#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <emmintrin.h>
char sum_array(const uint8_t a[], uint64_t n)
{
    __m128i vsum = _mm_set1_epi32(0);
    uint64_t sum;
    uint64_t i;

    for (i = 0; i < n; i += 16)
    {
        __m128i v = _mm_load_si128((__m128i *)&a[i]);
        vsum = _mm_add_epi8(v, vsum);
    }
    sum = _mm_cvtsi128_si64(_mm_add_epi8(vsum, _mm_srli_si128(vsum, 8)));
    return
        (0xff & (sum >> 0*8)) +
        (0xff & (sum >> 1*8)) +
        (0xff & (sum >> 2*8)) +
        (0xff & (sum >> 3*8)) +
        (0xff & (sum >> 4*8)) +
        (0xff & (sum >> 5*8)) +
        (0xff & (sum >> 6*8)) +
        (0xff & (sum >> 7*8));
}

int main(int argc, char** argv) {
    if (argc < 1) {
        printf("Too few arguments!");
        return 0;
    }
    const char *memblock;
    int fd;
    struct stat sb;

    fd = open(argv[1], O_RDONLY);
    fstat(fd, &sb);
    uint64_t filesize = sb.st_size;
    printf("Size: %lu\n", filesize);

    memblock = mmap(NULL, filesize, PROT_READ, MAP_SHARED | MAP_POPULATE, fd, 0);
    if (memblock == MAP_FAILED) {
        printf("oh no it failed\n");
        return 0;
    }
    printf("The answer is: %d\n", sum_array(memblock, filesize));
    return 0;
}