diff --git a/std/math.d b/std/math.d
index 3b6d161acba..cf50a1d1ce4 100644
--- a/std/math.d
+++ b/std/math.d
@@ -6837,12 +6837,30 @@ Unqual!(Largest!(F, G)) pow(F, G)(F x, G y) @nogc @trusted pure nothrow
         {
             // Result is real only if y is an integer
             // Check for a non-zero fractional part
-            if (y > -1.0 / real.epsilon && y < 1.0 / real.epsilon)
+            enum maxOdd = pow(2.0L, real.mant_dig) - 1.0L;
+            static if (maxOdd > ulong.max)
             {
-                long w = cast(long)y;
-                if (w != y)
+                // Generic method, for any FP type
+                if (floor(y) != y)
                     return sqrt(x); // Complex result -- create a NaN
-                if (w & 1) sign = -1.0;
+
+                const hy = ldexp(y, -1);
+                if (floor(hy) != hy)
+                    sign = -1.0;
+            }
+            else
+            {
+                // Much faster, if ulong has enough precision
+                const absY = fabs(y);
+                if (absY <= maxOdd)
+                {
+                    const uy = cast(ulong) absY;
+                    if (uy != absY)
+                        return sqrt(x); // Complex result -- create a NaN
+
+                    if (uy & 1)
+                        sign = -1.0;
+                }
             }
             x = -x;
         }
@@ -6920,6 +6938,15 @@ Unqual!(Largest!(F, G)) pow(F, G)(F x, G y) @nogc @trusted pure nothrow
     assert(isIdentical(pow(0.0, 6.0), 0.0));
     assert(isIdentical(pow(-0.0, 6.0), 0.0));
 
+    // Issue #14786 fixed
+    immutable real maxOdd = pow(2.0L, real.mant_dig) - 1.0L;
+    assert(pow(-1.0L,  maxOdd) == -1.0L);
+    assert(pow(-1.0L, -maxOdd) == -1.0L);
+    assert(pow(-1.0L, maxOdd + 1.0L) == 1.0L);
+    assert(pow(-1.0L, -maxOdd + 1.0L) == 1.0L);
+    assert(pow(-1.0L, maxOdd - 1.0L) == 1.0L);
+    assert(pow(-1.0L, -maxOdd - 1.0L) == 1.0L);
+
     // Now, actual numbers.
     assert(approxEqual(pow(two, three), 8.0));
     assert(approxEqual(pow(two, -2.5), 0.1767767));