Stress test CPU with int ops + ld + st

cpu_burn.cc

// g++ -std=c++20 cpu_burn.cc -o cpu_burn

#include <bits/stdc++.h>

constexpr size_t maxn = 192;
constexpr int iter = 1024 * 1024;

std::array<std::jthread, maxn> threads;
std::array<unsigned int, (1024ull * 1024 * 512)> data;

// NOTE(Ali): Ripped from GLSL shader code
constexpr unsigned int pcg_hash(unsigned int input) {
  unsigned int state = input * 747796405u + 2891336453u;
  unsigned int word = ((state >> ((state >> 28u) + 4u)) ^ state) * 277803737u;
  return (word >> 22u) ^ word;
}

int main(int argc, char *argv[]) {
  assert(argc == 2);

  const size_t n = std::min(maxn, static_cast<size_t>(atol(argv[1])));

  std::generate(std::begin(data), std::end(data),
                [i = 0ull]() mutable { return i++; });

  std::generate_n(std::begin(threads), n, [n, i = 0ull]() mutable {
    return std::jthread(
        [n](size_t tid) {
          size_t offset = tid * data.size() / n;
          for (int j = 0; j < iter; j++) {
            for (size_t i = 0; i < data.size() / n; i++) {
              // NOTE(Ali): Who needs rand amirite?
              auto &d = data[offset + i];
              d = pcg_hash(d);
            }
          }
        },
        i++);
  });
}