Fix mul128 for x86

algorithm/cryptonight.c

@@ -126,6 +126,40 @@ void CNKeccak(uint64_t *output, uint64_t *input)
 	memcpy(output, st, 200);
 }
 
+#if !defined(_WIN64) && !defined(__amd64__)
+static inline uint64_t hi_dword(uint64_t val) {
+	return val >> 32;
+}
+
+static inline uint64_t lo_dword(uint64_t val) {
+	return val & 0xFFFFFFFF;
+}
+
+static inline uint64_t mul128(uint64_t multiplier, uint64_t multiplicand, uint64_t* product_hi) {
+	// multiplier   = ab = a * 2^32 + b
+	// multiplicand = cd = c * 2^32 + d
+	// ab * cd = a * c * 2^64 + (a * d + b * c) * 2^32 + b * d
+	uint64_t a = hi_dword(multiplier);
+	uint64_t b = lo_dword(multiplier);
+	uint64_t c = hi_dword(multiplicand);
+	uint64_t d = lo_dword(multiplicand);
+
+	uint64_t ac = a * c;
+	uint64_t ad = a * d;
+	uint64_t bc = b * c;
+	uint64_t bd = b * d;
+
+	uint64_t adbc = ad + bc;
+	uint64_t adbc_carry = adbc < ad ? 1 : 0;
+
+	// multiplier * multiplicand = product_hi * 2^64 + product_lo
+	uint64_t product_lo = bd + (adbc << 32);
+	uint64_t product_lo_carry = product_lo < bd ? 1 : 0;
+	*product_hi = ac + (adbc >> 32) + (adbc_carry << 32) + product_lo_carry;
+
+	return product_lo;
+}
+#else
 static inline uint64_t mul128(uint64_t a, uint64_t b, uint64_t* product_hi)
 {
 	uint64_t lo, hi;
@@ -138,6 +172,7 @@ static inline uint64_t mul128(uint64_t a, uint64_t b, uint64_t* product_hi)
 
 	return lo;
 }
+#endif
 
 #define BYTE(x, y)		(((x) >> ((y) << 3)) & 0xFF)
 #define ROTL32(x, y)	(((x) << (y)) | ((x) >> (32 - (y))))