🚀 Optimized entropy functions

[serkor1]

📚 What?

This PR optimizes all entropy functions. The run-time on 200 x 1e6 matrices have been decreased significantly, see below:

Before optimization (Without OpenMP)

Iterations	Runtime (sec)	Garbage Collections [gc()]	gc() pr. second	Memory Allocation (MB)
100	2.5	0	0	0

After optimization (Without OpenMP)

Iterations	Runtime (sec)	Garbage Collections [gc()]	gc() pr. second	Memory Allocation (MB)
100	0.86	0	0	0

[ellipsis-dev]

👍 Looks good to me! Reviewed everything up to 81d28c7 in 1 minute and 16 seconds

More details

• Looked at 1074 lines of code in 27 files
• Skipped 1 files when reviewing.
• Skipped posting 12 drafted comments based on config settings.

1. src/classification_CrossEntropy.h:49

• Draft comment:
  Using #pragma omp simd and #pragma omp parallel for together can lead to inefficiencies. Consider using only #pragma omp parallel for for the outer loop parallelization.
• Reason this comment was not posted:
  Decided after close inspection that this draft comment was likely wrong and/or not actionable:
  The combination of parallel for and simd here actually makes sense - they operate at different levels of parallelism. Parallel for splits work across threads, while simd enables vectorization within each thread. The operations in the inner loops (simple additions and multiplications) are perfect candidates for SIMD vectorization. The comment appears to be incorrect about inefficiencies.
  I could be wrong about the interaction between these OpenMP directives. There might be specific hardware architectures where this combination is problematic.
  While there might be edge cases, the code follows a common and well-established pattern of combining thread-level and SIMD parallelism. The operations are simple enough that vectorization should be beneficial.
  The comment should be deleted as it suggests removing SIMD directives that are actually beneficial for performance when properly combined with parallel for as done in this code.

2. src/classification_CrossEntropy.h:62

• Draft comment:
  Using #pragma omp simd and #pragma omp parallel for together can lead to inefficiencies. Consider using only #pragma omp parallel for for the outer loop parallelization.
• Reason this comment was not posted:
  Marked as duplicate.

3. src/classification_CrossEntropy.h:123

• Draft comment:
  Using #pragma omp simd and #pragma omp parallel for together can lead to inefficiencies. Consider using only #pragma omp parallel for for the outer loop parallelization.
• Reason this comment was not posted:
  Marked as duplicate.

4. src/classification_Entropy.h:44

• Draft comment:
  Using #pragma omp simd and #pragma omp parallel for together can lead to inefficiencies. Consider using only #pragma omp parallel for for the outer loop parallelization.
• Reason this comment was not posted:
  Marked as duplicate.

5. src/classification_Entropy.h:57

• Draft comment:
  Using #pragma omp simd and #pragma omp parallel for together can lead to inefficiencies. Consider using only #pragma omp parallel for for the outer loop parallelization.
• Reason this comment was not posted:
  Marked as duplicate.

6. src/classification_Entropy.h:117

• Draft comment:
  Using #pragma omp simd and #pragma omp parallel for together can lead to inefficiencies. Consider using only #pragma omp parallel for for the outer loop parallelization.
• Reason this comment was not posted:
  Marked as duplicate.

7. src/classification_RelativeEntropy.h:50

• Draft comment:
  Using #pragma omp simd and #pragma omp parallel for together can lead to inefficiencies. Consider using only #pragma omp parallel for for the outer loop parallelization.
• Reason this comment was not posted:
  Marked as duplicate.

8. src/classification_RelativeEntropy.h:65

• Draft comment:
  Using #pragma omp simd and #pragma omp parallel for together can lead to inefficiencies. Consider using only #pragma omp parallel for for the outer loop parallelization.
• Reason this comment was not posted:
  Marked as duplicate.

9. src/classification_CrossEntropy.h:13

• Draft comment:
  All Rcpp functions and classes should be namespace qualified. For example, use Rcpp::NumericVector instead of NumericVector. This applies throughout the C++ code.
• Reason this comment was not posted:
  Comment was on unchanged code.

10. src/classification_Entropy.cpp:14

• Draft comment:
  All Rcpp functions and classes should be namespace qualified. For example, use Rcpp::NumericVector instead of NumericVector. This applies throughout the C++ code.
• Reason this comment was not posted:
  Marked as duplicate.

11. src/classification_Entropy.h:11

• Draft comment:
  All Rcpp functions and classes should be namespace qualified. For example, use Rcpp::NumericVector instead of NumericVector. This applies throughout the C++ code.
• Reason this comment was not posted:
  Marked as duplicate.

12. src/classification_RelativeEntropy.h:14

• Draft comment:
  All Rcpp functions and classes should be namespace qualified. For example, use Rcpp::NumericVector instead of NumericVector. This applies throughout the C++ code.
• Reason this comment was not posted:
  Marked as duplicate.

Workflow ID: wflow_L0sjpnWlN4E1W2HS

---

You can customize Ellipsis with 👍 / 👎 feedback, review rules, user-specific overrides, quiet mode, and more.