win: libm: add implementation for round/fmin/fmax and variants

round and fmin/fmax are relatively commonly used functions but were not
implemented anywhere in the tcc Windows distribution package. Newer mingw(64)
math.h stil doesn't include these implementations.

Add C implementations for these functions and place it as inline functions at
win32/include/tcc/tcc_libm.h - which is already included from math.h .

The code is mostly taken from musl-libc rs-1.0 (MIT) [*],

musl-libc: http://git.musl-libc.org/cgit/musl/tree/src/math?h=rs-1.0
license:   http://git.musl-libc.org/cgit/musl/tree/COPYRIGHT?h=rs-1.0

Potential enhancements:
- Check how many useful libm implementations are still missing and consider
  adding them (some of them already work via the MS runtime).
- Consider putting libm implementations in an actual libm.a file, or add a dummy
  one such that build processes which try to link with libm will not fail.

win32/include/tcc/tcc_libm.h

@@ -65,6 +65,7 @@ __CRT_INLINE int __cdecl __fpclassifyl (long double x) {
   return __fpclassify(x);
 }
 
+
 /* signbit */
 
 __CRT_INLINE int __cdecl __signbit (double x) {
@@ -81,6 +82,118 @@ __CRT_INLINE int __cdecl __signbitl (long double x) {
   return __signbit(x);
 }
 
+
+/* fmin*, fmax* */
+
+#define TCCFP_FMIN_EVAL (isnan(x) ? y :                                      \
+                         isnan(y) ? x :                                      \
+                         (signbit(x) != signbit(y)) ? (signbit(x) ? x : y) : \
+                         x < y ? x : y)
+
+__CRT_INLINE double __cdecl fmin (double x, double y) {
+  return TCCFP_FMIN_EVAL;
+}
+
+__CRT_INLINE float __cdecl fminf (float x, float y) {
+  return TCCFP_FMIN_EVAL;
+}
+
+__CRT_INLINE long double __cdecl fminl (long double x, long double y) {
+  return TCCFP_FMIN_EVAL;
+}
+
+#define TCCFP_FMAX_EVAL (isnan(x) ? y :                                      \
+                         isnan(y) ? x :                                      \
+                         (signbit(x) != signbit(y)) ? (signbit(x) ? y : x) : \
+                         x < y ? y : x)
+
+__CRT_INLINE double __cdecl fmax (double x, double y) {
+  return TCCFP_FMAX_EVAL;
+}
+
+__CRT_INLINE float __cdecl fmaxf (float x, float y) {
+  return TCCFP_FMAX_EVAL;
+}
+
+__CRT_INLINE long double __cdecl fmaxl (long double x, long double y) {
+  return TCCFP_FMAX_EVAL;
+}
+
+
+/* *round* */
+
+#define TCCFP_FORCE_EVAL(x) do {            \
+if (sizeof(x) == sizeof(float)) {           \
+  volatile float __x;                       \
+  __x = (x);                                \
+} else if (sizeof(x) == sizeof(double)) {   \
+  volatile double __x;                      \
+  __x = (x);                                \
+} else {                                    \
+  volatile long double __x;                 \
+  __x = (x);                                \
+}                                           \
+} while(0)
+
+__CRT_INLINE double __cdecl round (double x) {
+  union {double f; uint64_t i;} u = {x};
+  int e = u.i >> 52 & 0x7ff;
+  double y;
+
+  if (e >= 0x3ff+52)
+    return x;
+  if (u.i >> 63)
+    x = -x;
+  if (e < 0x3ff-1) {
+    /* raise inexact if x!=0 */
+    TCCFP_FORCE_EVAL(x + 0x1p52);
+    return 0*u.f;
+  }
+  y = (double)(x + 0x1p52) - 0x1p52 - x;
+  if (y > 0.5)
+    y = y + x - 1;
+  else if (y <= -0.5)
+    y = y + x + 1;
+  else
+    y = y + x;
+  if (u.i >> 63)
+    y = -y;
+  return y;
+}
+
+__CRT_INLINE long __cdecl lround (double x) {
+  return round(x);
+}
+
+__CRT_INLINE long long __cdecl llround (double x) {
+  return round(x);
+}
+
+__CRT_INLINE float __cdecl roundf (float x) {
+  return round(x);
+}
+
+__CRT_INLINE long __cdecl lroundf (float x) {
+  return round(x);
+}
+
+__CRT_INLINE long long __cdecl llroundf (float x) {
+  return round(x);
+}
+
+__CRT_INLINE long double __cdecl roundl (long double x) {
+  return round(x);
+}
+
+__CRT_INLINE long __cdecl lroundl (long double x) {
+  return round(x);
+}
+
+__CRT_INLINE long long __cdecl llroundl (long double x) {
+  return round(x);
+}
+
+
 /*******************************************************************************
   End of code based on MUSL
 *******************************************************************************/