FORCE_INLINE __m128i _mm_aesdec_si128()

in common/checksum/sse2neon.h [8885:8951]

• 50 lines of code
• 3 McCabe index (conditional complexity)

FORCE_INLINE __m128i _mm_aesdec_si128(__m128i a, __m128i RoundKey)
{
#if defined(__aarch64__)
    static const uint8_t inv_shift_rows[] = {
        0x0, 0xd, 0xa, 0x7, 0x4, 0x1, 0xe, 0xb,
        0x8, 0x5, 0x2, 0xf, 0xc, 0x9, 0x6, 0x3,
    };
    static const uint8_t ror32by8[] = {
        0x1, 0x2, 0x3, 0x0, 0x5, 0x6, 0x7, 0x4,
        0x9, 0xa, 0xb, 0x8, 0xd, 0xe, 0xf, 0xc,
    };

    uint8x16_t v;
    uint8x16_t w = vreinterpretq_u8_m128i(a);

    // inverse shift rows
    w = vqtbl1q_u8(w, vld1q_u8(inv_shift_rows));

    // inverse sub bytes
    v = vqtbl4q_u8(_sse2neon_vld1q_u8_x4(_sse2neon_rsbox), w);
    v = vqtbx4q_u8(v, _sse2neon_vld1q_u8_x4(_sse2neon_rsbox + 0x40), w - 0x40);
    v = vqtbx4q_u8(v, _sse2neon_vld1q_u8_x4(_sse2neon_rsbox + 0x80), w - 0x80);
    v = vqtbx4q_u8(v, _sse2neon_vld1q_u8_x4(_sse2neon_rsbox + 0xc0), w - 0xc0);

    // inverse mix columns
    // multiplying 'v' by 4 in GF(2^8)
    w = (v << 1) ^ (uint8x16_t) (((int8x16_t) v >> 7) & 0x1b);
    w = (w << 1) ^ (uint8x16_t) (((int8x16_t) w >> 7) & 0x1b);
    v ^= w;
    v ^= (uint8x16_t) vrev32q_u16((uint16x8_t) w);

    w = (v << 1) ^ (uint8x16_t) (((int8x16_t) v >> 7) &
                                 0x1b);  // multiplying 'v' by 2 in GF(2^8)
    w ^= (uint8x16_t) vrev32q_u16((uint16x8_t) v);
    w ^= vqtbl1q_u8(v ^ w, vld1q_u8(ror32by8));

    // add round key
    return vreinterpretq_m128i_u8(w) ^ RoundKey;

#else /* ARMv7-A NEON implementation */
    /* FIXME: optimized for NEON */
    uint8_t i, e, f, g, h, v[4][4];
    uint8_t *_a = (uint8_t *) &a;
    for (i = 0; i < 16; ++i) {
        v[((i / 4) + (i % 4)) % 4][i % 4] = _sse2neon_rsbox[_a[i]];
    }

    // inverse mix columns
    for (i = 0; i < 4; ++i) {
        e = v[i][0];
        f = v[i][1];
        g = v[i][2];
        h = v[i][3];

        v[i][0] = SSE2NEON_MULTIPLY(e, 0x0e) ^ SSE2NEON_MULTIPLY(f, 0x0b) ^
                  SSE2NEON_MULTIPLY(g, 0x0d) ^ SSE2NEON_MULTIPLY(h, 0x09);
        v[i][1] = SSE2NEON_MULTIPLY(e, 0x09) ^ SSE2NEON_MULTIPLY(f, 0x0e) ^
                  SSE2NEON_MULTIPLY(g, 0x0b) ^ SSE2NEON_MULTIPLY(h, 0x0d);
        v[i][2] = SSE2NEON_MULTIPLY(e, 0x0d) ^ SSE2NEON_MULTIPLY(f, 0x09) ^
                  SSE2NEON_MULTIPLY(g, 0x0e) ^ SSE2NEON_MULTIPLY(h, 0x0b);
        v[i][3] = SSE2NEON_MULTIPLY(e, 0x0b) ^ SSE2NEON_MULTIPLY(f, 0x0d) ^
                  SSE2NEON_MULTIPLY(g, 0x09) ^ SSE2NEON_MULTIPLY(h, 0x0e);
    }

    return vreinterpretq_m128i_u8(vld1q_u8((uint8_t *) v)) ^ RoundKey;
#endif
}