Optimize the fixup for +-0 by using better classification for this case

and by using a table lookup to avoid a branch when this case occurs.
On i386, this saves 1-4 cycles out of about 64 for non-large args.

1 files changed, 4 insertions, 2 deletions

diff --git a/lib/msun/src/s_rintl.c b/lib/msun/src/s_rintl.c
index b2a85b6..7f3b444e 100644
--- a/lib/msun/src/s_rintl.c
+++ b/lib/msun/src/s_rintl.c
@@ -44,6 +44,7 @@ shift[2] = {
 #error "Unsupported long double format"
 #endif
 };
+static const float zero[2] = { 0.0, -0.0 };
 
 long double
 rintl(long double x)
@@ -74,7 +75,8 @@ rintl(long double x)
 	 * If the result is +-0, then it must have the same sign as x, but
 	 * the above calculation doesn't always give this.  Fix up the sign.
 	 */
-	if (x == 0.0L)
-		return (sign ? -0.0L : 0.0L);
+	if (u.bits.exp < BIAS && x == 0.0L)
+		return (zero[sign]);
+
 	return (x);
 }