Etliche Evolutionsstufen später habe ich dann das folgende Programm ins Rennen geschickt:

unsigned long mem0s(unsigned char *s, unsigned long l) {
        unsigned long r=0, i=0, mem_read;


        while(i+8
                mem_read=(unsigned long)(((unsigned long*)(s*));

                asm volatile (

                        "not %1, %1 \n\t"
                        "ctpop %1, %1 \n\t"
                        "addq %0, %1 \n\t":

                        "=r"(r): "r"(mem_read)
                );

                i += 8;
        }

        while(i
                mem_read=s;

                asm volatile (

                        "not %1, %1 \n\t"
                        "and %1, 255, %1 \n\t"
                        "ctpop %1, %1 \n\t"
                        "addq %0, %1 \n\t":

                        "=r"(r): "r"(mem_read)
                );

                i += 1;
        }

        return r;
}