NEWS

---

DDRTree2 0.2.0

Major Improvements

• OpenMP Parallelization - 11x performance improvement

  • Parallelized soft assignment computation (primary bottleneck)
  • Achieved 11x speedup on real-world data (1000 genes × 5536 cells)
  • Execution time reduced from 296.83s to 27.00s
  • Throughput increased from 19 to 205 cells/second
  • No memory overhead or numerical differences

• Code Optimizations

  • Efficient matrix broadcasting using Eigen's replicate()
  • Pre-allocated matrices to reduce memory allocations
  • Cleaner, more maintainable code structure

• Build System Improvements

  • Added Makevars for automatic OpenMP compilation
  • Conditional compilation for backward compatibility
  • Thread count detection and reporting

Documentation

• Added comprehensive performance documentation:
  • OPENMP_OPTIMIZATION_REPORT.md: Technical details and analysis
  • PERFORMANCE_SUMMARY.txt: Quick reference guide
  • benchmark_real_data.R: Reproducible benchmark script

Notes

• This is a performance-optimized fork of CRAN v0.1.5
• Original algorithm and core functionality unchanged
• Maintains identical numerical results
• Backward compatible - falls back to serial execution if OpenMP unavailable
• Scales linearly with dataset size
• Maintainer: Billsfriend [email protected]
• Original Authors: Xiaojie Qiu, Cole Trapnell, Qi Mao, Li Wang

Usage

Control thread count for optimal performance:

# Set before loading the package
Sys.setenv(OMP_NUM_THREADS = 8)
library(DDRTree2)

Or via shell:

export OMP_NUM_THREADS=8
Rscript your_script.R

---

DDRTree2 0.1.5

Bug Fixes

• Fixed a problem where DDRTree would return different results on repeated runs given the same inputs. The problem was in two places: kmeans and irlba. We now call irlba with deterministically initialized eigenvectors and kmeans with deterministically selected rows of the input.

---

DDRTree2 0.1.4

Bug Fixes

• Fixed a build error triggered by recent versions of GCC using the C++14 standard