Port SIMD code to ARM NEON

[lemire]

Initially, SIMD code supports only x64 processors, but it should be simple to add support for ARM NEON as well.

[aqrit]

Possible approach for parsing of 8 UTF-16 digits (assumes little endian). Note: I know very little about ARM or NEON.

uint16x8_t x = vld1q_u16(src);
x = vsubq_u16(x, vdupq_n_u16(0x0030));
uint8x8_t m = vqmovn_u16(vcgtq_u16(x, vdupq_n_u16(0x0009)));
if(vget_lane_u64(vreinterpret_u64_u8(m), 0) == 0) {	// all 8 chars are digits
	static const uint32_t mul1[4] = {
		0x03E80064, 0x0000A0001, // 100 + (1000 << 16), 1 + (10 << 16)
		0x03E80064, 0x0000A0001
	};
	uint32x4_t v = vmulq_u32(vreinterpretq_u32_u16(x), vld1q_u32(mul1));
	uint32x2_t t = vpaddl_u16(vshrn_n_u32(v, 16));
	uint32_t a = vget_lane_u32(t, 0);
	uint32_t b = vget_lane_u32(t, 1);
	return (a * 10000) + b;
}