This is the mail archive of the mailing list for the glibc project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[PATCH] Remove unnecessary multiplication in __mpexp


In __mpexp, B is first reduced by multiplying with RADIXI (2^-24) and
then multiplied by 2 till it crosses 1/2.  It's easier (i.e. less
number of instructions) to just multiply by 2 till it crosses half of
the mp_no radix (2^23).  Verified that there are no regressions
resulting from this on x86_64 and ppc64 and that it also reduces an
instruction on x86_64 and 4 instructions on ppc64.  This is just a
micro-cleanup and hence has no noticeable performance impact.  OK to


	* sysdeps/ieee754/dbl-64/mpexp.c (__mpexp): Remove unnecessary

diff --git a/sysdeps/ieee754/dbl-64/mpexp.c b/sysdeps/ieee754/dbl-64/mpexp.c
index aca6bf6..811fb46 100644
--- a/sysdeps/ieee754/dbl-64/mpexp.c
+++ b/sysdeps/ieee754/dbl-64/mpexp.c
@@ -103,14 +103,14 @@ __mpexp (mp_no *x, mp_no *y, int p)
     a *= RADIXI;
   for (; i > EX; i--)
     a *= RADIX;
-  b = X[1] * RADIXI;
+  b = X[1];
   m2 = 24 * EX;
-  for (; b < HALF; m2--)
+  for (; b < HALFRAD; m2--)
       a *= TWO;
       b *= TWO;
-  if (b == HALF)
+  if (b == HALFRAD)
       for (i = 2; i <= p; i++)

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]