Add a "fork" of musl as //fusl.

fusl/src/math/fmod.c

Index: fusl/src/math/fmod.c
diff --git a/fusl/src/math/fmod.c b/fusl/src/math/fmod.c
new file mode 100644
index 0000000000000000000000000000000000000000..6849722bac50e477b9210f57f0271dcdabd8ae0c
--- /dev/null
+++ b/fusl/src/math/fmod.c
@@ -0,0 +1,68 @@
+#include <math.h>
+#include <stdint.h>
+
+double fmod(double x, double y)
+{
+	union {double f; uint64_t i;} ux = {x}, uy = {y};
+	int ex = ux.i>>52 & 0x7ff;
+	int ey = uy.i>>52 & 0x7ff;
+	int sx = ux.i>>63;
+	uint64_t i;
+
+	/* in the followings uxi should be ux.i, but then gcc wrongly adds */
+	/* float load/store to inner loops ruining performance and code size */
+	uint64_t uxi = ux.i;
+
+	if (uy.i<<1 == 0 || isnan(y) || ex == 0x7ff)
+		return (x*y)/(x*y);
+	if (uxi<<1 <= uy.i<<1) {
+		if (uxi<<1 == uy.i<<1)
+			return 0*x;
+		return x;
+	}
+
+	/* normalize x and y */
+	if (!ex) {
+		for (i = uxi<<12; i>>63 == 0; ex--, i <<= 1);
+		uxi <<= -ex + 1;
+	} else {
+		uxi &= -1ULL >> 12;
+		uxi |= 1ULL << 52;
+	}
+	if (!ey) {
+		for (i = uy.i<<12; i>>63 == 0; ey--, i <<= 1);
+		uy.i <<= -ey + 1;
+	} else {
+		uy.i &= -1ULL >> 12;
+		uy.i |= 1ULL << 52;
+	}
+
+	/* x mod y */
+	for (; ex > ey; ex--) {
+		i = uxi - uy.i;
+		if (i >> 63 == 0) {
+			if (i == 0)
+				return 0*x;
+			uxi = i;
+		}
+		uxi <<= 1;
+	}
+	i = uxi - uy.i;
+	if (i >> 63 == 0) {
+		if (i == 0)
+			return 0*x;
+		uxi = i;
+	}
+	for (; uxi>>52 == 0; uxi <<= 1, ex--);
+
+	/* scale result */
+	if (ex > 0) {
+		uxi -= 1ULL << 52;
+		uxi |= (uint64_t)ex << 52;
+	} else {
+		uxi >>= -ex + 1;
+	}
+	uxi |= (uint64_t)sx << 63;
+	ux.i = uxi;
+	return ux.f;
+}