Tag
The article explains why false sharing alignment on x64 should be 128 bytes instead of the typical 64 bytes, due to Intel's Sandy Bridge spatial prefetcher that loads cache lines in pairs. It provides reasoning and a benchmark demonstrating the improvement.