CRC16.hh

Go to the documentation of this file.

#ifndef CRC16_HH
#define CRC16_HH

#include <cstddef>
#include <cstdint>
#include <initializer_list>

namespace openmsx {

struct CRC16Lut { // replace with std::array in c++17
        uint16_t a[8][0x100];
};
inline constexpr CRC16Lut calcTable()
{
        CRC16Lut result = {};
        for (unsigned i = 0; i < 0x100; ++i) {
                uint16_t x = i << 8;
                for (int j = 0; j < 8; ++j) {
                        x = (x << 1) ^ ((x & 0x8000) ? 0x1021 : 0);
                }
                result.a[0][i] = x;
        }
        for (unsigned i = 0; i < 0x100; ++i) {
                uint16_t c = result.a[0][i];
                for (unsigned j = 1; j < 8; ++j) {
                        c = result.a[0][c >> 8] ^ (c << 8);
                        result.a[j][i] = c;
                }
        }
        return result;
}

class CRC16
{
public:
        explicit constexpr CRC16(uint16_t initialCRC = 0xffff)
                : crc(initialCRC)
        {
        }

        constexpr void init(uint16_t initialCRC)
        {
                crc = initialCRC;
        }

        constexpr void init(std::initializer_list<uint8_t> list)
        {
                crc = 0xffff;
                for (auto& val : list) {
                        update(val);
                }
        }

        constexpr void update(uint8_t value)
        {
                // Classical byte-at-a-time algorithm by Dilip V. Sarwate
                crc = (crc << 8) ^ tab.a[0][(crc >> 8) ^ value];
        }

        constexpr void update(const uint8_t* data, size_t size)
        {
                // Based on:
                //   Slicing-by-4 and slicing-by-8 algorithms by Michael E.
                //   Kounavis and Frank L. Berry from Intel Corp.
                //   http://www.intel.com/technology/comms/perfnet/download/CRC_generators.pdf
                // and the implementation by Peter Kankowski found on:
                //   http://www.strchr.com/crc32_popcnt
                // I transformed the code from CRC32 to CRC16 (this was not
                // trivial because both CRCs use a different convention about
                // bit order). I also made the code work on bigendian hosts.

                unsigned c = crc; // 32-bit are faster than 16-bit calculations
                                  // on x86 and many other modern architectures
                // calculate the bulk of the data 8 bytes at a time
                for (auto n = size / 8; n; --n) {
                        c = tab.a[7][data[0] ^ (c >>  8)] ^
                            tab.a[6][data[1] ^ (c & 255)] ^
                            tab.a[5][data[2]] ^
                            tab.a[4][data[3]] ^
                            tab.a[3][data[4]] ^
                            tab.a[2][data[5]] ^
                            tab.a[1][data[6]] ^
                            tab.a[0][data[7]];
                        data += 8;
                }
                // calculate the remaining bytes in the usual way
                for (size &= 7; size; --size) {
                        c = uint16_t(c << 8) ^ tab.a[0][(c >> 8) ^ *data++];
                }
                crc = c; // store back in a 16-bit result
        }

        constexpr uint16_t getValue() const
        {
                return crc;
        }

private:
        static constexpr CRC16Lut tab = calcTable();

        uint16_t crc;
};

} // namespace openmsx

#endif