value R is an diff --git a/sysdeps/ieee754/dbl-64/lgamma_product.c b/sysdeps/ieee754/dbl-64/lgamma_product.c index d956575bc7..fa89337ebc 100644 --- a/sysdeps/ieee754/dbl-64/lgamma_product.c +++ b/sysdeps/ieee754/dbl-64/lgamma_product.c @@ -18,37 +18,7 @@ #include <math.h> #include <math_private.h> -#include <float.h> - -/* 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); -#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 -} +#include <mul_split.h> /* Compute the product of 1 + (T / (X + X_EPS)), 1 + (T / (X + X_EPS + 1)), ..., 1 + (T / (X + X_EPS + N - 1)), minus 1. X is such that diff --git a/sysdeps/ieee754/dbl-64/x2y2m1.c b/sysdeps/ieee754/dbl-64/x2y2m1.c index 96078888a7..75052159e1 100644 --- a/sysdeps/ieee754/dbl-64/x2y2m1.c +++ b/sysdeps/ieee754/dbl-64/x2y2m1.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 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. */ |