This project is dedicated to the study and optimization of the algorithm for converting a Gregorian date to an ISO 8601 week number. The primary focus is on performance across legacy architectures and the impact of various compiler optimization levels.

Study in new_test_cases

This subdirectory contains a detailed comparison of four algorithm implementations. All tests were conducted using the i486-linux-musl-g++ cross-compiler (GCC 11.2) for the i586 architecture.

Tested Functions:

1. Original (version_original.cpp):

   • A classic implementation using standard if-else blocks and logical checks.
   • Contains multiple branches, which can lead to pipeline stalls on older processors.

2. V1_EarlyReturn (version_v1.cpp):

   • Optimized using "early returns" to exit the function as soon as the result is known.
   • Reduces the amount of code executed in simple cases but retains logical transitions.

3. V2_BitOps (version_v2.cpp):

   • Optimized through bitwise operations (shifts, masks).
   • Minimizes branching by replacing logic with arithmetic and bit manipulation.

4. V4_MathMask (version_v4.cpp):

   • Extreme optimization completely avoiding branches (branchless).
   • Uses mathematical masks (the 0 or 1 result of logical comparisons) and multiplications to select the final value without if statements.
   • Demonstrated the best performance results, especially at lower optimization levels.

Testing Parameters:

• Standard: C++20
• Architecture: -march=i586 -mtune=i686
• Flags: -O1, -O2, -O3, -fno-omit-frame-pointer, -static
• Data: 1,000,000 iterations on random dates within the range 1800-3000 AD.

Results

All reports and visualizations are located in the new_test_cases/ folder:

• results.csv: Summary table (Average, Median, Min, P95, P99).
• i586_benchmark_charts_combined.png: Comparative charts for execution time and speedup.

---

This project was prepared as part of a study on C++ micro-optimizations.