Function bytes_to_bits

cpp/src/arrow/compute/util.cc:264–310

}

void bytes_to_bits(int64_t hardware_flags, const int num_bits, const uint8_t* bytes,
                   uint8_t* bits, int bit_offset) {
  bits += bit_offset / 8;
  bit_offset %= 8;
  if (bit_offset != 0) {
    uint64_t bits_head;
    int bits_in_first_byte = std::min(num_bits, 8 - bit_offset);
    bytes_to_bits(hardware_flags, bits_in_first_byte, bytes,
                  reinterpret_cast<uint8_t*>(&bits_head));
    uint8_t mask = (1 << bit_offset) - 1;
    *bits = static_cast<uint8_t>((*bits & mask) | (bits_head << bit_offset));

    if (num_bits > bits_in_first_byte) {
      bytes_to_bits(hardware_flags, num_bits - bits_in_first_byte,
                    bytes + bits_in_first_byte, bits + 1);
    }
    return;
  }

  int num_processed = 0;
#if defined(ARROW_HAVE_RUNTIME_AVX2) && defined(ARROW_HAVE_RUNTIME_BMI2)
  if ((hardware_flags & CpuInfo::AVX2) && CpuInfo::GetInstance()->HasEfficientBmi2()) {
    // The function call below processes whole 32 bit chunks together.
    num_processed = num_bits - (num_bits % 32);
    avx2::bytes_to_bits_avx2(num_processed, bytes, bits);
  }
#endif
  // Process 8 bits at a time
  constexpr int unroll = 8;
  for (int i = num_processed / unroll; i < num_bits / unroll; ++i) {
    uint64_t bytes_next = util::SafeLoad(&reinterpret_cast<const uint64_t*>(bytes)[i]);
    bytes_next &= 0x0101010101010101ULL;
    bytes_next |= (bytes_next >> 7);  // Pairs of adjacent output bits in individual bytes
    bytes_next |= (bytes_next >> 14);  // 4 adjacent output bits in individual bytes
    bytes_next |= (bytes_next >> 28);  // All 8 output bits in the lowest byte
    bits[i] = static_cast<uint8_t>(bytes_next & 0xff);
  }
  int tail = num_bits % unroll;
  if (tail) {
    uint64_t bytes_next = SafeLoadUpTo8Bytes(bytes + num_bits - tail, tail);
    bytes_next &= 0x0101010101010101ULL;
    bytes_next |= (bytes_next >> 7);  // Pairs of adjacent output bits in individual bytes
    bytes_next |= (bytes_next >> 14);  // 4 adjacent output bits in individual bytes
    bytes_next |= (bytes_next >> 28);  // All 8 output bits in the lowest byte
    bits[num_bits / 8] = static_cast<uint8_t>(bytes_next & 0xff);
  }
}

bool are_all_bytes_zero(int64_t hardware_flags, const uint8_t* bytes,
                        uint32_t num_bytes) {