diff --git a/binaries/aarch64/ip2net b/binaries/aarch64/ip2net
index 633d34b7..ae734298 100755
Binary files a/binaries/aarch64/ip2net and b/binaries/aarch64/ip2net differ
diff --git a/binaries/arm/ip2net b/binaries/arm/ip2net
index 87523ed0..34cac9a4 100755
Binary files a/binaries/arm/ip2net and b/binaries/arm/ip2net differ
diff --git a/binaries/freebsd-x64/ip2net b/binaries/freebsd-x64/ip2net
index 45785964..efc65e7a 100755
Binary files a/binaries/freebsd-x64/ip2net and b/binaries/freebsd-x64/ip2net differ
diff --git a/binaries/mips32r1-lsb/ip2net b/binaries/mips32r1-lsb/ip2net
index 4d7c29ee..9af01b3c 100755
Binary files a/binaries/mips32r1-lsb/ip2net and b/binaries/mips32r1-lsb/ip2net differ
diff --git a/binaries/mips32r1-msb/ip2net b/binaries/mips32r1-msb/ip2net
index 94df0eb0..a4481acb 100755
Binary files a/binaries/mips32r1-msb/ip2net and b/binaries/mips32r1-msb/ip2net differ
diff --git a/binaries/mips64r2-msb/ip2net b/binaries/mips64r2-msb/ip2net
index e86926de..28f63ceb 100755
Binary files a/binaries/mips64r2-msb/ip2net and b/binaries/mips64r2-msb/ip2net differ
diff --git a/binaries/ppc/ip2net b/binaries/ppc/ip2net
index 7f471bac..7225e3b2 100755
Binary files a/binaries/ppc/ip2net and b/binaries/ppc/ip2net differ
diff --git a/binaries/x86/ip2net b/binaries/x86/ip2net
index 46cd7efd..19e81fcc 100755
Binary files a/binaries/x86/ip2net and b/binaries/x86/ip2net differ
diff --git a/binaries/x86_64/ip2net b/binaries/x86_64/ip2net
index 73464a86..de1dfe88 100755
Binary files a/binaries/x86_64/ip2net and b/binaries/x86_64/ip2net differ
diff --git a/ip2net/ip2net.c b/ip2net/ip2net.c
index dede7e79..60316bda 100644
--- a/ip2net/ip2net.c
+++ b/ip2net/ip2net.c
@@ -102,7 +102,14 @@ static inline const struct in6_addr *mask_from_bitcount6(uint32_t zct)
 }
 
 
-// result = a & b
+/*
+// this is "correct" solution for strict aliasing feature
+// but I don't like this style of coding
+// write what I don't mean to force smart optimizer to do what it's best
+// it produces better code sometimes but not on all compilers/versions/archs
+// sometimes it even generates real memcpy calls (mips32,arm32)
+// so I will not do it
+
 static void ip6_and(const struct in6_addr *a, const struct in6_addr *b, struct in6_addr *result)
 {
 	uint64_t a_addr[2], b_addr[2];
@@ -112,6 +119,24 @@ static void ip6_and(const struct in6_addr *a, const struct in6_addr *b, struct i
 	a_addr[1] &= b_addr[1];
 	memcpy(result->s6_addr, a_addr, 16);
 }
+*/
+
+// YES, from my point of view C should work as a portable assembler. It must do what I instruct it to do.
+// that's why I disable strict aliasing for this function. I observed gcc can miscompile with O2/O3 setting if inlined and not coded "correct"
+// result = a & b
+#if defined(__GNUC__) && !defined(__llvm__)
+__attribute__((optimize ("no-strict-aliasing")))
+#endif
+static void ip6_and(const struct in6_addr *a, const struct in6_addr *b, struct in6_addr *result)
+{
+#ifdef __SIZEOF_INT128__
+	// gcc and clang have 128 bit int types on some 64-bit archs. take some advantage
+	*((unsigned __int128*)result->s6_addr) = *((unsigned __int128*)a->s6_addr) & *((unsigned __int128*)b->s6_addr);
+#else
+	((uint64_t*)result->s6_addr)[0] = ((uint64_t*)a->s6_addr)[0] & ((uint64_t*)b->s6_addr)[0];
+	((uint64_t*)result->s6_addr)[1] = ((uint64_t*)a->s6_addr)[1] & ((uint64_t*)b->s6_addr)[1];
+#endif
+}
 
 static void rtrim(char *s)
 {