Skip to content

Latest commit

 

History

History
50 lines (37 loc) · 3.45 KB

README.md

File metadata and controls

50 lines (37 loc) · 3.45 KB
Raw

Benchmarking Unpacking Methods for IEnumerable<T>

This project compares the performance of various methods for unpacking IEnumerable<T> objects into 1D and 2D arrays. The goal is to find the fastest and most efficient way to flatten and extract the data.

Methods Compared

  1. UnpackMeNormalForLoop: Simple for loop that copies the data to the output arrays.
  2. UnpackMeSimd: Utilizes SIMD instructions for parallelization.
  3. UnpackMeCopyUnaligned: Uses Unsafe.CopyBlockUnaligned for unaligned memory copy of blittable types.
  4. UnpackMeCopyUnalignedParallelFor: Combines Unsafe.CopyBlockUnaligned with parallel processing using Parallel.For.
  5. UnpackMeCopyUnalignedTaskScheduler: Similar to UnpackMeCopyUnalignedParallelFor, but uses a task scheduler.
  6. UnpackMeParallelFor: Utilizes Parallel.For for parallelization without Unsafe.CopyBlockUnaligned.
  7. UnpackMeParallelForNested: Uses nested Parallel.For loops for parallelization.
  8. UnpackMeGenerated: Generated method using source generators.
  9. UnpackMeParallelForMoreData: UnpackMeParallelFor with a larger dataset.
  10. UnpackMeGeneratedMoreData: UnpackMeGenerated with a larger dataset.
  11. UnpackMe2Generated: Another variation of the generated method.

Benchmark Results

Method Mean Error StdDev Median
UnpackMeNormalForLoop 5.939 ms 0.0656 ms 0.0581 ms 5.944 ms
UnpackMeSimd 4.065 ms 0.0789 ms 0.0811 ms 4.062 ms
UnpackMeCopyUnaligned 4.173 ms 0.0736 ms 0.0688 ms 4.163 ms
UnpackMeCopyUnalignedParallelFor 2.833 ms 0.0309 ms 0.0289 ms 2.836 ms
UnpackMeCopyUnalignedTaskScheduler 17.016 ms 0.2761 ms 0.2583 ms 16.920 ms
UnpackMeParallelFor 2.942 ms 0.0523 ms 0.0463 ms 2.931 ms
UnpackMeParallelForNested 26.351 ms 0.5210 ms 0.8560 ms 26.280 ms
UnpackMeGenerated 2.847 ms 0.0216 ms 0.0202 ms 2.852 ms
UnpackMeParallelForMoreData 6.440 ms 0.2149 ms 0.6338 ms 6.340 ms
UnpackMeGeneratedMoreData 7.105 ms 0.1412 ms 0.3960 ms 7.194 ms
UnpackMe2Generated 7.112 ms 0.1413 ms 0.4100 ms 7.225 ms

The fastest method is UnpackMeCopyUnalignedParallelFor, which combines parallel processing with unaligned memory copy for blittable types. It achieves a mean execution time of 2.833 ms.

Fastest Method: UnpackMeCopyUnalignedParallelFor

The UnpackMeCopyUnalignedParallelFor method achieves the best performance by:

  1. Using Parallel.For for parallel processing.
  2. Utilizing Unsafe.CopyBlockUnaligned for fast unaligned memory copy of blittable types (int and float).
  3. Employing unsafe code and pointers for direct memory access.

The method converts the input IEnumerable<T> to an array, creates output arrays, and uses Parallel.For to process each element. It copies scalar members directly and uses Unsafe.CopyBlockUnaligned for array members of blittable types. Non-blittable types are copied element-wise.

Conclusion

The UnpackMeCopyUnalignedParallelFor method emerges as the fastest approach for unpacking IEnumerable<T> objects into arrays. However, consider factors such as data size, memory usage, and maintainability when choosing an unpacking method for your specific application.