fast wc -l

wcl.c

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

#include <unistd.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <sys/types.h>

#define BUF_SIZE 1048576

extern unsigned long memchrcount(void *mem, int c, size_t size);

int main(int argc, char **argv)
{
    unsigned long count = 0;
    ssize_t bytes_read;
    void *buf;
    int fd = open(argv[1], O_RDONLY);
    if (fd < 0)
    {
        perror("Cannot open file");
        return 1;
    }
    if (posix_memalign(&buf, 16, BUF_SIZE))
    {
        fputs("Cannot allocate memory\n", stderr);
        close(fd);
        return 1;
    }
    memset(buf, 0, BUF_SIZE);
    while ((bytes_read = read(fd, buf, BUF_SIZE)) > 0)
    {
        count += memchrcount(buf, '\n', BUF_SIZE);
    }
    close(fd);
    free(buf);
    printf("%lu\n", count);
    return 0;
}

wcl.s

.type memchrcount, @function
memchrcount:
    mov %esi, %eax
    sal $8, %esi
    or %esi, %eax
    sal $8, %esi
    or %esi, %eax
    sal $8, %esi
    or %esi, %eax
    movd %eax, %xmm5
    movsldup %xmm5, %xmm5
    movddup %xmm5, %xmm5
    movaps %xmm5,%xmm4
    movaps %xmm5,%xmm6
    movaps %xmm5,%xmm7
    xor %eax, %eax
    xor %r8, %r8
    xor %r9, %r9
    xor %r10, %r10
    xor %r11, %r11
    add %rdi,%rdx
    jmp .L2
.p2align 5
.L1:
    movaps 0(%rdi), %xmm0
    movaps 16(%rdi), %xmm1
    movaps 32(%rdi), %xmm2
    movaps 48(%rdi), %xmm3
    pcmpeqb %xmm4, %xmm0
    pcmpeqb %xmm5, %xmm1
    pcmpeqb %xmm6, %xmm2
    pcmpeqb %xmm7, %xmm3
    pmovmskb %xmm0, %r8
    pmovmskb %xmm1, %r9
    pmovmskb %xmm2, %r10
    pmovmskb %xmm3, %r11
    popcnt %r8, %r8
    popcnt %r9, %r9
    popcnt %r10, %r10
    popcnt %r11, %r11
    add %r8, %r9
    add %r10, %r11
    add %r9, %rax
    add %r11, %rax
    add $64, %rdi
.L2:
    cmp %rdi, %rdx
    jne .L1
    ret