Number parsing at a gigabyte per second

With disks and networks providing gigabytes per second, parsing decimal numbers from strings becomes a bottleneck. We consider the problem of to nearest binary floating-point value. The general requires variable-precision arithmetic. However, we need at most 17 digits represent 64-bit standard (IEEE 754). Thus, can significand with single word. By combining precomputed tables, compute number using as few one or two multiplications. Our implementation be several times faster than conventional functions present in C libraries on modern systems (Intel, AMD, ARM, POWER9). work is available open source software used by major such Apache Arrow Yandex ClickHouse. Go library has adopted version our approach.