#ifdef CPU_STRESS_TEST
    #define PRIME_LIMIT 100000
    static uint8_t sieve[PRIME_LIMIT];
    memset(sieve, 1, sizeof(sieve));
    sieve[0] = sieve[1] = 0;
    for (int i = 2; i * i < PRIME_LIMIT; i++)
    {
      #ifdef CPU_STRESS_TEST_WITH_MEMCPY
      memcpy(&gDdrBufA[i], &i, sizeof(i));
      #endif
      if (sieve[i])
      {
        for (int j = i * i; j < PRIME_LIMIT; j += i)
          sieve[j] = 0;
      }
    }
    volatile int prime_count = 0;
    for (int i = 0; i < PRIME_LIMIT; i++)
      prime_count += sieve[i];
#endif

#ifdef DDR_STRESS_TEST
    // 20 iterations x 128KB each way = ~5MB of DDR traffic.
    #define DDR_COPY_ITERS 20
    {
      // Seed bufA so the compiler can't optimize away the copies
      for (int j = 0; j < (int)DDR_STRESS_SIZE; j++)
        gDdrBufA[j] = (uint8_t)(j * 0x5A + 0x12);
      CacheP_wb(gDdrBufA, DDR_STRESS_SIZE, CacheP_TYPE_ALLD);
      CacheP_inv(gDdrBufA, DDR_STRESS_SIZE, CacheP_TYPE_ALLD);
      CacheP_inv(gDdrBufB, DDR_STRESS_SIZE, CacheP_TYPE_ALLD);
      for (int i = 0; i < DDR_COPY_ITERS; i++)
      {
        memcpy(gDdrBufB, gDdrBufA, DDR_STRESS_SIZE);
        memcpy(gDdrBufA, gDdrBufB, DDR_STRESS_SIZE);
      }
      CacheP_wb(gDdrBufA, DDR_STRESS_SIZE, CacheP_TYPE_ALLD);
      CacheP_wb(gDdrBufB, DDR_STRESS_SIZE, CacheP_TYPE_ALLD);
    }
#endif