Math: enlarge epsilon value for long doubles on MSVC.

They are internally treated as 64-bit doubles, so there's no difference.

src/Magnum/Math/Test/TypeTraitsTest.cpp

@@ -68,9 +68,27 @@ struct {
     long double get(long double) const { return c; }
     long double getStep(long double) const { return cStep; }
 } EqualsZeroData[EqualsZeroDataCount] = {
-    {"", -3.141592653589793f, 5.0e-5f, -3.141592653589793, 5.0e-14, -3.141592653589793l, 5.0e-17l},
+    {"", -3.141592653589793f, 5.0e-5f, -3.141592653589793, 5.0e-14, -3.141592653589793l,
-    {"small", 1.0e-6f, 5.0e-6f, -1.0e-15, 5.0e-15, 1.0e-18l, 5.0e-18l},
+        #ifndef _MSC_VER
-    {"large", 12345.0f, 0.2f, 12345678901234.0, 0.2, -12345678901234567.0l, 0.2l},
+        5.0e-17l
+        #else
+        5.0e-14
+        #endif
+    },
+    {"small", 1.0e-6f, 5.0e-6f, -1.0e-15, 5.0e-15, 1.0e-18l,
+        #ifndef _MSC_VER
+        5.0e-18l
+        #else
+        5.0e-15
+        #endif
+    },
+    {"large", 12345.0f, 0.2f, 12345678901234.0, 0.2,
+        #ifndef _MSC_VER
+        -12345678901234567.0l,
+        #else
+        -12345678901234.0l,
+        #endif
+        0.2l},
 };
 
 }

src/Magnum/Math/TypeTraits.h

@@ -58,9 +58,17 @@ for more headroom.
 
 They have "at least" 18 significant digits of precision, taking one digit less
 for more headroom.
+
+@attention On MSVC the precision is the same as for doubles, because
+    they are internally the same type. Source:
+    https://msdn.microsoft.com/en-us/library/9cx8xs15.aspx
 */
 #ifndef LONG_DOUBLE_EQUALITY_PRECISION
+#ifndef _MSC_VER
 #define LONG_DOUBLE_EQUALITY_PRECISION 1.0e-17l
+#else
+#define LONG_DOUBLE_EQUALITY_PRECISION 1.0e-14l
+#endif
 #endif
 
 namespace Magnum { namespace Math {