go-bp

Binary packing in Go

Features / TODO list

• Scalar implementation
• Fuzzing against the C++ implementation
• SIMD implementation
• Extend to AVX256 (ymm registers)
• Regular fuzz tests
• Benchmarks

How this works?

• Integers are in 128 mini blocks
• Each block contains 16 mini blocks (16 * 128 = 2048)
• Alignment to 128 bit boundaries for SIMD to work its magic
• AVX256 needs alignment to 256 bit boundaries
• Sorted input of integers, no duplicates