randomx_reciprocal is slow

[tevador]

This function is not optimized:

RandomX/src/reciprocal.c

Lines 47 to 72 in d3c9648

	uint64_t randomx_reciprocal(uint64_t divisor) {
	assert(divisor != 0);
	const uint64_t p2exp63 = 1ULL << 63;
	uint64_t quotient = p2exp63 / divisor, remainder = p2exp63 % divisor;
	unsigned bsr = 0; //highest set bit in divisor
	for (uint64_t bit = divisor; bit > 0; bit >>= 1)
	bsr++;
	for (unsigned shift = 0; shift < bsr; shift++) {
	if (remainder >= divisor - remainder) {
	quotient = quotient * 2 + 1;
	remainder = remainder * 2 - divisor;
	}
	else {
	quotient = quotient * 2;
	remainder = remainder * 2;
	}
	}
	return quotient;
	}

There is an optimized version inside the ARM64 JIT compiler implemented by @SChernykh . It would be better to optimize randomx_reciprocal rather than duplicating the code inline.

[SChernykh]

#284