Add some funcs with fast approximation for double as well

Actually all of them can be done with /fp:fast, but I don't bother enabling it

src/KKdLib/prj/math.hpp

@@ -7,6 +7,7 @@
 
 #include "../default.hpp"
 #include <xmmintrin.h>
+#include <emmintrin.h>
 
 namespace prj {
     inline int32_t extract_sign(float_t x) {
@@ -40,4 +41,36 @@ namespace prj {
         else
             return ceilf(x);
     }
+
+    inline int32_t extract_sign(double_t x) {
+        return _mm_movemask_pd(_mm_set_sd(x)) & 0x01;
+    }
+
+    inline double_t ceil(double_t x) {
+        int64_t x_int = (int64_t)x;
+        if (x_int != 0x8000000000000000 && (float_t)x_int != x)
+            x = (float_t)(x_int + !extract_sign(x));
+        return x;
+    }
+
+    inline double_t floor(double_t x) {
+        int64_t x_int = (int64_t)x;
+        if (x_int != 0x8000000000000000 && (float_t)x_int != x)
+            x = (float_t)(x_int - extract_sign(x));
+        return x;
+    }
+
+    inline double_t roundf(double_t x) {
+        if (x >= 0.0f)
+            return floor(x + 0.5f);
+        else
+            return ceil(x - 0.5f);
+    }
+
+    inline double_t truncf(double_t x) {
+        if (x >= 0.0f)
+            return floor(x);
+        else
+            return ceil(x);
+    }
 }