math: fix spurious underflow in erff and erf

The code relied on the final x + c*x to be done via an fma, otherwise
the intermediate c*x could underflow for tiny (almost subnormal) x.

Use explicit fmaf like elsewhere (this code is not expected to be
fast when fma is not inlined, but at least it should be correct).

Author: nsz-arm

math/erf.c

@@ -47,7 +47,7 @@ erf (double x)
 	{ /* a < 2^(-28).  */
 	  if (ia < 0x00800000)
 	    { /* a < 2^(-1015).  */
-	      double y =  x + TwoOverSqrtPiMinusOne * x;
+	      double y =  fma (TwoOverSqrtPiMinusOne, x, x);
 	      return check_uflow (y);
 	    }
 	  return x + TwoOverSqrtPiMinusOne * x;

math/erff.c

@@ -45,7 +45,7 @@ erff (float x)
 	{ /* |x| < 2^(-28).  */
 	  if (unlikely (ia12 < 0x040))
 	    { /* |x| < 2^(-119).  */
-	      float y = x + TwoOverSqrtPiMinusOne * x;
+	      float y = fmaf (TwoOverSqrtPiMinusOne, x, x);
 	      return check_uflowf (y);
 	    }
 	  return x + TwoOverSqrtPiMinusOne * x;