Cygwin: math: Properly propagate input NANs in a few functions

While the C99 standard doesn't explicitly require this, the standard
says it is recommended (F.9.13).

Signed-off-by: Martin Storsjö <martin@martin.st>

winsup/cygwin/math/acosh.def.h

@@ -50,7 +50,12 @@ __FLT_TYPE
 __FLT_ABI(acosh) (__FLT_TYPE x)
 {
   int x_class = fpclassify (x);
-  if (x_class == FP_NAN || x < __FLT_CST(1.0))
+  if (x_class == FP_NAN)
+    {
+      errno = EDOM;
+      return x;
+    }
+  else if (x < __FLT_CST(1.0))
     {
       errno = EDOM;
       return __FLT_NAN;

winsup/cygwin/math/erfl.c

@@ -243,6 +243,9 @@ long double erfcl(long double a)
 	if (isinf (a))
 		return (signbit(a) ? 2.0 : 0.0);
 
+	if (isnan (a))
+		return (a);
+
 	x = fabsl (a);
 
 	if (x < 1.0L)

winsup/cygwin/math/lgammal.c

@@ -216,7 +216,7 @@ long double __lgammal_r(long double x, int* sgngaml)
 	*sgngaml = 1;
 #ifdef NANS
 	if (isnanl(x))
-		return(NANL);
+		return x;
 #endif
 #ifdef INFINITIES
 	if (!isfinitel(x))

winsup/cygwin/math/log.def.h

@@ -56,6 +56,8 @@ __FLT_ABI(log) (__FLT_TYPE x)
       errno = ERANGE;
       return -__FLT_HUGE_VAL;
     }
+  else if (x_class == FP_NAN)
+    return x;
   else if (signbit (x))
     {
       errno = EDOM;
@@ -63,7 +65,5 @@ __FLT_ABI(log) (__FLT_TYPE x)
     }
   else if (x_class == FP_INFINITE)
     return __FLT_HUGE_VAL;
-  else if (x_class == FP_NAN)
-    return __FLT_NAN;
   return (__FLT_TYPE) __logl_internal ((long double) x);
 }

winsup/cygwin/math/pow.def.h

@@ -121,9 +121,13 @@ __FLT_ABI(pow) (__FLT_TYPE x, __FLT_TYPE y)
     return __FLT_CST(1.0);
   else if (x_class == FP_NAN || y_class == FP_NAN)
     {
-      rslt = (signbit(x) ? -__FLT_NAN : __FLT_NAN);
+      if (x_class == FP_NAN) {
         errno = EDOM;
-      return rslt;
+        return x;
+      } else {
+        errno = EDOM;
+        return y;
+      }
     }
   else if (x_class == FP_ZERO)
     {

winsup/cygwin/math/tgammal.c

@@ -272,7 +272,7 @@ long double __tgammal_r(long double x, int* sgngaml)
 	*sgngaml = 1;
 #ifdef NANS
 	if (isnanl(x))
-		return (NANL);
+		return x;
 #endif
 #ifdef INFINITIES
 #ifdef NANS