• Made non-assembly up to 40% faster.
• AVX512 can use multiple goroutines for lower latency + higher individual throughput.
• AVX512 5-9% faster.
• All code faster with user defined goroutines and high concurrency. Up to 8x faster due to less cache evictions.
• CPUID detects AMD CPUs with hyperthreading/multiple threads/core.
• CPUID detects AMD per CCX L3 cache size.
• Use L1 cache size to set minimum split size.
• Tests/benchmarks can disable specific assembly types.