How can Assembly Optimization Boost the Performance of a Positional Popcount Algorithm on Bytes?

How to Optimise this 8-bit Positional Popcount using Assembly?

The provided implementation of __mm_add_epi32_inplace_purego in Go is suboptimal due to the expensive passing of [8]int32 arrays. To improve performance, it is recommended to pass a pointer to the array instead.

However, the question goes beyond optimizing this specific function and explores the optimization of the inner loop using assembly for a positional population count algorithm on bytes.

Assembly Optimization

The provided assembly code offers two варианты of the positional population count algorithm:

• 32 Bytes at a Time without CSA (Constant Sum Adder)
• 96 Bytes at a Time with CSA

Improvements Introduced

The assembly code utilizes various techniques to improve performance:

• Prefetching: Prefetches data ahead to reduce cache misses.
• Vectorization: Employs SIMD (Single Instruction Multiple Data) instructions to process multiple bytes simultaneously.
• Pop Count Intrinsics: Uses intrinsic functions to count the population of bits efficiently.
• Carry-out Optimization: Takes advantage of the carry-out of shifted values to perform efficient population counting.
• 96-Byte Variant with CSA: Implements a technique called Constant Sum Addition to reduce the number of operations and improve performance by up to 30%.

Performance Benchmarks

Benchmarks show that the assembly optimizations result in significant performance improvements compared to a naive reference implementation in pure Go:

• Reg (32-byte variant): Up to 4998.53 MB/s
• RegCSA (96-byte variant with CSA): Up to 16053.40 MB/s

Full Source Code

The complete source code for both assembly variants can be found on GitHub. The code also includes a portable library that can be used for both variants in any Go program.

Conclusion

By implementing the positional population count algorithm in assembly, significant performance gains can be achieved. The provided assembly code utilizes various optimizations to maximize throughput. For further details and examples, please refer to the GitHub repository.

The above is the detailed content of How can Assembly Optimization Boost the Performance of a Positional Popcount Algorithm on Bytes?. For more information, please follow other related articles on the PHP Chinese website!