AVX2 strlen()

avx2_strlen.c

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <immintrin.h>

#define LENGTH 1000000010

/*
  coins syspro strlen()
  http://www.coins.tsukuba.ac.jp/~syspro/2015/shui/quizzes2.html#exam207
*/

size_t my_strlen1(const char *s)__attribute__((noinline));
size_t my_strlen2(const char *s)__attribute__((noinline));
size_t my_strlen3(const char *s)__attribute__((noinline));

size_t my_strlen1(const char *s)
{
        unsigned int i = 0;
        while (s[i++] != '\0');
        return --i;
}

size_t my_strlen2(const char *s)
{
        unsigned int i = 0;
        unsigned int a;
        unsigned int *p;

        p = (unsigned int *)s;
        a = *p;

        while (1) {
                if (!(a & 0x000000ff)) {
                        return i * 4;
                } else if (!(a & 0x0000ff00)) {
                        return i * 4 + 1;
                } else if (!(a & 0x00ff0000)) {
                        return i * 4 + 2;
                } else if (!(a & 0xff000000)) {
                        return i * 4 + 3;
                }
                a = p[++i];
        }
}

size_t my_strlen3(const char *s)
{
        unsigned int i = 0;
        const __m256i *p;
        __m256i mask, zero;

        p = (__m256i *) s;
        zero = _mm256_setzero_si256();

        while (1) {
                mask = _mm256_cmpeq_epi8(*p, zero);
                if (!_mm256_testz_si256(mask, mask)) {
                        return (i * 32) + my_strlen1((char *)p);
                }
                i++;
                p++;
        }
}

int main(void)
{
        char *str;
        FILE *fp;

        clock_t c1, c2, c3, c4, t;
        unsigned int l1, l2, l3, l4;

        if ((str = calloc(LENGTH, sizeof(char))) == NULL) {
                perror("calloc");
                exit(1);
        }
        if ((fp = fopen("samplestring", "r")) == NULL) {
                perror("fopen");
                exit(1);
        }
        fread(str, sizeof(char), LENGTH, fp);
        fclose(fp);

        t = clock();
        l1 = strlen(str);
        c1 = clock() - t;

        t = clock();
        l2 = my_strlen1(str);
        c2 = clock() - t;

        t = clock();
        l3 = my_strlen2(str);
        c3 = clock() - t;

        t = clock();
        l4 = my_strlen3(str);
        c4 = clock() - t;

        free(str);

        printf("strlen():     %u, %u\n", l1, (unsigned int)c1);
        printf("my_strlen1(): %u, %u\n", l2, (unsigned int)c2);
        printf("my_strlen2(): %u, %u\n", l3, (unsigned int)c3);
        printf("my_strlen3(): %u, %u\n", l4, (unsigned int)c4);

        return 0;
}
/*
[s0911454@borage08]~$ clang -mavx2 -O3 strlen.c

[s0911454@borage08]~$ ./a.out
strlen():     1000000000, 61636
my_strlen1(): 1000000000, 391436
my_strlen2(): 1000000000, 177409
my_strlen3(): 1000000000, 57086

[s0911454@borage08]~$ uname -a
Darwin borage08.coins.tsukuba.ac.jp 13.3.0 Darwin Kernel Version 13.3.0: Tue Jun  3 21:27:35 PDT 2014; root:xnu-2422.110.17~1/RELEASE_X86_64 x86_64 i386 iMac14,2 Darwin
*/