The interview lasted about 100 minutes.

First, I was asked four competitive programming questions in the order of increasing difficulty. The difficulty level was similar to Codeforces Div2 C/D/E. No coding was involved; I just needed to explain the solution verbally. This part of the interview lasted about 50-55 minutes, and then I was asked questions about my past projects, which were in C++. Here I needed to explain the objective of the project, various implementation choices, and the reasoning behind them. This lasted about 20 minutes. Then I was asked questions on operating systems, parallel programming, and computer organization/architecture.

To give an idea of the difficulty level of questions on systems, here are a few sample questions -

An operating system supports 48-bit virtual addresses for machines with 4GB of physical memory. The OS uses a single-level page table with a page size of 4 KB. How much memory in bytes is required for the page table (excluding permission bits)? Is there any issue with it? What is the page table size if we use a 4-level page table that uses 9 bits of virtual address for each level and the remaining 12 bits as page offset? Suppose we change the page size to 2 MB and use a 3-level page table that uses 9 bits of virtual address for each level and the remaining 21 bits as page offset. What are the advantages of this design over the earlier 4-level design? Any Disadvantages?
You are given an N*M grid. There is a lock associated with each cell of the grid. There are N*M threads that operate on the grid; each thread will operate on a unique cell of the grid. To operate on a cell, the corresponding thread has to acquire locks of the neighboring cells (that share a side), do some processing, and then release the locks. Design an algorithm that is deadlock-free and argue about its correctness.

Let N be a large number. Let A[N][N], B[N], C[N] and D[N][N] be arrays that use a row-major layout. Optimize the following code snippets

for(i = 0; i < N; i++)
    for(j = 0; j < N; j++) {
        B[i] += A[i][j];
        C[j] += A[j][i];
    }
   
for(i = 0; i < N; i++)
    for(j = 0; j < N; j++) {
        D[j][i] = A[i][j] * A[j][i];
    }

Explain the behavior of the following program (run on a little-endian machine)

#include <iostream>
#define N 100

struct my_array {
    long long arr[N];
    char name[4] = "LOL";
    int initialised = 0;
};

void init(my_array & array) {
    for(int i = 0; i <= N; i++) array.arr[i] = 1;
    return;
}

int main() {
    my_array array;
    while(!array.initialised) {
        array.initialised = 1;
        init(array);
    }

    std::cout << "Initialisation Done\n";
}