math: add scalar erff.

In round-to-nearest mode the maximum error is 1.09 ULP.

Compared to glibc-2.28 erff: throughput is about 2.2x better,
latency is about 1.5x better on some AArch64 cores (on random
input in [-4,4]).

There are further optimization and quality improvement opportunities.

1 files changed, 14 insertions, 0 deletions

diff --git a/math/math_errf.c b/math/math_errf.c
index 07154c5..771d982 100644
--- a/math/math_errf.c
+++ b/math/math_errf.c
@@ -64,3 +64,17 @@ __math_invalidf (float x)
   float y = (x - x) / (x - x);
   return isnan (x) ? y : with_errnof (y, EDOM);
 }
+
+/* Check result and set errno if necessary.  */
+
+HIDDEN float
+__math_check_uflowf (float y)
+{
+  return y == 0.0f ? with_errnof (y, ERANGE) : y;
+}
+
+HIDDEN float
+__math_check_oflowf (float y)
+{
+  return isinf (y) ? with_errnof (y, ERANGE) : y;
+}