NVIDIA Software Engineer Interview Experience — Phone Screen on a High-Performance String Class

NV phone screen

I was asked to implement a high-performance class to handle large strings, with functions like cmp, cat, substr, and so on. I created a simple class that uses a char array to store short strings. First I was asked to write the constructor:

const BUFF_SZIE = 128;
class myString {
private:
    char buf[BUFF_SZIE];
    size_t length;
    char *ptr;
public:
    myString(char *s, size_t len) {
        length = len;
        if (len < BUFF_SZIE) {
            strncpy(buf, s, len);
            buf[len] = 0;
        } else {
            ptr = (char *)malloc(len+1);
            if (ptr == nullptr) {
                throw "not enough memory";
            }
            memcpy(ptr, s, len);
            *(ptr+len) = 0;
        }
    }
}

• Q: For strncpy(buf, s, len), characters get copied one at a time — how can this be sped up? A: Use memcpy.
• Q: Is that the same as strncpy? A: ? — cast the buffer to integers, which is 4x faster.
• Q: For the cmp function, why is comparing short strings (<256 bytes) faster than comparing long strings (ignoring the difference in length)? A: ? — short strings sit in the CPU cache.
• Q: If BUFF_SZIE == 1, what is the size of myString? A: 12 bytes on a 32-bit machine, 24 bytes on a 64-bit machine.
• Q: If BUFF_SZIE == 8, but the strings are only 10-15 characters, how can the size of the class be reduced? A: Drop char *ptr and use the buffer itself to hold the pointer — this needs heap allocation (malloc) — a union can be used.