Replacing a 32-bit loop counter with 64-bit introduces crazy performance deviations with _mm_popcnt_u64 on Intel CPUs

Replacing a 32-bit loop counter with 64-bit: The Mystery of Crazy Performance Deviations with _mm_popcnt_u64 on Intel CPUs 😱💻

Are you ready for a tech mystery that will blow your mind? We're diving deep into the world of performance deviations when replacing a 32-bit loop counter with a 64-bit one, specifically when using _mm_popcnt_u64 on Intel CPUs. 🤔💥

The Benchmark: ⌛️🔬

Let's set the stage. Our goal is to find the fastest way to popcount large arrays of data. To measure performance, we create a buffer of random data with a user-specified size in megabytes. We then use the x86 popcount intrinsic (_mm_popcnt_u64) in an unrolled loop to count the bits. To get an accurate result, we repeat the popcount 10,000 times. The only difference we make is the type of the loop variable (unsigned or uint