implement rint / rintf via lrint (#36)

* implement rint / rintf via lrint

* optimize

* prefer 32-bit regs for reinterpret cast

* clean unnecessary parentesis

* fixed

source/m3_config_platforms.h

@@ -168,10 +168,27 @@ typedef short           i16;
 typedef uint8_t         u8;
 typedef int8_t          i8;
 
+static inline float rintf( float arg ) {
+  union { float f; uint32_t i; } u;
+  u.f = arg;
+  uint32_t ux = u.i & 0x7FFFFFFF;
+  if (UNLIKELY(ux == 0 || ux > 0x5A000000)) {
+    return arg;
+  }
+  return (float)lrint(arg);
+}
+
+static inline double rint( double arg ) {
+  union { double f; uint32_t i[2]; } u;
+  u.f = arg;
+  uint32_t ux = u.i[1] & 0x7FFFFFFF;
+  if (UNLIKELY((ux == 0 && u.i[0] == 0) || ux > 0x433FFFFF)) {
+    return arg;
+  }
+  return (double)lrint(arg);
+}
 //TODO
-static inline float rintf( float arg ) { return .0f; }
-static inline double rint ( double arg ) { return .0; }
-static inline uint64_t strtoull ( const char* str, char ** endptr, int base ) { return 0; }
+static inline uint64_t strtoull( const char* str, char ** endptr, int base ) { return 0; }
 
 #endif